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
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
Man, if these 4090 benchmarks are real, this generational jump is going to be more consequential than the 980Ti arrival (or the 1080Ti for that matter)
As in, it might really be worth upgrading despite owning the current gen, instead of the more typical skipping a generation or two.
I'm strongly considering getting the 4090, and yet I have the 3090. Already been planning how I'm going to get the funds, so by the holidays I should be able to afford it. By then official benchmarks will come out and that will help me decide if its worth it or not. That is probably the only good thing about them being so expensive; time to save and time to think.
I had a 1080Ti and managed to restrain myself from getting a 2080Ti, lol, but got a 3090 just a couple months after it launched (well, EVGA's awesome wait list got me a 3090). I'm seriously considering the 4090 as well at launch.
Well...there will not be any EVGA 4090...ever. They have ended their partnership with Nvidia. That really sucks, they are my favorite. They will still make power supplies and other components, but not GPUs.
Current 3090 owners have 3 options:
-Buy a brand new 4090 which I believe will offer double 3090 performance.
-Buy a 2nd 3090 instead and get the double performance on the cheap. 3090s are going to be in a free fall in pricing, and that could make sense for some people. I can picture them dropping to as low as $600 in the next couple weeks. Bonus option: you can use the Nvlink on the 3090 to pool their VRAM, which is a very compelling option many may overlook. To be clear, you only pool texture data with Iray, not mesh data. But even so, that can free up a lot of data.
I don't think it will be easy to run a 3090+4090 combo, due to the huge size and power draw of the 4090. They are talking about 4 slot cards here, in many PC builds that would take up too much room to keep anything else installed!
- The 3rd option would be to get the 4080 16gb, which will be faster than a 3090. But this means less VRAM. Still, the 4080 should be smaller and course, cheaper. If you don't use 16gb of VRAM then it doesn't really matter if it packs less than your 3090.
And to repeat, I am predicting the 4090 will do 45-50 iterations per second, which may vary depending on the power draw and clocks of the model. A high clocked one might break 50.
The 4080 will be a solid 30% weaker than the 3090, a much bigger gap than the 3080 to 3090. But that still means it will hit 34 or 35 in our bench.
The 4070 should also beat the 3090, not by much. I'm guessing about 25, unless it is bottlenecked by something.
We shall see how accurate that is. Though it may be a long wait since nobody benches Iray besides us.
Just curious. I posted a scene test with my 3080 Ti in DS4.20 last May 31, on page 31.
That test doesn't show up in the chart on the first page. There are no 3080 Ti's there, so I'm wondering if you feel I did the test wrongly, poor math, or whatever.
So far I love this card, tho I still have to adjust the fan settings better. It gets immediately up to 80c+ and stays there during renders. I need to add a fan to the top of the case, as well, I think...
And will probably lower voltages once I feel smart enough;-)
It's faster than I ever expected anyway!
Cheers.
The forums broke the formatting of the posts, and raydant hasn't had time or desire to go back and post everything on the chart. I have considered making a chart away from the Daz site and then linking it to the forums. But alas, that is a big task, and I just haven't got around to it.
Oh that's right! Thanks for responding, outrider42.
I had mentioned the formatting problem in the area below the main chart at the top, but he thought it was a problem with my browsers...
A lot is changing now, so maybe the chart has become a bit too heavy and outdated, in part at least, anyway.
Thanks again!
Yeah - not enough hours in the day is pretty much the short of it. Plus, much of the ultimate objective of this thread (to scientifically document the performance metrics of Iray rendering across full product lines of rendering hardware) has already been accomplished with the existing data. Nvidia GPUs perform almost scarily consistently between cards of the same model and scale up performance-wise in the overall product stack for that generation of GPU extremely predictably. All you really need is 2-3 fully documented runs of this benchmark on a specific GPU model (like a 3080ti) to know really all you will ever need to know about how well it performs in rendering - barring extremely major changes being made to how the Iray render engine itself works.
Once 40 series GPUs start rolling out en masse, I'm considering transplanting the existing table data/benchmark summaries found at the beginning of this thread to some new posts, and starting over with fresh data collected using only recent Daz Studio/Iray versions to make things more meaningful to current/prospective Daz users today.
And the formatting problems you see are due to the DAZ forum's user posts still being internally formatted using deprecated HTML4 tags, despite virtually no modern browsers being backwards compatible with anything before HTML5.
45-50 would be unreal.
At the other end of this conversation is the eiree silence regarding Lovelace in the Titan class/former Quadro grade cards.Rumor is the 4090 Ti may offer 48 GB & fully unlocked AD102. This would be good, but I have some concerns. First is the time it took Nvidia to put 3090 Ti to market. Are we going to wait another year for a Ti variant? Then there is the money grab towards the Pro users in the interim, where the fully unlocked AD102 may not come out until next spring (perhaps in the form of an AD6000 card or something similar?) with the 4090 Ti sometime later. Nvidia has piles of pro-grade GPU inventory they still need to unload too, so my guess is it will be next year before we really see the top performing Lovelace cards come to market.
With the ETH merge and other market factors working against Nvidia, I am bracing for some improvisation. It would be great if it the pressure induces a Ti this fall, opposite some pro-grade alternatives.
Watched the Nvidia Keynote (aka omniverse sales pitch) earlier this week. Sounds like 4090 drop will be 10/12 @ $1599 and ADA 48GB in the pro lineup this December. Nvidia DLSS 3.0 will help games, but who knows how long it will be before we see similar enhancements for IRAY. One thing I was happy to hear was Jensen often talking about Siemens, a major contributor to the commercial use of IRAY. This is always a good indicator as it points to Nvidia's stakeholder interest in sustaining driver functionality for IRAY.
With this info in hand, I may wait until the 48GB cards drop; December is not that far away.
And maybe by then Daz will have Lovelace support, too, LOL.
Indeed, I looked at some more info and some video games only had I believe a 60 or 70% increase, so not everything doubled. The thing that caught my attention was the little remote control racer demo they showed. Jenson said the 4090 was 4 times faster playing this game. There are 2 things to unpack on this stat. The demo was heavily ray traced, far more than most games. That is a plus for Iray as it shows what the 4090 can really do. However this number also appears to include DLSS turned on. So the true frame rate might have been just 2-3 times faster, it is hard to say. DLSS has nothing to do with Iray, so it doesn't help any. However, Iray has a denoiser, and my hope is we at least get an improved denoiser. If the denoiser gets a big improvement that kind of changes everything, as you may not need to fully render everything. That would save more time than raw rendering power.
The RTX 6000 is limited to 300 Watts. It is the full die, unlike the 4090, but dropping 150 Watts will still be a performance hit. It may be possible the card can be tuned. I don't know how well they tune these cards in the factory. I keep bring it up, but it seems to be relevant to multiple posts, Igor's Lab took a 3090ti from 450 Watts to 300 Watts and only lost 10% performance at most. We cannot be sure if Lovelace would behave this way, but I think it would be possible. So if the RTX 6000 is well tuned it might only drop 10% from the top die. I believe the 4090 is cut down slightly more than the 3090 was compared to full A6000. So that could be a factor. It will be curious to see how the new AD6000 fairs compared to the 4090.
The 4090 has no Nvlink unlike the 3090. So it is not possible to pool its VRAM with a 2nd card. So that may be important to some people.
System/Motherboard: Asus Prime Z690-A
CPU: Intel i9-12900KS @ 5.7Ghz/3.4Ghz (if left at defaults)
GPU: Asus nVidia RTX3070 TUF @ 1770Mhz
System Memory: Corsair DDR5 64GB @ 6000Mhz
OS Drive: Samsung EVO 980 PRO 1TB
Asset Drive: WD Green SN350 2TB
Power Supply: EVGA P3 1200W
Operating System: Windows 11 21H2
Nvidia Drivers Version: 517.40
Daz Studio Version: 4.20.0.17
Optix Prime Acceleration: N/A
Benchmark Results
DAZ_STATS
IRAY_STATS
Rendering Performance: 01800 / 10.09 iterations per second
Loading Time: 1:9.475 - 3:2.23
Killer rig. That is a nice setup for working in Daz.
I am having a tough time reading your load time though. What is your total and device rendering times? Is your load time 2.23 seconds?
4090s @ 600 watts each is going to be hard to stack multiple GPUs. Ignoring the obscene space requirements, 2x 4090s could spike near (or over) 1200 watts.
If the performance drop is only -10% for A6000, you can make up for it with 2 or 3 cards.
4090s don't use 600 Watts. You should check out the Der8auer review. The power draw of the 4090 has surprised almost everyone who reviews it.
For starters the 4090 almost never hits the 450 Watt mark. But more over, he did a very basic power drop and got the card to run near 300 Watts without losing much performance. This was such a revelation that he was stunned, wondering why Nvidia cranked the power up so much on this card for so little gain! (As I and many others are as well, this is just so weird.)
This chart below is for the Time Spy benchmark, which is gaming focused, yes, but when I run this benchmark myself I always use MORE power than ever do rendering Iray. So this bench is more punishing than even Iray. What this chart shows is the power target. All DerBauer did was change the GPU power target, no fancy undervolting or anything. What he found was crazy. The blue line shows the power draw, the red line shows the score. Notice how the highlight area shows scores extremely close to the default power draw score...but look at the power draw. At 329 Watts the score only dropped to 117.08 from 123.49. Not to mention the default setting only pulled 422 Watts.
So I feel very comfortable that rendering Iray the power draw will be below 420 Watts. But not only that, the user can simply adjust the power target down a little, save a ton of power, and still get crazy performance.
Thus you can absolutely run 2 of these without any issues...well aside from the massive 4 slot design probably getting in the way. I don't know if you can even put 2 of these in the same standard case. It would be close. But if they can physically fit, they will run fine.
Moreover, the testing by GamersNexus showed that the transient spikes on the Founder's 4090 to not be as high as expected. The spikes were only around 40% or so, nothing like the 100% spikes that Ampere could present at times. It is clear that the Founder's 4090 is well designed, and might be over engineered.
I would check out the GamersNexus review as well, as they really dive into the transient spikes more than anyone else. They were the outlet most concerned about these spikes before launch, but the 4090 performed better than expected in that area.
Considering that virtually all recent gen motherboard platforms feature their primary x16 PCI-E slot in the 2nd physical slot position from the top (to make room for an M.2 NVME SSD between it and the CPU socket) with a 2nd x16 slot not appearing until the 6th physical slot position (keeping in mind that full size ATX only accomodates 8 physical slots total) it is extremely likey that unless you go Founders Edition or custom watercooling, you will be limited to a single GPU on a current gen system for fitment reasons alone.
To the best of my knowledge the closest thing to current gen that supports truly flexible PCI-E x16 slot spacing is Threadripper PRO. And between exorbitant pricing and lack of forward-looking features like PCI-E gen 5, DDR 5, and (imo absolutely essential for a GPU compute/rendering rig) iGPU support, sinking that sort of money into it is something I myself simply havven't been able to stomach. Yet.
Seems my aging Z370/8700K (with its easy 3 full size GPU capacity) is destined to live on for another day or two. In a recent video, Level1Tech's Wendell was lamenting on the lack of any server class (which is to say, minimal spacing between physical PCI-E x16 slots) motherboard implementations for current gen Intel or AMD. I can't agree more. If only there weren't just dozens of others who feel likewise...
Great thread.... and its cool to see some of the 4090 predictions made here seem to be spot on.
Thanks. :)
Yeah, I have no idea where I got that number from. Maybe I was on drugs or something. :). Copied and pasted from the log file.
Reloaded it and got this:
Finished asset load (open): 1m 55.827s
Total Rendering Time: 3 minutes 35.78 seconds
It would pretty much have to be a Founder's card, since the others are all so massive, or water cooled. The water cooled card that Paul's Hardware showed was just 2 slots. You just have to figure out where to put the radiators, LOL.
The FE cards could work because of how they push air. One fan pushes the air upwards. But the other fan pushes most air out the back like a blower style cooler. So if two of these cards are stacked, a lot of the air gets pushed out the back like blowers do. So I think it is doable if the user has enough space in their case. I think this user would want to turn the power down some so avoid heating up the room so much.
I posted this in the 4000 reveal thread, but it is important here, too.
The new Daz Studio 4.21 still has the same build of Iray that dates back to July 2022. This version of Iray uses CUDA 11.2.2.
The problem is that CUDA 11.8 adds support for Lovelace, and Daz Iray does not have 11.8 yet.
So it looks unlikely that Lovelace will work with Daz Studio for now. I could be wrong, and I would be happy to be wrong, but the signs point to requiring a new CUDA to use Lovelace. Given that DS just shipped an update yesterday, it could be a while before any update supporting Lovelace comes out. The main release often takes quite a while. The beta may fair better, but even in the best case situations it takes at least a month or two for the beta to add support.
Of course if anyone has a 4090 and can confirm, that would be great. And it would also be great if Daz could say something on the subject. Like "Hey, we have a new Iray build in testing..." or something like that.
But I reiterate what I said earlier, anybody who buys a 4090 right now needs to be aware it may not work in Daz just yet. You need to be prepared for that possibility and have a backup GPU you can use just in case. Again, I would love to be wrong, but I think it is important to provide this warning until we get a confirmation.
Since we're all waiting for the 4090s, I wondered how well the 3090 I have actually responds to lowering the power limit. Because I hate noise, I undervolted my 3090 and recently also started lowering the power limit. And I've used the benchmark scene from this thread to create a graph showing what the optimal settings are for my particular card:
(I scaled both Iterations and Watt per Iterations by 10, so the actual values are what is displayed in the image divided by 10)
And it turns out the optimal power limit is 68-70%. You can see in the graph that at that point the magenta line is lowest, meaning I get the most Iray iterations for the least amount of power (and therefore least amount of heat/noise). But again keep in mind this is not stock settings, but my undervolt settings. However I tried stock settings briefly, and the only difference it made was that the card consumed more power, and rendered a bit slower (maybe a temperature issue, I have not measured that but I set the limit to 83C and it might have hit that, though I don't think so).
System Configuration
System/Motherboard: X570 AORUS ELITE
CPU: AMD Ryzen 9 5900X @ 3700 Mhz
GPU: Gainward Phantom GS Nvidia RTX 4090
System Memory: 64 Gb
OS Drive: SSD 2 Tb
Power Supply: 1000W
Operating System: Windows 10 family
Nvidia Drivers Version: 522.25 Game Ready
Daz Studio Version: 4.21.0.54 Public build Beta 64bits
Optix Prime Acceleration: On
Benchmark Results
Total Rendering Time: 1 minutes 23.12 seconds
IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 4090): 1800 iterations, 1.466s init, 79.788s render
Iteration Rate: 22,56 iterations per second
Loading Time: 3,33 seconds
Are these numbers correct?
If they are, then clearly there is a problem because the 4090 here only matches a 3090ti at best. So that is honestly a terrible result.
But on the bright side at least card is functional! That is tremendous news as I was really concerned it would not work at all. But still, this shows that we really badly need an update for Iray and Daz Studio that makes these cards work properly. While it may have worked for the benchmark scene we have, it may have problems in other scenes.
For example, does dforce work on the 4090? Dforce always needs to be recompiled for new hardware. There is also the denoiser, and any number of other plugins. There could be glitches and issues until we get a proper update.
https://www.daz3d.com/forums/discussion/596086/rtx-4090-support#latest
I saw some Blender benchmarks with a 4090 that were nearly twice the performance of a 3090. With the 4090 power draw being around 450 watts and performance decrease being minimal for lowering the power limit, it seems that two or three water cooled 4090s can be put in one system. Many of the 4090s turn into very small cards once you put a water block on them. Three 4090s with a 400 watt power limit still leaves power left over for the CPU and the rest of the system if you use a 1600 watt power supply.
Well, I do keep forgetting how terribly slow Daz 4.20+ is. I just tested 4.21 with my 3090 and it is much slower than that slow 4090 score. This is the 4.21 beta.
2022-10-13 19:34:41.979 [INFO] :: Total Rendering Time: 2 minutes 7.34 seconds
2022-10-13 19:35:30.407 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1800 iterations, 1.173s init, 124.887s render
That comes out to just 14.413 iterations per second. That is just...sad. That is the worst personal time I have had with my 3090. It is like Iray just keeps getting slower! So all the power that the 4090 brings ends up being wasted.
The 4090 right now is about 64% faster with current drivers. Interestingly this lines up with a lot video game performance charts. Maybe an update will help. I don't know. I don't know what the Iray team is doing, and I have to wonder if they even know what they are doing. Seriously.
However, I still have Daz 4.16 installed. So I ran a test with it.
2022-10-13 19:46:31.019 Total Rendering Time: 1 minutes 36.57 seconds
2022-10-13 19:46:33.940 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1800 iterations, 1.820s init, 92.826s render
And indeed, there it is. I am back to 19.391 iterations per second by using the older version of Daz Studio. So it remains that one of the best upgrades you make to Daz is to downgrade it, LOL.
Unfortunately you cannot do that with Lovelace. At least I don't think you can. Since it works with Daz now, maybe it could work in an older version? Wouldn't that be something?
Hmm... at least it is working and faster. I keep thinking Iray gets slower because some better features but I don't know what that would be, nor why those features would slow down anything if we're using the same old scene. Try 4.20? Is it just a beta thing?
I let it overwrite the 4.20 beta I had since I don't use a lot anyway. I mainly use 4.16 for everything unless I specifically require a feature from the new versions like volumetrics.
I did a whole bunch of testing with 4.20 a while back, it is on the forums somewhere, I made a thread. There are times when 4.20 was as fast or faster. But in my testing that was rare. Most of the time 4.20 was slower. Sometimes it was disturbingly slower, and I mean like greater than 50% slower.
I can't explain why, because the resulting pics were similar. If anything, my pics from 4.16 looked better IMO. 4.20 often took longer per iteration, and on top of that it ran more iterations. This creates a huge double blow that can cause renders to take much, much longer than 4.16. I could not find a factor that led to this. I tried no HDRI, only HDRI, mesh lights, and all kinds of things and I couldn't find something that made sense. The one scene that rendered faster in 4.20 was a bar scene. I can't recall which at this moment. The reason it rendered faster was that 4.20 in this case took a lot fewer iterations than 4.16.
That is the one thing I kept finding, 4.20 was doing a lot more iterations in those renders. But then that one case in the bar it used less. So go figure. In case anyone thinks the iterations must have done more work than 4.16, that is absolutely not true. I capped the iteration count to be equal and when I did, often times 4.20 was grainier, the pixels were not as converged.
There some people in the forums who swear 4.20 has been just as fast for them. Maybe they are doing the same kind of stuff from that bar scene. I don't know what it is about that scene that is different from the others. I tried glass in other scenes.
I can't figure it out. I already tried reinstalling 4.20 before, and I just fresh installed 4.21. So it can't be that. When I want to render, I just stick to 4.16.
I can explain his result because I also got RTX 4090.
Card does not pull more than 285 W when rendering in Daz Studio (tried both 4.16.0.3 Release and 4.21.0.5 Beta) which means the Iray code is not optimized yet for new architecture.
dForce is broken completely in 522.25 Game Ready and Studio driver (OpenCL related errors).
Finally I have to correct you on one thing -- RTX 4090 can absolutely pull 600W.
My card has default 450W power limit which can be raised to 600W using nvidia-smi -pl 600 command. I have tested mine, and I got it to pull 565W in Furmark Stress Test in 1080p.
WARNING: Only try changing the limit to 600W if you have at least 850W power supply (more if you have 12th or 13th gen Intel CPU), preferably with the new 12VHPWR connector (which guarantees 100% power excursions for 100 us intervals) or if you have a really high quality recent power supply like say Seasonic SSR-850 with 4x 8-pin PCI-E cables connected to the adapter cable you get with the card. Needless to say, not all cards and VBIOS-es are going to support such a large power increase (mine is ASUS TUF Gaming GeForce RTX 4090 OC Edition 24GB GDDR6X so it does), and you are going to want a really big and well ventilated case if you plan running that for more than a few minutes at a time.