Now that WU's allocate less than 900MB, is it possible to relax the GPU memory requirement to 1GB ?

I'll try to do it today and then watch if I get any new WU errors on the project.

Now that WU's allocate less than 900MB, is it possible to relax the GPU memory requirement to 1GB ?

I'll try to do it today and then watch if I get any new WU errors on the project.

Did this ever get done?

It was done. Current requirement is 800 MB GPU memory to run a task.

Hi,

Just wondering what the actual system requirements are...

Today I added 2 machines (S3PC-S04 and S3PC-S25) to this project, both machines are running 1 GPU WU each, and seems to run out of memory; after about 30 secs to a minute, the WU gets status "Waiting for memory".
Both machines have 8GB system memory, and running Ubuntu, which does not take much resources.
Based on BOINC memory settings, the GPU application takes more then 6GB (75%) of system memory...? Ridiculous!

If these are the real specifications, then application need some serious optimization/redesign, this is not workable.

Hi,

Just wondering what the actual system requirements are...

Today I added 2 machines (S3PC-S04 and S3PC-S25) to this project, both machines are running 1 GPU WU each, and seems to run out of memory; after about 30 secs to a minute, the WU gets status "Waiting for memory".
Both machines have 8GB system memory, and running Ubuntu, which does not take much resources.
Based on BOINC memory settings, the GPU application takes more then 6GB (75%) of system memory...? Ridiculous!

If these are the real specifications, then application need some serious optimization/redesign, this is not workable.

Presently the GPU workunits require 8GB memory. If you do not have enough RAM, you can increase the pagefile size and enable more pagefile usage in BOINC to compensate for the shortfall (this may require a restart of the system to take effect). Beware though that this will read and write an enormous amount of data. If you have a hard drive it will be unlikely to keep up and would be very taxing on it, and an SSD will incur a significant number of writes. On my system with a GTX 1650 and 8GB RAM, I had 28TBW over the month or so I ran Amicable Numbers tasks.

Regards,
Kellen

Presently the GPU workunits require 8GB memory. If you do not have enough RAM, you can increase the pagefile size and enable more pagefile usage in BOINC to compensate for the shortfall (this may require a restart of the system to take effect). Beware though that this will read and write an enormous amount of data. If you have a hard drive it will be unlikely to keep up and would be very taxing on it, and an SSD will incur a significant number of writes. On my system with a GTX 1650 and 8GB RAM, I had 28TBW over the month or so I ran Amicable Numbers tasks.

Regards,
Kellen

Hi Kellen,

Thanks for your reply.

I increased swap size to 32GB; after using (almost) all system memory I see about 1.5GB of swap is used, then WU again errors (255: Unknown error code).
For now I give up and move GPU's to different project, one which runs without any hassle...

I don't understand why the GPU version uses that much system memory, as the CPU version uses much less (~ 2.8GB)...?

Because GPU version has to store all prime numbers < 10¹¹ and it can't use compact bit array as CPU version does.

Why do the GPU tasks take so long? On my laptop running a 1660ti after 8 hours the prediction is 10 more days to finish. Ridiculous for a GPU task. I aborted it.

Why do the GPU tasks take so long? On my laptop running a 1660ti after 8 hours the prediction is 10 more days to finish. Ridiculous for a GPU task. I aborted it.

The only thing I can think of is are you leaving a cpu core free just for the gpu to use? My Nvidia 1050Ti with 4gb and the same driver you are using of onboard memory is doing tasks in: 4,580.39 seconds and getting 6,836.19 credits for each one. The only main difference is I'm using Linux and you are using Windows but that should NOT matter at all.

I know this is a low priority - I just got my (modified) MacPro up on Amic, and CPU tasks are running now. It appears there is no Mac version of the GPU applications... :-/

(Yeah, I know, I may be the ONLY one on this project with a Mac with an AMD card in it...)

Tern, thanks for asking for this. It is now extremely difficult to use Mac/AMD rigs to crunch Einstein tasks. So, I have moved my rigs over here. With the rollout of M-series Macs, I hope that Amicable continues to support those of us who are crunching using Intel/AMD Macs!

I know this is a low priority - I just got my (modified) MacPro up on Amic, and CPU tasks are running now. It appears there is no Mac version of the GPU applications... :-/

(Yeah, I know, I may be the ONLY one on this project with a Mac with an AMD card in it...)

I finally figured out how to compile OpenCL version for MacOS and added AMD and NVIDIA OpenCL application versions for MacOS 10.7 or newer: https://sech.me/boinc/Amicable/apps.php

Thank you so much!

I just restarted crunching Amicable Numbers and have been reading some of the message postings.
I have 2 EVGA 3090ti's / 2 EVGA 3090's and 1 AMD 7900 XTX.

The AMD card produces (2.7M/day) 95% usage a 1/3 of what the EVGA 3090's produce (5.5-6.0M/day) 85-90% usage on average.

To be clear - what suggestions do you have that might get the AMD card into the same range of output?

---

Hi Farscape:

Based on what other AMD 7900 XTX crunchers are achieving, I think you are getting the expected throughput:
https://sech.me/boinc/Amicable/gpu_list.php

The average throughput of this GPU is 248.3 seconds/task. This translates into ~2.4 M credits/day.

AMD GPUs don't appear to perform as well as NVIDIA GPUs at Amicable Numbers. My 6800 XTs are averaging approximately 1.1 M credits/day and my 5700 XTs are averaging approximately 600 K credits/day.

Happy Crunching!

Tigers_Dave

I just restarted crunching Amicable Numbers and have been reading some of the message postings.
I have 2 EVGA 3090ti's / 2 EVGA 3090's and 1 AMD 7900 XTX.

The AMD card produces (2.7M/day) 95% usage a 1/3 of what the EVGA 3090's produce (5.5-6.0M/day) 85-90% usage on average.

To be clear - what suggestions do you have that might get the AMD card into the same range of output?

---

Hi Tiger-Dave - thanks for the reply.

I guess that I almost expected this information!

We date, the only project that I have used where AMD kicked booty on the nVidia cards was on Milkyway GPU work. Since then, Einstein is producing better numbers than here at Amicable - just not as good as the nVidia cards...

Thanks once again for it response!

Tim
aka Fascape

---

Hi Tim:

I agree - my AMD GPUs generated a 20-25% higher RAC at Einstein @ Home than here at Amicable Numbers. However, in the third quarter of 2023, Einstein ceased supplying data for one of their apps and the only alternative app for AMD GPUs on Macs wouldn't run reliably. That's when I committed 100% of my GPUs and their effort to Amicable.

Good luck with your crunching.

Dave

--

Hi Tiger-Dave - thanks for the reply.

I guess that I almost expected this information!

We date, the only project that I have used where AMD kicked booty on the nVidia cards was on Milkyway GPU work. Since then, Einstein is producing better numbers than here at Amicable - just not as good as the nVidia cards...

Thanks once again for it response!

Tim
aka Fascape

---