Folding Forum

Try to pause CPU slot and see if GPUs get more PPD. If true then you need to reduce CPU slots CPUs count.

It's also preferred to run Linux for folding.
My RTX 2080 Ti gets 2-2.2M PPD on PCIE 3.0 8x, but only 1.8M PPD at PCIE 3.0 1x (or PCIE 2.0 4x, same thing, from what I read).
If you're running windows, it's normal to see a few percent lower (1,6-1,8M PPD).

Try to run everything at PCIE 3.0 speeds.
Or, if you're running at PCIE 2.0 speeds, try getting at least an 8x speed slot.

foldy wrote:Try to pause CPU slot and see if GPUs get more PPD. If true then you need to reduce CPU slots CPUs count.

I've currently got it set up for 2 slots, 1x 16 and 1x 12 leaving 2x untouched but I will pause and see if it makes any difference! Do I need to do anything specific to "park" a CPU core? Other than leave it out of the FAH config?

MeeLee wrote:It's also preferred to run Linux for folding.
My RTX 2080 Ti gets 2-2.2M PPD on PCIE 3.0 8x, but only 1.8M PPD at PCIE 3.0 1x (or PCIE 2.0 4x, same thing, from what I read).
If you're running windows, it's normal to see a few percent lower (1,6-1,8M PPD).

Try to run everything at PCIE 3.0 speeds.
Or, if you're running at PCIE 2.0 speeds, try getting at least an 8x speed slot.

Everything is at PCIE 3.0 speeds but I haven't tried the Linux route yet.

Sac87 wrote: I've currently got it set up for 2 slots, 1x 16 and 1x 12 leaving 2x untouched but I will pause and see if it makes any difference! Do I need to do anything specific to "park" a CPU core? Other than leave it out of the FAH config?

Just don't assign a slot for the CPU.
If your CPU is running in hyper threading mode, there is a chance it might limit the speed of the most powerful GPUs (like the 2080 Tis), so I'd try to turn off HT in the bios, and see if PPD is any higher.

If not, and you want to maximize PPD count, you could assign a cpu slot taking a limited amount of threads.
The formula for windows is CPU thread count (eg 8) minus GPU count (eg: 2) minus Windows OS thread (1)
For multi core CPU counts beyond 6 cores, I'd assign an extra CPU core. I've noted that the folding happens a little better that way.

For Linux, it's a lot easier.
CPU - GPU count.
Eg: 8 CPU threads, 2 GPUs, that gives you access to a full 6 threads in Linux.

However, do know that folding on CPU is less efficient.

MeeLee wrote:

Sac87 wrote: I've currently got it set up for 2 slots, 1x 16 and 1x 12 leaving 2x untouched but I will pause and see if it makes any difference! Do I need to do anything specific to "park" a CPU core? Other than leave it out of the FAH config?

Just don't assign a slot for the CPU.
If your CPU is running in hyper threading mode, there is a chance it might limit the speed of the most powerful GPUs (like the 2080 Tis), so I'd try to turn off HT in the bios, and see if PPD is any higher.

If not, and you want to maximize PPD count, you could assign a cpu slot taking a limited amount of threads.
The formula for windows is CPU thread count (eg 8) minus GPU count (eg: 2) minus Windows OS thread (1)
For multi core CPU counts beyond 6 cores, I'd assign an extra CPU core. I've noted that the folding happens a little better that way.

For Linux, it's a lot easier.
CPU - GPU count.
Eg: 8 CPU threads, 2 GPUs, that gives you access to a full 6 threads in Linux.

However, do know that folding on CPU is less efficient.

Interesting idea... I'll disable Hyperthreading and see if anything changes!
I am definitely interested in maximizing my PPD so I will probably try and table math your formula out. I have 32 cores total.

I woke up this morning and one GPU was almost at 1.9 Mil so we're making some progress!

I did not see any difference by turning off Thread Optimization or not folding with the CPUs. 1.5 Mil and 1.4 Mil when I woke up this morning.

Sac87 wrote:I did not see any difference by turning off Thread Optimization or not folding with the CPUs. 1.5 Mil and 1.4 Mil when I woke up this morning.

That indicates your CPUs are fast enough to feed your GPUs, as configured by F@H. Might as well let them fold.

You'll only see the fastest times and highest points under linux, that will add roughly 10% to PPD

I may have to learn how to do Linux then. Can I do it under VM software?

Also, does it hurt having the cards SLI enabled? Or does it not change anything?

SLI makes no difference. Not sure about running the gpu's under VM software, you would need to be able to pass through the drivers and i'm not sure you can do that with regular geforce cards.

I have seen reports of GPU drivers not working when running under some VM software. The guest operating system doesn't receive a dedicated device but rather a simulated GPU driver. I might depend on which VM software you happen to use.

I recommend you install the appropriate drivers in whatever native OS is being booted and that you run FAH in that OS.

If you discover what does/does_not work, please report back.

Aside from dual booting, there's always an option to install a cheap 32-64 GB SSD in your system, and boot from UEFI with F11, to your stock OS disk, or the Linux disk.

After several hours of manually overclocking, I have managed to break 2Mil on each card. I had a feeling the auto overclock was leaving a lot on the table. Turns out it was!

Mission Accomplished! Thanks so much to everyone who helped tweak my system. Happy camper!

Yes, 2 mil is with an overclock. Especially if you limit the power by almost 100 Watts.

Sac87 wrote:After several hours of manually overclocking, I have managed to break 2Mil on each card. I had a feeling the auto overclock was leaving a lot on the table. Turns out it was!

Mission Accomplished! Thanks so much to everyone who helped tweak my system. Happy camper!

Expect to see large variations between projects, on my RTX 2080ti I see a range of 1.6 to 2.2 MPPD. A sudden drop off may just be a change in project type/ size etc rather than a system issue.