Anyone else running Core Ultra cpu folding?

[notyettoday]

Hello all,

I was surprised to see there wasn't any discussion about the Intel Core Ultra cpus. I suppose with GPU folding being much more efficient that makes sense. I decided to grab a 265K with some 7600mhz ram, So far I'm observing 550k-800k estimates in the v7 client on 19 cores, which is close to my 7950x rig, impressive. Power usage is hovering in the 160w area on average according to HWinfo. I'd be curious to hear other's results!

[Joe_H]

People have not been commenting here much beyond gripes about Intel processors recently. The big-little configuration is not well utilized by the CPU folding core code. The Windows scheduler also has its quirks about dealing with big-little, often scheduling the folding threads on the efficiency cores instead of the performance cores. Utilizing the efficiency cores tends to slow down WU processing.

It would be interesting to see what processing on just the 8 performance cores gets for PPD. Under Windows that may take using a third party utility such as Process Lasso to get the CPU folding core threads assigned to the performance cores.

[Xeon_phi_guy]

I tuned mine (200s boost spec) and OC’ed the E cores. I’m doing 1.588m ppd on average so far. It’s a machine.

[Mactin]

It would be interesting to see what processing on just the 8 performance cores gets for PPD. Under Windows that may take using a third party utility such as Process Lasso to get the CPU folding core threads assigned to the performance cores.

I have the same comment with my M4 Mac, how to fold on the 4 performance core and not the 6 efficiency ?
My own "testing" gives better results with 10 than 4, so I do it with 10.

[muziqaz]

Intel, as Joe, said, have not been doing well with their big.shittle architecture in folding.
AMD is ruling this space.
9950x on Linux averages 1.8m PPD, but goes up to 3.3m
9950x3d on Windows averages 1.6m PPD, but goes up to 3.1m
7950x3d on Windows averages 1.2m PPD
5950x on Windows is averaging 614k, but goes up to 1.4m
5800x3d on Linux averages 642k, but goes up to 1m

[Xeon_phi_guy]

Intel, as Joe, said, have not been doing well with their big.shittle architecture in folding.
AMD is ruling this space.
9950x on Linux averages 1.8m PPD, but goes up to 3.3m
9950x3d on Windows averages 1.6m PPD, but goes up to 3.1m
7950x3d on Windows averages 1.2m PPD
5950x on Windows is averaging 614k, but goes up to 1.4m
5800x3d on Linux averages 642k, but goes up to 1m

Running a CPU WU right now on my 265k machine. 1,512,349PPD on project 12475. 2:48TPF on Windows 11

[Joe_H]

It would be interesting to see what processing on just the 8 performance cores gets for PPD. Under Windows that may take using a third party utility such as Process Lasso to get the CPU folding core threads assigned to the performance cores.

I have the same comment with my M4 Mac, how to fold on the 4 performance core and not the 6 efficiency ?
My own "testing" gives better results with 10 than 4, so I do it with 10.

As best as I can tell, the macOS scheduler will use the performance cores for folding if you set the thread count to match the number of them available. With as many as 6 economy cores running, that may be enough to make up for the performance core threads waiting for the other threads.

Personally on my M1 Max system I set the thread count to 6 (out of 8 performance cores) and leave it at that. I might get a bit more performance out of a higher setting, but that leaves it snappy for my interactive use.

[muziqaz]

Intel, as Joe, said, have not been doing well with their big.shittle architecture in folding.
AMD is ruling this space.
9950x on Linux averages 1.8m PPD, but goes up to 3.3m
9950x3d on Windows averages 1.6m PPD, but goes up to 3.1m
7950x3d on Windows averages 1.2m PPD
5950x on Windows is averaging 614k, but goes up to 1.4m
5800x3d on Linux averages 642k, but goes up to 1m

Running a CPU WU right now on my 265k machine. 1,512,349PPD on project 12475. 2:48TPF on Windows 11

This is the family of projects where my "up to" numbers come in

[Xeon_phi_guy]

Intel, as Joe, said, have not been doing well with their big.shittle architecture in folding.
AMD is ruling this space.
9950x on Linux averages 1.8m PPD, but goes up to 3.3m
9950x3d on Windows averages 1.6m PPD, but goes up to 3.1m
7950x3d on Windows averages 1.2m PPD
5950x on Windows is averaging 614k, but goes up to 1.4m
5800x3d on Linux averages 642k, but goes up to 1m

Running a CPU WU right now on my 265k machine. 1,512,349PPD on project 12475. 2:48TPF on Windows 11

This is the family of projects where my "up to" numbers come in

Good data point. I was using the LAR database, on the same project (ending in 475) my CPU is ~5% worse than a 9950X3D. You weren’t kidding about the “up to”, some of them absolutely scream on Team red.

[toTOW]

I tuned mine (200s boost spec) and OC’ed the E cores. I’m doing 1.588m ppd on average so far. It’s a machine.

Set the number of folding thread to match the number of your performance threads. Use Process Lasso to automatically park Fahcore_a8 and Fahcore_a9 to performance threads.

Forget about efficiency cores, for FAH, they are a waste of silicium.

[muziqaz]

I tuned mine (200s boost spec) and OC’ed the E cores. I’m doing 1.588m ppd on average so far. It’s a machine.

Set the number of folding thread to match the number of your performance threads. Use Process Lasso to automatically park Fahcore_a8 and Fahcore_a9 to performance threads.

Forget about efficiency cores, for FAH, they are a waste of silicium.

The problem is that Intel is reducing P core count more and more. Sometimes it is better to fold on E cores, since there are many more of them.