85% TDP Throttling Question

[wuffy68]

Back in 2015, I asked a question about equipment wear-n-tear of running the client on Full ... I think at the time, Dr. Pande actually answered saying the client only runs the hardware up to ~85% TDP. Now I can't find the conversation, but I just want to verify ... in v7.6.21 is there any built-in throttling other than CPU and GPU sensed critical temperature? Understandably, folding on GPUs cant really use full TGP since it doesn't use that much board memory, but it would be interesting to know how to answer when others outside the community ask the same question. Cheers!

PS: if anyone has a link to a similar thread where this is already answered - that works for me.

[FaaR]

Generally "no" I would say, and particularly in case of GPUs, as there's no software-side throttle as such on them - when they go, they go, at whatever speed they're configured to go at. We have some knobs and levers to turn in the driver control panels these days (power limit in particular, also undervolting to some degree - this is black magic though as it can cause system instability - and clock speed target, but power limit generally works the best) which can help to reduce GPU power draw greatly.

I remember BOINC supporting some sort of high-res CPU usage feature long ago, I'm not sure if that feature is still in there. You could set it at say, 85%, and it would load the CPU cores up to 85% and no more. Never saw anything similar for F@H though. On the other hand, we can use Ryzen Master or Intel's ETU instead to set hard limits on how high the CPU will clock. This will be system-wide though, so will affect all software and not just F@H. Oh, and you can also limit the number of CPU cores you fold on - that will also help to bring down power usage and temperatures.

That said, assuming your system is properly configured and outfitted with sufficient cooling I wouldn't worry. I've run plenty folding on all of my previous rigs since 2008 and never suffered a hardware failure because of it (anecdotal evidence, so unfortunately worthless! lol) Only reason I'm not currently folding is that my old AMD Vegas are simply too long in the tooth and can't run power efficiently compared to today's GPUs - and particularly not with these electricity prices we have in Europe right now.

Hoping to be able to upgrade this winter though. Looking at an all-AMD rig for the first time since AMD K6-III/Rage 128 Pro! (Except AMD hadn't bought ATI yet back then...)

[JimboPalmer]

One method of throttling your CPU in software is the light, medium, full setting.
Light = total threads / 2
Medium = total threads - 1
Full = total threads

If you had 6 CPU threads, then medium would be about 85%.

There are finer detail settings for setting CPU threads as well, but each thread is only on or off. And GPUs are only on or off.

[wuffy68]

Thanks JimboPalmer and Faar ... that helps.

Yea, I've run some old hardware ~24/7 for six years (stopping only for vacations and occasional cleaning) with no issues other than fan bearing failure towards the end of the six years. I have no problem with wear at that rate - small price to pay, but it's a question (more like comment) that I often get when I introduce folding to others... "huh, how long will my video card survive?!" ... I guess, again ... with only a portion of the memory being utilized, the GPU boards never truly run at 100% "TGP" while folding unless you're trying to do something else like streaming or light game play at the same time. And yea, using Afterburner or similar can bring the overall utilization down if one wants to be more savvy about it.

Good luck putting together the new AMD rig (at least consumer electronics haven't felt the bite of inflation) ... I'm hoping to upgrade one of mine with a used 3090TI (currently I have a 1080Ti that seems to have gotten air bubbles in the water cooling (had it mounted at an odd angle)). For winter it would still make a great heater as-is if I can shake the air up to the radiator. For better or worse, I live in a gas fracking state in the US, so our electric bill hasn't changed much (yet) except they did restructure our peak and off-peak rates. Much less expensive off-peak (good for charging EVs), much MORE expensive during peak (AC and ... folding).

[JimboPalmer]

I can recommend a Generac solar inverter - battery combo, it seems to be able to bias to peak rates.

[BobWilliams757]

Folding will stress a system more than most benchmarks, but to me that's a good thing. If it isn't stable, it isn't stable and using F@H will find flaws other benchmarks might not.

That being said, I've run into the same concerns trying to get people to use their rigs for folding. "I've heard it will burn my system up!" and other such crap. I guess these people never use their systems near capacity for extended periods of time. They will baby them until they become obsolete.

I've run a lot of my systems for extended periods, even before I was folding. The only thing I've found regarding wear is that fans do have a somewhat limited life, and of course you tend to at least slowly add to storage usually over time. Fans are easy, and storage is cheap. I would much rather run my gear while before it becomes obsolete for daily use. Systems these days are so capable that even reasonably low end stuff can and will do much more than the average user will ever ask of it.

My only real considerations on our PC is power consumption and sound levels. Since it's generally on 24/7 and sits close to the desk with the monitor, I don't want any more noise than is needed. As such it's a single fan case, and it still leaves plenty of overhead on the cooling if it were ever needed. I do power limit the GPU, but more to just have it operating in a window of efficiency vs any concerns over sound levels or longevity. For CPU folding I do usually limit using all cores to overnight, or just set the cores lower so it doesn't make much noise if folding while we are on the PC for other things. Often my folding is taking place in background with two user profiles logged in. Even if I stream appointments and such folding only takes a small hit (10-15% maybe) during that time that I'm on video.

[FaaR]

I have no problem with wear at that rate - small price to pay, but it's a question (more like comment) that I often get when I introduce folding to others... "huh, how long will my video card survive?!"

The video card itself will generally live much longer than its attached fans, if using multiple axial-type fans (these fans are often quite awful where quality is concerned, even on expensive high-end models.) Video cards using a single radial blower-style fan seem to be much more durable/longer service life, at the cost of not cooling as well, and - of course - increased noise levels (sometimes hugely so)... The reason for this may be that axial fans tend to re-ingest hot air coming out of the video card's own cooler, this heats up the fan bearings and slowly evaporates the lubricants in them over time which eventually causes fan failure. As blower coolers shove virtually all the hot air outside the computer's case, this issue doesn't happen there.

So if you can direct the airflow such that there's as little re-ingestion of hot air as possible, the video card will run cooler and the fans will endure longer. A roomy computer case also helps here, as tight cases (while neat and tidy) tend to bottle in and concentrate heat. An issue I've run into myself in my own Fractal Design Dimension C, where I have to tear off the front panel and all the air filters to let the GPUs breathe more unrestricted if I run Folding. Oh, and nerf the power limit to the ground, as otherwise the GPUs will cap out at 85C... Not really a problematic temperature for the chips themselves, it's more a fan longevity issue, plus as the Vega 64 uses HBM video memory if you go over 80C I think it is you'll get hit by a greatly increased refresh rate penalty which lowers the available bandwidth. At least these older-gen HBMs are quite finicky, probably a reason they never caught on in consumer products.

...Along with the huge cost and great complexity of the stacking process and mounting on the silicon interposer. These aspects are probably not AS severe anymore today - interposers are generally not a thing anymore as we have silicon bridges instead, and a bunch of other professional gear type products do use HBM, so stacking costs should have dropped as well. Still, GDDR is much simpler and generally just as fast, so will probably be a while yet until something like HBM makes a re-appearance in gaming GPUs.

... I guess, again ... with only a portion of the memory being utilized, the GPU boards never truly run at 100% "TGP" while folding unless you're trying to do something else like streaming or light game play at the same time.

Generally, running other stuff - like gaming - on the GPU in parallel to folding tend to disrupt performance, as you're now having additional data flows going through the chip and its various buffers and caches which compete with the folding compute task. You're also lighting up other sections of the chip which are never used in pure compute, like the geometry and texturing hardware and rasterizer pipelines for example, which draw additional power and dissipate additional heat, which may cap your clock speed.

On some platforms, gaming while folding may also cause random glitches which may cause the current work unit to fail, graphics driver to crash or entire PC BSODing (simply running Google Maps in a hardware accelerated browser window could be enough to make the system trip itself over); I saw that on my previous Intel Haswell-based rig. My current Intel Skylake-X rig on the other hand does not have such issues, or they're so incredibly rare I've simply never noticed a pattern.

My experience only covers what is now considered older hardware, but I don't expect things to have changed much on this front; computing hardware has limited ability to run multiple tasks on the same set of execution units. Of course, you would not want to game on older Nvidia GPUs in particular as chips previous to the Turing series did not support asynchronous compute, which can severely limit framerates...

But listen to me yakking on and on! lol