RTX 4090 is getting 26M PPD

[red454]

I just reviewed the RTX 4090 and I am seeing as high as 26 million PPD so far. Folding starts at the 31:15 mark.

Video review link https://youtu.be/mzgfWSEJzk8

[PaulTV]

Looking at your review, and the others that came online today, that's an insanely fast card. I'm so tempted to find myself an excuse to spend the (probably) two grand in euros... hopefully reason will win though

[red454]

Right - and 26 is the top for me so far. They way the WU bounce around, I may see higher. Have you seen anyone report what I am seeing, or higher?

[PaulTV]

Not specifically for F@h, but when looking at the raw power as tested by Jay, Steve and all they other youtubers,,, With the CUDA cores it's not surprising to see wild fluctuations on the PPD, even more than the 30 series. I imagine a card like the 4090 just has to fold for the alzheimers projects in order to utilize it properly.

[Breach]

Thanks for this! I am currently folding on 1x OC'ed 3090 and 1x OC'ed 3080 (ca. 15m PPD, but depends), and was thinking 4090 would be roughly about the same - apparently not so

@PaulTV, yeah, 4090 FE is 1.959 EUR in the EU... Coming out tomorrow.

[red454]

Not specifically for F@h, but when looking at the raw power as tested by Jay, Steve and all they other youtubers,,, With the CUDA cores it's not surprising to see wild fluctuations on the PPD, even more than the 30 series. I imagine a card like the 4090 just has to fold for the alzheimers projects in order to utilize it properly.

I'll focus on some Alzheimer's projects and see what PPD I get...

[BobWilliams757]

I was going to order a half a dozen of them, but if they aren't pulling at least 30M PPD why bother?

That's some serious folding power in that GPU. It's amazing how far things have come in a short period of time. I have to wonder when it will slow down on the development side of things.


As I said in other threads, I wouldn't be surprised if we have reached a point where the variations in PPD rises. With so many cores there will be more work units with too few atoms to really get the most out of the cards. We were already seeing this somewhat with some of the 3xxx cards, and I expect it to increase.

It would be interesting to see the high and low PPD's of the cards.

[Rusty2644]

26MPPD

I'm interested in how the 4090 performs with F@H. Please keep us updated!

I'm also curious what the most PPD per gpu $ is in the current environment w/ WU atom bottlenecks, POS Ethereum, etc...

Think I should sell some 30xx's to go get a 4090? Should I continue with my current set up this winter?
kWh is still a concern at house, even though we have solar.

[red454]

I'll report back with PPD info and power usage.

[BobWilliams757]

Some 4090 samples are showing up on the LAR Systems website statistics....

https://folding.lar.systems/gpu_ppd/bra ... e_rtx_4090

It could just be a single user at this point, as the numbers aren't rising up really quickly yet. Sometimes with that site you have to take things with a grain of salt, as you can't be sure what settings, other computer use, etc are taking place while folding.

But it looks like it would be fairly hard to slow down a 4090 even if you tried.

[red454]

I am seeing between 16m and 22m PPD now. Haven't seen the 26m since the first couple of days on the 4090.

[Alex_Atkin]

Someone on Linus Tech Tips forum did a quick test:
PL 100% = 370W - 23 MIL PPD Core Clock 2775mhz
PL 90% = 350W - 22 MIL PPD Core Clock 2750 - 2775mhz (wattage fluctuation high)
PL 80% = 350W - 22 MIL PPD Core Clock 2700 - 2750mhz (wattage fluctuation high)
PL 70% = 320W - 21 MIL PPD Core Clock 2500 - 2700mhz
PL 60% = 270W - 21 MIL PPD Core Clock 2400 - 2500mhz
PL 50% = 220W - 19 MIL PPD Core Clock 2100 - 2200mhz
PL 40% = 180W - 13 MIL PPD Core Clock 1300 - 1500mhz
PL 30% = 130w - 9 MIL PPD Core Clock 800 - 900mhz
https://linustechtips.com/topic/1458969 ... t=15615993

Will be interesting to see how this scales to 4080.

[red454]

I have a 4080 coming soon...

[BobWilliams757]

That's some fairly amazing efficiency at lower power limits. Even more amazing that it can actually run reliably as such low power levels.

[jchang6]

it's too bad that we do not have PPD for both fixed project (true) and specific unit (no control?) at each of several fixed frequencies instead of Power (W). The expectation is that performance per unit frequency falls as frequency increases even though performance does increase, just somewhat less than linear.
This is the memory latency effect. Latency between CPU and main/system memory is probably 65-75ns (depending). Latency between GPU and its memory is usually higher, 100ns?, because the special memory used is designed for high bandwidth, not lower latency.
FaH only uses a small fraction of the GPU memory. Its too bad we do not have a special GPU card with smaller low latency memory