Iray Starter Scene: Post Your Benchmarks!

1356742

Comments

  • IanTPIanTP Posts: 1,326
    edited December 1969

    My only PC :)

    Windows 7 Pro x64
    i7-4790 4GHz (8 cores)
    GTX750 2GB OC Edition
    16GB DDR3 1600MHz RAM

    Complete Scene to 90%
    CPU - 21m 22s
    GPU - 8m 23s
    Both - 7m 2s

    Scene less spheres 8 & 9 to 90%
    CPU - 3m 50s
    GPU - 1m 47s
    Both - 1m 26s

    Interestingly I had an issue with fireflies only on the edited scene, they came and went in the complete scene, possibly a greater number of iray iterations?

    test-gpu+cpu-edit.png
    400 x 520 - 183K
    test-gpu+cpu-comp.png
    400 x 520 - 168K
  • IanTPIanTP Posts: 1,326
    edited December 1969

    SimonJM said:
    Can we get a link to that atrium, First Bastion?

    That looks very much like the free Sponza model. I think it's at ShareCG and someone (from here?) did a .daz scene load for it.

    In fact you used it here.... http://simonjm.deviantart.com/art/Walk-in-sunlight-206930382 tho you didn't add a link to the model :(

  • Cayman StudiosCayman Studios Posts: 999
    edited December 1969

    I thought I'd repeat my render test with a Film ISO setting of 200 and a shutter speed of 1/256, thus simulating a faster film. I was hoping this would also result in a faster render - not so lucky! The render took just as long as for the ISO 100, 1/128 test, and there was no discernible difference in the image quality either.

    So I am wondering what the purpose of the ISO, F/Stop and Shutter Speed parameters is seeing as the effects they would normally engender have to be fashioned by other means anyway.

  • MDO2010MDO2010 Posts: 1,459
    edited March 2015

    My results:
    Intel Core i7-3610QM 2.30Ghz x8
    12 GB RAM
    NVIDIA GEFORCE GTX 670M 3GB
    Windows 7 Home 64-bit

    Full Scene
    CPU Only Render Time to 90%: 24m19s
    GPU Only Render Time to 90%: 10m28s
    Both Render Time to 90%: 7m44s

    Spheres 8 and 9 Removed
    CPU Only Render Time to 90%: 3m40s
    GPU Only Render Time to 90%: 2m17s
    Both Render Time to 90%: 1m46s

    Post edited by MDO2010 on
  • Dumor3DDumor3D Posts: 1,316
    edited December 1969

    Cayman said:
    I thought I'd repeat my render test with a Film ISO setting of 200 and a shutter speed of 1/256, thus simulating a faster film. I was hoping this would also result in a faster render - not so lucky! The render took just as long as for the ISO 100, 1/128 test, and there was no discernible difference in the image quality either.

    So I am wondering what the purpose of the ISO, F/Stop and Shutter Speed parameters is seeing as the effects they would normally engender have to be fashioned by other means anyway.

    They can be used as exposure settings just like on a real camera, so can lighten or darken a scene overall for instance, but will only under some circumstances speed up, or slow down render times. For instance in a dark room, a higher film speed if it brightens the room enough, should show a reduction in render times. If you totally blow out a scene with light, like mostly a white scene, it will render super fast. :) You most likely won't wind up with anything useful. ROTFL!

  • AnIronButterflyAnIronButterfly Posts: 243
    edited December 1969

    OK--rebench...Not going to re-do CPU-only, that didn't change.
    CPU&GPU;: Approx 5-6 Min
    GPU-only: approx 6.5 Min

    What changed? Buh-bye GTX 550 Ti... Huuullooo GTX 970...

  • SickleYieldSickleYield Posts: 7,318
    edited December 1969

    OK--rebench...Not going to re-do CPU-only, that didn't change.
    CPU&GPU;: Approx 5-6 Min
    GPU-only: approx 6.5 Min

    What changed? Buh-bye GTX 550 Ti... Huuullooo GTX 970...

    Congratulations!

  • kittenwyldekittenwylde Posts: 151
    edited December 1969

    Finally had a chance to really run the benchmark tests. Not impressive compared to what some people get, but I'm okay with what my 4 year old machine can do.

    specs:
    AMD Phenom II X6 1100T Black Edition 3.3Ghz
    Nvidia GTX 560 Ti 1GB
    16 GB RAM
    dual monitors, one at 1920x1080, one at 1280x1024
    Windows 7 Pro

    times:
    GPU + CPU - full image
    90% 5:08
    100% 12:57

    GPU only - full scene
    90% 7:50
    100% 21:06

    -minus sphere 8 & 9
    90% 1:52
    100% 17:59

    CPU only - minus sphere 8 & 9
    90% 4:48
    100% 1:02:53

    - full scene
    90% 26:03
    still at 90% at 40:44, gave up, I need to go to bed. ;)

    conclusion: Not as fast as some people's impressive builds, but not too bad. The time beats Lux hands down with both CPU and GPU doing their thing. Can't wait till I have some real time to play with this shiny new toy!
    And by the way, as always while rendering, I had my poor machine doing several other things as well. It might improve times if I didn't have the browser open, and wasn't running Celtx, or the random photo widget, or the random desktop backgrounds, or... you get the picture. My poor computer is used to getting shoved beyond what it should reasonably be expected to do.

    sy_benchmark.png
    400 x 520 - 158K
  • parrotdolphinparrotdolphin Posts: 109
    edited December 1969

    Gread thread! Here's my results:

    GeForce GTX 780 ti (2880 CUDA cores)
    Intel i7-4770K CPU @ 3.5 GHz

    GPU Only: 5 min 53 sec
    GPU+CPU: 5 min 24 sec

  • AnIronButterflyAnIronButterfly Posts: 243
    edited December 1969

    OK--rebench...Not going to re-do CPU-only, that didn't change.
    CPU&GPU;: Approx 5-6 Min
    GPU-only: approx 6.5 Min

    What changed? Buh-bye GTX 550 Ti... Huuullooo GTX 970...

    Congratulations!

    Thanks! Did some price matching from an accepted online source, at a local branch of a blue-and-yellow-big-box-store. :) Got a bit of a deal on the super-super-clocked version! It's sweetly humming away in my case!

  • Wendy_CarraraWendy_Carrara Posts: 19,876
    edited December 1969

    using GPU only, took 9mins 6 secs on my rig, not trying other options as THAT was too slow for my liking, I do animations and use faster choices.
    intel core i5-2500 CPU @ 3,30GH
    16.0 GB RAM
    GeForce GTX 760

  • SassyWenchSassyWench Posts: 601
    edited December 1969

    Update! Got me a GTX 970 4gb card :cheese:

    Previous results: Desktop= I7, 3770 3.40 GHz 4 Core - 90% = 20 min. and finished in 51 min.
    New ones: didn't see the 90% but finished in 6 min 54 seconds.

    Whoo hoo! LOL

    Now to figure out how to actually make good renders instead of just fast ones. :red:

  • Subtropic PixelSubtropic Pixel Posts: 1,588
    edited March 2015

    On my system, it took 1 minute, 45 seconds to get to 90% and 3 minutes, 40 seconds to get to 100%. I did not remove any spheres, nor change any other aspects of the scene file or its render options.

    I confirmed that indeed CPU and both GPUs were employed. My system went to 100% utilization during the render and my screen movements got really sluggish.

    Post edited by Subtropic Pixel on
  • ZarconDeeGrissomZarconDeeGrissom Posts: 5,283
    edited December 1969

    AMD FX-8350, ASUS M5A97, 32GB Ram.
    Cache and temp in 4GB DRAM-Drive. No Fake ram (aka no swapfile).
    GeForce 8600GT 512MB VDDR (CUDA, what's that, lol.)

    Total Rendering Time: 1 hours 56.47 seconds

    The render maxed out on iterations (5000 samples) before it completed the 95% convergence, as hinted by my screen-cap. Yes, I added a zero to the max time, as I am on CPU only.
    I should have had more faith in my computer I guess. :-/

    So the billion dollar question is. Dose "Architectural Sampler" and/or "Caustic Sampler" improve CPU rendering? They are off in this bench by default. And dose "Optix Prime Acceleration" do anything on CPU only mode? Sounds like a chance to do more testing.

    Zdg_SyBench_Render_1.jpg
    400 x 520 - 73K
    Clipboard02_crop1.png
    1432 x 662 - 230K
  • DAZ_SpookyDAZ_Spooky Posts: 3,100
    edited December 1969

    AMD FX-8350, ASUS M5A97, 32GB Ram.
    Cache and temp in 4GB DRAM-Drive. No Fake ram (aka no swapfile).
    GeForce 8600GT 512MB VDDR (CUDA, what's that, lol.)

    Total Rendering Time: 1 hours 56.47 seconds

    The render maxed out on iterations (5000 samples) before it completed the 95% convergence, as hinted by my screen-cap. Yes, I added a zero to the max time, as I am on CPU only.
    I should have had more faith in my computer I guess. :-/

    So the billion dollar question is. Dose "Architectural Sampler" and/or "Caustic Sampler" improve CPU rendering? They are off in this bench by default. And dose "Optix Prime Acceleration" do anything on CPU only mode? Sounds like a chance to do more testing.

    OptiX should probably be on, yes it does work in CPU only mode.

    Architectural Sampler and Caustic Sampler are for specific cases and should only be used on a case by case basis. For example an interior scene lit mostly by exterior sunlight, or looking for super accurate caustics. These are used by Architects for Previz and Jewelry store advertisements. Render time wise they are not cheap to activate. (The Splash screen with those on went 17+ hours and wasn't done, with those off, took around 2 hours on a system with 2 x K6000.)

  • BangBang Posts: 6
    edited December 1969

    GPU Only
    .log
    Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : CUDA device 0 (GeForce GTX 770): 5000 iterations, 24.014s init, 636.832s render
    Total Rendering Time: 11 minutes 3.55 seconds

  • ZarconDeeGrissomZarconDeeGrissom Posts: 5,283
    edited December 1969

    Morning y'all. As I'm on CPU only, and I was curious how some settings effected render times. I set the max samples of the original test scene to one-thousand, and recorded the times it took to get 1,000 iterations done.

    I have no idea what OptiX Acceleration is, tho apparently if your on an AMD or the Graphics card is shall we say 'incapable', it probably is best to leave it off. I had it checked, and my GPU, hoping in vain, that the graphics card would contribute something, That was a mistake.

    I also wanted to see the Bench scene in a tad more detail, so I set it up to run as I slept. That's how I do most things. I work on scenes while I'm awake doing lots of small test renders. Then if it is ready, I'll let it run at high quality as I sleep. Like SY, I have short patience with my computers, when I want to be using them, lol.

    I let this one run with the time and sample limits essentially off, so I could capture the settings not put in the log-file (Daz3d Pleas add them, pleas, pleas, pleas). I also decided that Fiery Genesis looked a lot better with “Crush Blacks” off (0.00) in this scene. The room tho, needs to be allot bigger to get that ambiance of endless space back. Something to try when I go to bed tonight.

    AMD FX-8350, ASUS M5A97, 32GB Ram.
    Cache and temp in 4GB Dram-Drive. No Fake ram (swapfile).
    GeForce 8600GT (CUDA, what's that, lol.)

    * For the sake of reporting benchmark times, I highly advise that you don’t change any settings when the scene is loaded. That way we get an accurate idea of what a computer (CPU and/or GPU) can actually do. This is a 'Benchmark', lets not be fudging the results y'all.

    IrayBench_Promo001_CrshBlks_000_r.png
    1000 x 1300 - 992K
    Time_ToDoSomething_LikeWakeingUp_005_9kiters_endedHere_crop1.png
    2000 x 1144 - 944K
    Time_ToDoSomething_LikeWakeingUp_004_90pcntConverged_crop1.png
    2000 x 1133 - 939K
    Time_ToDoSomething_LikeWakeingUp_003_StartedWinampHere_Crop1.png
    2000 x 1124 - 901K
  • ZarconDeeGrissomZarconDeeGrissom Posts: 5,283
    edited March 2015

    (SNIP)So the billion dollar question is. Dose "Architectural Sampler" and/or "Caustic Sampler" improve CPU rendering? They are off in this bench by default. And dose "Optix Prime Acceleration" do anything on CPU only mode? Sounds like a chance to do more testing.
    OptiX should probably be on, yes it does work in CPU only mode.

    Architectural Sampler and Caustic Sampler are for specific cases and should only be used on a case by case basis. For example an interior scene lit mostly by exterior sunlight, or looking for super accurate caustics. These are used by Architects for Previz and Jewelry store advertisements. Render time wise they are not cheap to activate. (The Splash screen with those on went 17+ hours and wasn't done, with those off, took around 2 hours on a system with 2 x K6000.)
    Wow, now I don't felt bad at all about my light fixture test run, lol. The man page only appeared to 'List' the stuff in the render tab (no definitions for them or anything. Just a list.), so I was not sure if that was like 'Shadow maps' vs ray-trace or not. Good to know. Thanks!

    I had shut off everything except Daz, and set the limit to 1k samples to look at times. It looks like AMD or Something somewhere in my computer, dose not like "OptiX Prime Acceleration" at all. added nearly 2 minutes per 1k iterations with SY's test scene at stock settings. That combined with instance Optimization, Memory/Speed had some really curious results.

    Preliminary test Truncated times, (AMD FX-8350, Win7 64bit, CPU only)
    12 minutes per 1k iterations. instance Optimation (Speed), OptiX (on)
    12 minutes per 1k iterations. instance Optimation (Memory), OptiX (on)
    11 minutes per 1k iterations. instance Optimation (Memory), OptiX (off)
    9 minutes per 1k iterations. instance Optimation (Speed), OptiX (off)

    Post edited by ZarconDeeGrissom on
  • ZarconDeeGrissomZarconDeeGrissom Posts: 5,283
    edited December 1969

    GPU Only
    .log
    Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : CUDA device 0 (GeForce GTX 770): 5000 iterations, 24.014s init, 636.832s render
    Total Rendering Time: 11 minutes 3.55 seconds
    Now I feel like an idiot, lol. Where did you get that, I see only finished time, and paths, lol.

    "O", my notepad window was to narrow. :red:

    Info_Is_On_Log_001.png
    1163 x 334 - 28K
  • TotteTotte Posts: 10,363
    edited March 2015

    Cake One said:
    Hi everyone
    well i downloaded your scene and did several test with it :

    First : Specs :

    Mac Pro 3,1 (upgraded to 5,1)
    2x 2,93 Ghz 6-core Xeon (12 physical cores + 12 logical cores : 24 cores)
    32 Gb 1333 MHz DDR3 ECC memory
    Nvidia GeForce GTX 780 - CUDA Cores: 2304 - Memory Detail: 6144 MB GDDR5
    System and Daz studio on SSD

    CPU + GPU : Total Rendering Time: 25 minutes 9.12 seconds
    GPU Only : 10 minutes to reach 90% Total Rendering Time: 24 minutes 24.79 seconds
    CPU Only : 10 minutes to reach 90% Total Rendering Time: 24 minutes 24.79 seconds

    And this is where i realized that i CAN'T render your scene with GPU (see, the numbers are identical because in fact, it only uses the CPU)
    I checked the logs and it's written like that (see end of post for log info):

    I have rendered several scenes with more than 1 Gen2 and didn't seem to have any error.
    I don't know why this specific scene doesn't want my geforce to be used...

    System : Yosemite
    System and Nvidia drivers up to date

    If anyone has any idea...?
    I can send logs if needed

    Thank you
    C.

    I have the same setup, almost, GTX 780 6GB RAM, MacPro2012 with 10.10.2, and I've discovered today after quite an extensive research that I cannot render the Bree Texture with GPU. Try this:
    Set preview-mode to OpenGL Textures (to prevent IRay to load the Bree texture in the preview.

    - Load G2F - Set the texture to V6 Belle (or any other texture except the default Bree one.
    - Render, check logs of GPU Render works
    - Now set the texture to Bree
    - Render (and you will see the memory errors on the GPU). I bet that there is an error in some of the textures that the normal jpeg-reader programs doesn't choke on, but, when the texture is to be downloaded to IRay, it looks like it will need a Gazillion Bytes, and it fails.

    Please do test as you have the same setup as I do, and things like this can be card related (how things are implemented in the card or the driver).

    Post edited by Totte on
  • Maboroshi DaikonMaboroshi Daikon Posts: 21
    edited March 2015

    Cayman said:
    Thanks for this, SickleYield.

    My system is over 5 years old: Intel Core i7-920 2.67 GHz, 4 Cores, 8 threads.

    CPU only (with spheres 8 & 9 removed): 28 mins to 90% convergence, 63 mins total.
    Iterations: 4751 (which I guess is about the same for everyone).

    I am impressed that some systems can do this in under 5 minutes. I wonder if I can jam one of these Nvidia cards onto my old Asus motherboard, because over an hour for a relatively small and simple scene is not really an attractive proposition for me. Presumably if it were four times as large (800 x 1040) it would take four times as long.

    I've got a similarly old system Intel Core i7-930, but with one difference. The 930 defaults to a 2.8Ghz clock, but I have mine running at 3.8Ghz with a fairly beefy air cooler (Cooler Masters Hyper 212 EVO). I could push it over 4Ghz, but I wanted to keep temps in the 70's under full load so I backed it off a little.

    Anyway, CPU only - Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : CPU (8 threads): 5000 iterations, 18.658s init, 2660.251s render. 44 Minutes 20 seconds.

    I've been "borrowing" the GTX 980 out of my gaming machine while I decide which GPU to put in my art machine. The 980 does:

    Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : CUDA device 0 (GeForce GTX 980): 5000 iterations, 19.855s init, 276.719s render. 4 minutes 36 seconds.

    This is with the newer 4.8.0.9 build and sphere's 8 and 9 hidden.

    EDIT:
    Forgot to mention that GTX 980 is the EVGA Superclocked version running at the stock speed it shipped at: 1266Mhz. It's supposed to Boost to 1367Mhz (according to the GPU-Z info), but watched the sensors, it ramped up to 1404 and pretty much stayed there the entire time. That may change on longer renders, though.

    Reading back through this it makes me sound like some sort of overclocking maniac, but I rarely do it and I usually don't buy factory overclocked cards either... things just happened to work out that way this time :)

    EDIT 2:
    Went ahead and ran the full scene (sphere 8 and 9 visible) on the 980 by itself just for completion as it might make it easier to compare scores against other cards.

    Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : CUDA device 0 (GeForce GTX 980): 5000 iterations, 19.783s init, 305.965s render
    Finished Rendering
    Total Rendering Time: 5 minutes 27.33 seconds

    Post edited by Maboroshi Daikon on
  • HaruchaiHaruchai Posts: 458
    edited March 2015

    Intel i7 4960x Hex core 3.6Ghz, Nvidia GTX Titan 6GB, Windows 7.

    Ran at the preload settings and dimensions.

    CPU & GPU - 5 minutes 10 seconds
    GPU only - 6 minutes 12 seconds
    CPU only - 23 minutes 21 seconds

    CPU & GPU with OptiX Prime Acceleration ticked - 3 minutes 54 seconds

    Guessing I won't be turning the GPU off very often :)

    Post edited by Haruchai on
  • ZarconDeeGrissomZarconDeeGrissom Posts: 5,283
    edited March 2015

    dminut, nice score. I noticed the new Beta version 4.8.0.9 did make allot of stuff faster, others the same. The time was around 50 minutes, tho it's not in the log now? "O" well I'll run the test again later. Shaving about ten minutes off of a render in CPU only mode just doesn't seem right to me, unless there was a drastic change to the 3delight interface from the last version (4.8.0.4).

    I'm currently digging up info on Iray compatible CUDA cards, apparently there is a version of CUDA's that doesn't work at all with Iray, others do, and so on.

    Post edited by ZarconDeeGrissom on
  • Eustace ScrubbEustace Scrubb Posts: 2,400
    edited December 1969

    I'm sure I'm not the only one to share plug-ins between editions of DS4. Has anyone tried Iray in one of the earlier editions?

  • DAZ_SpookyDAZ_Spooky Posts: 3,100
    edited March 2015

    dminut said:
    Cayman said:
    Thanks for this, SickleYield.

    My system is over 5 years old: Intel Core i7-920 2.67 GHz, 4 Cores, 8 threads.

    CPU only (with spheres 8 & 9 removed): 28 mins to 90% convergence, 63 mins total.
    Iterations: 4751 (which I guess is about the same for everyone).

    I am impressed that some systems can do this in under 5 minutes. I wonder if I can jam one of these Nvidia cards onto my old Asus motherboard, because over an hour for a relatively small and simple scene is not really an attractive proposition for me. Presumably if it were four times as large (800 x 1040) it would take four times as long.

    I've got a similarly old system Intel Core i7-930, but with one difference. The 930 defaults to a 2.8Ghz clock, but I have mine running at 3.8Ghz with a fairly beefy air cooler (Cooler Masters Hyper 212 EVO). I could push it over 4Ghz, but I wanted to keep temps in the 70's under full load so I backed it off a little.

    Anyway, CPU only - Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : CPU (8 threads): 5000 iterations, 18.658s init, 2660.251s render. 44 Minutes 20 seconds.

    I've been "borrowing" the GTX 980 out of my gaming machine while I decide which GPU to put in my art machine. The 980 does:

    Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : CUDA device 0 (GeForce GTX 980): 5000 iterations, 19.855s init, 276.719s render. 4 minutes 36 seconds.

    This is with the newer 4.8.0.9 build and sphere's 8 and 9 hidden.I will point out that overclocking with rendering is usually a bad combination. (Regardless of cooler.) It has a tendency to cause BSOD crashes, especially with onger renders and CPU's develop hot spots that are above crash threshold.

    Note Factory overclocked, like most Video Cards arrive is fine, the cooling is designed for the temperature.

    Post edited by DAZ_Spooky on
  • Maboroshi DaikonMaboroshi Daikon Posts: 21
    edited December 1969

    I will point out that overclocking with rendering is usually a bad combination. (Regardless of cooler.) It has a tendency to cause BSOD crashes, especially with onger renders and CPU's develop hot spots that are above crash threshold.

    Note Factory overclocked, like most Video Cards arrive is fine, the cooling is designed for the temperature.

    While all true, there are different degrees and different situations with different processors. In my case, I ran 24 hours with Prime95 under full load without crashing or getting numerical errors in the prime calculations. Many tweaks had to be made to get it to that point, so it's definitely not for the faint of heart :) And while it's not liquid cooling, the heatsink I used is huge (it uses an 120mm fan and can take a 2nd one if I needed it). It doesn't hurt that the i7-920 and 930 are really good overclockers.

    I originally did it to speed up my 3Delight renders after waffling on upgrading to an 8-core Haswell-E. Now, I'm glad I held off as a $600 Gtx 980 would be a much larger improvement in speed than the 3 grand I was looking at for the new system. I don't even use the CPU for my normal renders and tick only the GPU box now... not much point in adding the CPU when it would only save a minute or so anyway :)

    My only concern now is that I'm not sure how limiting 4GB of ram on a GPU is going to be. The Titan X is tempting, but I'm not sure if I'd just be wasting my money or not. With my current scenes, nothing gets over 2.6GB usage on the 980 so I'm probably ok. I'm just not sure what a 4GB Daz scene would look like.

  • Dumor3DDumor3D Posts: 1,316
    edited December 1969

    dminut said:

    My only concern now is that I'm not sure how limiting 4GB of ram on a GPU is going to be. The Titan X is tempting, but I'm not sure if I'd just be wasting my money or not. With my current scenes, nothing gets over 2.6GB usage on the 980 so I'm probably ok. I'm just not sure what a 4GB Daz scene would look like.

    I'm wondering the same thing... and actually testing at the moment. I sent a 1.2GB scene plus 12 people and hit 3.2gb. However Iray from what I'm seeing is superb at handling texture images. It really does seem to load a shared map only one time. It loves tiling. So this scene is pretty rough on OpenGL in texture mode. Pretty sluggish moving about on the GTX660s. It sort of presses the limits of scenes I like to use, but yet I didn't blow out the VRAM. But, again, this is going to come down to the textures. In theory, one outfit with a lot of huge maps could blow it out.

    I'm thinking of at some point adding a Titan X, just for those occasions that will certainly happen somewhere along the line, where 4 is not enough and then for most scenes, let the work horses be 980s with the Titan. I keep watching to see if something blows my CUDA core theory up. So far it seems to be the number of cores = a very linear gain in speed regardless of card specs.

  • ZarconDeeGrissomZarconDeeGrissom Posts: 5,283
    edited December 1969

    Dumor3D said:
    dminut said:

    My only concern now is that I'm not sure how limiting 4GB of ram on a GPU is going to be. The Titan X is tempting, but I'm not sure if I'd just be wasting my money or not. With my current scenes, nothing gets over 2.6GB usage on the 980 so I'm probably ok. I'm just not sure what a 4GB Daz scene would look like.

    I'm wondering the same thing... and actually testing at the moment. I sent a 1.2GB scene plus 12 people and hit 3.2gb. However Iray from what I'm seeing is superb at handling texture images. It really does seem to load a shared map only one time. It loves tiling. So this scene is pretty rough on OpenGL in texture mode. Pretty sluggish moving about on the GTX660s. It sort of presses the limits of scenes I like to use, but yet I didn't blow out the VRAM. But, again, this is going to come down to the textures. In theory, one outfit with a lot of huge maps could blow it out.

    I'm thinking of at some point adding a Titan X, just for those occasions that will certainly happen somewhere along the line, where 4 is not enough and then for most scenes, let the work horses be 980s with the Titan. I keep watching to see if something blows my CUDA core theory up. So far it seems to be the number of cores = a very linear gain in speed regardless of card specs. I asked the same exact question, given my potent 32bit experience :coolhmm:
    How many unique generation 6 HD figures, cloths and hair, can you cram into 4GB?
    DAZ_Spooky... http://www.daz3d.com/forums/viewreply/785791/
    to prevent cross-posting, lol.

  • ZarconDeeGrissomZarconDeeGrissom Posts: 5,283
    edited March 2015

    It looks like version 4.8.0.9 is a tad faster then the last version (4.8.0.4).

    4.8.0.4, Total Rendering Time: 1 hours 56.47 seconds
    Total Rendering Time: 1 hours 56.47 seconds

    4.8.0.9
    1.0 IRAY rend info : CPU (8 threads): 5000 iterations, 24.298s init, 3281.361s render
    Total Rendering Time: 55 minutes 7.93 seconds
    about 89% and change converged.

    I can think of two reasons for the improvement. The removal of diagnostic code from the Studio-to-3delight interface, or further optimization to that interface code. Either way, I'm happy with the results.

    I didn't run a test with spheres removed the first time, as I wanted to know how my computer would fare with figures in scenes, not just static props, lol. The same this time as well, tho now I'm curious. dminut had done a test on a system rather close to what I have. I have eight cores in this computer with an Integer-Unit for each of them, however they only have four Floating-Point-Units to share between all of them (Not quite 4+4 Hyper-threading, tho close). I will have to try the test again, without the SSS spheres (8 and 9).

    "O" FYI, this CPU is at stock 4.0GHz ALL the time, I disabled that boost thing (day one), for similar concerns voiced thus far. I love massive heat-sinks with lots of pipes, tho I don't do overclocking. I learned my lesson years ago with "Low Noise Amplifier" designs, in another life (1980s).

    AMD FX-8350, 32GB Ram, GeForce 8600GT (512MB), SSD Boot drive(C), SSD Programs drive (D), eight data drives, cache & temp in 4GB Ram-Drive.
    (EDIT)
    Now this is curious. While making the SSS spheres vanish from view in the scene, it dose NOT decrease the render time as considerably as I had thought. Also, the ONE factor, that differs considerably between having SSS and not in the scene, is NOT included in the log file. Percent Converged. It's all good tho, I'm having fun rendering stuff.

    No sphere 8 and 9.
    1.0 IRAY rend info : CPU (8 threads): 5000 iterations, 24.135s init, 3046.625s render
    Total Rendering Time: 51 minutes 12.69 seconds
    about 92% converged.
    (EDIT2)
    Good point noise.gate.mike, All three of these tests were dune with CPU, GPU, and “OptiX Prime Acceleration” checked off just to keep them consistent. forgot to mention that.
    The GPU is only CUDA 1.1, so it did nothing, if only slowing things down a clock-cycle or so initializing the render. It stayed at 0% load the entire time.

    Clipboard02_Crop1.png
    1001 x 744 - 237K
    Clipboard01_Crop1.png
    792 x 645 - 212K
    Post edited by ZarconDeeGrissom on
  • Shonen_BatoShonen_Bato Posts: 1
    edited March 2015

    CPU: Intel i7 4790k 4.0Ghz
    GPU: nVidia GTX 970 4G

    ___________________________________

    Default preload settings and dimensions.

    TESTS:

    1_ CPU + GPU: 5 min 41 sec

    2_ CPU + GPU with OptiX Accel: 3 min 51 sec

    3_ GPU Only: 6 min 13 sec

    4_ GPU Only with OptiX Accel: 4 min 04 sec

    5_ CPU Only (without spheres 8 & 9): 30 min 04 sec

    6_ CPU Only with OptiX Accel (without spheres 8 & 9): 31 min 53 sec

    EDIT: I'm not sure about TEST 5 & 6... i was browsing in Chrome at the same time.

    Post edited by Shonen_Bato on
Sign In or Register to comment.