Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
From iray rendering benchmark I got from a friend programmer
With my case I just plug the card in without any screws .. 6 second work it is as much work as with HD
Nope .. when you have open iray viewport calculations are done already and so the transfer , it is instant rendering the moment you hit the render button
2 identical GPU DON''T get 2x faster as that is not possible with iray software at this moment for that reason iray is number 3 on the fastest unbiased GPU rendering list
Yes, i think many people dont factor in that initial loading time that is the scene data transferring into your GPU's - and leading to believe that is causing a non-linear performance increase when adding second identical card and so on.
I did my own tests to test this. i did 1 Titan X alone, then 2 Titan X together. Then i also did 1 Titan X and CPU, and both Titan X plus CPU. When i did this, i did it on a small scene (~5 minutes) and on a big scene (over 1 hour). I did not use the log file to get the times. i sat at my computer and watched it loading the scene into the GPU, and then started my own stopwatch as soon as the scene started to render. In the case of the small scene, i sat and watched it until it finished to get the time. In the big scene, i was multitasking (using a third video card to power my displays), while watching the render progress on another monitor.
The results were just shy of twice as fast with both Titans working together. i dont remember the exact numbers, but the small scene was something like 4 minutes 50 seconds with both Titans and like 10 minutes and 3 seconds with just the one titan. The large scene was 1 hour and 18 minutes with both titans, and 2 hours 48 minutes with one titan.
So the performance increase for the extra card isnt quite 2x, but its very close. something like 1.95 times faster. Not sure if that becomes less efficient going to a third one though. I have a third card, but its not the same so will not give accurate comparison if i include it.
As for including the CPU - not worth it. They dont seem to play well together. From memory all the tests except one where i included the CPU were slower than without it (small scene, 1 titan + CPU was the one that was faster, by like 10 seconds over the same scene with just one titan alone)
I know that feeling.
It's like having Arnold Schwarzenegger inside of you! :D
As @PeterFulford noted, that looks like 4 Titans ARE rendering at 1/4 the time of a single. The discrepancies between the dual/triple and single/quad lead me to believe that the scheduler in the driver/Iray code isn't load-balancing the cards quite correctly. It's as if some parts of the scene are more complex, and the rest of the cards have to idle while they wait for those parts to complete (an iteration) before continuing. Not the best scheme, but easy to implement (it's just path-tracing different scene volumes per card.) But since some parts of the volume may be more complex, the other cards end up waiting on completions from the slower sections.
Without creating some very symmetrical scenes of various symmetries to try to figure out how the Iray scheduler is splitting up the work between the cards, it's all conjecture. But those numbers do seem to indicate something of that nature happening.
This is a bit disingenuous. GPU core clock speed DOES affect GPU render speed. But there are OTHER factors you have coming into play. Multiple cards (as evidenced by your time benchmarks) may be causing delays based on how the Iray scheduler is dividing the work between the cards. And architectural differences (Maxwell vs. Pascal) may cause even more differences.
Take a single Titan-X at 1000MHz core clock. Render the scene. Reload DS/whatever, Overclock the Titan-X to 1250MHz core clock, and up the memory clock by the same factor (25%). Render the same scene again. Monitor card temps to make sure no throttling occurs. If you don't get around a 25% speed-up (discounting setup/load time), something is VERY wrong with basic digital electronics theory and physics.......
I said there was no difference between my regular 1277Mhz speed and 1500Mhz with iray there is huge difference in speed between 1000Mhz and 1277Mhz I tested stuff since last year with all possible scenarios and I use separate non cuda card for monitors and I agree with you that clock speed does affect GPU rendering as every idiot know that and not for nothing I pay extra money for the super clocked version , but Autodesk stated there may be the software problem with GPU scaling that slow down the second card and the same issue maybe is here as I have no problem with other engines
but but but good news.. as Autodesk have confirmed that there was a GPU scaling issue with the software and they got patch to fix it, I was about to check whats new with the beta build of DS and guess what ? everything run faster, not 195% as that is little BS maybe Quadro cards that have special driver for iray but not GTX but it run almost 20% faster than 4.9.1 build , also the viewport move faster already on 100msec I can rotate the scene without been choppy before I needed 300msec for smooth rotation ( 10 genesis in the scene ) and I just got around 70% extra speed with second card , also the new Nvidia display driver not crashing anymore when using Interactive mode so I guess they did something good this time , maybe was the patch maybe by fixing the issue with interactive mode and the display driver they improved the performance , the Uberiray shader changed also ..but most important it improved and a lot , 4.9.1 was the worse build I used anyway so I am running on Beta from now.
You see I know people that have faster CPU and better rig than me with the same cards still they did not rendered faster , different shaders and light setup will affect the result too so comparison based on individual tests are questionable unless it is a real benchmark and based on megapixel per sec an not by counting the min and sec in DS .
My results are for my rig and my scene and there is no way that second GPU will give you 100% extra performance as that is not what Nvidia wanted you to have in first place, they created Quadro cards with special driver for iray and it cost double , GTX cards are not build for rendering even if a single card like Titan X render faster than Quadro and if GTX 1080 get the key for that ,we don't know , for Nvidia would be very bad business doing so as nobody would ever buy Quadro , Titan X or GTX 980Ti again unless they really need the VRAM , iray is also not ready for the new card anyway
Just wanted to post these, since nVidia has posted the specs for the 1070 online now.
nVidia 1070 GTX
1920 CUDA cores
1506MHz base core clock, 1683 boost core clock
8GB GDDR5 VRAM, 256-bit wide memory bus (8Gbps memory speed, 256GB/sec memory bandwidth)
Maximum resolution: 7680 x 4320 @60Hz
Graphics Card Power: 150W
Single 8-pin power connector.
So, slightly less than was predicted/speculated (128 less CUDA cores, and 100MHz slower clock speed.) But still WAY better than the 970 GTX (about 15% more cores, and 50% higher base core clock.)
@hphoenix yeah it fall exactly in the place with one prediction diagram I saw couple days ago plus it have more vram so better for the money
but my pressure is already released , my budget spent and no worries anymore
Just snapped one more EVGA Titan X SC Hybrid water cooled for $1099
with total of 9216 cudas at 1277Mhz it is going to be DS party on Friday lol
I was thinking long about and come to conclusion it would be better for me to keep my vram at 12GB
so I can use it also for rendering bigger scenes together with my twins and not just for GPU texturing and Photoshop work
I hope you guys find what is best for you , I am starting saving for the next year purchase as I want 4 of them in total but if the prices drop more maybe early in December
It getting harder to get good deal on the hybrids as the stock is so limited .
wow MECH4D you can have 10 gen figs in a scene cool I love doing big complex scenes or at least try to well most time they end up being I think and you do highly detailed renders that look real of course your excellent shaders and other content you've made helped too but having a good rig as you do to get the most out of them again also helps
stats have been posted of the gtx 970, 980, titan, 1070 and 1080 cards took screen shots still no news on how well work for 3d editing but those of you that understand these stats and what is more important for us maybe help us noobs to understand but the speed, cooling power savings and price sound good though though titan still has the cuda and vram advantage is that more important or do the new cards have more than enough for very large complex scenes and 3d creation - zbrush, mudbox, marvelous designer, substance painter extra - do these need programs need or benifit from more powerful cards as well as iray rendering at high settings? 1 or 2 1080s are a big saving over 1 or 2 titans either way going for two cards was thinking 4 1080s but now know that wont work or 3. Current prices haven't changed yet so whether or not the current prices for 2x 1080s $2762 and 2x Titans $3799 Australian will go up down or remain same don't know
Just curious as to what Titans you are looking at? those prices seem a little inflated...only 2 of the evga watercooled titans would be around that price. I am in australia aswell, and my two titans cost me 1600 australian each, so total 3200. they are the gigabyte xtreme version
well the max I can work with is 50 then it start to be choppy lol well at this moment upgrading my motherboard to 4 way SLI so my cards run optimal .
My Titans X can run 50 % faster than standard Titan X so 1080 will be 30% slower than my cards unless superclocked but still only in games that support it and we don't know even if iray is updated yet for this architecture to run optimal since it was created to support games and realtime VR, your viewport may spin faster but if the GPU scaling of the software do the same nobody knows . Remeber that power limitation in 1080 is 250W and that is how much it will use when superclocked , conlusion here ... i stay with my cards and 12GB , I got the last super clocked water cooled for $1099 usd , Titan X is so overprised outside US no matter the version
...agreed, the cost will most likely be higher (AMD's first HBM GPU is more expensive than a 980TI and only had 4 GB of video memory).
However HBM 2 will have a few major advantages to justify the price, like a smaller form factor (about half the length of a current generation unit), lower power consumption, a better pipeline between the GPU processor and video memory, and the potential for more video memory (very important for 3D rendering). The first commercial units to be released will no doubt be oin the Quadro line which could boast 32 GB of HBM 2 (The Quadro M6000 was recently upgraded to 24 GB GDDR5). From what I have gathered, the 1070/1080's advantages will primarily benefit the gaming market rather than CG production.
8 GB is "borderline" for my needs as a fair number of my scenes exceed that during rendering. If a scene dumps to the CPU, it doesn't matter how fast the clock speed is or how many threads the GPU has. This is why I am looking at the M6000 for now as it pretty much will handle anything I throw at it. It may not be as fast as the forthcoming Pascal/GDDR5X GPUs head to head, but not having to risk exceeding GPU memory is a major speed advantage as well.
Also as I understand, SLI only benefits gaming, not CG rendering.
Keep in mind though that for the price of a quadro m6000 with 24gb you could get 4 gtx 1080's, a new mother board, case, power supply, ram, cpu, and Octane render with the DS plugin and not have to worry about the amount of ram on your GPU at all. Plus you would have all of the other features available with Octane render 3x that aren't available with Iray like true volumetrics, motion blur, and hair (right now you would need the Carrara plugin for dynamic hair, but adding that would still kkeep you under the cost of the quadro m4000). You would also still be able to use Iray for smaller scenes.
Just a thought, I'm sure most people would prefer the convenience of Iray, but there are other more cost effective options available for rendering large scenes on your gpu, without breaking the bank on a something like the m6000. In fact, you could render large scenes quite quickly on a system with even 4 older 2 gb gtx 780 cards using Octane. Different strokes for different folks, but other, possibly more affordable and feature rich, options are available rather than investing ~$4,500 on a single video card.
Hey DustRider, is the DAZ Studio plugin for Octane working? It wasn't fully for the longest time. Seemed to be a problem for the guy working on it. Has everything been resolved? Does it convert Iray material to Octane or do we have to use those doggone nodes?
...however I would have only one third the video memory for full GPU rendering. I frequently have been seeing scene files top out at 10 GB and even higher (going into the virtual memory partition which is even slower).
As I have become quite accustomed to Iray (even more than 3DL, and far more than Lux now since I ditched R4 because of all the bugginess) and I am content to keep working in it. Personally, I like the workflow of being able to access the tools directly in Daz during setup rather than sending everything to, and flipping back an forth between the Daz viewport an another UI. As I do not work with dynamic hair (even though I have Carrara) that is not an issue since Daz does not support (nor offer) dynamic hair content. 24 GB of video memory would also support importing and rendering of my Garibaldi hair designs as .obj files.
My original "workstation II" plans called for x4 liquid cooled Titan X's which would have required a fairly hefty (and expensive) server grade PSU. The M6000 eliminates that requirement as well as the need for exotic cooling since it consumes less power than a Titan X or 980 TI (actually close to what of my old Fermi GPU does). Yes I'd have only 3072 threads but as Mec4D has pointed out, adding more GPUs does not increase render speed by 100% per unit. The best you render speed advantage can get with 4 GPUs is barely twice that of one. Where multiple GPUs do have a greater effect is on screen response which is why such setups are favoured for gaming rigs. As I do not do games, I don't have the need for that level of performance.
For my purpose, not having a scene drop from GPU memory while rendering is more important, 24 GB of VRAM would pretty much guarantee that.
...this was partly why I wasn't so impressed as it seemed everything (both the in the reveal and information on their site) primarily points to a boost for gaming performance rather than CG production. I'll wait until the dust settles and then see what comes out in the wash so to say. If they indeed release a Pascal based Quadro line using HBM 2 (particularly if the top end unit ends up with 32 GB), then the price of that M6000 may no longer be as big an issue.
I'm waiting to see if they will release a Pascal version of a Tesla 'compute' card...
...they already have, its the P100 with 16 GB of HBM 2 memory. Prices for indicvidual units have not been released yet as for now they are only available as part of Nvidia's DGX-1 deep learning supercomputer for use in AI research.
The M6000 and Titan X STANDARD are exactly the same chipset and have exactly the same number of cores, and exactly the same single precision compute capabilities + they usually have extra features like better colour depth and better driver for rendering with iray . rendering with iray don't mean how faster you finish, it is about how accurate the calculations are in your final render .
Having the same cards specification means nothing when rendering with iray , Nvidia decide here , if that was the case there will be angry companies and people at Nvidia's doors soon asking for refund and sue for misleading , why you think M6000 12GB cost 4 times the money and it is slower than Titan X 12GB? but they interact differently with iray software , so not expect 1080 to come even closer , it is not a Game it is rendering with raytraced light path and shader calculations not used in games that are still long behind photoreal real-time rendering .
only idiot would spend 5K in a one slower card where he or she can have 500% the speed with other card for the same money , it would be a scam .. but there is a secret key that unlock the power of rendering with iray and GTX cards don't have it . And I am not talking about speed in rendering , I am talking about raytraced calculations and light path that are not used in Games .
If your render use less speed with more iterations to finish your render it does not mean it is better or accurate render , 4000 iterations at 50 sec will look less accurate than 1800 iterations at 11 min rendering time with all correct samplers
4000 iterations may give you zero noises in 50 sec , but light paths will be not as accurate for realistic rendering and less photo real .
and that's how it works .
but we are not architects or scientists that need proper light calculations to the last photon , so we can go away with cheaper GTX that clean the noises faster , as usually nobody will miss the extra precision that professionals depending on that really need . And they don't use 1 M6000 in a PC and at last $50.000 worth of rendering power , you could not do that with 8 Titan X in one PC case .
comprende ?
MEC4D did post some (third party) Iray benchmark results that showed four cards being nearly linear:
1 x GTX Titan X 12GB - 100.5 s
2 x GTX Titan X 12GB - 62.25s
3 x GTX Titan X 12GB - 44.25s
4 x GTX Titan X 12GB - 26.15s
She has also mentioned that Iray is performing better in the new DS beta.
When fantasizing, you might as well fantasize big.
"It's not what you do with it, it's the size that counts." -- Arrogant Worms
Are there any charts showing the CUDA rendering performance improvement over Maxwell?
For Reality/LuxRender users, the performance is on par with the Fury cards from AMD.
My R9 Nano scores:
I love having a GPU speed up the rendering process, things on Reality are so much faster with OpenCL compared to Iray. I hope the GTX 1070 has performance similar to the R9 Nano - OpenCL and Iray + 8GB VRAM = great!
...again it is the memory size that I see as the biggest advantage. Currently, no other GPU on the market offers that level of video memory and probably will not for several years. Again, I don't do games so features like a hyper fast frame rate & overclocked memory speed are meaningless to me.
To me not having to be concerned about large high quality (as in gallery sized/quality output for printing) render jobs dumping to CPU mode and taking countless hours if not days to complete is worth the price. I used to paint in oils and create detailed pencil drawings before advancing arthritis took both away from me. I now look to 3D CG as their replacement and hold to the same exacting standards as I did when I worked with brushes and paints on canvas, or multiple weight graphite on fine Bristol.
It seems to me that, for someone with the cash to spend, Octane might be a better buy than multiple cards. Octane 3 is due out (over a week ago, according to their press release) which can offload textures and some geometry to system RAM. Surely that would cut the VRAM requirement significantly? If the reason for buying another GPU is memory rather than speed, then Octane might be the answer? The question of whether the plugin is working yet is obviously important in such a decision. I don't have Octane and I am finding the 4GB on my GTX970 quite limiting for IRay already (I've only had the card a few weeks).
...well in effect, video memory becomes "speed" provided the scene can be totally rendered on the GPU from start to finish. I'm not so much interested in trying to render a large complex scene scene in say, five minutes. Even if it took a couple hours, that is fine (instead of potentially a day or more in pure CPU mode). A number of my scenes have topped out at over 11 GB as I tend to use a lot of "in render" effects since my postwork ability is pretty poor (especially when it comes to layering and dealing with depth and and shadows). I know I have to "fake" volumetics in Iray, but have become pretty good at making it work for my needs.
As I mentioned, I prefer the workflow Daz/Iray offers compared to using a separate programme and UI. For example, with a more powerful GPU I will be able to easily check lighting, blocking, & such without having to stop and perform test renders by just switching to Iray View while still working on a scene. Having all my shader and surface controls right there in the surfaces tab is just more elegant. To me that makes it more intuitive.
Oh and to add, just read the post after this one and forgot to mention that I have a fairly extensive library of Iray shaders and utilities which would be of little use in Octane.