I am running both CPU and GPU WUs on my Win 11 rig. My GPU is a GeForce RTX 3060 (not a 3060i). When I take driver updates from Nvidia Experience I have choice between Gaming driver or Studio driver. Is there one choice better than the other for crunching e@h GPU WUs?
I am running both CPU and GPU WUs on my Win 11 rig. My GPU is a GeForce RTX 3060 (not a 3060i). When I take driver updates from Nvidia Experience I have choice between Gaming driver or Studio driver. Is there one choice better than the other for crunching e@h GPU WUs?
It should not make a difference as long as your system crunches 24/7 without crashing.
And you do not need to upgrade as long as what you are using works.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
If you routinely power limit your Nvidia gpus to 300 watts (rtx 3080 ti).
Is it a good idea to enable the "persistence" function so the GPU remembers that setting after every re-boot?
Does the persistence function also remember memory overclocks?
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
State of this type lasts from the time the driver initializes a GPU until the time the GPU is unititialized. This is the narrowest lifecycle, as the kernel driver itself is still loaded and may be managing other GPUs. The GPU typically initializes a GPU if a client application tries to access the GPU. The GPU is typically deinitialized after the last client exits.
Under Linux systems where X runs by default on the target GPU the kernel mode driver will generally be initalized and kept alive from machine startup to shutdown, courtesy of the X process. On headless systems or situations where no long-lived X-like client maintains a handle to the target GPU, the kernel mode driver will initilize and deinitialize the target GPU each time a target GPU application starts and stops. In HPC environments this situation is quite common. Since it is often desireable to keep the GPU initialized in these cases, NVIDIA provides two options for changing driver behavior: Persistence Mode (Legacy) and the Persistence Daemon.
It does not remember your clocks from the last session. With each restart of the computer you need to reset your clocks and power and fan settings.
Best to script it and run the script on every startup.
I am not sure if I should post this in the general CPU thread or this one (or Windows...).
Anyone have any ideas to remediate the discrepancy in the time it is taking our new host (https://einsteinathome.org/host/13073591) to complete a Gamma-ray pulsar binary search #1 work unit versus this host: https://einsteinathome.org/host/12940794 ? It is not making much sense to me. Here is some additional info I can provide:
The times do not change when I halt all other work on the CPU.
It does not look like the the work unit is switching cores as the work unit progresses.
Other work units that are CPU only are CRAZY fast (I am doing Mapping Cancer Markers in 45 min each x 115 of these work units).
Settings in NVIDIA settings are the same.
Both are using the same profile under E@H.
Both are A4500 GPUs.
Now, for the odd things:
It could be Windows 11 Workstation, but I have no way to know.
When I pull up HWMonitor, there are big differences in the following clocks (the Threadripper system is the first number below).
- Graphics: 1665 versus 1755 MHz
- Memory: 7600 versus 7800 MHz
- Video: 1480 versus 1550 MHz
This makes no sense to me. The GPU temps are the same on both (~75C), the power is the same (~200W). I have a hard time thinking that it is the AMD processors fault. Does this sound like a Windows 11 issue? It looks like when it offloads to the CPU (last ~10% of the work unit) it takes about 5 seconds, so it's not hanging there for long.
that's pretty perplexing. i don't think the clock speed differences are the sole cause of the performance difference (less than 100MHz clock speed diff wont change crunch time that much). the memory speed difference to me seems to suggest that they might be running is different P-states. but again the difference isnt that much, and not on the order that would explain the performance difference you're seeing.
have you tried swapping the GPUs between these systems to see if the clock speed anomalies or performance issues follow the card or stay with the system?
Sometimes Amd has motherboard drivers for windows.
I like Ian&Steves either clean reset the new TR gpus. Or clean swap the gpus.
Are the pci slots set to "auto" or to Gen4 or Gen3?
Are the cards showing how hard they are loaded? Any differences between systems.
Is the TR MB bios current?
Hth,
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Chooka wrote: Looking
)
I as well. Thanks.
I am running both CPU and GPU
)
I am running both CPU and GPU WUs on my Win 11 rig. My GPU is a GeForce RTX 3060 (not a 3060i). When I take driver updates from Nvidia Experience I have choice between Gaming driver or Studio driver. Is there one choice better than the other for crunching e@h GPU WUs?
Styx N Stones wrote: I am
)
wont make any difference for einstein
_________________________________________________________________________
It depends on which
)
It depends on which version(s) are stable.
It should not make a difference as long as your system crunches 24/7 without crashing.
And you do not need to upgrade as long as what you are using works.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Thanx...
)
Thanx...
If you routinely overclock
)
If you routinely power limit your Nvidia gpus to 300 watts (rtx 3080 ti).
Is it a good idea to enable the "persistence" function so the GPU remembers that setting after every re-boot?
Does the persistence function also remember memory overclocks?
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Read this doc. Driver
)
Read this doc. Driver Persistence
2.1. GPU Initialization Lifecycle
State of this type lasts from the time the driver initializes a GPU until the time the GPU is unititialized. This is the narrowest lifecycle, as the kernel driver itself is still loaded and may be managing other GPUs. The GPU typically initializes a GPU if a client application tries to access the GPU. The GPU is typically deinitialized after the last client exits.
State:
1.2. Linux
Under Linux systems where X runs by default on the target GPU the kernel mode driver will generally be initalized and kept alive from machine startup to shutdown, courtesy of the X process. On headless systems or situations where no long-lived X-like client maintains a handle to the target GPU, the kernel mode driver will initilize and deinitialize the target GPU each time a target GPU application starts and stops. In HPC environments this situation is quite common. Since it is often desireable to keep the GPU initialized in these cases, NVIDIA provides two options for changing driver behavior: Persistence Mode (Legacy) and the Persistence Daemon.
It does not remember your clocks from the last session. With each restart of the computer you need to reset your clocks and power and fan settings.
Best to script it and run the script on every startup.
I am not sure if I should
)
I am not sure if I should post this in the general CPU thread or this one (or Windows...).
Anyone have any ideas to remediate the discrepancy in the time it is taking our new host (https://einsteinathome.org/host/13073591) to complete a Gamma-ray pulsar binary search #1 work unit versus this host: https://einsteinathome.org/host/12940794 ? It is not making much sense to me. Here is some additional info I can provide:
The times do not change when I halt all other work on the CPU.
It does not look like the the work unit is switching cores as the work unit progresses.
Other work units that are CPU only are CRAZY fast (I am doing Mapping Cancer Markers in 45 min each x 115 of these work units).
Settings in NVIDIA settings are the same.
Both are using the same profile under E@H.
Both are A4500 GPUs.
Now, for the odd things:
It could be Windows 11 Workstation, but I have no way to know.
When I pull up HWMonitor, there are big differences in the following clocks (the Threadripper system is the first number below).
- Graphics: 1665 versus 1755 MHz
- Memory: 7600 versus 7800 MHz
- Video: 1480 versus 1550 MHz
This makes no sense to me. The GPU temps are the same on both (~75C), the power is the same (~200W). I have a hard time thinking that it is the AMD processors fault. Does this sound like a Windows 11 issue? It looks like when it offloads to the CPU (last ~10% of the work unit) it takes about 5 seconds, so it's not hanging there for long.
Any ideas?
that's pretty perplexing. i
)
that's pretty perplexing. i don't think the clock speed differences are the sole cause of the performance difference (less than 100MHz clock speed diff wont change crunch time that much). the memory speed difference to me seems to suggest that they might be running is different P-states. but again the difference isnt that much, and not on the order that would explain the performance difference you're seeing.
have you tried swapping the GPUs between these systems to see if the clock speed anomalies or performance issues follow the card or stay with the system?
_________________________________________________________________________
Different gpu driver
)
Different gpu driver versions?
Sometimes Amd has motherboard drivers for windows.
I like Ian&Steves either clean reset the new TR gpus. Or clean swap the gpus.
Are the pci slots set to "auto" or to Gen4 or Gen3?
Are the cards showing how hard they are loaded? Any differences between systems.
Is the TR MB bios current?
Hth,
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!