ffbr
-
Posts
193 -
Joined
-
Last visited
Content Type
Forums
Status Updates
Blogs
Events
Gallery
Downloads
Store Home
Everything posted by ffbr
-
I do not see the code for app_config.xml in your post. Interested in it!
-
Unfortunately I cannot on the laptops. Strict company policy (don't get me started on that...) to keep windows as the main OS led me to have to work with the WSL. Have to say that I am quite impressed of the results for most of the time.
-
I have 20 or so that escaped my bunkering (even at 0.01kos ...) before I blocked per time of day too.
-
So I have been experimenting with a P100 and the results appears quite disapointing. I am currently sitting at around 7-8 min per task on numbers. I wonder if the vcpu that comes with the instance is not starving the card.
-
What does the quorum means for the projects?
-
Wonder if BOINC run on the WSL. Anyone have tried it?
-
This is highly frustrating, my T4s also ran out of job during the night, such a waste. Glad the V100 are currently not available, they would probably ran out of jobs in 1h
-
Well ... didn't think to look at the log... So yes, numbers tell me to back off "too many upload in progress" Gonna set up another boinc instance
-
My system crunchin task for numbers ran out and I do not seems to be able to pick new ones. Any ideas?
-
So I might have access to 1 or 2 32 cores computation unit. On what should they be focused?
-
For the javelin I just spun up a T4 unfortunately cannot access better at the moment. Might have a P100 or a V100 free at some point but nothing certain (I also want to see how fast those spit the jobs for the fun of it). The T4 is at about 9min per job. I also need to figure out how to get the cpu working on a different project than the gpu (the T4 comes with 4vcores and I currently have an i7 on numbers instead of something more useful)
-
numbers bunkering is going quite well, already 84 task trying to upload . On average they take 8min on a 2070. Everything else is running rosetta
-
Another interesting thing about number limiting the upload lead to high project backoff, I am currently at 5hours+ Hum another thought, might be the 100s of task I had to cancel on my amd rigs ...
-
The task using my vega 56 and HD 6970 are taking longer than on cpu cores. I am gonna finish those and switch back to rosetta. For the javelin are the project switch for each throw? I avoid looking
-
It seems highly inneficient with amd cards... Around 1h for the cpu, 8min with a rtx 2070 and 45min on a vega 56
-
Ok numbers task are so small that even by limiting the upload rate they still are released. Trying to block boinc upload in the firewall now.
-
On my only nvidia laptop (RTX 2070) I got a bunch of cuda jobs, they seems to be the same as the cpu ones just taking 10min compared to the 1hour of the cpu ones Considering aborting the jobs on the other computers to make them run rosetta instead. Wonder the penalty for that. Might be spinning one or two T4 to help then, as they are on AWS its less annoying for me to setup than the IBM cloud ones
-
So apparently number likes nvidia gpus but not amds ones
-
Starting bunkering for javlin with 20 cores / 32threads Might bunker cloud instances for the marathon if I manage to find 1hour to set it up
-
Ok, good to know. At least I checked that my bunkering works, I limited both the upload rate and time frame allowed to 1 min per day. I requested 5 days worth of task but somehow do not get that much on one of my computer. Gonna prepare the instance(s), we only have some ibm cloud available and I am not so use to it.
-
Another question concerning bunkering, I see that my rosetta task have a deadline that is before the start of the marathon. How does that works?
-
Sorry if it's already in other answer, super busy at the moment with a days old baby. I got couple of idle instances that I can use for the Marathon, trying to find the best allocation in term of ram/disk space. What's a reasonable estimate in term of ram need (GB per thread?). And if I spun them now in bunkering mode, what's a rough need in storage (same GB per thread)?
-
I am seeing: - 9h on an i7-9750H - 6h on an i5-6600k - 1D on a ryzen5 2600
-
Gonna mainly go for the marathon (second girl arrived yesterday... not gonna be able to play with computers for a while.) Might be able to throw in a couple of AWS instances if projects don't need them, but haven"t had time yet to check docker or linux deployments. At home I have 16 cores ready to go! (and a couple of GPUs)
-
Yeah, was also my own damn fault not to be willing to go investigate why was that fan making noise ...
