I'm in need of some assistance. My PPD per 2080 Ti's are hovering around 1.2Mil-1.6Mil. I'm not certain if there is something wrong, but I am skeptical since I have seen so many posts of 2080 Ti's breaking 2.2Mil+. I do have my Threadripper folding as well. I have paused it for a few hours to see if that adjusted anything, but did not see a change.
I am using the stock FAH client settings out of the box.
I am quite positive (knowing me) that I have just overlooked something obvious you all will get a good laugh at.
System Specs:
Motherboard: GIGABYTE X399 AORUS XTREME Rev. 1
GPU: 2x ZOTAC GAMING GeForce RTX 2080 Ti AMP MAXX
CPU: AMD 2nd Gen Ryzen Threadripper 2950X
PSU: Thermaltake 1250W RGB PS-TPG-1250DPCTUS-T
RAM: G.SKILL Flare X Series 128GB DDR4 2933
FAH Details:
Version 7.5.1
Passkey is enabled.
10+ WUs have been completed.
Do I need to park two cores? If so, what is the recommended software to do this?
Neither of these are obvious or will get you laughed at, but they are worth trying.
While you posted a log, you did not post the most interesting part, the configuration. This allows us to see what the Client thinks your PC looks like (CPU, OS, RAM, etc.etc.) Our advice may well improve with better data.
It is possible you added hardware after you installed the Client. The F@H client does a wildly better job of finding hardware during installation. So consider finishing running WUs, un-installing F@H, deleting all work files, and reinstalling overwriting data. This should force the installer to rediscover the hardware. You will need to type in the name, team and passcode.
[As always, I am not affiliated with F@H, just a user like you.]
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
JimboPalmer wrote:Neither of these are obvious or will get you laughed at, but they are worth trying.
While you posted a log, you did not post the most interesting part, the configuration. This allows us to see what the Client thinks your PC looks like (CPU, OS, RAM, etc.etc.) Our advice may well improve with better data.
It is possible you added hardware after you installed the Client. The F@H client does a wildly better job of finding hardware during installation. So consider finishing running WUs, un-installing F@H, deleting all work files, and reinstalling overwriting data. This should force the installer to rediscover the hardware. You will need to type in the name, team and passcode.
[As always, I am not affiliated with F@H, just a user like you.]
Thanks for the quick reply!
I just got the system all set up yesterday night and after I was completed, I installed FAH. So it didn't see me install the cards. I can do an uninstall but where would I find the Work Files? I've checked the program directory but don't see anything in the FAH folder that looks like Work Orders.
I *think* I uploaded the interesting bits of the log to my original message.
I also uploaded pics showing the Advanced tab in Gpuz.
The underlying GROMACS code that F@H uses for CPU folding hates large primes, and multiples of large primes. with 32 CPUs (and 2 reserved for feeding the GPUs) you have 30 CPUs folding ("-nt 30") this would be ideal as it is 2 times 3 times 5, no large primes. There is some chance that you may run out of WUs (many researchers limit themselves to 16) if so, make two CPU Slots, one of 16 and one of 12. If you do not have issues with 30, leave it there.
One of your GPUs seems to be limited by voltage, how large is your Power Supply? Are both cards on the same chain of connectors?
You seem to be using 45% of the PCIE bus. Are both of these cards in x16 slots?
Other than thaat, I am in over my head and will let others give advice.
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
JimboPalmer wrote:The underlying GROMACS code that F@H uses for CPU folding hates large primes, and multiples of large primes. with 32 CPUs (and 2 reserved for feeding the GPUs) you have 30 CPUs folding ("-nt 30") this would be ideal as it is 2 times 3 times 5, no large primes. There is some chance that you may run out of WUs (many researchers limit themselves to 16) if so, make two CPU Slots, one of 16 and one of 12. If you do not have issues with 30, leave it there.
One of your GPUs seems to be limited by voltage, how large is your Power Supply? Are both cards on the same chain of connectors?
You seem to be using 45% of the PCIE bus. Are both of these cards in x16 slots?
Other than thaat, I am in over my head and will let others give advice.
Thanks for the reply! I added my system specs to the original post in case that helps troubleshooting.
Would it be better for me to just split the CPU right now?
As for the voltage limitation.. I'm using a 1,250W Thermaltake modular PSU (PS-TPG-1250DPCTUS-T). It is a single rail PSU but to be certain I adjusted the connectors around with no change. After some brief Googling... it seems that my graphics cards are already overclocked to their voltage limit. Perhaps I should dial back the OC?
Both cards are inserted into the PCIE x16 slots. I verified with the Nvidia control panel as well as HWiNFO64. Is there anything that could be automatically throttling me? I am on the Performance power plan, too.
If you are getting WUs at 30 threads, that should be the most points, two smaller slots will get less total points. I only mentioned it in case you suddenly notice the CPU slot idle.
I am no expert at OCing, but if you get curious, I would try stock to see how many PPD it gets. I would also try a single GPU, just to see if PCIE bandwidth is an issue.
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
Good news is FAH now recognizes my CPU properly while my GPU says "01:39:04:Enabled folding slot 01: PAUSED gpu:0:TU102 [GeForce RTX 2080 Ti Rev. A] M 13448 (not configured)"
Mostly, we have exceeded my skill level and need a real expert.
"01:39:36:WARNING:WU00:FS00:Changed SMP threads from 30 to 31 this can cause some work units to fail
01:39:36:WARNING:WU00:FS00:AS lowered CPUs from 31 to 30"
Because you only have one GPU now, it only reserves one CPU, then realizes that is a large prime, and goes back to reserving 2.
One GPU uses 33% of the PCIE bus, so there may have been some contention with 2.
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
Let's suppose you leave the number of CPUs set to 30. Depending on what projects happen to be assigning at the moment you request a new WU, you might get a WU that uses all 30 cores ... or maybe the projects that can be run on all 30 are in short supply. The Assignment Server may direct you to another Work Server which happens to have projets that are limite to CPU:16 and one of those WUs might be downloaded. FAHClient will recognize this situation and reallocate 16 CPU to that project, leaving the 14 other CPU idle. When this WU is completed, you'll be assigned whatever project that can use as many as possible of your 30 threads.
By creating two slots, CPU:16 and CPU:12 you'll increase the average number of CPUs that will be active for any situation when you can't get one WU that uses all 30..
From the GPU side, it all looks good.
It appears to be installed in a full size slot, at gen 3.0.
The Tis can even pull 1.8mil in Linux on a pcie 3.0 1x slot (with riser).
Which means they'll do the same in windows on a 2x to 4x slot.
You're sending 250Watt average to the GPU, which also looks normal.
GPU is running at 75 C, which is a bit high.
While this is not the real issue at hand, I'd recommend to pull the power consumption down to a Max of 200 watt. 180 watt preferably.
It may be counterintuitive, dropping performance on your GPU to increase the performance, but it actually won't significantly affect your PPD. In return, your card will run a lot cooler.
I would then try to run an overclocking utility, to pull the GPU frequency back up, once the card's real issue has been found.
One discrepancy I immediately noted, was that your GPUs run at extremely low clockrates. The average clockrate mine runs at is 1995 to 2040 MHz, while yours is running at 1350-1410 MHz.
The GPUs came Fromm Zotac already Overclocked so I’ll need to undo their clocks. What software do you use to OC your 2080s?
My GPUs are set around 1350 base clock with a “Boost” of up to 1950 I believe. On the second page of Gpuz it shows what they’re currently running at (right? I could be wrong...)
Oh, ok. I thought this was the case. GPUz doesn't display boost frequency.
The 2 most popular free overclocking programs for NVidia cards are EVGA's XOC, and MSI's afterburner. Both are for Windows.
For Linux it'll be NVidia xconfig with cool-bits enabled.
Once you install one of these programs you can easily see the temperatures, frequency and voltages of the GPU and memory.