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
System/Motherboard: MSI MPG Z490 Carbon EK X
CPU: I9-10850K @ 3.6 ghz
GPU: MSI RTX 3090 x2 (No NVLINK)
System Memory: 64 GB Corsair Dominator Platinum DDR4-3466
OS Drive: Samsung 970 EVO SSD 1TB – M.2 NVMe
Asset Drive: Same
Operating System: Win 10 Pro, 1909
Nvidia Drivers Version: 460.89
Daz Studio Version: 4.14
Benchmark Results
2021-01-08 19:49:21.573 Finished Rendering
2021-01-08 19:49:21.595 Total Rendering Time: 51.39 seconds
2021-01-08 19:49:21.296 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2021-01-08 19:49:21.296 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 3090): 923 iterations, 1.546s init, 48.265s render
2021-01-08 19:49:21.296 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (GeForce RTX 3090): 877 iterations, 1.678s init, 47.745s render
Iteration Rate: 1800/51.39 = 35.0262 iterations per minute
Loading Time: 3.125 seconds
________________________
Sample with CPU Rendering Enabled:
2021-01-08 20:32:20.714 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2021-01-08 20:32:20.714 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 3090): 902 iterations, 1.885s init, 49.515s render
2021-01-08 20:32:20.717 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (GeForce RTX 3090): 873 iterations, 1.834s init, 48.947s render
2021-01-08 20:32:20.717 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CPU: 25 iterations, 1.545s init, 49.581s render
2021-01-08 20:32:20.769 Saved image: C:\Users\georg\Desktop\render with cpu and 2x 3090.jpg
2021-01-08 20:32:20.987 Finished Rendering
2021-01-08 20:32:21.018 Total Rendering Time: 53.48 seconds
______________________
I have a few concerns I was hoping others may help with. First, the overall time increased with the CPU enabled. My guess is this is in part because I am stuck at 8x PCIe per 3090 with this z490 board. My questions would be: will an NVLINK help? And should I move the 3090s over to a x16 capable board? My thoughts were that the higher clock on I9-10850K would make working in Daz easier, as compared to something like a i9-10980xe which has a slower clock, but seeing this I have to consider the extra lanes that would be opened up on the X299 boards and overall throughput. Would the wider PCIe lanes help while working in Daz as well?
This is actually not the case. The only time there is intensive loading of the PCI-E bus (with scene data) going to any of the graphics card being used for a render is during the usually seconds long Iray initilaization stage (highlighted in red in your results above.) Once that is done with, the only communication related to rendering that goes on between that GPU and anything else in the system is updated pixel values for the final image sent to from the GPU to CPU (essentially a series of large BMP files spaced out seconds or even minutes apart) and occasional command/control messages from the CPU to the GPU describing which iterations it should work on next. You can verify this by running GPU-Z on your system, switiching it to the "Sensors" tab for a GPU about to be used for rendering, initiating an Iray render utilizing that GPU, and then studying the graph for "Bus Interface Load".
The true reason why rendering with CPU + GPU(s) in a system like yours results in (marginally) slower rendering is because Iray is/was designed to work best in large-scale render-farm type deployments where multiple/many individual machines loaded with multiple identical GPUs or CPUs contributing to a single render is the norm. In such setups, a single CPU/GPU running at a drastically lower rate of performance than the rest is usually a sign of a faulty component. Consequently Iray employs a load-balancing mechanism whereby any singular rendering device found to be severely underperforming in comparsion to its peers has its in-process iterations re-assigned to other more capable devices and its own results dumped when eventually completed (this is to protect the integrity of the overall render since Iray has no way of knowing when or if that final completion comes.) In a smaller setup like yours where the CPU's rendering performance is greatly outclassed by either GPU, this results in a situation where specific iterations being worked on by the CPU are constantly being thrown out and re-assigned to one of the GPUs - resulting in the overall slightly slower render with CPU performance seen above (the CPU is essentially acting like a dead weight due to its relative tardiness.)
In a word: no. NVLink helps overcome vram size limitations for large scenes at the expense of rendering performance (from the limited testing I've seen, around 15% slower overall iirc.)
The only thing you really stand to gain from the higher lane count is a decrese in the Iray initilaization time. Which in the case of this benchmarking scene would mean an absolute best case scenario (taking your results from above) of approximately 0.9 seconds being shaved off of the overall rendering time. Actual production-level scenes will offer a greater speedup than that (due to their greater size than this limited benchmarking scene.) But its never bound to be more than a handful of seconds given how small (relatively speaking) the 24GB framebuffer of the 3090 is in comparsion to the data transfer speed of PCI-E 3.0 (much less 4.0) spread over 8+ lanes.
The reality with Iray is that there are no CPUs in existence even remotely capable of not indadvertantly slowing down (much less keeping up performance-wise with) latest generation high-end GPU hardware. In a 3090 based system like yours you are far better off disabling CPU contributions entirely (see the Render Settings > Advanced > Hardware settings in Daz Studio for the relevant controls.)
Using the CPU is only good for people who have weak GPUs. The 3090 is so fast that rendering with CPU actually slows it down. Even in the very best cases the CPU might add only a small gain. It is not worth it in most situations. That is why I always suggest not worrying about CPU when building a PC focused on Daz Studio and Iray. It is money better spent elsewhere. The CPU is only relevant if you use CPU intensive applications outside of Daz, and even then almost any modern CPU can handle most user's workloads. You do not need "balance" if Daz and Iray are all you use.
The Daz Studio app itself only runs on a single thread, so monster 16 core desktop CPUs mean very little to Daz. Single thread performance may help the Daz app run better, but once again almost any modern CPU can handle the app without any issues.
It would be extreme, but you could totally build a PC with two 3090s and use a 2 core Celeron from 5 years ago. It would not make the slightest difference in your GPU render times.
In some cases if you run out of GPU memory then the CPU is all you have left rendering. But in that situation, it would still be sensible to either modify the scene to get under the limit or to buy a large VRAM card, like the 3090. There is a benchmark here where a $2000 Threadripper just about matched a GTX 1080 in rendering. That was kind of impressive...but the 1080 was $500 back in 2016. And then the 3090 came out with its 24GB capacity, and every Ampere completely destroys those benchmarks. If you need more than 24GB then buy two 3090s and Nvlink them.
I tried comparing the difference between using PCIE 4.0 x16 and PCIE 3.0 X1 riser cables with PCIEx16 slots for multi GPU rendering in the same system and the difference in render times was minimal. The difference was probably within margin of error. Its probably not worth if unless you use other applications that are very PCIE bandwidth intensive.
Pcie has pretty much never mattered for Iray. As you add more and more GPUs you will start to see very small diminishing returns as the number goes up. But that is only with extremes, it does not apply to 2 GPUs. Iray tends to scale extremely well across multiple GPUs. If you do see performance drops, you are more likely to be thermally limited, especially with all the heat these 3090s create. The more power used, the more heat. Two 3090s are going to be putting out heat equivalent to a portable space heater. So your case and cooling solution may need to be examined, not the motherboard.
Any build with multiple GPUs needs to be more careful about planning for the extra heat that extra GPUs will make, and 3090s take this to a whole new level. One 3090 has a TDP nearly 100 Watts more than the 1080ti from 2016. And that is Nvidia's 1st party version, 3rd party variants can go even higher and approach an insane 400 Watts when overclocked.
Thanks RayDAnt and outrider42 for points re: PCIe-bus. Puget systems also notes GPU rendering is rarely affected by fewer lanes. Gaming and video editing is another story. I am going to run some tests with both cards on 4x and then go from there.
Thermals are another topic. So long as temps are under 91 C, you are ok. Right?
Thermals are related, because they can directly effect performance. 91 seems very high. It might be technically OK, but it is the max acceptable, and is absolutely throttling. I read that by default the max on the 3090 is 83, which can manually changed to 91. That alone should tell you this is too high. This changes the throttling amount for high performance overclocking...we don't want that for Iray. Actually, the way Nvidia designs their GPUs they start throttling very early, like in the 50s. That's why pro overclockers use LN2 and why water cooling is a thing. I don't know the specifics for Ampere, but they use a complicated set of parameters that decides when to throttle. The temperature is of course the main factor, but so is the workload. And it just so happens that Iray is a pretty hard workload. So you combine running 100% with 91C and you will throttle.
You should be able to see this by observing the clock speed. It will start out at one number and drop as the card climbs to 91.
I would try to get them down to 83. But like I said above, you'll still be throttling even at that temp, just not as much.
I have two 1080tis. One of them has hit up to 84, but not in Iray. Many Daz users think Iray is the worst thing you can do on GPU. But gaming can actually run a GPU hotter than Iray, because gaming will run VRAM harder than Iray will. With Iray it only hits 74. The other 1080ti doesn't even get this warm, it stays in the 50s and 60s at all times. But my 1080tis use about 100 Watts less EACH. You are talking about 200 extra Watts of power in your system. That's a lot of heat that need to be ejected.
Interesting points about temps. Sadly, the system is fully apart right now. Waiting on a stronger power supply and some better adapters. The power draw is more than anticipated. Everything is back ordered right now and with shipping they are saying about (2) weeks, so everything is on hold for the moment. To protect the rest of the system in the interim, I am stuck on an old i5-3570K. Thankfully, you guys have talked me into keeping the z490. One less part to wait for..
I mention the 91 C as a theoretical ceiling only. Honestly, I have not been watching the temps. You can hear the fans spinning loud during renders so it is probably getting very hot in there. When I get it back together, I will add thermal logs to the benchmark notes using Open Hardware Monitor and share.
I use an app to manually control the GPU fans. I set a more aggressive fan curve than the default. This isn't often needed for a single GPU, but for two or more it really is, especially when they are 3090s. Without my app to control the fans, my warmer card will get hotter than 74. I think it would get to near 90, I can't remember exactly because I only did it once a long time ago. It is only because of the fan curve I set that it sticks to 74. It is louder, that is for sure, but it stays reasonably cool and that is more important.
The app I use is MSI Afterburner. This app can monitor most of your system as well, including CPUs, and provide a slew of data.
The case airflow is going to be vital for a multiple GPU build. Make sure when you put it back together that nothing blocks the air to the GPUs (like stuff in drive bays) and that the case is not too close to anything that may hinder its air. Try to keep the cables positioned so that every bit of air gets to the GPUs. If your case fans can be controlled, you can turn them up as well for rendering. Again much of this would be a bonus for most single GPU setups, but for two 3090s all of this becomes important, as you could be pushing 1000 Watts with that system. That is a legit space heater. Even the ambient temp of the room factors in. It will help your parts last longer, too.
I think CPU rendering is the worst thing you can do to a motherboard / CPU, especially laptops. That's almost certainly largely alleviated with any laptop bought in the last 5 years though.
Yes, I agree on all your points. Even in a water based PC, all of this is very important. This rig idles about 25 C. The curve is set to silent until around 40 C. It spins up some, so it is getting at least into the 40s. I will be sure to document actual temps during bencmarks and under load. The results will also help me fine tune the fan curves.
For a visual, here is a pic fom a few days ago. I cannot wait to start doing some Daz on it.
That is a thing of beauty.
DAZ 3D 4.15.0.2 x64 (GTX 1660 Super)
System Configuration
System/Motherboard: MSI Z370 PC PRO
CPU: Intel i7-8700 3.20 GHz @ stock
GPU: Zotac GTX 1660 Super AMP (GPU: +60mhz MEM: +775mhz Power: 105% in screenshot)
System Memory: 1x CORSAIR VENGEANCE 16GB DDR4 (CMK16GX4M1B3000C15) @ stock
OS Drive: Samsung PM981 NVMe 1024 GB
Asset Drive: Seagate 2TB (ST2000DL003-9VT116)
Operating System: Windows 10 Home Single 20H2
Nvidia Drivers Version: 461.09
Daz Studio Version: 4.15.0.2 Pro x64
Benchmark Results
DAZ_STATS
2021-01-19 13:36:56.560 Finished Rendering
2021-01-19 13:36:56.588 Total Rendering Time: 9 minutes 29.32 seconds
IRAY_STATS
2021-01-19 13:46:25.962 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2021-01-19 13:46:25.962 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce GTX 1660 SUPER): 1800 iterations, 6.072s init, 560.073s render
Iteration Rate: 1800/560.073 = 3.213
Loading Time: 569.32 - 560.073 = 8.59
Ok guys, time for rebuild the tests with the Rtx 30xx in the new Daz 4.15.0.2, please.
So on a whim I bought a 20
60 KO that just so happened to be returned at micro center today (since I cannot find a 3090 to replace my 2080ti. I saw other buys mixing up on here with oddball cards and getting better performance. Well a 2080ti and 2660 ko really... and I mean really like each other. I will post the final numbers when I finish tweaking. I am sitting right at 2 min flat down from 4 min. And my viewport is almost real-time. I would say cutting half the time off is significant,.. Now, my bet there is going to be a huge gotcha that I don't know about. If there is one... let me down easy and I will take it back. My thinking is the memory will default to the lowest card. I mean by that is 6gb and small jobs will scream but that's it. I am going to run some tests.... BUT DANG! Yes the memory and clock speeds have been pushed a bit and a nasty thrown together custom cooling loop. I can keep both cards over the 2100 mark all day long.
I really hope I did all this right. Because this is cool, and very quick.
2021-01-20 20:00:22.591 Total Rendering Time: 2 minutes 0.76 seconds
2021-01-20 20:00:22.609 Loaded image r.png
2021-01-20 20:00:22.645 Saved image: C:\Users\Jay_Jarrett\AppData\Roaming\DAZ 3D\Studio4 Public Build\temp\RenderAlbumTmp\Render 1.jpg
2021-01-20 20:02:10.819 Saved image: C:\Users\Jay_Jarrett\OneDrive\Documents\DAZ 3D\Studio\Render Library\all off.png
2021-01-20 20:02:11.027 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2021-01-20 20:02:11.027 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 2080 Ti): 1124 iterations, 1.250s init, 117.585s render
2021-01-20 20:02:11.027 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (GeForce RTX 2060): 676 iterations, 1.407s init, 117.178s render
what the hell?? 2 rtx 3090s and just using daz
that's really waste of money and stupid idea and it's overkill for daz and for all games, i guess it's rich people stuff then. :/
Its only a waste if I lack imagination.
Not if you are an animator. 2 3090s make certain things possible again if you keep to a schedule.
Computers as a whole are a total waste of time/money if you want to get technical about it.
I can't download the benchmark scene anymore, it says "Page Not Found".
Is it only me?
Was doing some thread re-formatting the other day, and something went awry. Try it now!
Having two graphics cards (particularly with that much memory) will ensure being able to opull off just about anything any rarely, if ever, having to be concerned with having too little of VRAM.
There's certainly no "overkill" happening here.
It works now, thanks!
Let's test this new Daz Studio 4.15.
Sadly I had to get rid of my 2060, but I guess my 2070 Super will go a bit better now.
System Configuration
System/Motherboard: Asus TUF X570
CPU: Ryzen 5 3600
GPU: RTX 2070 SUPER
System Memory: 64GB 3000Mhz
OS Drive: Sabrent Rocket Plus 1TB
Asset Drive: Sabrent Rocket Q 1TB
Operating System: Windows 10 10.0.190432
Nvidia Drivers Version: 460.89
Daz Studio Version: VERSION BITS
Benchmark Results
DAZ_STATS
2021-01-23 15:04:17.282 Finished Rendering
2021-01-23 15:04:17.312 Total Rendering Time: 4 minutes 19.34 seconds
IRAY_STATS
2021-01-23 15:05:16.525 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2021-01-23 15:05:16.525 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 2070 SUPER): 1800 iterations, 2.229s init, 254.546s render
Iteration Rate: 7.07 iterations per second
Loading Time: 4.79 seconds
For reference, in Daz 4.14 2070S+2060 achieved 10.53 iter/s.
In that case, the 2070S did 61.2% of the work (1103 iterations out of 1800), so 6.45 iter/s. Other benchmarks in 4.14 with the 2070S had an average of 6.49.
So, I saw an improvement of +9.6% from Daz Studio alone, yet another time: that's great!
The same GPU, in a similar system, used to give me 5.82 iter/s in December 2019 (Daz 4.12): that's a 21.5% improvement from software optimization only!
Great work guys.
Hi, what version of the benchmark are people using here?
Is this from the first page of this thread?
https://www.daz3d.com/forums/discussion/341041/daz-studio-iray-rendering-hardware-benchmarking/p1
Yes - see this link.
Thanks a lot. Probably will need to use it on the new computer.
Does anybody use Daz Studio on dual Intel CPU on Windows 10 Pro?
I have had opportunity to make a tests a couple years ago on dual Intel Xeon CPUs,
but could not get Daz Studio to use the second CPU on Windows 7 Pro at that time.
I am still hoping, that in future one could use rendering in iray on multiple CPUs
and have much more RAM available for the scenes.