work unit benchmark testing

[whocrazy]

Hi there.
I'm getting a new computer soon.
it's an Intel - BX8071513700F
Intel Core i7 13700F (Base: 2.1GHz, Turbo: 5.2GHz / 30MB / LGA1700 / 16 Core / 24 Thread / 65 W / Raptor Lake)
The current computer I'm folding on is about 15 years old.
It's an AMD Phenom(tm) II X4 955 Processor
CPU ID: AuthenticAMD Family 16 Model 4 Stepping 2
CPUs: 4
I am currently folding Project: 16977 (Run 5, Clone 1022, Gen 508)
I am really interested if there is a way to test the same work unit on different machines to see how fast it goes, or if anyone can tell me if I describe a machine's specs on here, if anyone can tell me how fast a certain WU would run on it.
How fast would project 16977 work units run on the new machine VS. the old one?
This current work unit would take about a day on the old Phenom machine.
Thanks.

[PaulTV]

Hola,

Basically, FAH isn't a benchmarking tool.

There is no way to get a the exact same job on another machine. Jobs are assigned by the central server based on machine specs and optionally preferences for a particular cause. The preferences may help to get jobs from the same project, but you would get a different run/clcone/gen. As your old and new machines will have quite different specs, it will be hard to find a GPU based project that would run on both the old and new machine. If you're mostly looking at CPU based folding, setting the same number of CPU threads (in FAH 7 language: cores) will help.

Some people try to use FAH for tweaking hardware. As FAH really needs stable hardware to calculate accurate results, using FAH to test over-/underclocking or -volting is discouraged. Example: an unstable GPU may show artifacts / glitches in a video game, which for a game is annoying, but for FAH it would result in possibly hard to detect errors in the FAH calculations, and that'd harm the science.

On https://folding.lar.systems/projects/ you can find PPD results for various projects and hardware to compare. However, those numbers greatly depend on hardware models (e.g. factory overclocked or not), settings, OS, other uses of the computer, internet speed even, etc, etc, so your mileage may vary, as they say.

[bikeaddict]

Intel's 12th and 13th gen models that are hybrid big.LITTLE design with performance and efficiency cores are not ideal for FAH. Because of the way the GROMACS library works, a WU will run only as fast as the slowest core.

It's possible to use tools like Process Lasso (Windows) or taskset (Linux) to set affinity so the folding process only runs on the performance cores. But performance is still likely to be slower and less power efficient than the newest Ryzen processors. I think a Ryzen 7900X or 7900 would be a better choice than an Intel i7. My i7-13700K does 450-650K PPD on Linux with taskset to use just p-cores and uses about 200-240W of power.

[whocrazy]

Hi Paul TV
I'm not planning on doing any overclocking, or work unit and CPU tampering or anything like that, I was just hoping that, because I'll be using a newer machine to do the folding, that I could do the same project and return each work unit quicker, also, I won't be folding GPU units just yet, only CPU ones.

[JimboPalmer]

Welcome to Folding@Home!

The servers will sent different complexity Work Units based on the capabilities of your CPU. (Number of usable threads and talent of its SIMD math unit, Last I knew SSE 2.1, AVX, and AVX2 were supported) I am guessing your new CPU is 'better'.

For GPU folding, there is FAHBench. I am not aware of benchmark for CPU folding. http://fahbench.github.io/overview.html

[toTOW]

FAHBench can run on CPU platform ...

[whocrazy]

Hi. Can you please tell me how to tell fah how many cores I want it to use? and which option I should set on the commandline, or the config.sml.
Thanks