New RTX3xxx cards

[ap1978]

Ordered at launch. I'll probably get it late October or some time in November.

[foldy]

Heard the RTX 3080 founders edition gets loud under full load. Custom boards from other vendors maybe are more silent. For folding only these RTX 3000 GPUs look good although gamers may wait until some RTX 3080 ti with 16GB+ VRAM arrive.

[bruce]

https://www.youtube.com/watch?v=nJBggUfYozY

very good video explaining how the capability of a core and therefore what can be called a core has changed since Fermi.
...
I can't wait to see if this is actually beneficial and whether FAH can make use of the immensely wide chip that Ampere appears to be.

In the early days, FAH's throughput was always limited mostly by the limits of the FP32 processing. A few hundred Fermi cores did whatever they could do with the work given to them. The core count was always the limitation.

Now the limits that are encountered are due to the simplicity of the atomic struture of the assigned proteins. Modern GPUs now have exceptionally large GFLOPS rating but now performance is limited by how fast the data can get to the cores. Small proteins can't use all those cores all of the time.

Unfortunately a lot of the current COVID proteins are small and won't be effectively loading all the cores. You might get better benchmrking from some of the larger proteins.

[ajm]

Watching the different benchmarks that now start to come out, it looks like the 3080 is about 20% above the 2080ti in raw performance, which is not especially exciting considering its wattage. There's of course a very significant progress in price/performance, hence the huge success of this launch, and for gamers RTX & DLSS 2.0 will make a much bigger difference, eventually, but... well. The 3090 should reach some 50% more performance than the 2080ti, which starts to be interesting, but all in all, this 3xxx Series looks more and more like a souped-up 2xxx Series, rather than a real technology leap.

[HaloJones]

Most of the tests that have been done are related to games performance. If you look for tests related to compute - blender, vray, optix - the numbers are much higher.

[ajm]

There indeed are some nice results to show in this general realm: https://compubench.com/compare.jsp?benc ... TX+2080+Ti

But OpenCL... 20% https://hothardware.com/reviews/nvidia- ... iew?page=2

[ap1978]

"The RTX 3090 allegedly scores 249153 points in the CUDA benchmark, compared to 215361 of RTX 3080. That is an increase of +15.7%."

Source: https://videocardz.com/newz/even-more-nvidia-geforce-rtx-3090-benchmarks-leak

[MeeLee]

Well... I dabbled at trying to get a 3080 ordered this morning. Yeah, that didn't work out with the websites crashing and apparently low supply... o-well TIme is on my side for getting one at some point.

Nvidia GPUs be like:

[Mxyzptlk]

Well... I dabbled at trying to get a 3080 ordered this morning. Yeah, that didn't work out with the websites crashing and apparently low supply... o-well TIme is on my side for getting one at some point.

Nvidia GPUs be like:

Pretty much sums it yo

[MeeLee]

In terms of performance, I am on the fence on buying an RTX3080.
I have a feeling that the difference between the 3080 and 3090 is going to be very minimal in folding, due to the smaller atom WUs.
But even if the WUs were optimized for both GPUs, the difference is about 20-25% between cards (although that 20% scales in PPD, resulting in perhaps 50-75% higher PPD max).

I'll need to read up more on TDPs, as some of the reference design 3080s can go as high as 350W, and the 3090s surpass 400W.
And the only report of a 180W WU we have here on the forum, might have been one that wasn't fully utilizing the 3080.

It seems in terms of performance/watt we're seeing a lower improvement in efficiency than I hoped the RTX 2000 series GPUs would do.
Though not a lot of research has been done about power capping them.

[kiore]

Although the promise of the RTX 3090 seems incredible, the current price I see is off putting and the power consumption and likely heat generation look like a serious headache for users. I suspect the later iterations of the 3080, ti or whatever they call the higher binned chips may be the sweet spot once again.
This time I also suspect that F@H is better prepared and poised to optimize this leap in performance more than I have seen previously and the time lag that is only reasonable is looking much shorter this time. Eternally optimistic I think the new core versions and soon to be released new driver versions will drag the most out of current cards and hopefully maximize this new generation as it emerges.

[MeeLee]

Although the promise of the RTX 3090 seems incredible, the current price I see is off putting and the power consumption and likely heat generation look like a serious headache for users. I suspect the later iterations of the 3080, ti or whatever they call the higher binned chips may be the sweet spot once again.
This time I also suspect that F@H is better prepared and poised to optimize this leap in performance more than I have seen previously and the time lag that is only reasonable is looking much shorter this time. Eternally optimistic I think the new core versions and soon to be released new driver versions will drag the most out of current cards and hopefully maximize this new generation as it emerges.

What I like about the reference designs of the 3080 and 3090, is that they send most of their heat out the back of the PC case.
Unlike the 3Fan design, which generally works well under 200W TDP, many of the 3rd party designs will not have a sufficient cooler for the GPU.
The 3090 though, is a monster compared to anything before. So the fans are slightly larger.
But the 3080's reference cooling is actually superior to a 2080's cooling.
Another thing Jayz2cents did, was put a case fan in the center of these GPUs.
Only necessary for when you'd want to overclock them. I don't think folding will see 400W on the 3090, but could if it was overclocked.

From what I understood from the design, was that the 2 fans cool the GPU. While the passive heat sinks in the middle cool the VRAM.
When overclocking the VRAM, they mentioned it's best to cool the GPU with an extra fan.
Doesn't have to be a big fan, just an 80 to 120mm fan, operating at 5V (~1100-1400rpm, 1Watt), connected to maybe a case fan header on the motherboard.
That fan would allow for higher performance, due to cooling the design by an additional few watts, also clearing any extracted heat hovering around the GPU, from reentering the GPU fans.

[gunnarre]

Some of the rear fan air does pass the middle heat sinks. Also, when you have good airflow in the case, that will move heat off those finns. More detailed airflow analysis here: https://www.youtube.com/watch?v=VVaGRtX80gI

[PantherX]

FYI, here's some data for comparison only, Linux with Project 11765, time for 1% in seconds:
73s - 1080Ti OpenCL
57s - 1080Ti CUDA
49s - 2080Ti OpenCL
39s - 2080Ti CUDA
36s - 3080 OpenCL
31s - 3080 CUDA
30s - 3090 OpenCL
26s - 3090 CUDA

[MeeLee]

FYI, here's some data for comparison only, Linux with Project 11765, time for 1% in seconds:
73s - 1080Ti OpenCL
57s - 1080Ti CUDA
49s - 2080Ti OpenCL
39s - 2080Ti CUDA
36s - 3080 OpenCL
31s - 3080 CUDA
30s - 3090 OpenCL
26s - 3090 CUDA

Cuda doesn't seem to scale very well.
This is when 2 identical GPUs plugged in, and load sharing between them via CUDA?
Seems to work well for the GTX series, but the higher the GPU, the lower the performance gains, it seems.