Loading...
> >

OT: New Nvidia Cards to come in RTX and GTX versions?! RTX Titan first whispers.

1…910121415…27

Comments

  • MendomanMendoman Posts: 404
    edited August 2018

    Here's a link to an article I found about RTX. It's from chaos group ( V-ray developers ): https://www.chaosgroup.com/blog/what-does-the-new-nvidia-rtx-hardware-mean-for-ray-tracing-gpu-rendering-v-ray

     

    In that article there's also little example about Nvlink, and they also talk about geforce cards:

    "NVLink is a technology that allows two or more GPUs to be connected with a bridge and share data extremely fast. This means that each GPU can access the memory of the other GPU and programs like V-Ray GPU can take advantage of that to render scenes that are too large to fit on a single card. Traditionally when rendering on multiple graphics cards, V-Ray GPU duplicates the data in the memory of each GPU, but with NVLink the VRAM can be pooled. For example, if we have two GPUs with 11GB of VRAM each and connected with NVLink, V-Ray GPU can use that to render scenes that take up to 22 GB. This is completely transparent to the user — V-Ray GPU automatically detects and uses NVLink when available. So, while in the past doubling your cards only allowed you to double your speed, now with NVLink you can also double your VRAM.


    NVLink was introduced in 2016 and V-Ray GPU was the first renderer to support it officially in V-Ray 3.6 and newer versions. Until now, the technology has only been available on professional Quadro and Tesla cards, but with the release of the RTX series, NVLink is also available on gaming GPUs - specifically on the GeForce RTX 2080 and GeForce RTX 2080 Ti. Connecting two cards with NVLink requires a special NVLink connector, which is sold separately."

    Post edited by Mendoman on
  • bluejauntebluejaunte Posts: 1,990
    Mendoman said:

    Here's a link to an article I found about RTX. It's from chaos group ( V-ray developers ): https://www.chaosgroup.com/blog/what-does-the-new-nvidia-rtx-hardware-mean-for-ray-tracing-gpu-rendering-v-ray

     

    In that article there's also little example about Nvlink, and they also talk about geforce cards:

    "NVLink is a technology that allows two or more GPUs to be connected with a bridge and share data extremely fast. This means that each GPU can access the memory of the other GPU and programs like V-Ray GPU can take advantage of that to render scenes that are too large to fit on a single card. Traditionally when rendering on multiple graphics cards, V-Ray GPU duplicates the data in the memory of each GPU, but with NVLink the VRAM can be pooled. For example, if we have two GPUs with 11GB of VRAM each and connected with NVLink, V-Ray GPU can use that to render scenes that take up to 22 GB. This is completely transparent to the user — V-Ray GPU automatically detects and uses NVLink when available. So, while in the past doubling your cards only allowed you to double your speed, now with NVLink you can also double your VRAM.


    NVLink was introduced in 2016 and V-Ray GPU was the first renderer to support it officially in V-Ray 3.6 and newer versions. Until now, the technology has only been available on professional Quadro and Tesla cards, but with the release of the RTX series, NVLink is also available on gaming GPUs - specifically on the GeForce RTX 2080 and GeForce RTX 2080 Ti. Connecting two cards with NVLink requires a special NVLink connector, which is sold separately."

    Yeah I linked this earlier and we discussed it. Ultimately I wasn't convinced that they got this right, because they were testing on Quadros and they may have just thought that the gaming cards will have the same full featured NVLINK too. After what outrider just posted though (Our friends at OTOY say that they have been told the gaming NVLink is fully functional), the hope lives I guess smiley

  • MendomanMendoman Posts: 404

    @bluejaunte Oops, sorry. I must have missed that :/

  • billyben_0077a25354billyben_0077a25354 Posts: 771
    edited August 2018
    Taoz said:
    bluejaunte said:
    ebergerly said:
    Here's what the NVIDIA website says: "The GeForce RTX NVLink™ bridge connects two NVLink SLI-ready graphics cards with 50X the transfer bandwidth of previous technologies. This means you can count on super-smooth gameplay at maximum resolutions with ultimate visual fidelity in GeForce RTX 2080 Ti and 2080 graphics cards." And nothing about memory stacking.

    Hmm ok, so the SLI term is here to stay? Guess I had the wrong idea. SLI has such negative connotation, would have thought they would replace it. But maybe the underlying tech is still SLI, just much faster with NVLINK.

    SLI means "Scalable Link Interface" which sounds like a rather generic term which can be used in many different contexts.

    No, SLI is supposed to mean Scan-Line Interleave.  Unfortunately NVidia put 3DFX out of bussiness and completely ruined the term SLI.  Back in the day I ran Voodo & Voodo 2 cards (first 2 Voodo's then upgraded to 2 Voodoo 2 cards) in SLI and it was nirvana on an OG 21" CRT.  Boy have times changed.

    Post edited by billyben_0077a25354 on
  • outrider42outrider42 Posts: 3,679
    edited August 2018
    kyoto kid said:

    ...still not buying it until I see actual proof, not just what "someone says". 

    Were that true that would mean, a 4 slot NVLink bridge should allow for 44 GB with 4 2080 Ti's.

     

    You can't do 4 way NVLink on gaming regardless. All the card listings I have seen only show 2 way NVLink support. So no you cannot do a 44 gb setup with 4 2080ti's. You can only do 2 for 22 gb. There are multiple sources that show 2 way support on 2080ti and 2080, including the PNY page linked before. Also, while the order page does not mention any features of NVLink, the cards do not discuss their features much, either. On Nvidias own page, you do not see any mention of Tensor cores. When you look up most preorders, they only show CUDA core counts, and no core counts for Tensor and RT. That would seem to be a pretty important thing to leave out.

    It makes sense to me that NVLink would stack. For one, Nvidia killed NVLink on the 2070, so you cannot do this too cheaply. On the 2080 and 2080ti, these are cards that are $800 and $1200 each, so even at the very cheapest, you are still spending about as much as a single Quadro might cost, and you are capped at only 2. If NVLink is simply SLI with a bit higher bandwidth, that seems really basic. That Nvidia charges over $450 for Quadro NVLink is not relevant when Quadros themselves are $10,000. Quadros can be 5 to 10 times more expensive, so of course their NVLink would also be 5 to 10 times more expensive. And the gaming version has a big restriction, since it only supports 2 cards.

    Interestingly, I have found a number of people claiming that NVLink is under NDA. That is rather odd, but that would explain why none of the manufacturers are responding to questions about NVLink. There are forum posts on EVGA and Nvidia asking this question. I have asked this question directly on Nvidia's forum. Days later there is still no response. So I starting to think there really is a NDA in place.

    This link should work, I was on mobile before. This is actually a different tweet anyway, directly from OTOY themselves. Scroll the replies.

    https://twitter.com/OTOY/status/1034599997583572998

    The other tweet

    https://twitter.com/AaronCovrett/status/1034459411060809729

    Post edited by outrider42 on
  • bluejauntebluejaunte Posts: 1,990

    What I don't get it why they would be secretive about this. Now is the time for all the marketing hype in the world, why would they not mention that the memory stacks for gamer cards? This is really the main thing that makes me sceptical. If there is an NDA then even more so. Why would they want to shut people up about a very positive feature. Unless that feature isn't there at all.

  • kyoto kidkyoto kid Posts: 41,845
    edited August 2018

     

     

    outrider42 said:
    kyoto kid said:

    ...still not buying it until I see actual proof, not just what "someone says". 

    Were that true that would mean, a 4 slot NVLink bridge should allow for 44 GB with 4 2080 Ti's.

     

    You can't do 4 way NVLink on gaming regardless. All the card listings I have seen only show 2 way NVLink support. So no you cannot do a 44 gb setup with 4 2080ti's. You can only do 2 for 22 gb. There are multiple sources that show 2 way support on 2080ti and 2080, including the PNY page linked before. Also, while the order page does not mention any features of NVLink, the cards do not discuss their features much, either. On Nvidias own page, you do not see any mention of Tensor cores. When you look up most preorders, they only show CUDA core counts, and no core counts for Tensor and RT. That would seem to be a pretty important thing to leave out.

    It makes sense to me that NVLink would stack. For one, Nvidia killed NVLink on the 2070, so you cannot do this too cheaply. On the 2080 and 2080ti, these are cards that are $800 and $1200 each, so even at the very cheapest, you are still spending about as much as a single Quadro might cost, and you are capped at only 2. If NVLink is simply SLI with a bit higher bandwidth, that seems really basic. That Nvidia charges over $450 for Quadro NVLink is not relevant when Quadros themselves are $10,000. Quadros can be 5 to 10 times more expensive, so of course their NVLink would also be 5 to 10 times more expensive. And the gaming version has a big restriction, since it only supports 2 cards.

    Interestingly, I have found a number of people claiming that NVLink is under NDA. That is rather odd, but that would explain why none of the manufacturers are responding to questions about NVLink. There are forum posts on EVGA and Nvidia asking this question. I have asked this question directly on Nvidia's forum. Days later there is still no response. So I starting to think there really is a NDA in place.

    This link should work, I was on mobile before. This is actually a different tweet anyway, directly from OTOY themselves. Scroll the replies.

    https://twitter.com/OTOY/status/1034599997583572998

    The other tweet

    https://twitter.com/AaronCovrett/status/1034459411060809729

    ...well then why for GeForce is it being sold in 2, 3, and 4 slot versions like SLI bridges were instead of just a single pair of widgets like for Quadro and Tesla cards? You need two bridges between a pair of cards to get the full benefit of the link.
    Post edited by kyoto kid on
  • kyoto kidkyoto kid Posts: 41,845
    bluejaunte said:

    What I don't get it why they would be secretive about this. Now is the time for all the marketing hype in the world, why would they not mention that the memory stacks for gamer cards? This is really the main thing that makes me sceptical. If there is an NDA then even more so. Why would they want to shut people up about a very positive feature. Unless that feature isn't there at all.

    ...+1
  • outrider42outrider42 Posts: 3,679
    kyoto kid said:

     

     

    outrider42 said:
    kyoto kid said:

    ...still not buying it until I see actual proof, not just what "someone says". 

    Were that true that would mean, a 4 slot NVLink bridge should allow for 44 GB with 4 2080 Ti's.

     

    You can't do 4 way NVLink on gaming regardless. All the card listings I have seen only show 2 way NVLink support. So no you cannot do a 44 gb setup with 4 2080ti's. You can only do 2 for 22 gb. There are multiple sources that show 2 way support on 2080ti and 2080, including the PNY page linked before. Also, while the order page does not mention any features of NVLink, the cards do not discuss their features much, either. On Nvidias own page, you do not see any mention of Tensor cores. When you look up most preorders, they only show CUDA core counts, and no core counts for Tensor and RT. That would seem to be a pretty important thing to leave out.

    It makes sense to me that NVLink would stack. For one, Nvidia killed NVLink on the 2070, so you cannot do this too cheaply. On the 2080 and 2080ti, these are cards that are $800 and $1200 each, so even at the very cheapest, you are still spending about as much as a single Quadro might cost, and you are capped at only 2. If NVLink is simply SLI with a bit higher bandwidth, that seems really basic. That Nvidia charges over $450 for Quadro NVLink is not relevant when Quadros themselves are $10,000. Quadros can be 5 to 10 times more expensive, so of course their NVLink would also be 5 to 10 times more expensive. And the gaming version has a big restriction, since it only supports 2 cards.

    Interestingly, I have found a number of people claiming that NVLink is under NDA. That is rather odd, but that would explain why none of the manufacturers are responding to questions about NVLink. There are forum posts on EVGA and Nvidia asking this question. I have asked this question directly on Nvidia's forum. Days later there is still no response. So I starting to think there really is a NDA in place.

    This link should work, I was on mobile before. This is actually a different tweet anyway, directly from OTOY themselves. Scroll the replies.

    https://twitter.com/OTOY/status/1034599997583572998

    The other tweet

    https://twitter.com/AaronCovrett/status/1034459411060809729

     

    ...well then why for GeForce is it being sold in 2, 3, and 4 slot versions like SLI bridges were instead of just a single pair of widgets like for Quadro and Tesla cards? You need two bridges between a pair of cards to get the full benefit of the link.

    That is not what that means. The "slots" you are speaking of refer to the length of the link, not how many links it has. Remember that many of these Turing cards are more than 2 slots thick, many are 2.75 slots thick.

    Take a look at this and you will understand. Also note that the Quadro 5000 NVLink is the same speed as the gaming link. So this is a clear indicator that the bandwidth is there to pool VRAM, and I believe that the gaming link is basically the lower end 5000 link in a gaming package.

     

    bluejaunte said:

    What I don't get it why they would be secretive about this. Now is the time for all the marketing hype in the world, why would they not mention that the memory stacks for gamer cards? This is really the main thing that makes me sceptical. If there is an NDA then even more so. Why would they want to shut people up about a very positive feature. Unless that feature isn't there at all.

    Neither do I. A lot of things are under NDA right now, including many gaming benchmarks. But if Nvidia is intentially hiding negative information that they do not stack, the blowback from angry people who preordered 2 cards would be VERY nasty. They already are facing blowback for the prices and other things. Even though Nvidia has total control of the market right now, this would potentially do real harm to their brand and possibly even result in a few lawsuits. I am not speculating that Nvidia is doing this, I am only saying that -IF- they are, they will face serious consequences for such actions. I don't think they can be that dumb. AMD might not compete much anymore, but Intel's GPU is looming soon. If people become dissenchanted with Nvidia, they will be much more willing to jump to whatever AMD and Intel bring out.

  • kyoto kidkyoto kid Posts: 41,845
    edited August 2018
    ...well why then have those multiple slot widths if you are not linking more than two cards? I also still do not believe that Nvidia would allow their base level cards to offer the same advantage as their high priced pro series. Two 2080 Ti's with a 79$ bridge offering almost as much VRAM as and more CUDA, Tensor, and RTX cores for less than half the proce of their 6,300$ RTX 6000 doesn't make sense marketing wise just as the original 12 GB M6000 and Titan-X didn't.
    Post edited by kyoto kid on
  • TaozTaoz Posts: 10,236
    billyben_0077a25354 said:
    Taoz said:
    bluejaunte said:
    ebergerly said:
    Here's what the NVIDIA website says: "The GeForce RTX NVLink™ bridge connects two NVLink SLI-ready graphics cards with 50X the transfer bandwidth of previous technologies. This means you can count on super-smooth gameplay at maximum resolutions with ultimate visual fidelity in GeForce RTX 2080 Ti and 2080 graphics cards." And nothing about memory stacking.

    Hmm ok, so the SLI term is here to stay? Guess I had the wrong idea. SLI has such negative connotation, would have thought they would replace it. But maybe the underlying tech is still SLI, just much faster with NVLINK.

    SLI means "Scalable Link Interface" which sounds like a rather generic term which can be used in many different contexts.

    No, SLI is supposed to mean Scan-Line Interleave.  Unfortunately NVidia put 3DFX out of bussiness and completely ruined the term SLI.  Back in the day I ran Voodo & Voodo 2 cards (first 2 Voodo's then upgraded to 2 Voodoo 2 cards) in SLI and it was nirvana on an OG 21" CRT.  Boy have times changed.

    It seems that it originally meant that, but in an NVidia context it now appears to mean the other thing:

    "Scalable Link Interface (SLI) is a brand name for a multi-GPU technology developed by Nvidia for linking two or more video cards together to produce a single output. SLI is a parallel processing algorithm for computer graphics, meant to increase the available processing power."

    https://en.wikipedia.org/wiki/Scalable_Link_Interface

    "Scan-Line Interleave (SLI) from 3dfx is a method for linking two (or more) video cards or chips together to produce a single output. It is an application of parallel processing for computer graphics, meant to increase the processing power available for graphics. SLI from 3dfx was introduced in 1998 and used in the Voodoo2 line of graphics accelerators. However, the original Voodoo Graphics card and the VSA 100 were also SLI-capable. Nvidia Corporation reintroduced the SLI acronym in 2004 (though it now stands for Scalable Link Interface) and intends for it to be used in modern computer systems based on the PCI Express bus."

    http://www.thefullwiki.org/Scan_Line_Interleave

  • kyoto kidkyoto kid Posts: 41,845
    edited August 2018
    ...BTW, I was being silly about the 44 GB. I know how SLI worked. The NVLink for the GeForce cards will improve crosstalk speed between the cards, that is a given, and yes, an improvement in performance. Now if a 599$ set of NVLink bridges would allow for memory stacking, that would make sense but a single 79$ one, no.
    Post edited by kyoto kid on
  • bluejauntebluejaunte Posts: 1,990

    Do Quadros still have different drivers? It may very well be just the drivers preventing memory stacking.

  • Ghosty12Ghosty12 Posts: 2,080
    edited August 2018

    Well here is something to be concerned about if true, in how Nvidia are wanting by the looks of things to control how the preview/reviews of their new cards are done and for people to sign NDA's..

    Post edited by Ghosty12 on
  • kyoto kidkyoto kid Posts: 41,845
    bluejaunte said:

    Do Quadros still have different drivers? It may very well be just the drivers preventing memory stacking.

    ....yes they do.

  • kyoto kidkyoto kid Posts: 41,845
    edited August 2018
    ghosty12 said:

    Well here is something to be concerned about if true, in how Nvidia are wanting by the looks of things to control how the preview/reviews of their new cards and for people to sign NDA's..

    ...hmm, kind of reminds me of Intel trying to quash Linux benchmarks of their spectre patches. 

    Considering the two companies often work together, doesn't surprise me.

    Post edited by kyoto kid on
  • outrider42outrider42 Posts: 3,679
    kyoto kid said:
    ...well why then have those multiple slot widths if you are not linking more than two cards? I also still do not believe that Nvidia would allow their base level cards to offer the same advantage as their high priced pro series. Two 2080 Ti's with a 79$ bridge offering almost as much VRAM as and more CUDA, Tensor, and RTX cores for less than half the proce of their 6,300$ RTX 6000 doesn't make sense marketing wise just as the original 12 GB M6000 and Titan-X didn't.

    Because the cards are so fat! Like I said, many of the cards coming are 2.75 slots wide. With that size, you cannot use a 2 slot wide link, because the two cards wont fit that close together. Now there are cards that will be only 2 slots wide, but my guess here is that Nvidia felt 2 slots was too close together for proper cooling. AIB partners are still free to make their own, so you may find a 2 slot link from a 3rd party at a later date.

    If Nvidia is really doing what that video claims, there will be serious backlash. You don't get away with making gamers angry. Many are already extremely skeptical and things are set to explode if Turing is not all its cracked up to be. This could be a really wild launch. You might have gamers in full pitchfork mode while people who use ray tracers like we do rejoice. Which in the end, might mean Turing still sells very well.
  • kyoto kidkyoto kid Posts: 41,845

    ...the gaming community needs faster communication between cards for smoother frame rates, not more VRAM.  The NVLink for GeForce cards will deliver the former, but not the latter,.  We 3D enthusiasts are still a very small niche in comparison to the gaming community or professional 3D production studios, which is why I don't see them offering memory stacking in their consumer line.

    Also, if these cards are indeed that wide, it would require people to purchase a new compatible MB if they want to use more than one card as pretty much all MBs today can at best handle dual width cards. 

  • Ghosty12Ghosty12 Posts: 2,080
    kyoto kid said:
    ghosty12 said:

    Well here is something to be concerned about if true, in how Nvidia are wanting by the looks of things to control how the preview/reviews of their new cards are done and for people to sign NDA's..

    ...hmm, kind of reminds me of Intel trying to quash Linux benchmarks of their spectre patches. 

    Considering the two companies often work together, doesn't surprise me.

    Yeah would not be surprised as I think Nvidia know they have no real competition at the moment..

  • ColinFrenchColinFrench Posts: 649
    edited August 2018
    outrider42 said:
    If Nvidia is really doing what that video claims, there will be serious backlash.

    For those wanting to go directly to the text, the info is at [H]ardOCP here.

    That site has done previous reporting on Nvidia's frankly anti-consumer practices and is effectively blacklisted by the company, which makes life tough for a tech review site, but they're sticking to their guns.

    Post edited by ColinFrench on
  • GatorGator Posts: 1,319
    edited August 2018

    Pretty interesting...  Without RTX support, there's a pretty good generational performance boost, but not unlike the past.  We're told 35-45% performance increase with traditional gaming performance.

    https://wccftech.com/nvidia-rtx-2080-ti-2080-2070-are-40-faster-vs-pascal-in-gaming/

    We have big a difference compared to previous generations, now the prices for the same lineup jumped - giving you the same or inferior performance in traditional gaming for the dollar.  This page has a poll, 80% think it's not worth it.  The new features of RTX sound great, but people recognize if it isn't supported it's of no help right now.  This is solidifying my opinion that pre-sales are going to be weak.  

    Now, for us for rendering use Turing is much more exciting.  smiley

    Post edited by Gator on
  • ebergerlyebergerly Posts: 3,255
    I suppose another way to look at it is that any person on the planet with a network connection can become a hardware evaluator. And given the increasing complexity of the hardware and software and drivers it becomes easier to post numbers that might misrepresent actual performance due to reviewer error or misunderstanding. And since that can directly impact revenue, what would you do if you were them?
  • bluejauntebluejaunte Posts: 1,990
    ebergerly said:
    I suppose another way to look at it is that any person on the planet with a network connection can become a hardware evaluator. And given the increasing complexity of the hardware and software and drivers it becomes easier to post numbers that might misrepresent actual performance due to reviewer error or misunderstanding. And since that can directly impact revenue, what would you do if you were them?

    It's not that hard. Run a game with the same settings on various videocards and compare the performance. If the game runs X faster on the new hardware and it is not impressive because Y feature didn't exist in games yet then that is irrelevant to people until such a feature gets implemented. If one or two review sites somehow mess up and misrepresent performance then there will still be 100 other sites getting it right. Only a few sceptics who were never gonna buy in the first place will claim that everyone got it wrong except those few.

  • ebergerlyebergerly Posts: 3,255
    edited August 2018
    Especially with totally new technology like RTX, making sure people are using appropriate/updated drivers is real important for stuff like this. As well as understanding the status of the different games or renderers in terms of their support for the hardware and drivers. I'm assuming this will change over time. And even stuff like understanding what NVLink really means and what features it includes with the different GPU versions. And even stuff like the system used for the tests, making sure there arent other issues affecting performance. Its not about just plugging in a card and running a game. And keep in mind first impressions are super important, so small mistakes up front can be very damaging long term.
    Post edited by ebergerly on
  • ebergerlyebergerly Posts: 3,255
    In fact, I wouldn't be surprised if the NDA and approved reviewer list, etc, were just a temporary thing for the first 6 months or a year until the technology gets established and fully supported.
  • ebergerlyebergerly Posts: 3,255
    Another thought...a non-skeptic might even view this as a great move for us users. Considering all new hardware architecture and new CUDA and Physx and NGX and Optix and Iray and games and drivers, personally I'd love to be sure I'm getting accurate and up-to-date performance numbers that aren't missing some key considerations that only NVIDIA would be aware of. There's so many new components and variables that it seems almost overwhelming.
  • GatorGator Posts: 1,319
    ebergerly said:
    Another thought...a non-skeptic might even view this as a great move for us users. Considering all new hardware architecture and new CUDA and Physx and NGX and Optix and Iray and games and drivers, personally I'd love to be sure I'm getting accurate and up-to-date performance numbers that aren't missing some key considerations that only NVIDIA would be aware of. There's so many new components and variables that it seems almost overwhelming.

    Eh, I'd rather know.  Good to read from my link a few posts above from Nvidia a basic metric with relevance, 35-45% better performance in "traditional gaming", in other words, not using any RTX enhancements.  Pretty safe you can bank on a very similar improvement with Iray for now, without any use of the RTX tech.

    I will wait for benchmarks of course.  I water cool, so I want to wait to see water cooling solutions too.  I am in a good spot, my rendering performance is pretty good and what little gaming I do I can do at 4K max detail... don't need any more performance for gaming.  

    It will be really interesting to see when we get RTX support in Iray with Daz and what kind of performance improvements we see with complicated shaders like SSS and skin.

  • nicsttnicstt Posts: 11,715
    outrider42 said:
    kyoto kid said:

    ...still not buying it until I see actual proof, not just what "someone says". 

    Were that true that would mean, a 4 slot NVLink bridge should allow for 44 GB with 4 2080 Ti's.

     

    You can't do 4 way NVLink on gaming regardless. All the card listings I have seen only show 2 way NVLink support. So no you cannot do a 44 gb setup with 4 2080ti's. You can only do 2 for 22 gb. There are multiple sources that show 2 way support on 2080ti and 2080, including the PNY page linked before. Also, while the order page does not mention any features of NVLink, the cards do not discuss their features much, either. On Nvidias own page, you do not see any mention of Tensor cores. When you look up most preorders, they only show CUDA core counts, and no core counts for Tensor and RT. That would seem to be a pretty important thing to leave out.

    It makes sense to me that NVLink would stack. For one, Nvidia killed NVLink on the 2070, so you cannot do this too cheaply. On the 2080 and 2080ti, these are cards that are $800 and $1200 each, so even at the very cheapest, you are still spending about as much as a single Quadro might cost, and you are capped at only 2. If NVLink is simply SLI with a bit higher bandwidth, that seems really basic. That Nvidia charges over $450 for Quadro NVLink is not relevant when Quadros themselves are $10,000. Quadros can be 5 to 10 times more expensive, so of course their NVLink would also be 5 to 10 times more expensive. And the gaming version has a big restriction, since it only supports 2 cards.

    Interestingly, I have found a number of people claiming that NVLink is under NDA. That is rather odd, but that would explain why none of the manufacturers are responding to questions about NVLink. There are forum posts on EVGA and Nvidia asking this question. I have asked this question directly on Nvidia's forum. Days later there is still no response. So I starting to think there really is a NDA in place.

    This link should work, I was on mobile before. This is actually a different tweet anyway, directly from OTOY themselves. Scroll the replies.

    https://twitter.com/OTOY/status/1034599997583572998

    The other tweet

    https://twitter.com/AaronCovrett/status/1034459411060809729

    "yes two cards gives you 22 GB if NVlink works on 2080 as it does on Tesla/Quadro"

    Interesting, but we still seem to be in the dark about RAM sharing - IF

  • nicsttnicstt Posts: 11,715
    edited August 2018

    Looking at what Nvidia are doing and how they're behaving, think I'll vote with my wallet.

    Edit

    In fareness to Nvidia, I've beein doing a bit of investigation.

    This site 'claimed' to have a lawyer to interview. Whilst I'm inclined to trust their site, it is something I have to take on faith.

    The NDA Nvidia want signing is: in a nutshell, nothing special. It doesn't indefinitely gag the journalists or prevent them from writing/recording a detrimental review.

    Post edited by nicstt on
  • GatorGator Posts: 1,319

    Wow, great find nicstt - I called that we wouldn't get memory pooling with the GeForce gaming series.  I would LOVE to be wrong, sounds like I may be!  laugh

     

     

    2x_RTX_cards.JPG
    964 x 1644 - 324K
Sign In or Register to comment.