Should F@H abandon CPU folding?

[timberwolf]

Understand that this was NOT someone at Folding@Home considering dropping CPU folding.

Some user using their own data was trying to interpolate for the rest of us.

You need not be dismayed that one person has a question. Especially as the informed answers supported your choice anyway.

Core_a4 supports all x86 CPUs back to Willamette and works reliably.

Core_a7 seems stable, and can support CPUs almost as old, but prefers CPUs starting at Haswell. A7 has issues with older OSs, more than older CPUs.

Core_21 is doing well on Nvidia from Keplar, and AMD from GCNv2, there is some issue with the new DRNA chip, but I think that is 'just' a driver update.

Folding@Home is working on a new Core_22, but I doubt it will abandon any GPUs. I think it is just better science.

Thank you very much as I wasn't sure what the differences between each core was.
Can you ilaborate on DRNA chip? Are these the new AMD GPUs (Navi 5700 or 5700 XT)?

[JimboPalmer]

Yes, they replaced Graphics Core Next, the AMD structure since about 2012. So far the driver does not support F@H or other OpenCL applications. I hope it is just a matter of time.

[bruce]

The original question was based on an assumption which is FALSE. Not all projects are suited for all hardware. FAH runs best when a certain amount of project assignment tuning is performed -- and CPU do have their place. Older (smaller) GPUs also have a place.

If I understand well this: https://stats.foldingathome.org/os and conservatively assume that in average a CPU requires 30W of power and a GPU 250W, it looks like CPU folding consumes 25% of total CPU+GPU electricity to provide 1% of total x86 TFlops. This is quite a huge waste of energy!

The complexity of the proteins that need to be analyzed vary considerably. One simple parameter that can easily be determined is the number of atoms. (See column 3 of https://apps.foldingathome.org/psummary.) The smallest proteins contain 5000 to 10000 atoms. The most complex contain 300k to 400k atoms. i,e.- a couple orders of magnitude.

The number of parallel threads (shaders in GPUs and threads on CPUs, though they're not exactly the same things) also varies by several orders of magnitude.

it's very clear (to me, at least. that a protein with lots of atoms will run more efficiently on hardware the processes many calculations in parallel and a protein with fewer atoms will run very well on hardware that processes fewer operations in parallel.

Now look at url=viewtopic.php?p=308730#p308730]Subject: Tiny projects killing PPD on big GPUs[/url] and Subject: Please allow smaller WUs on FAHClient.

How many watts are required to process 5000 atoms on a RTX 2080 Ti?

[Luscious]

I've been thinking about this question for a while now but was wondering where to post - will BIGADV ever return as an option for CPU slots?

It was understandable to get rid of it a few years back, but CPU's have long gone past quad core and even 12 cores. For a modern server running dual Rome chips for example (128 cores/256 threads), BIGADV could generate some serious points and reignite the interest in CPU-heavy setups, and as a result give a boost to those WU as well. More so as dual Rome setups lack the PCIe for running multiple graphics cards as they are favored for their CPU core count instead.

Obviously, the requirements for BIGADV would have to be changed from what they were before, but even a bottom floor limit of 18C/36T would cover a large swath of the CPU selection available today, including high end enthusiast stuff and dual socket workstations. In fact it would be relatively easy to deploy a dual 18 core system, assign 24 threads per CPU and still have CPU headroom to spare for multiple GPU's. But for a lack of any impetus to fold with CPU that same rig may only get dual 6 cores installed and fold with GPU's only.

[Joe_H]

I've been thinking about this question for a while now but was wondering where to post - will BIGADV ever return as an option for CPU slots?

From my understanding, the answer is a definite no. PG was pretty clear about that when they ended BigADV. Advances in GPU folding since then have made it even less likely to return.

[MeeLee]

There are still case scenarios where CPU folding makes sense.
Like, I'm experimenting with FAH Windows Binaries, to run them on systems where administrator privileges block these programs from installing, or where the user simply prefers not to install a program at all.
So far Client works well on CPU (using CMD).

Not to mention, Apple/Macs, as well as Intel systems using an IGP, still don't have a GPU option, I believe.

These systems will continue to fold for a while.

There might be a possibility that in the near future, most GPUs will be able to fold at double precision.
When that happens, CPU folding might become less and less of an option.

[JimboPalmer]

There might be a possibility that in the near future, most GPUs will be able to fold at double precision.

All GPUs that fold now, spend some time folding a Double Precision.

Inherently, it is twice as slow as Single Precision. To preserve the markup for 'Workstation' cards, most consumer cards add even more slowdown, as much as 1/32 Single Precision speed. But if a GPU does not support DP, F@H does not work at all on it.

[bruce]

There are still case scenarios where CPU folding makes sense.

My point, exactly.

...Not to mention, Apple/Macs, as well as Intel systems using an IGP, still don't have a GPU option, I believe.

The chance of ever folding on an iGPU is a pretty remote possibility, but also not impossible. One important fact, is that it could help fill a speed gap between bottom tier GPUs and high end CPUs but high-end CPUs (with AVX) seem to be filling in this gap quite well.

As far as the original question is concerted, what would be the point in eliminating CPU folding? The biggest fact in favor of CPU folding: It's extremely close to a zero-cost option from FAH's perspective. (Pretty much a zero maintenance option.) Some work gets done that wouldn't otherwise be done -- and if you don't like it, then don't use it.

FAH's development plans include an upcoming Core22, which will enhance production on high-end/mid-range hardware, increasing the PPD spread across the hardware spectrum so CPU folding will SEEM even less of a threat to those with lots of points.

[RichardGiddens]

Like Timberwolf, I am a small fish in a large pond but even the smallest fish make their contribution.

[JiiPee]

If I understand well this: https://stats.foldingathome.org/os and conservatively assume that in average a CPU requires 30W of power and a GPU 250W, it looks like CPU folding consumes 25% of total CPU+GPU electricity to provide 1% of total x86 TFlops. This is quite a huge waste of energy!

So shouldn't F@H start thinking about abandonning CPU folding and rather than the "I am one in a million" campaign that aims to increase the number of folders in general, try to target gamers who own GPUs (often hi-end GPUs moreover)?

Energy wasting sure is something to think about. Lets take my setup as example. GPU is burning 220W and CPU is burning 120W so total is 340W when I fold with both and that is just telemetry info, you gotta add VRM loss and PSU loss into that so total is higher. Now if I was folding 24/7/365, I would waste lot of energy. But I fold only when I need to turn on heating anyway, direct electricity heating here.
So folding like that, I don't really waste energy, because I would use that same energy anyway to heat room. Now back to that 24/7/365.. At summer time it get so hot here that I would really like to have AC to cool down this place even without folding, like last summer there was few weeks when inside temp hit 30C and even over that few days. Think about if I was also folding same time, I sure would need AC so that would be double waste of energy because extra stress for AC to dump out that extra heat what folding would cause.

Now that I have upgrade my cumputer, I have MB what have temp sensor connectors, I have already ordered some probes and my plan is to make small script what will turn off and on folding depending what is room temp, what I get with probe. I'm gonna use hwinfo alert feature to feed temp readings to script. I would have liked to use something what doesn't need to be running background all a time, but I haven't found any windows monitoring software what could read sensors from command line and after done simply exit.
Anyway with that setup I get nice automation and my room temp wont go too high.

[MeeLee]

If I understand well this: https://stats.foldingathome.org/os and conservatively assume that in average a CPU requires 30W of power and a GPU 250W, it looks like CPU folding consumes 25% of total CPU+GPU electricity to provide 1% of total x86 TFlops. This is quite a huge waste of energy!

So shouldn't F@H start thinking about abandonning CPU folding and rather than the "I am one in a million" campaign that aims to increase the number of folders in general, try to target gamers who own GPUs (often hi-end GPUs moreover)?

Energy wasting sure is something to think about. Lets take my setup as example. GPU is burning 220W and CPU is burning 120W so total is 340W when I fold with both and that is just telemetry info, you gotta add VRM loss and PSU loss into that so total is higher. Now if I was folding 24/7/365, I would waste lot of energy. But I fold only when I need to turn on heating anyway, direct electricity heating here.
So folding like that, I don't really waste energy, because I would use that same energy anyway to heat room. Now back to that 24/7/365.. At summer time it get so hot here that I would really like to have AC to cool down this place even without folding, like last summer there was few weeks when inside temp hit 30C and even over that few days. Think about if I was also folding same time, I sure would need AC so that would be double waste of energy because extra stress for AC to dump out that extra heat what folding would cause.

Now that I have upgrade my cumputer, I have MB what have temp sensor connectors, I have already ordered some probes and my plan is to make small script what will turn off and on folding depending what is room temp, what I get with probe. I'm gonna use hwinfo alert feature to feed temp readings to script. I would have liked to use something what doesn't need to be running background all a time, but I haven't found any windows monitoring software what could read sensors from command line and after done simply exit.
Anyway with that setup I get nice automation and my room temp wont go too high.

Why not set the PC up in an empty room with window open, a balcony, or in the attic?
Either keep the heat in that room, or leave a window open, just enough for the heat to escape?
There's no rule stating that the PC needs to vent the heat inside the house...