1080ti efficiency sweet spot?

[Endgame124]

Has anyone done any analysis on where the power limit sweet spot is on a 1080ti? I rearranged some systems (read decommissioned a q9650 and A10-5800K) and moved a 1660 Super to my 2700X host with a titanium power supply. Unfortunately, that power supply is only 650 watt, and was running over 650 watts with both Video cards running F@H at 100% and with 14 Rosetta threads. I set both cards to 65% and dropped to 13 Rosetta threads which dropped the power to around 540 watt, but now I’m wondering if 65% is the right amount. I saw a drop of about 400k (2.4 mill to 2 mil) ppd after making the change.

Should I cut more power to the cards? Bump them back up a little? Thoughts?

 

[Assimilator1]

Was the original 650w draw from the wall? If so, you were below the PSUs max (DC) output rated power, although not by much! (at ~611w, assuming 94% eff at 100% load for 230v, are you on 115v though?).

 

[biodoc]

On F@H, I run my 1080 Ti's at 180 watts which is 72% of 250 watts. I don't have the data recorded, but I monitored TPF while reducing the power limit on the GPU by 10% increments.

I'm surprised your power draw at the wall is so high on that system. At 100% power, the GPUs should be drawing a total of 375 watts (1080 Ti: 250 watts, 1660 super: 125 watts). At 65% power, the GPUs should be drawing around 243 watts. At 65% power to the GPUs, power draw at the wall is 540 watts which means the CPU/MB is drawing 540-243 or around 300 watts. Do you know the power draw at the wall when you suspend FAH on the GPUs?

 

[Endgame124]

Was the original 650w draw from the wall? If so, you were below the PSUs max (DC) output rated power, although not by much! (at ~611w, assuming 94% eff at 100% load for 230v, are you on 115v though?).

I'm on 115v, and the peak I saw at the wall was 657. I didn't think about the fact the wall power would still put me below the max of the power supply, but you're right, I was under the limit

On F@H, I run my 1080 Ti's at 180 watts which is 72% of 250 watts. I don't have the data recorded, but I monitored TPF while reducing the power limit on the GPU by 10% increments.

I'm surprised your power draw at the wall is so high on that system. At 100% power, the GPUs should be drawing a total of 375 watts (1080 Ti: 250 watts, 1660 super: 125 watts). At 65% power, the GPUs should be drawing around 243 watts. At 65% power to the GPUs, power draw at the wall is 540 watts which means the CPU/MB is drawing 540-243 or around 300 watts. Do you know the power draw at the wall when you suspend FAH on the GPUs?

when I get a few minutes, I'll try suspending the GPUs and see what the power looks like. Beyond the GPUs and CPU, I've got 32GB ram, a corsair H150i, 9x 120mm fans, 2x 1TB Samsung 970pros, and 1x 2TB samsung 860 Evo ssd.

 

[biodoc]

If you're interested in reducing the power draw of the 2700x, lowering the PPT works well for me on Zen2. Depending on the processor and MB bios, it may or may not be possible.

Lowering PPT.

Good tutorial for reducing ryzen power consumption.

 

[Endgame124]

If you're interested in reducing the power draw of the 2700x, lowering the PPT works well for me on Zen2. Depending on the processor and MB bios, it may or may not be possible.

Lowering PPT.

Good tutorial for reducing ryzen power consumption.

I didn't think changing the PPT was an option for the 2000 series of chips. I'll look into it, as any further power reduction puts me further into the sweet spot on the PSU.

Edit - a very quick google search on the Asus Strix X470 has me thinking that PPT is available, at least if I update the bios (I'm likely still running the launch bios). Additionally, the chip is certainly running the boost clock as far as it will go all core stock (4.1ghz), as I'm seeing clock speed between 4.05 and 4.1 ghz watching cpu-x, so I'm almost certainly hitting 100W or more on the CPU alone. Something to play with when I have time, which is in short supply these days.

 

[StefanR5R]

I'm seeing clock speed between 4.05 and 4.1 ghz watching cpu-x,

Perhaps something like 3.5 GHz would be a better compromise between wasting energy in the CPU, and feeding the GPUs appropriately.

 

[Endgame124]

Power at the wall with F@H suspended peaks at 240 watts, with dips into the 220 range. Suspending F@H actually pushes the clockspeed up on the CPUs to just under 4.2Ghz so the real CPU system draw with F@H running is probably a little lower. My current bios rev doesn't have any PPT settings (its from 2018), and there is a new bios from 4 days ago - I will update in the not too distant future.

I'll probably leave it at this speed until I can set aside some time to flash the bios and play with any new features bios wise.

The below article says the 1080 is roughly at its best efficiency at 60% power, so 65% on my 1080ti is probably close without spending any real effort dialing it in.

NVIDIA GEFORCE GTX 1080 Folding@Home Review (Part 2)

Welcome back. In the last article, I found that the GeForce GTX 1080 is an excellent graphics card for contributing to Stanford University’s charitable distributed computing project Folding@H…

greenfoldingathome.com

 

[StefanR5R]

he below article says the 1080 is roughly at its best efficiency at 60% power,

Good article.

I wonder if the sweet spot shifts when folding under Linux instead of Windows.

One more factor to keep in mind is cooling: Good cooling allows the card BIOS to operate the GPU at a slightly higher ratio of clock to Voltage.

 

[Assimilator1]

I'm on 115v, and the peak I saw at the wall was 657. I didn't think about the fact the wall power would still put me below the max of the power supply, but you're right, I was under the limit

Ah ok, so in which case your rig is pulling about 591w (peak), although it's likely to be a little less, as that figure is for 100% load, when 591w is a 91% load. You could probably make a more accurate estimation if you can find it's power efficiency curve.