Daz Studio Iray - Rendering Hardware Benchmarking

outrider42

LenioTG said:

I've made this Google Sheet for me, and it's public.

It has all entries for this monstrous 4.14 update, and it quickly compares the iteration rate for all the GPUs.

Link: https://docs.google.com/spreadsheets/d/1wkuVTjE9LDrq_YHstg_iJy6HUO5eQTNlXGMB25mfk44/

I'm attaching the charts comparing the various GPUs.

I know it's not 100% precise (you can't see if the GPU was overclocked, some entries are extrapolated from multi-GPU setup where the total result differs from the sum of the single ones, you don't know in what system the GPU was etc.), but I needed a quick a clear comparison to compare the GPU prices and all.

You know what would be real cool, would be to have a pre 4.14 bar on the same bar, just a different color. Sort of like how some hardware reviews have 0.1% lows overlaid on top of their average frame rates. This would demonstrate the difference between them. Just a thought. I may try that when I get home.

LenioTG

You know what would be real cool, would be to have a pre 4.14 bar on the same bar, just a different color. Sort of like how some hardware reviews have 0.1% lows overlaid on top of their average frame rates. This would demonstrate the difference between them. Just a thought. I may try that when I get home.

Yes it would be cool, and maybe it could let us extrapolate the performance of the yet untested GPUs in 4.14!
But there's still no 3070 benchmark around, for any version of Daz.
Feel free to use any data you want from the Google Sheet :D

RayDAnt

outrider42 said:

RayDAnt said:

colcurve said:

do I understand the figures on the first page right and 3060ti vs 3080 is about 15% difference only? but price is like 500 vs 800?

As of this post, the only benchmark result we have for the 3080 was done on Daz Studio version 4.12.2.051 Beta. Whereas The onky 3060Ti and multiple of the 3090 results were done on DS version 4.14.0.008 Beta or newer. Since there was a major across-the-board rendering performance uplift for all graphics cards during that time, that 3080 result (nor the majority of the rest of the results compiled in this thread) is not as fast as it should be in comparison to those other 3000 series cards. We need someone to test/re-test a 3080 and post there resutls while running DS 4.14.0.008 or later here to know for sure how it really matches up.

I think we may need to change up the table a bit to better differentiate 4.14 from the previous versions. The performance change is just too dramatic, and anybody who casually looks at the chart may be mislead by the skewed results between different versions. 

While DS versions are listed in the second chart, the big main table does not. I do not think we need to have every version of DS on there, since performance never really changed between them. But I think would should have a column on the chart just for 4.14 and even make it a different color so that it stands out from the rest. As well as a note at the top of the chart (since many people will not read notes at the bottom) describing what we have found with 4.14 so far. Every bench before 4.14 is now simply out of date.

Added a separate column to the main results table to differentiate between pre and post 4.14.0.008 results. Was hoping to find another way since adding columns to that chart ups the raw ascii character count of the post, and the Daz Forums have a fixed per-post character limit (the Forum UI also ins't conducive to many-columned posts...) But that ended up being the most graceful way to keep the information already here legible. Eventually I am going to reach a point where that character limit is reached (either for the main summary chart post or one of the individual device summary chart posts) and I'll have to start culling information from somewhere. Most likely the combined device benchmarks (eg. the Titan RTX + Titan RTX entry) since - if there's one thing all the testing in this thread has proven time and again - performance with multi-GPU setups and Iray really is almost linear. Meaning that all you really need to know is how fast a single Titan RTX is in order extrapolate how fast N number of Titan RTXs will be to within a very close margin.

 

LenioTG said:

Hi RayDAnt, is it possible that you have missed a few benchmarks?

I had tested RTX 2060, RTX 2070 Super and Ryzen 5 3600 in all combinations, both in 4.12, 4.14 Beta and 4.14, but I don't see them in the updated charts.

There are a number of posts from the pre-Ampere release that I have still yet to incorporate into the summarized results (hence the increasingly complex wording of the first of the notes in red.) Eventually I will get them in there.

RayDAnt

System Configuration
System/Motherboard: Gigabyte Z370 Aorus Gaming 7
CPU: Intel i7-8700K @ stock (MCE enabled)
GPU: Nvidia Titan RTX @ stock (watercooled)
System Memory: Corsair Vengeance LPX 32GB DDR4 @ 3000Mhz
OS Drive: Samsung Pro 970 512GB NVME SSD
Asset Drive: Sandisk Extreme Portable SSD 1TB
Operating System: Windows 10 Pro version 20H2
Nvidia Drivers Version: 460.89
Optix Prime Acceleration: N/A

Benchmark Results - Titan RTX (TCC Mode)

Daz Studio Version: 4.14.0.010 Pro x64
Total Rendering Time: 3 minutes 27.40 seconds
CUDA device 0 (TITAN RTX): 1800 iterations, 6.069s init, 198.796s render
Iteration Rate: (1800 / 198.796) = 9.055 iterations per second
Loading Time: ((0 + 180 + 27.40) - 198.796) = 8.604 seconds

Daz Studio Version: 4.14.1.028 Beta x64
Total Rendering Time: 3 minutes 18.60 seconds
CUDA device 0 (TITAN RTX): 1800 iterations, 6.839s init, 189.597s render
Iteration Rate: (1800 / 189.597) = 9.498 iterations per second
Loading Time: ((0 + 180 + 18.60) - 189.597) = 9.003 seconds

---

It's been a while since I've submitted anything myself. Going by my last entry (7.899 IPS) that's a performance  increase of 1.104 iterations per second or about 14% with the latest post 4.14.0.008 releases on the Titan RTX.

outrider42

Good work, I'm sure plenty of people appreciate your efforts. I'd say if you reach character limits to drop CPUs and old GPUs, especially ones that haven't had any recent posts. They could go into an archived post if you have space for one. If not, a link to separate archive thread.

11212003_7b775cb26b

System Configuration
System/Motherboard: Gigabyte Aorus X399 Pro
CPU: AMD Ryzen Threadripper 1920X @ stock
GPU: NVIDIA Geforce RTX 3090 @ stock
System Memory: HyperX Fury 64GB DDR4 @ 2666 MHz
OS Drive: Kingston A2000 SSD 1TB
Asset Drive: Samsung 970 EVO 1TB
Operating System: Windows 10 Pro version 2004 build 19041.685
Nvidia Drivers Version: 460.89
Daz Studio Version: 4.14.0.10 x64

Benchmark Results
2020-12-18 10:56:43.666 Finished Rendering
2020-12-18 10:56:43.704 Total Rendering Time: 1 minutes 47.1 seconds

2020-12-18 10:57:06.178 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
2020-12-18 10:57:06.178 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 7.797s init, 96.058s render

Iteration Rate: 18.739 iterations per second
Loading Time: 11.042 seconds

imaginateca

First RTX 3070 benchmark. No CPU used for rendering. Just GPU.

System Configuration
System/Motherboard: ASUS TUF B450 Gaming Plus II
CPU: Ryzen 7 3700x 4.4GHz (stock)
GPU: Zotac Gaming GeForce RTX 3070 Twin Edge OC 8GB GDDR6 1755 MHz (stock)
System Memory: Corsair Vengeance LPX 16 GB (2 x 8 GB) DDR4 3200 MHz C16 (stock)
OS Drive: 256GB SSD
Asset Drive: Same
Operating System: Windows 10 Pro
Nvidia Drivers Version: 460.79
Daz Studio Version: 4.14.0.10 Pro Edition 64bit
Optix Prime Acceleration: N/A (Daz Studio 4.12.1.086 or earlier only)

Benchmark Results
2020-12-20 15:15:38.830 Finished Rendering
2020-12-20 15:15:38.863 Total Rendering Time: 2 minutes 38.21 seconds
2020-12-20 15:15:46.547 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
2020-12-20 15:15:46.548 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3070):      1800 iterations, 1.900s init, 153.632s render
Iteration Rate: (1800/153.632) = 11,71 iterations per second
Loading Time: ((0*3600 + 2*60 + 38.21) - 153.632) = 4.578 seconds

imaginateca

nonesuch00 said:

Well, looking at the numbers for the 3060 TI they are about 50 seconds slower than the 3090s numbers I've seen in this thread.

I don't see no 3080 or 3070 numbers though. I'm more interested in a 3070's numbers though from a personal affordability perspective. Did I miss those posts? I know those cards are the 2 most difficult to get.

I have posted a 3070 benchmark: My 3070 card was some milisecs faster than the only one 3080 benchmark published (my time was 2 minutes 38.21 secons, while the 3080's time was 2 minutes 38.53 seconds).

15,49 seconds faster than the 3060TI benchmark.

And 89,16 seconds faster than the best 2080TI benchmark published (this is a huge difference if you plan to render DAZ videos)

I suppose that the better result compared to the 3080 is because the different version of Daz Studio used in the two benchmarks (4.14 in 3070 vs 4.12 in 3080).

I hope this info helps you to decide what to buy. I think that the 3070 is a good choice for 549 EUR (it is an OC version, the no-OC version is 519 EUR).

Cheers!

nonesuch00

imaginateca said:

nonesuch00 said:

Well, looking at the numbers for the 3060 TI they are about 50 seconds slower than the 3090s numbers I've seen in this thread.

I don't see no 3080 or 3070 numbers though. I'm more interested in a 3070's numbers though from a personal affordability perspective. Did I miss those posts? I know those cards are the 2 most difficult to get.

I have posted a 3070 benchmark: My 3070 card was some milisecs faster than the only one 3080 benchmark published (my time was 2 minutes 38.21 secons, while the 3080's time was 2 minutes 38.53 seconds).

15,49 seconds faster than the 3060TI benchmark.

And 89,16 seconds faster than the best 2080TI benchmark published (this is a huge difference if you plan to render DAZ videos)

I suppose that the better result compared to the 3080 is because the different version of Daz Studio used in the two benchmarks (4.14 in 3070 vs 4.12 in 3080).

I hope this info helps you to decide what to buy. I think that the 3070 is a good choice for 549 EUR (it is an OC version, the no-OC version is 519 EUR).

Cheers!

Thanks! So it's basically between the 3070 & 3080 for me. I think I'll get the 3080 because I don't want to regret not having enough video RAM. 

musicaRoca

System/Motherboard: AS Rock B450
CPU: AMD Ryzen 7 3700X 8-Core Processor 3.60 GHz
GPU: RTX 2070 Super
System Memory: Corsair Vengeance RGB Pro 32GB (2x16GB) DDR4 3200MHz C16 
Asset Drive: 2TB TOSHIBA HDWD120
Operating System: Windows 10 Pro - 20H2 - 19042.685
Nvidia Drivers Version: 460.89 StudioVersion
Daz Studio Version: 4.14.0.010

Benchmark Results
Total Rendering Time: 4 minutes 44.80 seconds

CUDA device 0 (GeForce RTX 2070 SUPER): 1627 iterations, 3.070s init, 278.470s render
CPU: 173 iterations, 1.906s init, 278.595s render

Iteration Rate: (1800 / 284,8)  6.320 iterations per second
Loading Time: ((0 + 240  + 44.8) - 278.595) = 6.205  seconds

and 

Total Rendering Time: 4 minutes 52.76 seconds
CUDA device 0 (GeForce RTX 2070 SUPER): 1800 iterations, 2.054s init, 288.112s render
Iteration Rate: (1800 / 292.76) = 6.148 iterations per second
Loading Time: ((0 + 240 + 52.76) - 288.112) = 4,648 seconds

outrider42

nonesuch00 said:

imaginateca said:

nonesuch00 said:

Well, looking at the numbers for the 3060 TI they are about 50 seconds slower than the 3090s numbers I've seen in this thread.

I don't see no 3080 or 3070 numbers though. I'm more interested in a 3070's numbers though from a personal affordability perspective. Did I miss those posts? I know those cards are the 2 most difficult to get.

I have posted a 3070 benchmark: My 3070 card was some milisecs faster than the only one 3080 benchmark published (my time was 2 minutes 38.21 secons, while the 3080's time was 2 minutes 38.53 seconds).

15,49 seconds faster than the 3060TI benchmark.

And 89,16 seconds faster than the best 2080TI benchmark published (this is a huge difference if you plan to render DAZ videos)

I suppose that the better result compared to the 3080 is because the different version of Daz Studio used in the two benchmarks (4.14 in 3070 vs 4.12 in 3080).

I hope this info helps you to decide what to buy. I think that the 3070 is a good choice for 549 EUR (it is an OC version, the no-OC version is 519 EUR).

Cheers!

Thanks! So it's basically between the 3070 & 3080 for me. I think I'll get the 3080 because I don't want to regret not having enough video RAM. 

I'm waiting on that 3080ti myself. The 3080ti, if true, is basically what I wanted all along. 20gb at $1,000 to be exact. Of course availablity will be a big concern, but I think Nvidia is ramping up production now that their lineup is taking shape. They now know what AMD is doing, so they can begin their counter attack. I think the 3080ti and potential 3070ti will be produced in much higher numbers, as the original 3080 quietly gets phased out. Well, not totally phased out, but the emphasis will be on the "ti" models.

Nvidia signed a big contract with Samsung to expand production. How fast that kicks in is uncertain, but there will be a lot more GPUs available in the future, and a possible refresh.

nonesuch00

outrider42 said:

nonesuch00 said:

imaginateca said:

nonesuch00 said:

Well, looking at the numbers for the 3060 TI they are about 50 seconds slower than the 3090s numbers I've seen in this thread.

I don't see no 3080 or 3070 numbers though. I'm more interested in a 3070's numbers though from a personal affordability perspective. Did I miss those posts? I know those cards are the 2 most difficult to get.

I have posted a 3070 benchmark: My 3070 card was some milisecs faster than the only one 3080 benchmark published (my time was 2 minutes 38.21 secons, while the 3080's time was 2 minutes 38.53 seconds).

15,49 seconds faster than the 3060TI benchmark.

And 89,16 seconds faster than the best 2080TI benchmark published (this is a huge difference if you plan to render DAZ videos)

I suppose that the better result compared to the 3080 is because the different version of Daz Studio used in the two benchmarks (4.14 in 3070 vs 4.12 in 3080).

I hope this info helps you to decide what to buy. I think that the 3070 is a good choice for 549 EUR (it is an OC version, the no-OC version is 519 EUR).

Cheers!

Thanks! So it's basically between the 3070 & 3080 for me. I think I'll get the 3080 because I don't want to regret not having enough video RAM. 

I'm waiting on that 3080ti myself. The 3080ti, if true, is basically what I wanted all along. 20gb at $1,000 to be exact. Of course availablity will be a big concern, but I think Nvidia is ramping up production now that their lineup is taking shape. They now know what AMD is doing, so they can begin their counter attack. I think the 3080ti and potential 3070ti will be produced in much higher numbers, as the original 3080 quietly gets phased out. Well, not totally phased out, but the emphasis will be on the "ti" models.

Nvidia signed a big contract with Samsung to expand production. How fast that kicks in is uncertain, but there will be a lot more GPUs available in the future, and a possible refresh.

Double the RAM for $300 more is well worth the wait if it comes by July 1st. After that and it becomes a case of too much lost productivity as these current 30X0 series cards can do bundles more help to us then past cards could. I've been considering saving another 3 - 4 months until Jul 1st and even getting a 3090 for $1500 (assuming the price gouging ends by Jul 2021 but who knows). When the card can finally do what one needs then it's best to get one with the sufficient RAM to avoid it becoming a paperweight in 2 years.

I guess there is not even a 2021 quarter prediction though for when a 3080 TI becomes available though is there? 

artphobe

why is a 2070 super faster than a 2080 non super

2080 non super has more cuda cores

outrider42

The tests that are posted are from different versions of Daz Studio and different versions of Iray. It is best if you are able to compare directly. At least one of the 2080 scores comes from Iray before it even had full RTX support, so the RT cores were not being used at all. That would certainly skew results. It would be great if some of our 2080 and 2070 owners ran new benchmarks in 4.14.

At any rate, even if it turns out a 2070 S does beat a 2080, this is not so unprecedented. The 1070ti could match and even beat a 1080. If you check Nvidia's info, the 1070ti actually had a higher CUDA version than the 1080. I suspect this enabled it to perform the way it did.

But wait, there's still more! The 2060 KO is a very interesting chip. It was a cut down 2080 chip instead of a 2070. Nvidia disabled the cores to match the proper 2060 CUDA and RT counts, but some 2080 features were still active, allowing the 2060 KO to perform surprisingly well at non gaming tasks. I can't remember if that effected Iray or not, but the point here is that sometimes they do things that enable a card to defy their specs.

junk

Using Daz 4.14.0.10
NVidia driver 457.30 
CPU: Ryzen 5600x overclocked to 4.7GHz
Memory: 64GB DDR-3600
Windows 10 Pro
GPU: EVGA GeForce RTX 3090 XC3 ULTRA HYBRID GAMING overclocked GPU by 200+, DDR6x by 1225Mz

FROM THE LOG FILE:
2020-12-25 21:42:36.896 Total Rendering Time: 1 minutes 29.57 seconds
2020-12-25 21:42:42.664 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
2020-12-25 21:42:42.664 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.957s init, 85.436s render

Test was completed following everything that I could see to a T. Viewport set to "Texture shaded", Computer rebooted, File opened directly without being opened previousely.  GPU reached 56 celsius during the test thanks to the water cooling.

umut.mslhoglu

System Configuration
System/Motherboard: H81M-E33
CPU: Intel Core i7-4790K
GPU: GeForce GTX 1070 ROG Strix 
System Memory: Samsung 250GB 850 540MB / 520MB
Asset Drive: WD Green 2 TB
Operating System: Windows 10 P
Nvidia Drivers Version: 460.89
Daz Studio Version: 4.14

 

Benchmark Results
2020-12-26 20:05:29.463 Finished Rendering
2020-12-26 20:05:29.494 Total Rendering Time: 10 minutes 20.19 seconds

2020-12-26 20:05:43.502 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

2020-12-26 20:05:43.502 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1070): 1800 iterations, 3.000s init, 614.348s render

Iteration Rate: 1800/10,2 = 176,4
Loading Time: [(0 + 600 + 20.19) - (614.348)] = 5,842 seconds

junk

outrider42 said:

But wait, there's still more! The 2060 KO is a very interesting chip. It was a cut down 2080 chip instead of a 2070. Nvidia disabled the cores to match the proper 2060 CUDA and RT counts, but some 2080 features were still active, allowing the 2060 KO to perform surprisingly well at non gaming tasks. I can't remember if that effected Iray or not, but the point here is that sometimes they do things that enable a card to defy their specs.

My friend bought the 2060 KO and we tried it out in Daz Studio.  I can't say that we saw much of an improvement.. maybe 3% or so but we were really trying hard to be fair during testing.  Maybe someone else had different results. 

I forgot to mention that he also had a 2060, then bought the 2060 KO afterwards so were able to do the tests.

outrider42

I suppose Iray doesn't utilize those features. That's a shame, of note Blender is listed as one app that does benefit with the KO. There are software where 2060 KO performs way above what a 2060 should. In fact a lot better. Gamers Nexus reported an incredible 26% to 47% uplift in some non gaming applications over the original 2060. They did a full video on what they found.

There is one other thing, according to GN, not all 2060 KOs actually use the 2080 die. It is basically silicon lottery as to if you get one. So there is a possibly that maybe your friend got the standard 2060 die. The only way to know for certain would be to take card apart and examine it. 

I suppose that could also be a warning to anyone trying to buy a KO hoping for the 2080 die.

junk

outrider42 said:

I suppose Iray doesn't utilize those features. That's a shame, of note Blender is listed as one app that does benefit with the KO. There are software where 2060 KO performs way above what a 2060 should. In fact a lot better. Gamers Nexus reported an incredible 26% to 47% uplift in some non gaming applications over the original 2060. They did a full video on what they found.

There is one other thing, according to GN, not all 2060 KOs actually use the 2080 die. It is basically silicon lottery as to if you get one. So there is a possibly that maybe your friend got the standard 2060 die. The only way to know for certain would be to take card apart and examine it. 

I suppose that could also be a warning to anyone trying to buy a KO hoping for the 2080 die.

That could be very true.  We did see that video and is the reason he purchased the KO when he already has the 2060.  I think he shopped at a specific place that guaranteed they had the 2080 neautured GPU but I really don't know in the end.

outrider42

Did your friend try any other apps that are confirmed to run faster on a KO? I think these KOs might have ran hotter than a normal 2060, too. They used the much larger 2080 die, so they were not nearly as efficient as normal 2080s. They used more energy and thus created more heat. So comparing that to the original 2060 could provide a clue if not running those apps. 

The one big downside is that 6GB of VRAM is a bit tight for Daz content these days. But it is an interesting piece of hardware.

junk

outrider42 said:

Did your friend try any other apps that are confirmed to run faster on a KO? I think these KOs might have ran hotter than a normal 2060, too. They used the much larger 2080 die, so they were not nearly as efficient as normal 2080s. They used more energy and thus created more heat. So comparing that to the original 2060 could provide a clue if not running those apps. 

The one big downside is that 6GB of VRAM is a bit tight for Daz content these days. But it is an interesting piece of hardware.

My friend is all about benchmarks.... only.  He buys GPU's like they're water and leaves them in PC's that he doesn't turn on anymore.  :(  No gaming, no productivity based programs, nothing but benchmarks...  3DMark, Passmark, Geekbench, Superposition, on and on.  Oh wait, he did install the trial of Shadow of the Tomb Raider for the benchmark (which is subjective anyways depending upon your screen resolution and monitor refresh rate). 

From what I can recall he tried other benchmarks and felt it was faster in a specific Blender benchmark he uses once in a while.  But honestly I can't remember the details.

skyeshots

System/Motherboard: MSI MPG Z490 Carbon EK X 
CPU: I9-10850K @ 3.6 ghz 
GPU: MSI RTX 3090
System Memory: 32 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
2020-12-31 23:56:13.198 Finished Rendering
2020-12-31 23:56:13.224 Total Rendering Time: 1 minutes 35.86 seconds
2020-12-31 23:56:12.947 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
2020-12-31 23:56:12.947 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.410s init, 92.448s render

Iteration Rate: 1800/95.86 = 18.7773 iterations per minute
Loading Time: 3.412 seconds

 

RayDAnt

junk said:

outrider42 said:

Did your friend try any other apps that are confirmed to run faster on a KO? I think these KOs might have ran hotter than a normal 2060, too. They used the much larger 2080 die, so they were not nearly as efficient as normal 2080s. They used more energy and thus created more heat. So comparing that to the original 2060 could provide a clue if not running those apps. 

The one big downside is that 6GB of VRAM is a bit tight for Daz content these days. But it is an interesting piece of hardware.

My friend is all about benchmarks.... only.  He buys GPU's like they're water and leaves them in PC's that he doesn't turn on anymore.  :(  No gaming, no productivity based programs, nothing but benchmarks...  3DMark, Passmark, Geekbench, Superposition, on and on.  Oh wait, he did install the trial of Shadow of the Tomb Raider for the benchmark (which is subjective anyways depending upon your screen resolution and monitor refresh rate). 

From what I can recall he tried other benchmarks and felt it was faster in a specific Blender benchmark he uses once in a while.  But honestly I can't remember the details.

GPU-Z can tell you both which GPU die a card has and even what revision it is. I have access to a 2060 KO Ultra in a streaming system I put together for work back in November, and this is GPU-Z has to say about it:

Although there is indeed technically no guarantee that a 2060 KO will be built from a cut down TU104 die rather than the usual TU106, the practical reality is that there is virtually no chance of the KO line ever being switched over to the TU106 since it makes much more sense to use a 2060-capable TU106 die in a standard 2060. The KO line was conceived of as a way to make the most of subpar TU104 stock. Running out of supply (for whatever reason) of subpar TU104 stock would be better grounds for discontinuing the KO line entirely rather than retooling it to use a different GPU die (when that specific configuration already exists as an established product.)

As for the baseline performance anomalies seen with the KO, here's a quick and dirty description I sent to GamersNexus last February (my, how time flies) of the underlying tecnical differences in play:

Per Nvidia's parlance, all a Turing generation GPU die has to do in order to qualify as a 2060 level product is contain exactly 30 functional Streaming Multiprocessors (the underlying building block of modern Nvidia GPUs - not individual Cuda cores since those aren't capable of operating on independent workloads) Where these 30 SMs exist physically and how they interconnect with each other/additional processing units isn't ever specified. Meaning that there are lots of potential configurations on a given GPU die generation that could lead to varrying performance figures from identically classed cards.

In short, a TU104 cut down to a 2060's overall Cuda core count ends up having a greater number of accompanying processing cores and data pipelines per Cuda core than what you will ever get from the TU106. Hence the noticeable performance boost seen with KO cards under some workloads having less to do with traditional gaming graphics processing (where raw Cuda core count isn't necessarily the most important thing.)

chrislb

I was having issues with my EVGA RTX 3090 FTW3 card and the manufacturer offered to give me a full refund if I returned the card.  Luckily I was able to buy a MSI Gaming X Trio at MSRP + Tax locally.  I installed the EVGA 520 watt RTX 3090 BIOS on the Gaming X Trio(stock BIOS is 380 watts) and ran the benchmark to see if there was any difference.  I think I actually had a higher iteration rate with the EVGA RTX 3090, but that was also a differnet driver version and was run before the latest Windows 10 updates.

System Configuration
System/Motherboard: MSI MEG X570 ACE
CPU: AMD R9 3950X @ Stock with PBO +200
GPU: MSI Gaming X Trio RTX 3090 with EVGA 520 watt BIOS
System Memory: Corsair Vengeance RGB Pro 64 GB @ 3600 MHz CAS18
OS Drive: 1TB Sabrent Rocket NVMe 4.0 SB-ROCKET-NVMe4-1TB
Asset Drive: XPG SX 8100 NVMe SSD
Operating System: Windows 10 Pro version 2004 Build 19042.685
Nvidia Drivers Version: 460.89
Daz Studio Version: 4.14.0.10

Benchmark Results: One MSI RTX 3090 Gaming X Trio with EVGA 520 watt BIOS. No CPU rendering.
2021-01-05 02:25:25.828 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Maximum number of samples reached.
2021-01-05 02:25:26.344 Finished Rendering
2021-01-05 02:25:26.391 Total Rendering Time: 1 minutes 29.8 seconds

2021-01-05 02:25:30.391 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

2021-01-05 02:25:30.391 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 0.857s init, 85.751s render

Iteration Rate: (1800 / 85.751s) = 20.991 iterations per second
Loading Time: (89.8 seconds) - 85.751s) = 4.049 seconds

LenioTG

If you're interested, I'm still updating the Google Sheet about Daz 4.14 Iray benchmarks: https://docs.google.com/spreadsheets/d/1wkuVTjE9LDrq_YHstg_iJy6HUO5eQTNlXGMB25mfk44/

chrislb

I'm trying to run the benchmark in a system with three RTX 3090 cards, but Daz only ever detects two of them.  Windows and other applications see all three.

chrislb

Since I coudln't get three cards to work, I tried the MSI and EVGA card together.

 

System Configuration

System/Motherboard: MSI MEG X570 ACE

CPU: AMD R9 3950X @ Stock with PBO +200

GPU: MSI Gaming X Trio RTX 3090 with MSI 450 watt BIOS and EVGA RTX 3090 FTW3 with EVGA 450 watt BIOS

System Memory: Corsair Vengeance RGB Pro 64 GB @ 3600 MHz CAS18

OS Drive: 1TB Sabrent Rocket NVMe 4.0 SB-ROCKET-NVMe4-1TB

Asset Drive: XPG SX 8100 NVMe SSD

Operating System: Windows 10 Pro version 2004 Build 19042.685

Nvidia Drivers Version: 460.89

Daz Studio Version: 4.14.0.10

 

Benchmark Results: MSI Gaming X Trio RTX 3090 with MSI 450 watt BIOS and EVGA RTX 3090 FTW3 with EVGA 450 watt BIOS No CPU rendering.

2021-01-05 14:10:46.967 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Maximum number of samples reached.

2021-01-05 14:10:47.526 Finished Rendering

2021-01-05 14:10:47.578 Total Rendering Time: 53.22 seconds

 

2021-01-05 14:10:50.573 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

2021-01-05 14:10:50.573 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 895 iterations, 1.554s init, 48.862s render

2021-01-05 14:10:50.574 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce RTX 3090): 905 iterations, 1.170s init, 48.394s render

 

Iteration Rate: (1800 / 48.862s) = 36.838 iterations per second

Loading Time: (53.22 seconds) - 48.862s) = 4.358 secondss

trahcan_ceab2168e6

System Configuration
System/Motherboard: MSI X470 gaming pro
CPU: AMD Ryzen 7 2700x
GPU: GeForce® RTX 2070 SUPER™ GAMING OC 3X 8G (rev. 1.0)
System Memory: G Skill F4-3000C16-16GISB 32768 Mbyte DDR 4 @ 1066.4 MHz
OS Drive: Samsung Evo 970 1 TB
Asset Drive:  Samsung Evo 970 1 TB
Operating System: Microsoft Windows 10 64 bit (Version 10.0.19041.685)
Nvidia Drivers Version: 27.21.14.5730 (GeForce 457.30) 
Daz Studio Version: DAZ Studio 4.14.0.8 Pro Edition 64-bit
Optix Prime Acceleration: OptiX version 7.1.0

Benchmark Results
Total Rendering Time: 4 minutes 29.84 seconds
01800 iterations / 266.652s
Iteration Rate: (CUDA device 0 (GeForce RTX 2070 SUPER) (01800 iterations / 266.652s) = 6.750 iterations per second
Loading Time: 3.188 seconds

trahcan_ceab2168e6

System Configuration
System/Motherboard: MSI X470 gaming pro
CPU: AMD Ryzen 7 2700x
GPU: GeForce® RTX 2070 SUPER™ GAMING OC 3X 8G (rev. 1.0)
System Memory: G Skill F4-3000C16-16GISB 32768 Mbyte DDR 4 @ 1066.4 MHz
OS Drive: Samsung Evo 970 1 TB
Asset Drive:  Samsung Evo 970 1 TB
Operating System: Microsoft Windows 10 64 bit (Version 10.0.19041.685)
Nvidia Drivers Version: 27.21.14.5730 (GeForce 457.30) 
Daz Studio Version: DAZ Studio 4.14.1.38 Pro Edition 64-bit Bublic Build
Optix Prime Acceleration: OptiX version 7.1.0

Benchmark Results
Total Rendering Time: 4 minutes 17.25 seconds
01800 iterations / 253.962 s
Iteration Rate: (CUDA device 0 (GeForce RTX 2070 SUPER) (01800 iterations / 253.962s) = 7.087 iterations per second
Loading Time: 3.288 s

 

trahcan_ceab2168e6

System Configuration
System/Motherboard: MSI X470 gaming pro
CPU: AMD Ryzen 7 2700x
GPU: GeForce® RTX 2070 SUPER™ GAMING OC 3X 8G (rev. 1.0)
System Memory: G Skill F4-3000C16-16GISB 32768 Mbyte DDR 4 @ 1066.4 MHz
OS Drive: Samsung Evo 970 1 TB
Asset Drive:  Samsung Evo 970 1 TB
Operating System: Microsoft Windows 10 64 bit (Version 10.0.19041.685)
Nvidia Drivers Version: 27.21.14.5730 (GeForce 457.30) 
Daz Studio Version: DAZ Studio 4.14.1.38 Pro Edition 64-bit Bublic Build
Optix Prime Acceleration: OptiX version 7.1.0

Benchmark Results
Total Rendering Time: 4 minutes 7.50 seconds
01800 iterations / 245.354s
Iteration Rate: (CUDA device 0 (GeForce RTX 2070 SUPER) + AMD Ryzen 7 2700x  (01800 iterations / 245.354s) = 7.336  iterations per second
Loading Time: 2.146 s