Windows 10 vs Windows 11 PPD?

[the garynator]

Still trying to figure out if the MSI RTX 4070 Super Gaming X Slims are magic or if there's something else effecting PPD on my system. We've pretty much eliminated PCIE bandwidth as the issue as both his MSI cards do the same ppd even with one in a PCIE 4.0 x4 slot, and mine is currently running in a PCIE 3.0 x8 slot. @KeeperOfTheButch 's 2 MSIs are putting out 13+ Million ppd each consistently whereas the Asus RTX 4070 Super Dual in my system only does about 11.5m max (might have seen 12m on certain wu's with overclock), and regularly gets around 10-11m, and as low as 6-7m on bad wus. Thing is, temps are comparable and while the MSIs have higher core clock speed, I seem to be able to overclock more so can run similar gpu clocks.

We ran a PNY 2 slot 4070 super in the new dell xeon workstation he picked up for dedicated folding and while it wasn't quite as high of ppd, it was quite close (~12.9 vs ~13.5m) and he just switched both msi cards into it to see if they still do the same PPD as they do in his main rig (new ryzen system). I figure this test will put the PCIe 4.0 x4 vs PCIe 3.0 x8 question (possible latency differences or something) to bed if they perform the same as in his main rig.

Only other thing I can think of that would be nerfing my ppd is that I'm still running Windows 10 on my machine and he's running Windows 11 on both of his. Has anyone compared PPD output on newer cards between these two OS's? I really would rather not upgrade to Windows 11 if I don't have to lol.

Saw that the default Win 10 scheduler might be causing performance issues (which would make sense if Win 11 is better), but it's quite an old thread with older cards so might give this a try on my machine.

https://www.reddit.com/r/Folding/comments/k2r01j/520_folding_speed_boost_for_free

Any thoughts, suggestions, or ideas anyone?

Post magically merged: May 20, 2024

One other theory I'm currently testing is that maybe I'm running up against TDP and it's pulling clock speed. @KeeperOfTheButch isn't running any memory overclock on his cards. I just ran a TDP average in gpuz on my card with +125 on gpu and 600 on memory vs just 125 on gpu and it appears that TDP average dropped by like 10w. So I'm going to test with just the max gpu overclock I can run stable and see what kind of difference that makes. Still don't think this alone would account for 2-3 million PPD difference though, but might be part of it.

 

[dfonda]

So my Win 10 does better. It isn't used for anything but fah. 2 4070 Ti's.

The Win 11 (4070 Ti Super and a 4070 Super) has faster GPU's but it does get used by my wife otherwise if not used they pretty much match 25 K per PC.

 

[the garynator]

So my Win 10 does better. It isn't used for anything but fah. 2 4070 Ti's.

The Win 11 (4070 Ti Super and a 4070 Super) has faster GPU's but it does get used by my wife otherwise if not used they pretty much match 25 K per PC.

interesting. I tried turning on hardware accelerated scheduling and hasn't seemed to make a difference other than windows correctly reporting gpu utilization lol.

on that note, preliminary results show the PNY card is now outperforming the MSI cards now that it's in @KeeperOfTheButch 's main PC. it's doing ~14m ppd and the MSI cards are doing ~12.5 and ~12m in the Xeon Dell. Same overclocks. The PNY on the same projects as the MSI cards was outperforming by about the same amount. I wonder if F@H is sending out WUs based on system specs....hard to believe that PCIe gen would effect the ppd by such a significant amount even when bandwidth should be the same (i.e. pcie 3.0 x8 vs pcie 4.0 x4 or pcie 3.0 x16 vs pcie 4.0 x8).

 

[Flamethrower1972]

I have windows 11 and I for the moment have only one rig I think when my other is built, I might run that gpu at night only and game during the day.

 

[the garynator]

FYI, got some wus over the last few days that we're putting out closer to what we have been seeing on @KeeperOfTheButch 's rigs. I did a couple changes that seem to have helped. I think my card was bumping up against tdp limits. I pulled the memory overclock down to 0, upped the GPU clock to +150 and turned the fans up a bit to bring temps down a bit lower. Saw 13.5m on certain wus for the first time, and seems to be doing 12-12.5m on about half of the other wus, and 10.5-11.5 on the rest.

Working on a way of getting log file data parsed out so I can compare apples to apples with the other machines, but it's probably going to be slow going as I'm doing it manually at the moment and idk when/if I'll get time to automate it.

 

[Flamethrower1972]

FYI, got some wus over the last few days that we're putting out closer to what we have been seeing on @KeeperOfTheButch 's rigs. I did a couple changes that seem to have helped. I think my card was bumping up against tdp limits. I pulled the memory overclock down to 0, upped the GPU clock to +150 and turned the fans up a bit to bring temps down a bit lower. Saw 13.5m on certain wus for the first time, and seems to be doing 12-12.5m on about half of the other wus, and 10.5-11.5 on the rest.

Working on a way of getting log file data parsed out so I can compare apples to apples with the other machines, but it's probably going to be slow going as I'm doing it manually at the moment and idk when/if I'll get time to automate it.

I can't wait what numbers you come up with.

 

[HayesK]

all my gpu have memory underclocked, -500 on the 4000 series and -400 on the 3000 series. the older cards have-200 or -100 on the memory. All the gpu frequency have overclock and power limits set. All folding hosts run linux, mostly Ubuntu 20.04, a few Ubuntu 22.04. I use HFM and FAH Advanced Control running on windows computers to monitor the linux GPU.

 

[the garynator]

all my gpu have memory underclocked, -500 on the 4000 series and -400 on the 3000 series. the older cards have-200 or -100 on the memory. All the gpu frequency have overclock and power limits set. All folding hosts run linux, mostly Ubuntu 20.04, a few Ubuntu 22.04. I use HFM and FAH Advanced Control running on windows computers to monitor the linux GPU.

Thanks for chiming in, that is very helpful!!

 

[freeagent]

My 4070Ti with a tune will do ~16m PPD on Zen 3. +150 and +1466 mem, so core speed at ~3GHz.