7 replies to this topic

[Johan]

Hi all,

Which version of Folding@Home would you recommend for my PC, to get the most points per day out of it?

Possibly relevant information:
• CPU: AMD Athlon 64 3200+, overclocked to 2.46 GHz
• RAM: 1.5 GB PC3200 DDR-1
• Video card: Geforce 7300 GS
• OS: Windows XP Pro (32-bit)

Thanks.

[dnamechanic]

Which version of Folding@Home would you recommend for my PC, to get the most points per day out of it?

...

Thanks for your interest, and for providing your system info, it is relelant.

Probably, your best option is the Windows Uniprocessor (classic) Client Installation.

FYI, a couple of high performace clients and their requirements:

- To fold on an NVIDIA GPU, it must be CUDA enabled.

Didn't see your card in the list.

- To fold SMP work units requires a dual-core or multiple-core processor.

[Johan]

Thanks a lot for the info. I will be buying a new computer shortly, and will certainly look for a CUDA-enabled video card. Perhaps the Geforce GTX 260 (896MB) or the GTS 250 (1GB) would be a good choice? Not too expensive either.

Edited by Johan, 13 July 2009 - 09:25 PM.

[dnamechanic]

... Perhaps the Geforce GTX 260 (896MB) or the GTS 250 (1GB) would be a good choice? Not too expensive either.

Yes, good choices.

Currently, for folding memory (GDDR) is not an issue.

The largest GDDR RAM usage that I have seen to date by folding work units just slightly exceeds 100 Mbytes. In addition, GDDR3 memory consumes some electrical power (thus releasing heat).

Of course there are applications (or will be applications that may benefit from increased GDDR).

In future Stanford may gen up work units that require increased GDDR RAM .

[Johan]

Is there any advantage in terms of points per day when running two clients simultaneously on a single-core CPU, for example the systray client and the console client, compared to running only one of them?

Edited by Johan, 14 July 2009 - 09:10 PM.

[dnamechanic]

Is there any advantage in terms of points per day when running two clients simultaneously on a single-core CPU, for example the systray client and the console client, compared to running only one of them?

The short answer is: no advantage.

Since the above response provides minimum info, and an inquiring mind usually prefers more detail:

- The Systray client and the console client are the same as far as actual folding, the display and user control are different.

Given that the two clients are essentially the same then is it better or worse to run two or more clients simultaneously?

The short answer is: It is usually worse.

Since the above response provides minimum info, and an inquiring mind usually prefers more detail:

The reason that it is usually worse is that each process (client executable and related cores) must share a single CPU. This gives rise to context switching. Context switching is usually an inefficient process, thus running two or more clients yields less performance than a single client.

How much worse?

According to my measurements (made a couple of years back).

- On AMD about 0.5 % to 1 % worse, depending on work units.

- On Intel (Dothan class processors), ~5% worse.

In the real world, exceptions are prevalent:

- For an Intel CPU with Hyper-Threading (the latter Pentium 4 class processors, i7 processors, & the Atom), there is a considerable gain from running an excess of clients.

Hyper-Threading provides a gain in throughput by reducing processor stalls (time waiting for data). Perhaps oversimplified; Hyper-threading uses the time that might be wasted in context switching by having a pre-filled pipeline (from the 2nd process, or other processes). ready for any available CPU cycles.

[Johan]

Thanks for the extensive explanation. Looks like I'll be running one client, then (systray).

Oh, and another question: Does Folding@Home "performance" depend on RAM memory size to some extent, or is it mostly the CPU?

[dnamechanic]

... Does Folding@Home "performance" depend on RAM memory size to some extent, or is it mostly the CPU?

F@h performance is mostly CPU related.

Folding is a CPU intensive process. The work units keep the CPU fully occupied. The priority is set so that the CPU stops folding when you use the computer.

Above a certain amount, increasing main memory size doesn't help. It's been a while since examining a regular client RAM requirements, but would guess no current uniprocessor* work unit requires more than ~100-150 MBytes. This is in addition to regular OS and other applications' RAM needs. Many folding work units require much less than 100 MBytes RAM.

Some earlier AMD processors showed appreciable performance gains from fast RAM (mainly because those AMD CPUs had relatively small cache sizes).

CPU cache size can make a considerable difference on several types of work units. In general, the larger the CPU cache, the better the folding performance.


* SMP work units can use more system RAM. Each F@h SMP client spawns at least 4 processes, with each process utilizing more than 100 MBytes.

Edited by dnamechanic, 15 July 2009 - 11:54 PM.