Helpful to adjust core/memory clock and power limits?
Moderator: Site Moderators
Forum rules
Please read the forum rules before posting.
Please read the forum rules before posting.
-
- Posts: 62
- Joined: Mon Apr 13, 2020 11:47 am
Helpful to adjust core/memory clock and power limits?
Is it helpful to lower core clock, raise memory clock, and lower power limits for folding at home like it is in crypto mining? I'm not trying to increase folding performance, but if I can lower power usage while maintaining the same folding performance, then that would be worthwhile.
Re: Helpful to adjust core/memory clock and power limits?
I'll chip in a few of my thoughts
memory - do not clock higher than rated. It will only make your system more unstable and prone to crashing.
cpu - at the very least disable turbo/overdrive. When the processor turbos/overdrives it consumes disproportionally more power than the performance gain, it also produces more noise as the fans have to work harder to remove the additional heat. Typically I run my Ryzen at 90% (3.34GHz) using the power management controls built into Windows 10.
cpu slot - By default the FAHClient will assign a cpu core/thread to each graphics card and grab the rest for the cpu slot. Something I usually recommend is adjusting the size of your cpu slot to leave at least 1 free cpu core (2 cpu threads for hyperthreading processors) to allow the operating system room to breath.
Graphics card - MSI Afterburner. Typically I power limit my GPUs to 70%, I think I get something like 80-90% performance. I certainly do not let them boost. Again, it's a large power draw for a small performance gain.
memory - do not clock higher than rated. It will only make your system more unstable and prone to crashing.
cpu - at the very least disable turbo/overdrive. When the processor turbos/overdrives it consumes disproportionally more power than the performance gain, it also produces more noise as the fans have to work harder to remove the additional heat. Typically I run my Ryzen at 90% (3.34GHz) using the power management controls built into Windows 10.
cpu slot - By default the FAHClient will assign a cpu core/thread to each graphics card and grab the rest for the cpu slot. Something I usually recommend is adjusting the size of your cpu slot to leave at least 1 free cpu core (2 cpu threads for hyperthreading processors) to allow the operating system room to breath.
Graphics card - MSI Afterburner. Typically I power limit my GPUs to 70%, I think I get something like 80-90% performance. I certainly do not let them boost. Again, it's a large power draw for a small performance gain.
-
- Site Moderator
- Posts: 6349
- Joined: Sun Dec 02, 2007 10:38 am
- Location: Bordeaux, France
- Contact:
Re: Helpful to adjust core/memory clock and power limits?
Don't overclock video memory ... it will only increase the probability of errors for only a minor gain ...
The easiest ans safest way to save power is to play with power limit of the GPU.
The easiest ans safest way to save power is to play with power limit of the GPU.
Re: Helpful to adjust core/memory clock and power limits?
I presume you're talking about folding on an Nvidia GPU?
If so,
I'd have to argue different from aetch.
The memory is rated at 15Gbps, but from the factory tuned to 13-14Gbps.
So in many cases you can overclock the memory by 500/1000Mhz (depending on the program you use).
If the program says 7Ghz, you can overclock by 500Mhz max. If it says 13 or 14Gbps, you can overclock by up to 1000 Mhz.
The speed of folding is affected between 2,5-10%. The GPU will run hotter.
On the other hand, lowering the memory speed will result in higher stable GPU core clocks (of a few tens of Mhz).
I did some testing, and some of my RTX 2000 GPUs could run the memory up to 15800Mhz before crashing, and only about 14900Mhz or so, before showing glitches in games and occasionally error on FAH;
Anything beyond 14500-14750Mhz didn't really affect folding performance noticeably, so I would recommend if you do overclock, to not go too far off the 14-14,25Gbps speeds.
My GPUs are now 3 years old almost, so I would presume modern GPUs can safely OC higher.
With folding you can overclock the memory quite a bit, because not all memory is used (usually 500MB to 2GB at most). Which means most of the other memory is passive, and it helps keeping those active memory modules cooler, and run higher speeds.
Generally you don't need to enable turbo boost for GPU folding, so long your CPU is at least a 2Ghz for RTX 2000 and RTX 3050/3060 GPUs, and a 3Ghz CPU for RTX 2080Ti and RTX 3060 and up GPUs.
GPU settings:
On GTX 1000 series GPUs generally you'd be overclocking the cr@p out of them!
On RTX 2000 series GPUs generally you'd want to lower the TDP by about 20%, and overclock the GPU to reach close to the same as stock boost frequencies.
I have no experience with RTX 3000 series GPUs, but from what I read, they run pretty close to their lowest TDP. Lowering power will lower performance.
Not sure how much you can overclock them, but you probably can do a little.
If so,
I'd have to argue different from aetch.
The memory is rated at 15Gbps, but from the factory tuned to 13-14Gbps.
So in many cases you can overclock the memory by 500/1000Mhz (depending on the program you use).
If the program says 7Ghz, you can overclock by 500Mhz max. If it says 13 or 14Gbps, you can overclock by up to 1000 Mhz.
The speed of folding is affected between 2,5-10%. The GPU will run hotter.
On the other hand, lowering the memory speed will result in higher stable GPU core clocks (of a few tens of Mhz).
I did some testing, and some of my RTX 2000 GPUs could run the memory up to 15800Mhz before crashing, and only about 14900Mhz or so, before showing glitches in games and occasionally error on FAH;
Anything beyond 14500-14750Mhz didn't really affect folding performance noticeably, so I would recommend if you do overclock, to not go too far off the 14-14,25Gbps speeds.
My GPUs are now 3 years old almost, so I would presume modern GPUs can safely OC higher.
With folding you can overclock the memory quite a bit, because not all memory is used (usually 500MB to 2GB at most). Which means most of the other memory is passive, and it helps keeping those active memory modules cooler, and run higher speeds.
Generally you don't need to enable turbo boost for GPU folding, so long your CPU is at least a 2Ghz for RTX 2000 and RTX 3050/3060 GPUs, and a 3Ghz CPU for RTX 2080Ti and RTX 3060 and up GPUs.
GPU settings:
On GTX 1000 series GPUs generally you'd be overclocking the cr@p out of them!
On RTX 2000 series GPUs generally you'd want to lower the TDP by about 20%, and overclock the GPU to reach close to the same as stock boost frequencies.
I have no experience with RTX 3000 series GPUs, but from what I read, they run pretty close to their lowest TDP. Lowering power will lower performance.
Not sure how much you can overclock them, but you probably can do a little.
Re: Helpful to adjust core/memory clock and power limits?
@MeeLee
What tool(s) are you using to verify a successful GPU overclock?
I don't believe graphics artefacts or FAH errors, or lack thereof, are a good indicator that the GPU is stable.
I have tried to find tools which specifically test the stability and accuracy of CUDA but have so far failed.
Where are your folding machines housed? I got the impression from previous posts that they were housed in remote locations suggesting the likes of a datacenter or a co-location so noise, and very likely power consumption, are not factors.
What tool(s) are you using to verify a successful GPU overclock?
I don't believe graphics artefacts or FAH errors, or lack thereof, are a good indicator that the GPU is stable.
I have tried to find tools which specifically test the stability and accuracy of CUDA but have so far failed.
Where are your folding machines housed? I got the impression from previous posts that they were housed in remote locations suggesting the likes of a datacenter or a co-location so noise, and very likely power consumption, are not factors.
Re: Helpful to adjust core/memory clock and power limits?
For Linux, use nvidia-smi to set power.
I personally use Nvidia X-Server for overclocking and fan speed (GUI).
The features will enable after doing a coolbits=28.
For Windows, you can use any overclocking tool from any major card manufacturer.
MSI Afterburner, EVGA precision software (XOC or X1), PNY Velocity X, Gigabyte easytune etc...
If FAH shows no errors, that means the GPU is running stable.
When playing games, artifacts appearing after an overclock, usually indicate the GPU is not stable.
FAH taxes the GPU more than any games. You can run games at higher overclock values, with minor glitches, while a FAH WU would just crash with even the slightest glitch.
My racks are located on my balcony. At peak, I was consuming ~2,5kW of power.
Since then I've retired, and now only do Boinc atm. Because of system instability.
GPU servers are best run in cold environments. The temperatures over here reach well in the 90s during summer (Farenheit).
wha
I also am waiting for a better time to run more efficient hardware. The RTX 3090 is a great card, but too expensive for FAH (or boinc) for me.
So I wait for technology to bring down the prices, and may do a repurchase when GPUs become affordable again (like what happened when the RTX 2000 series were released).
I personally use Nvidia X-Server for overclocking and fan speed (GUI).
The features will enable after doing a coolbits=28.
For Windows, you can use any overclocking tool from any major card manufacturer.
MSI Afterburner, EVGA precision software (XOC or X1), PNY Velocity X, Gigabyte easytune etc...
If FAH shows no errors, that means the GPU is running stable.
When playing games, artifacts appearing after an overclock, usually indicate the GPU is not stable.
FAH taxes the GPU more than any games. You can run games at higher overclock values, with minor glitches, while a FAH WU would just crash with even the slightest glitch.
My racks are located on my balcony. At peak, I was consuming ~2,5kW of power.
Since then I've retired, and now only do Boinc atm. Because of system instability.
GPU servers are best run in cold environments. The temperatures over here reach well in the 90s during summer (Farenheit).
wha
I also am waiting for a better time to run more efficient hardware. The RTX 3090 is a great card, but too expensive for FAH (or boinc) for me.
So I wait for technology to bring down the prices, and may do a repurchase when GPUs become affordable again (like what happened when the RTX 2000 series were released).
Re: Helpful to adjust core/memory clock and power limits?
@MeeLee
My biggest takeaway is that your systems are outside, exposed to the weather, and they're unstable.
My biggest takeaway is that your systems are outside, exposed to the weather, and they're unstable.
Re: Helpful to adjust core/memory clock and power limits?
Yeah, for now.aetch wrote:@MeeLee
My biggest takeaway is that your systems are outside, exposed to the weather, and they're unstable.
I wouldn't recommend it, even though you save a lot of AC cost in the process.
I only run a bit of GPU loads in the winter nowadays.
But outside air is not good for pc systems.
Lots of floating carbon in the air, as well as other metals that seem to corrode electronics, unless you live in a forest somewhere.
My systems are still working, but I'm not actively doing FAH or a lot of boinc anymore because of this.
I do still have 24 Atomic Pi units doing CPU work, and they seem impervious to the outside particles.
They're industrial boards, so they may have been designed to different standards than motherboards made for desktop pcs.
One thing I do nowadays, is trade a little bit of cooling capabilities to combat pollution, by coating the motherboards with a layer of epoxy.
It seems to be very successful in preventing short circuits, and keeping the motherboards look and perform like new, while thin layers of it may prevent capacitors to cool as well as without the layer.
But doesn't seem to be a problem, so long I don't spray any in the PCIE or ram slots, or on the VRMs or CPU die.
Re: Helpful to adjust core/memory clock and power limits?
If you want to run a system outside, it sounds like you either need to run them fully passive, or to put some very good filters on them, or run them in their own little micro-climate that has a heat exchanger to the outside. I've heard of some people doing that by having big radiators (possibly repurposed from car climate systems) both inside and outside of a box, thus allowing heat exchange while having the box hermetically sealed against air. The latter solution will run hotter and on slightly lower clocks than if it was allowed to run in the open air, but at least there will be no dust accumulating on the components; only the outside radiator picks up dust.
Online: GTX 1660 Super + occasional CPU folding in the cold.
Offline: Radeon HD 7770, GTX 1050 Ti 4G OC, RX580
Re: Helpful to adjust core/memory clock and power limits?
What would actually work best, is a water cooling system using a heat pump. Much more efficient than AC units, and dump the heat outside.
But that's $$$$.
As far as using passive systems outside, they still get full of carbon. My balcony has a mosquito net, and the carbon/particles still go through them.
My system uses a large box fan, using a Hepa filter (the spunge looking thing), but it still seems to go through. Perhaps a carbon air filter is better than a hepa one?
Second best is to use indoor air (in case of hot outside air) to cool a room like a garage, and have the hot air vent out the ceiling/roof.
This would create an air pressure differential in your house, sucking in outside air that needs to be cooled, and still adds to overall cooling cost.
But that's $$$$.
As far as using passive systems outside, they still get full of carbon. My balcony has a mosquito net, and the carbon/particles still go through them.
My system uses a large box fan, using a Hepa filter (the spunge looking thing), but it still seems to go through. Perhaps a carbon air filter is better than a hepa one?
Second best is to use indoor air (in case of hot outside air) to cool a room like a garage, and have the hot air vent out the ceiling/roof.
This would create an air pressure differential in your house, sucking in outside air that needs to be cooled, and still adds to overall cooling cost.
Re: Helpful to adjust core/memory clock and power limits?
What works best varies from person to person.
I get the impression that for most here "best" is just a computer tucked away in a corner of a room that isn't too noisy, doesn't need exotic cooling and isn't too heavy on the electric bill.
I have seen that a few here run folding machines in place of room heaters in the colder months.
This is pretty much describes myself as well.
For me, running a system outside isn't even an option. I have neither the secure location nor the climate, living in "sunny" Scotland.
BTW, I think you're hepa filter might not be a hepa filter.
I believe they're meant to be a fine cloth type fabric, not a sponge.
I get the impression that for most here "best" is just a computer tucked away in a corner of a room that isn't too noisy, doesn't need exotic cooling and isn't too heavy on the electric bill.
I have seen that a few here run folding machines in place of room heaters in the colder months.
This is pretty much describes myself as well.
For me, running a system outside isn't even an option. I have neither the secure location nor the climate, living in "sunny" Scotland.
BTW, I think you're hepa filter might not be a hepa filter.
I believe they're meant to be a fine cloth type fabric, not a sponge.
Re: Helpful to adjust core/memory clock and power limits?
In winter when folding-electricity's waste output of heat helps heat my house, I run all my rigs at full speed and at proven stable overclocks (GPU only). In summer, I tend to remove the overclocks and even turn off some cards. All depends on your ability to pay the bills.
single 1070