My 3 video card experience

I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

 

So, those are my observations.

«1

Comments

  • nicsttnicstt Posts: 7,628

    I'm sorry, but there is something else affecting it in your particular setup; core for core, each generation has been better; certainly 7, 9 10; I can't comment on 20 but it seems so too.

     

  • Yeah, I also disagree.  I saw a big difference in the generations.  Going from Titan X Maxwell to Pascal there was about a 60% improvement IIRC.

    Doesn't seem the price tag will be worth the extra performance, not yet anyways.  Maybe we'll see a nice jump in improvement when (if?) Iray supports RTX.

  • nickalamannickalaman Posts: 176
    areg5 said:

    I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

    1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

    2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

    3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

    4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

     

    So, those are my observations.

    From my tests what i was able to deduce is that there is modest gains between generations, the biggest gain between generation is speed of the GPU. Problem with mixing old cards and new is that iray tends to slow all cards down to the speed of the slowest card. That's why it looks like adding a new gen card does not make much difference if your mixing it with an old gen card.

     

     

  • areg5areg5 Posts: 379
    areg5 said:

    I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

    1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

    2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

    3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

    4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

     

    So, those are my observations.

    From my tests what i was able to deduce is that there is modest gains between generations, the biggest gain between generation is speed of the GPU. Problem with mixing old cards and new is that iray tends to slow all cards down to the speed of the slowest card. That's why it looks like adding a new gen card does not make much difference if your mixing it with an old gen card.

     

     

    I'm not saying there's no difference at all, I'm saying that any difference is marginal.

  • areg5areg5 Posts: 379
    areg5 said:
    areg5 said:

    I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

    1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

    2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

    3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

    4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

     

    So, those are my observations.

    From my tests what i was able to deduce is that there is modest gains between generations, the biggest gain between generation is speed of the GPU. Problem with mixing old cards and new is that iray tends to slow all cards down to the speed of the slowest card. That's why it looks like adding a new gen card does not make much difference if your mixing it with an old gen card.

     

     

    I'm not saying there's no difference at all, I'm saying that any difference is marginal.  And I have not seen newer cards throttled down when mixed with older cards.

     

  • surodysurody Posts: 63
    edited January 7
    areg5 said:
    areg5 said:

    I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

    1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

    2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

    3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

    4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

     

    So, those are my observations.

    From my tests what i was able to deduce is that there is modest gains between generations, the biggest gain between generation is speed of the GPU. Problem with mixing old cards and new is that iray tends to slow all cards down to the speed of the slowest card. That's why it looks like adding a new gen card does not make much difference if your mixing it with an old gen card.

     

     

    I'm not saying there's no difference at all, I'm saying that any difference is marginal.

    I got 2 rigs, 1 has 2x 980ti and the newer has 2x 1080ti. If you say the performance boost is marginal it's probably because you are mixing generations. My newer build with 2x 1080ti is about 40% faster in my tests compared to the 980ti rig. Besides that it isn't recommended to use more than 2 consumer cards in SLI. The performance drop is too high and the RAM doesn't stack. The performance dropoff seems to be fixed with NVLink and the 2000 generation though.
    Post edited by surody on
  • joseftjoseft Posts: 169

    What is your definition of marginal? It can be a very subjective word. You should have posted the render times. Improving a 10 second render by 10% is an improvement of 1 second, most people would call that marginal. Shaving 1 second off an already fast render is not a big deal. But a 10% improvement on a 2 hour render is almost 20 minutes faster, which suddenly becomes much more attractive does it not?

    As others have mentioned, most render engines dont operate at their best with mixed generations of GPU, iRay i think is known to be the worst at it. 

    Looking at your 3 GPU configurations in OctaneBench, i think it illustrates the difference not mixing card generations makes. 

    1x 780ti + 2x 980ti - OB 325

    2x 980ti + 1080ti - OB 390

    3x 1080ti - OB 552

    To drive that home a little more, 3x 980ti has the same bench result as 2x 980ti + 1080ti. So in Octane, having that 1080ti over another 980 card makes 0 difference in render speed. This would at the very least be the same in iRay, if not worse. 

  • namffuaknamffuak Posts: 2,747
    areg5 said:

    I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

    1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

    2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

    3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

    4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

     

    So, those are my observations.

    From my tests what i was able to deduce is that there is modest gains between generations, the biggest gain between generation is speed of the GPU. Problem with mixing old cards and new is that iray tends to slow all cards down to the speed of the slowest card. That's why it looks like adding a new gen card does not make much difference if your mixing it with an old gen card.

    I keep seeing people say this but I never see it.

     

    Real times - 1 g8f, with outfit, default lighting

    980TI:           4 minutes 4.5 seconds, 731 iterations

    1080TI          2 minutes 43.9 seconds, 735 iterations

    Both            1 minute 49.9 seconds, 715 iterations

    For all three renders, gpu utilization was in the 95 to 97% range and both cards ran at the maximum GPU core and memory clock speed as reported by GPU-Z. I've never had either card throttle back when both were in use except in those cases where the scene was over 6 GB and the 980TI got dropped.

     

    I'd very much like to see GPU-Z graphs showing a faster card being throttled to match a slower one. Remember, Nvidia does NOT recommend using SLI with Iray.

  • joseftjoseft Posts: 169
    edited January 7
    namffuak said:
    areg5 said:

    I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

    1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

    2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

    3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

    4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

     

    So, those are my observations.

    From my tests what i was able to deduce is that there is modest gains between generations, the biggest gain between generation is speed of the GPU. Problem with mixing old cards and new is that iray tends to slow all cards down to the speed of the slowest card. That's why it looks like adding a new gen card does not make much difference if your mixing it with an old gen card.

    I keep seeing people say this but I never see it.

     

    Real times - 1 g8f, with outfit, default lighting

    980TI:           4 minutes 4.5 seconds, 731 iterations

    1080TI          2 minutes 43.9 seconds, 735 iterations

    Both            1 minute 49.9 seconds, 715 iterations

    For all three renders, gpu utilization was in the 95 to 97% range and both cards ran at the maximum GPU core and memory clock speed as reported by GPU-Z. I've never had either card throttle back when both were in use except in those cases where the scene was over 6 GB and the 980TI got dropped.

     

    I'd very much like to see GPU-Z graphs showing a faster card being throttled to match a slower one. Remember, Nvidia does NOT recommend using SLI with Iray.

     

    The concept of the render engine not utilising mixed generations to their maximum would not be visibile through hardware usage monitoring. I remember reading an article that explained this much much better than i ever could ages ago, im trying to remember where it was from so i can find it again.

    Its not that the render engine throttles any hardware, its more how the engine uses the power available to it from two components that are different at the firmware level. Think about how render engines need to be updated to support new generation cards as they are released, that indicates that there is something different going on in the engine's code depending on the generation of card. How that translates to any performance loss is probably too technical for most people (myself included), but that article i referred to did explain it in a way that most can understand 

    Post edited by joseft on
  • namffuaknamffuak Posts: 2,747
    joseft said:
    namffuak said:
    areg5 said:

    I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

    1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

    2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

    3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

    4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

     

    So, those are my observations.

    From my tests what i was able to deduce is that there is modest gains between generations, the biggest gain between generation is speed of the GPU. Problem with mixing old cards and new is that iray tends to slow all cards down to the speed of the slowest card. That's why it looks like adding a new gen card does not make much difference if your mixing it with an old gen card.

    I keep seeing people say this but I never see it.

     

    Real times - 1 g8f, with outfit, default lighting

    980TI:           4 minutes 4.5 seconds, 731 iterations

    1080TI          2 minutes 43.9 seconds, 735 iterations

    Both            1 minute 49.9 seconds, 715 iterations

    For all three renders, gpu utilization was in the 95 to 97% range and both cards ran at the maximum GPU core and memory clock speed as reported by GPU-Z. I've never had either card throttle back when both were in use except in those cases where the scene was over 6 GB and the 980TI got dropped.

     

    I'd very much like to see GPU-Z graphs showing a faster card being throttled to match a slower one. Remember, Nvidia does NOT recommend using SLI with Iray.

     

    The concept of the render engine not utilising mixed generations to their maximum would not be visibile through hardware usage monitoring. I remember reading an article that explained this much much better than i ever could ages ago, im trying to remember where it was from so i can find it again.

    Its not that the render engine throttles any hardware, its more how the engine uses the power available to it from two components that are different at the firmware level. Think about how render engines need to be updated to support new generation cards as they are released, that indicates that there is something different going on in the engine's code depending on the generation of card. How that translates to any performance loss is probably too technical for most people (myself included), but that article i referred to did explain it in a way that most can understand 

    My suspicion is that the real impact is on the cpu side of Iray as it integrates the output of multiple disparate speed gpus. With the possibility of a faster cpu tending to lessen the impact of the disparate speed. I think I'm seeing a bit of this, but I don't have the proper instrumentation currently to be sure - and running an I7 5930K at 3.5 GHz doesn't help on this short render. I may play with this a bit if I can find the time.

  • surodysurody Posts: 63
    namffuak said:
    joseft said:
    namffuak said:
    areg5 said:

    I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

    1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

    2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

    3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

    4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

     

    So, those are my observations.

    From my tests what i was able to deduce is that there is modest gains between generations, the biggest gain between generation is speed of the GPU. Problem with mixing old cards and new is that iray tends to slow all cards down to the speed of the slowest card. That's why it looks like adding a new gen card does not make much difference if your mixing it with an old gen card.

    I keep seeing people say this but I never see it.

     

    Real times - 1 g8f, with outfit, default lighting

    980TI:           4 minutes 4.5 seconds, 731 iterations

    1080TI          2 minutes 43.9 seconds, 735 iterations

    Both            1 minute 49.9 seconds, 715 iterations

    For all three renders, gpu utilization was in the 95 to 97% range and both cards ran at the maximum GPU core and memory clock speed as reported by GPU-Z. I've never had either card throttle back when both were in use except in those cases where the scene was over 6 GB and the 980TI got dropped.

     

    I'd very much like to see GPU-Z graphs showing a faster card being throttled to match a slower one. Remember, Nvidia does NOT recommend using SLI with Iray.

     

    The concept of the render engine not utilising mixed generations to their maximum would not be visibile through hardware usage monitoring. I remember reading an article that explained this much much better than i ever could ages ago, im trying to remember where it was from so i can find it again.

    Its not that the render engine throttles any hardware, its more how the engine uses the power available to it from two components that are different at the firmware level. Think about how render engines need to be updated to support new generation cards as they are released, that indicates that there is something different going on in the engine's code depending on the generation of card. How that translates to any performance loss is probably too technical for most people (myself included), but that article i referred to did explain it in a way that most can understand 

    My suspicion is that the real impact is on the cpu side of Iray as it integrates the output of multiple disparate speed gpus. With the possibility of a faster cpu tending to lessen the impact of the disparate speed. I think I'm seeing a bit of this, but I don't have the proper instrumentation currently to be sure - and running an I7 5930K at 3.5 GHz doesn't help on this short render. I may play with this a bit if I can find the time.

    Linus says that you shouldn't mix GPUs with different memory capacities. He doesn't say any specifics why but I would assume because the system is throttled down to the 6GB of the 980ti in this case and doesn't utilize the 1080ti memory.

  • surody said:
    namffuak said:
    joseft said:
    namffuak said:
    areg5 said:

    I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

    1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

    2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

    3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

    4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

     

    So, those are my observations.

    From my tests what i was able to deduce is that there is modest gains between generations, the biggest gain between generation is speed of the GPU. Problem with mixing old cards and new is that iray tends to slow all cards down to the speed of the slowest card. That's why it looks like adding a new gen card does not make much difference if your mixing it with an old gen card.

    I keep seeing people say this but I never see it.

     

    Real times - 1 g8f, with outfit, default lighting

    980TI:           4 minutes 4.5 seconds, 731 iterations

    1080TI          2 minutes 43.9 seconds, 735 iterations

    Both            1 minute 49.9 seconds, 715 iterations

    For all three renders, gpu utilization was in the 95 to 97% range and both cards ran at the maximum GPU core and memory clock speed as reported by GPU-Z. I've never had either card throttle back when both were in use except in those cases where the scene was over 6 GB and the 980TI got dropped.

     

    I'd very much like to see GPU-Z graphs showing a faster card being throttled to match a slower one. Remember, Nvidia does NOT recommend using SLI with Iray.

     

    The concept of the render engine not utilising mixed generations to their maximum would not be visibile through hardware usage monitoring. I remember reading an article that explained this much much better than i ever could ages ago, im trying to remember where it was from so i can find it again.

    Its not that the render engine throttles any hardware, its more how the engine uses the power available to it from two components that are different at the firmware level. Think about how render engines need to be updated to support new generation cards as they are released, that indicates that there is something different going on in the engine's code depending on the generation of card. How that translates to any performance loss is probably too technical for most people (myself included), but that article i referred to did explain it in a way that most can understand 

    My suspicion is that the real impact is on the cpu side of Iray as it integrates the output of multiple disparate speed gpus. With the possibility of a faster cpu tending to lessen the impact of the disparate speed. I think I'm seeing a bit of this, but I don't have the proper instrumentation currently to be sure - and running an I7 5930K at 3.5 GHz doesn't help on this short render. I may play with this a bit if I can find the time.

    Linus says that you shouldn't mix GPUs with different memory capacities. He doesn't say any specifics why but I would assume because the system is throttled down to the 6GB of the 980ti in this case and doesn't utilize the 1080ti memory.

    He's talking about gaming, not rendering.

  • areg5areg5 Posts: 379
    surody said:
    areg5 said:
    areg5 said:

    I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

    1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

    2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

    3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

    4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

     

    So, those are my observations.

    From my tests what i was able to deduce is that there is modest gains between generations, the biggest gain between generation is speed of the GPU. Problem with mixing old cards and new is that iray tends to slow all cards down to the speed of the slowest card. That's why it looks like adding a new gen card does not make much difference if your mixing it with an old gen card.

     

     

    I'm not saying there's no difference at all, I'm saying that any difference is marginal.

     

    I got 2 rigs, 1 has 2x 980ti and the newer has 2x 1080ti. If you say the performance boost is marginal it's probably because you are mixing generations. My newer build with 2x 1080ti is about 40% faster in my tests compared to the 980ti rig. Besides that it isn't recommended to use more than 2 consumer cards in SLI. The performance drop is too high and the RAM doesn't stack. The performance dropoff seems to be fixed with NVLink and the 2000 generation though.

    SLI?  It's not recommended to use SLI with Iray.  None of my cards are in SLI.  SLI makes the 2 cards act as one, and as I understand it (which could be wrong) would essentially half the number of CUDA cores.  SLI is great for gaming.  It's not meant for rendering.  I would see what your times are running each rig without SLI.  Also, for scenes that use more than 6 gigs of VRAM, I wasn't mixing generations because the 980Ti won't run at that load, so I was using only the 1080Ti's.

    Now, one of things that will throttle the card down is the temp.  The 980Ti runs much hotter than the 1080Ti, and it throttles back when it reaches a certain temperature.  This can be avoided by having a very well cooled case.

  • nicsttnicstt Posts: 7,628

    I suppose the best test would be a scene using just one card from each generation; throttle the faster card to the speed of the slower. Monitor the render to see if the speed matches the throttle.

    divide the total time in seconds by the cuda cores, and then multiply the time of one cuda core by the number of cuda cores in the card with the greatest number. That should give a reasonable comparrison if performed a few times. Once might be sugestive but by no means conclusive due to random factors affecting one render.

  • areg5areg5 Posts: 379
    nicstt said:

    I suppose the best test would be a scene using just one card from each generation; throttle the faster card to the speed of the slower. Monitor the render to see if the speed matches the throttle.

    divide the total time in seconds by the cuda cores, and then multiply the time of one cuda core by the number of cuda cores in the card with the greatest number. That should give a reasonable comparrison if performed a few times. Once might be sugestive but by no means conclusive due to random factors affecting one render.

    That's a lot of work!  The bottom line for me is that I just haven't seen an enormous difference, and the primary benefit of the new cards is more VRAM and cooler temps.

  • namffuaknamffuak Posts: 2,747
    areg5 said:
    surody said:
    areg5 said:
    areg5 said:

    I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

    1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

    2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

    3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

    4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

     

    So, those are my observations.

    From my tests what i was able to deduce is that there is modest gains between generations, the biggest gain between generation is speed of the GPU. Problem with mixing old cards and new is that iray tends to slow all cards down to the speed of the slowest card. That's why it looks like adding a new gen card does not make much difference if your mixing it with an old gen card.

     

     

    I'm not saying there's no difference at all, I'm saying that any difference is marginal.

     

    I got 2 rigs, 1 has 2x 980ti and the newer has 2x 1080ti. If you say the performance boost is marginal it's probably because you are mixing generations. My newer build with 2x 1080ti is about 40% faster in my tests compared to the 980ti rig. Besides that it isn't recommended to use more than 2 consumer cards in SLI. The performance drop is too high and the RAM doesn't stack. The performance dropoff seems to be fixed with NVLink and the 2000 generation though.

    SLI?  It's not recommended to use SLI with Iray.  None of my cards are in SLI.  SLI makes the 2 cards act as one, and as I understand it (which could be wrong) would essentially half the number of CUDA cores.  SLI is great for gaming.  It's not meant for rendering.  I would see what your times are running each rig without SLI.  Also, for scenes that use more than 6 gigs of VRAM, I wasn't mixing generations because the 980Ti won't run at that load, so I was using only the 1080Ti's.

    Now, one of things that will throttle the card down is the temp.  The 980Ti runs much hotter than the 1080Ti, and it throttles back when it reaches a certain temperature.  This can be avoided by having a very well cooled case.

    A well-ventilated case is a must for more than one of these cards. Another highly recommended tweak is to use custom fan profiles; the default is set up for gaming, where the card hits 80 to 90% gpu utilization intermittently and for short time periods. I've run both the 1080TI and 980TI at 97% gpu for hours at a time with th hotter of the two (980ti) staying a good 10 degrees below the throttling point.

  • surodysurody Posts: 63
    areg5 said:
    surody said:
    areg5 said:
    areg5 said:

    I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

    1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

    2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

    3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

    4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

     

    So, those are my observations.

    From my tests what i was able to deduce is that there is modest gains between generations, the biggest gain between generation is speed of the GPU. Problem with mixing old cards and new is that iray tends to slow all cards down to the speed of the slowest card. That's why it looks like adding a new gen card does not make much difference if your mixing it with an old gen card.

     

     

    I'm not saying there's no difference at all, I'm saying that any difference is marginal.

     

    I got 2 rigs, 1 has 2x 980ti and the newer has 2x 1080ti. If you say the performance boost is marginal it's probably because you are mixing generations. My newer build with 2x 1080ti is about 40% faster in my tests compared to the 980ti rig. Besides that it isn't recommended to use more than 2 consumer cards in SLI. The performance drop is too high and the RAM doesn't stack. The performance dropoff seems to be fixed with NVLink and the 2000 generation though.

    SLI?  It's not recommended to use SLI with Iray.  None of my cards are in SLI.  SLI makes the 2 cards act as one, and as I understand it (which could be wrong) would essentially half the number of CUDA cores.  SLI is great for gaming.  It's not meant for rendering.  I would see what your times are running each rig without SLI.  Also, for scenes that use more than 6 gigs of VRAM, I wasn't mixing generations because the 980Ti won't run at that load, so I was using only the 1080Ti's.

    Now, one of things that will throttle the card down is the temp.  The 980Ti runs much hotter than the 1080Ti, and it throttles back when it reaches a certain temperature.  This can be avoided by having a very well cooled case.

    What's so bad about SLI with IRAY? I did a test now. One without SLI and with SLI. The render time is exactly the same. Both tests needed exactly 101 seconds to reach 95%, same scene, same settings.

  • surody said:
    areg5 said:
    surody said:
    areg5 said:
    areg5 said:

    I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

    1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

    2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

    3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

    4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

     

    So, those are my observations.

    From my tests what i was able to deduce is that there is modest gains between generations, the biggest gain between generation is speed of the GPU. Problem with mixing old cards and new is that iray tends to slow all cards down to the speed of the slowest card. That's why it looks like adding a new gen card does not make much difference if your mixing it with an old gen card.

     

     

    I'm not saying there's no difference at all, I'm saying that any difference is marginal.

     

    I got 2 rigs, 1 has 2x 980ti and the newer has 2x 1080ti. If you say the performance boost is marginal it's probably because you are mixing generations. My newer build with 2x 1080ti is about 40% faster in my tests compared to the 980ti rig. Besides that it isn't recommended to use more than 2 consumer cards in SLI. The performance drop is too high and the RAM doesn't stack. The performance dropoff seems to be fixed with NVLink and the 2000 generation though.

    SLI?  It's not recommended to use SLI with Iray.  None of my cards are in SLI.  SLI makes the 2 cards act as one, and as I understand it (which could be wrong) would essentially half the number of CUDA cores.  SLI is great for gaming.  It's not meant for rendering.  I would see what your times are running each rig without SLI.  Also, for scenes that use more than 6 gigs of VRAM, I wasn't mixing generations because the 980Ti won't run at that load, so I was using only the 1080Ti's.

    Now, one of things that will throttle the card down is the temp.  The 980Ti runs much hotter than the 1080Ti, and it throttles back when it reaches a certain temperature.  This can be avoided by having a very well cooled case.

    What's so bad about SLI with IRAY? I did a test now. One without SLI and with SLI. The render time is exactly the same. Both tests needed exactly 101 seconds to reach 95%, same scene, same settings.

    In the past nVidia has strongly advised against it, and people have reported issues. Those may have been resolved, but I'd still be inclined to turn it off (if i had two GPUs) since it apparently does no good.

  • surodysurody Posts: 63
    surody said:
    areg5 said:
    surody said:
    areg5 said:
    areg5 said:

    I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

    1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

    2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

    3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

    4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

     

    So, those are my observations.

    From my tests what i was able to deduce is that there is modest gains between generations, the biggest gain between generation is speed of the GPU. Problem with mixing old cards and new is that iray tends to slow all cards down to the speed of the slowest card. That's why it looks like adding a new gen card does not make much difference if your mixing it with an old gen card.

     

     

    I'm not saying there's no difference at all, I'm saying that any difference is marginal.

     

    I got 2 rigs, 1 has 2x 980ti and the newer has 2x 1080ti. If you say the performance boost is marginal it's probably because you are mixing generations. My newer build with 2x 1080ti is about 40% faster in my tests compared to the 980ti rig. Besides that it isn't recommended to use more than 2 consumer cards in SLI. The performance drop is too high and the RAM doesn't stack. The performance dropoff seems to be fixed with NVLink and the 2000 generation though.

    SLI?  It's not recommended to use SLI with Iray.  None of my cards are in SLI.  SLI makes the 2 cards act as one, and as I understand it (which could be wrong) would essentially half the number of CUDA cores.  SLI is great for gaming.  It's not meant for rendering.  I would see what your times are running each rig without SLI.  Also, for scenes that use more than 6 gigs of VRAM, I wasn't mixing generations because the 980Ti won't run at that load, so I was using only the 1080Ti's.

    Now, one of things that will throttle the card down is the temp.  The 980Ti runs much hotter than the 1080Ti, and it throttles back when it reaches a certain temperature.  This can be avoided by having a very well cooled case.

    What's so bad about SLI with IRAY? I did a test now. One without SLI and with SLI. The render time is exactly the same. Both tests needed exactly 101 seconds to reach 95%, same scene, same settings.

    In the past nVidia has strongly advised against it, and people have reported issues. Those may have been resolved, but I'd still be inclined to turn it off (if i had two GPUs) since it apparently does no good.

    I used SLI longterm now while rendering and I never faced any issues to be honest. At least not with the 1000+ generation.

  • areg5areg5 Posts: 379
    namffuak said:
    areg5 said:
    surody said:
    areg5 said:
    areg5 said:

    I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

    1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

    2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

    3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

    4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

     

    So, those are my observations.

    From my tests what i was able to deduce is that there is modest gains between generations, the biggest gain between generation is speed of the GPU. Problem with mixing old cards and new is that iray tends to slow all cards down to the speed of the slowest card. That's why it looks like adding a new gen card does not make much difference if your mixing it with an old gen card.

     

     

    I'm not saying there's no difference at all, I'm saying that any difference is marginal.

     

    I got 2 rigs, 1 has 2x 980ti and the newer has 2x 1080ti. If you say the performance boost is marginal it's probably because you are mixing generations. My newer build with 2x 1080ti is about 40% faster in my tests compared to the 980ti rig. Besides that it isn't recommended to use more than 2 consumer cards in SLI. The performance drop is too high and the RAM doesn't stack. The performance dropoff seems to be fixed with NVLink and the 2000 generation though.

    SLI?  It's not recommended to use SLI with Iray.  None of my cards are in SLI.  SLI makes the 2 cards act as one, and as I understand it (which could be wrong) would essentially half the number of CUDA cores.  SLI is great for gaming.  It's not meant for rendering.  I would see what your times are running each rig without SLI.  Also, for scenes that use more than 6 gigs of VRAM, I wasn't mixing generations because the 980Ti won't run at that load, so I was using only the 1080Ti's.

    Now, one of things that will throttle the card down is the temp.  The 980Ti runs much hotter than the 1080Ti, and it throttles back when it reaches a certain temperature.  This can be avoided by having a very well cooled case.

    A well-ventilated case is a must for more than one of these cards. Another highly recommended tweak is to use custom fan profiles; the default is set up for gaming, where the card hits 80 to 90% gpu utilization intermittently and for short time periods. I've run both the 1080TI and 980TI at 97% gpu for hours at a time with th hotter of the two (980ti) staying a good 10 degrees below the throttling point.

    I use the EVGA Precision XOC sometimes.  I used to use it all the time to keep the 980Ti's cooler.  With the 1080Ti it doesn't seem necessary.  My case is very well ventilated.  My mobo, however, really jams those PCIex slots close together.  Not too badly between the first and second cards, but the second and third are nearly touching.  So the card in the middle is always the hottest, but it hasn't gone over 78 degrees even after a very long render.

  • areg5areg5 Posts: 379
    surody said:
    surody said:
    areg5 said:
    surody said:
    areg5 said:
    areg5 said:

    I thought some of you might be interested in my experience running 3 cards.  I've been doing it for a while, and I recently added my third 1080Ti to replace a 980 Ti.  Here are some obsertvations:

    1.  My first 3 card setup was 980TiX2/780 Ti.  Now, you're going to hear some noise out there regarding the CUDA generation, i.e. how they aren't comparable one generation to the next, the newest generation is always faster, and that sort of thing.  If there is a difference, it's miniscule.  For a scene that results in less than 3 gig VRAM, my first setup was really, really fast.  There was virtually no difference in that build and my current 1080Tix3 build in running the Sickleyield Benchmark.  I have found that it just doesn't matter if the Cuda cores are Keppler, Maxwell or Paschal.  The key is now, and has always been, the number of CUDA cores not the generation.  Some might disagree and point to benchmarks showing a difference.  In practical, daily use, there isn't one.

    2.  For a while I was running 980Tix2/1080Ti.  Again, basically just as fast as my current build.

    3.  The main benefit of my current build is that I can do more complex scenes without the feared GPU crash to CPU.  I had gotten very used to doing big scenes, and deleting much of what was not in the viewport so it would fit the 6 GIG capacity of the 980Ti's.  Now, I have yet to do that, although having a smaller scene does make it load faster.

    4.  From what I have read about the new 2080Ti's is that there are more cores than the 1080Ti (4352 vs. 3584).  So it should be a bit faster than the 1080Ti.  The question boils down to this:  are 800 more cores worth the ridiculous price tag?  The cheapest one on the EVGA site sells for 1000 bucks.  I just got my latest 1080Ti on EBAY for 580.  The newer card is way too expensive IMHO.  What about the newer tech behind the cores?  That part doesn't matter.  Really, the new card has raytracing ability, That may be an issue with some games, but not with running IRAY.

     

    So, those are my observations.

    From my tests what i was able to deduce is that there is modest gains between generations, the biggest gain between generation is speed of the GPU. Problem with mixing old cards and new is that iray tends to slow all cards down to the speed of the slowest card. That's why it looks like adding a new gen card does not make much difference if your mixing it with an old gen card.

     

     

    I'm not saying there's no difference at all, I'm saying that any difference is marginal.

     

    I got 2 rigs, 1 has 2x 980ti and the newer has 2x 1080ti. If you say the performance boost is marginal it's probably because you are mixing generations. My newer build with 2x 1080ti is about 40% faster in my tests compared to the 980ti rig. Besides that it isn't recommended to use more than 2 consumer cards in SLI. The performance drop is too high and the RAM doesn't stack. The performance dropoff seems to be fixed with NVLink and the 2000 generation though.

    SLI?  It's not recommended to use SLI with Iray.  None of my cards are in SLI.  SLI makes the 2 cards act as one, and as I understand it (which could be wrong) would essentially half the number of CUDA cores.  SLI is great for gaming.  It's not meant for rendering.  I would see what your times are running each rig without SLI.  Also, for scenes that use more than 6 gigs of VRAM, I wasn't mixing generations because the 980Ti won't run at that load, so I was using only the 1080Ti's.

    Now, one of things that will throttle the card down is the temp.  The 980Ti runs much hotter than the 1080Ti, and it throttles back when it reaches a certain temperature.  This can be avoided by having a very well cooled case.

    What's so bad about SLI with IRAY? I did a test now. One without SLI and with SLI. The render time is exactly the same. Both tests needed exactly 101 seconds to reach 95%, same scene, same settings.

    In the past nVidia has strongly advised against it, and people have reported issues. Those may have been resolved, but I'd still be inclined to turn it off (if i had two GPUs) since it apparently does no good.

    I used SLI longterm now while rendering and I never faced any issues to be honest. At least not with the 1000+ generation.

    Ok.  Here's what SLI does:  it links 2 cards together which makes them run faster for gaming.  It's a real slick thing for gaming.  For Iray, what it does is make the program see both of those cards as one card.  Specifically, it makes Iray see the CUDA cores of one card.  You're running 2 cards, but bear in mind all the rendering power comes from the CUDA cores.  What I would do is a test render ... a complex one ... with SLI.  Then disable SLI and render again.  Compare the times.  Use whatever is faster.

  • Hello, i didn't know where to post so i will do it here. 

     

    My current PC is i7 6800k, gpu 1080 evga hybrid with water cooler, 32 gb Ram and asus deluxe II x99. 1200 PSU Corsair.

     

    My question is, since 1080 series is pretty much done, what gpu can i add to this composition, can i add a 1070 or 2080 or something else? Will it work? Is it better to use SLI or not. Just for rendering in Daz Studios 4.11

     

    Thank you.

  • areg5areg5 Posts: 379

    I

    Hello, i didn't know where to post so i will do it here. 

     

    My current PC is i7 6800k, gpu 1080 evga hybrid with water cooler, 32 gb Ram and asus deluxe II x99. 1200 PSU Corsair.

     

    My question is, since 1080 series is pretty much done, what gpu can i add to this composition, can i add a 1070 or 2080 or something else? Will it work? Is it better to use SLI or not. Just for rendering in Daz Studios 4.11

     

    Thank you.

    You don't want to use SLI.  SLI is great for gaming but isn't made for IRAY.  It essentially makes the 2 connected cards act like one, meaning that fewer of the CUDA cores are rendering.  Some of the other comments here denigrate mixing cards, but I've been mixing cards for years.  Works great!  Your goal is to have as many CUDA cores working for you as possible, and ideally as much VRAM as possible.  That being said, I would go with the 1080Ti.  You can get one on Ebay these days for a reasonable price.  I bought 2 cards there last year, one for $800 and one for 780.  Just bought one a couple of weeks ago for $580.  I only get EVGA cards because they have a 3 year transferrable warranty.  The 2080 has more CUDA cores, but the same VRAM.  Nice, but the price has to come down before it would really be cost effective.   I 've always been a fan of slightly older tech.  If the 1080 was out, I got 980's.  That sort of thing.  I did take the plunge with the 1080Ti when it was still "the" card, but only because I wanted more VRAM for bigger scenes.  It's worth pointing out that if your scenes are not typically larger than 6 gig VRAM, that the 980Ti is a great card, and I would bet you could get one for about $400 bucks.

  • https://www.asus.com/Motherboards/X99-DELUXE-II/

     

    I don't know if my Mobo can be used without SLI. I mean i am not a tech guy, that's why i ask here. Money isn't a problem here, i wanted to buy another 1 maybe 2 gpu 1080 evga hybrid with water cooler but there are no 1080 left, i mean like it disappeared from market. Now i have to find alternative for it .That is why i am asking, if i can make it possible to put one or maybe another gpu without SLI with my current Mobo and if i can, what gpu's can i get 2080 or 1080ti or 1070 ti ...

  • areg5areg5 Posts: 379

    https://www.asus.com/Motherboards/X99-DELUXE-II/

     

    I don't know if my Mobo can be used without SLI. I mean i am not a tech guy, that's why i ask here. Money isn't a problem here, i wanted to buy another 1 maybe 2 gpu 1080 evga hybrid with water cooler but there are no 1080 left, i mean like it disappeared from market. Now i have to find alternative for it .That is why i am asking, if i can make it possible to put one or maybe another gpu without SLI with my current Mobo and if i can, what gpu's can i get 2080 or 1080ti or 1070 ti ...

    SLI is an option, but you turn it on and off using the nvidia control panel.  If that's your mobo above, it would be fine.  You can find 1080Ti's on Ebay.  That's where I got all of mine.  Hybids run about 10 degrees cooler than the regular ones, but the 1080 runs cool in a well wentelated case.

  • surodysurody Posts: 63
    edited January 9

    Hello, i didn't know where to post so i will do it here. 

     

    My current PC is i7 6800k, gpu 1080 evga hybrid with water cooler, 32 gb Ram and asus deluxe II x99. 1200 PSU Corsair.

     

    My question is, since 1080 series is pretty much done, what gpu can i add to this composition, can i add a 1070 or 2080 or something else? Will it work? Is it better to use SLI or not. Just for rendering in Daz Studios 4.11

     

    Thank you.

     

    It really depends on how long you want to use your cards and for what specifically you are using them. If you are able to wait for the 2000+ generation to drop in price I maybe would do so and hope that Daz/IRAY supports the raytracing cores the next few months.

    Post edited by surody on
  • areg5areg5 Posts: 379
    surody said:

    Hello, i didn't know where to post so i will do it here. 

     

    My current PC is i7 6800k, gpu 1080 evga hybrid with water cooler, 32 gb Ram and asus deluxe II x99. 1200 PSU Corsair.

     

    My question is, since 1080 series is pretty much done, what gpu can i add to this composition, can i add a 1070 or 2080 or something else? Will it work? Is it better to use SLI or not. Just for rendering in Daz Studios 4.11

     

    Thank you.

     

    It really depends on how long you want to use your cards and for what specifically you are using them. If you are able to wait for the 2000+ generation to drop in price I maybe would do so and hope that Daz/IRAY supports the raytracing cores the next few months.

    Personally, at this point what I would be looking for in a new card would be even more VRAM.  I doubt raytracing will be a part of Iray any time soon, if at all.  The newest generation does have 800 more CUDA, so it should be a bit faster from that standpoint.  Cost wise, I don't like what nvidia has done.  When the 980Ti was new, it was around 700 I think.  When the 1080ti was new, around 800.  Now you're looking at 1000 for the cheapest 2080ti.  That tells me that when the next one comes out, the 2080ti comes down to what, 800?  900?  That's a lot of money.  The performance difference might not be worth it.

  • I have Corsair 780t as my towercase, I got corsair h115i for my CPU, and the cooler from my GPU that means top is for CPU and back is for GPU. i still have space for front but idk if that is a good ideea to remove the towercase coolers just to put another 1 maybe 2 coolers for my gpu.

  • areg5areg5 Posts: 379

    I have Corsair 780t as my towercase, I got corsair h115i for my CPU, and the cooler from my GPU that means top is for CPU and back is for GPU. i still have space for front but idk if that is a good ideea to remove the towercase coolers just to put another 1 maybe 2 coolers for my gpu.

    How many fans are in that case?  It looks well ventilated from what I see online.  I'm completely aircooled, and it works fine.  The 1080Ti's run pretty cool as is.  I don't think the hybrid is a big advantage.  I know the whole rationale behind watercooling, and I think it was a bigger issue for the older cards that could run sort of hot, but not for the newer ones.

  • My first gpu is GDDR5X and the 2080 series is GDDR6X it isn't an issue i think right?

Sign In or Register to comment.