Folding Forum

Ability of score points rating of the the same hardware type only

Hi!

As I know the hardware used to folding can be video cards and CPUs.

I fold with CPU only (it is fanless mini pc). All points I got originated from CPU folding. In my folding condition/mode I have near 15K per day.
Some users has typical modern video card and made 600K points per day. So I need 40 days to fold, some others need only 1 day to fold and points are the same. But our efforts there is a huuuge gap. Rift. Canyon.

I wonder how competitive does my HW and chosen folding mode with other's CPU only originated points comparing of another users with their points got from the same type (CPU hardware) only.

Is it able to compare the user scores of the same hardware type between each other?

If to compare overall points got from any hardware I fell that I compare efforts of my mini PC to PCs of others like efforts of scooter to racing cars.
Effectiveness likes GPU so much, but "Grammy Award" of efforts itself of constantly folding has a CPU only hardware.

Are there total score metrics to compare points got of the the same hardware type (not theoretical PPD, not practical PPD of best day or any single day, but overall efforts/effectiveness/total score got as folding results in average day/period of time) between the same origin: (inside of group of HW with CPU only scored points) and in independent way (inside of HW of GPU only scored points)?

If it is not exists for now, please discuss to add it to plans of further development.

For example you can introduce HW attribute in new client API of result submitting procedure: "CPU", "GPU", ...
And in server stats GUI let the users to switch HW scored from sources: overall/total, CPU only, GPU only, ...

I feel need of such HW-type score splitting/fragmentation. That score filtering/grouping functionality is the way to compare efforts itself inside of the same HW type group: (a mouse with other mouses) and (a cheetah with other cheetahs) without mixing them together.
To show effectiveness/score within the same HW type.

Thank you!

Get a threadripper. They crunch about as fast as a GPU.

You are going to get wildly different results.

At one point, I had 55 PCs doing CPU folding on my account, I am down to 8 now. While none of my PCs can make 15k per day, in total they exceed that.

Some folks only fold during a small part of the day, others fold 24/7.

There are 3 general types of CPUs folding. CPUs between 2001 and 2012 may be using SSE2 to solve 4 operations at once per thread.I have both Pentiums and Core 2 Duos doing SSE2.

Intel CPUs since 2012 use AVX-256 to do 8 operations per thread at once. AVX offers smarter operations, so they are more than twice as effective as SSE2.

AMD CPUs from 2012 to the New Zen2 (3000 series desktops) Also use AVX but more slowly so they also do 4 operations per thread. Zen2 is like Intel, 8 operations at once.

Should you want to feel more equal to GPUs, I can suggest a Threadripper 3990x running Linux, which has 128 threads each doing 8 operations at once. It will be very competitive with GPUs.

https://www.amd.com/en/products/cpu/amd ... pper-3990x

It may do 1 million points per day

While you can request any enhancement you want, the researchers tend to prioritize those which do more science.

wezh wrote:...Is it able to compare the user scores of the same hardware type between each other?...

There was FAHBench (viewtopic.php?f=38&t=24225) but it wasn't updated to FahCore_22. However, they are plans to do it once FahCore_22 version 0.0.6 is released, FahCore_22 CUDA version is released and then FAHBench is next. There's no ETA.

wezh wrote:...If it is not exists for now, please discuss to add it to plans of further development....

Historically, one of the pinch point has been funding for the project which means that there was a single developer who has looked after most of the code base for F@H. Thus, the most bang for bucks for science was chosen. They are plans to make certain aspects of the code base open source to allow the F@H Community to contribute to it freeing up the developer to focus on other aspects. There are open issues for the development team to consider:
https://github.com/FoldingAtHome/fah-issues/issues/1479
https://github.com/FoldingAtHome/fah-issues/issues/1504