[H] issued challenge - Stress testing Amicable Numbers for DC-Vault inclusion

I'll see what's happening at the time; already spent a lot of cycles on this project after it opened.
 
For this project, the CPU work is multi-threaded. It originally paid exceptionally well compared to other CPU projects. I have not watched the CPU earnings since the GPU application came out. So, it is probably worth running on both.
 
I ran at full tilt on my gpus for a day but stopped a couple of hours ago. No issues identified that would impact it being vault worthy.

Only concerns for me were the cache limit of 20 tasks per gpu (less than an hour's worth on 1080 Tis) and the increasing cpu allocation required.
 
Have you posted any request to up the limit of work units to cache?
 
They might increase the limits if you did. I do not know of why they limit it so low now as is.
 
I'm testing with this app_config file below.

I'll let it run for a white before doing any changes. Anyone is doing similar testing?

With two tasks per GPU, the GPU utilization is hovering around 34% to 52% as shown by MSI afterburner


upload_2017-6-14_21-30-40.png


<app_config>
<app>
<name>amicable_2_64</name>
<plan_class>opencl</plan_class>
<gpu_versions>
<gpu_usage>.50</gpu_usage>
<cpu_usage>.25</cpu_usage>
</gpu_versions>
</app>
</app_config>
 
Should I switch over to this? I'm sitting cpu at citizen science and gpu @ Einstein
 
If you want to help test, then sure. I'm pretty confident it is fine for the Vault. I did it more as a formality to give others a chance to verify before adding it.
 
Hmm, looks like I've to try this app_config to see if this is better than the one I'm using now. I'm not hitting close to 100% GPU utilization.
 
Let us know how that goes. If it still doesn't max it out, up the number of work units in the app_config and see if it increases the utilization.
 
The project administrator mentioned that it now needs 8 task per GPU for 100% for 1080Ti. I'll try that.

I remembered it wasn't like that earlier on in the project.
 
Ok. With 8 tasks per GPU, finally got 100% but chewing up 8 CPU threads!

I'll let it run for half a day and see what I got in terms of PPD.

BTW, i'm running with power limit (at 50%) as it is getting hot here but gains on crunching efficiency. Save electricity and lower temp (barely 60°C, which I like it for 24/7 type of operation).

upload_2017-6-16_7-20-41.png
 
Damm, the project administrator just added a new application task that seems to take one task per GPU at 100%. Have to re-edit the app_config and retry.

upload_2017-6-16_7-45-41.png
 
He just added an option to select which application tasks to run. The double check one runs very fast with one task per GPU, like 30 secs per task but all credits are still pending. Maybe low scoring points.

Ok. I gave up. Need to take a break.

upload_2017-6-16_8-23-29.png
 
It is so much faster to run the double check task application as shown above and only uses 1 CPU thread instead of 8.

I estimated based on the last 100 validated tasks, PPD is about 1M on my 1080 running with 50% (~100W) power limit.

With 4 tasks per GPU, I could only get about avg 465K PPD but if scale to 8 tasks per GPU and using 8 CPU threads, this may be slightly lower than the double check applicaiton.
 
You must log in or register to reply here.
Back
Top