High disk activity running Gamma on GPU (but only one of them)?!

Peter Hucker
Peter Hucker
Joined: 12 Aug 06
Posts: 765
Credit: 397,601,612
RAC: 12,861
Topic 222364

On one computer, I have a 4GB AMD Radeon RX 560 GPU running 2 gamma tasks simultaneously.  I noticed the disk light flashing a lot, so I checked the Windows 10 task manager.  The two gamma tasks are writing 1MB/second between them to the system drive.

On another computer, I have three 3GB AMD Radeon R9 280X GPUs running 2 gamma tasks each simultaneously.  These are only writing 0.2MB/second per card, a fifth of the amount, on cards that do that tasks 2.5 times quicker.

So that's the equivalent of 12.5 times more disk activity than I would expect on the slower card.

It's not enough disk activity to cause a problem, as it's an SSD, I was just wondering why it's doing it.

Peter Hucker
Peter Hucker
Joined: 12 Aug 06
Posts: 765
Credit: 397,601,612
RAC: 12,861

Still doing it, but only now

Still doing it, but only now and then.  Once it starts doing it, it does it with every Gamma task until a reboot.  Any ideas?

Mike Hewson
Mike Hewson
Moderator
Joined: 1 Dec 05
Posts: 6,414
Credit: 209,057,798
RAC: 14,375

Hypothesis : one possible

Hypothesis : one possible deduction is that the gamma tasks are not equivalent b/w the slower and the faster cards ie. they systematically differ in say, ( astronomical ) source template frequency. It's generally true that higher source frequencies require more templates to be tested per given frequency interval. Ergo more disk accesses to write out results/candidates? You could test this by examining the task names via Boinc Manager, as the naming scheme generally include/imply a source template frequency.   

Cheers, Mike.

I have made this letter longer than usual because I lack the time to make it shorter ...

... and my other CPU is a Ryzen 5950X :-) Blaise Pascal

Peter Hucker
Peter Hucker
Joined: 12 Aug 06
Posts: 765
Credit: 397,601,612
RAC: 12,861

Mike Hewson

Mike Hewson wrote:

Hypothesis : one possible deduction is that the gamma tasks are not equivalent b/w the slower and the faster

Does b/w mean "between"?

Mike Hewson wrote:

cards ie. they systematically differ in say, ( astronomical ) source template frequency. It's generally true that higher source frequencies require more templates to be tested per given frequency interval. Ergo more disk accesses to write out results/candidates? You could test this by examining the task names via Boinc Manager, as the naming scheme generally include/imply a source template frequency.   

Cheers, Mike.

It's not doing the high disk access just now, but the list of Gammas on my machines have a similar range of frequencies on the one with the slow card and the ones with the faster cards.  If the 8 digit number is the frequency, they all range from about 24260000 to 35740000.  The start of the task name is always LATeah1063L44_444.0_0_0.0_ on all machines.

I'll assume that my slow card gets higher frequency ones occasionally.  Admittedly it's the only machine I'd be likely to spot high disk access on.  The others I only look at when I need to.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.