| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: GPU Support Forum
Thread: CL_DEVICE parameters: static or dynamic? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 4
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
My understanding is that except for the CL_DEVICE_MAX_CLOCK_FREQUENCY parameter (which may be changed by the cruncher), all other CL_DEVICE parameters are supposed to have static values.
----------------------------------------CL_DEVICE_MAX_MEM_ALLOC_SIZE: x MByte CL_DEVICE_GLOBAL_MEM_SIZE: y MByte For my HD7770, x used to have a static value of 512, and y used to have a static value of 1024. Lately though, I've observed that the values for the x, and for the y for my HD7770 are changing from one done HCC-GPU-WU to another. I've seen x:y pairs of 308:593, 308:602, 301:605, 287:501, 244:418, 310:601, 249:429, and 512:605 (x=default of 512: y not-equal to default of 1024) as examples. For a Tahiti GPU, as I've seen in my wingWU's result-log, x is static with a value of 512, and the y is also static with a value of 2048. Are the above CL_DEVICE parameter supposed to be static/unchanging? What may have accounted for the dynamic values indicated for the above parameters for my HD7770? Is my HD7770 broken? I continue to have valid doneWUs though or was that just luck else some trouble lies in wait for my GPU-card? ; ; andzgridPost#760 ; [Edit 2 times, last edit by Former Member at Dec 25, 2012 6:26:49 PM] |
||
|
|
armstrdj
Former World Community Grid Tech Joined: Oct 21, 2004 Post Count: 695 Status: Offline Project Badges: 
|
The Global Memory size I would think would be static. The Max allocation size is a percentage of the Global size and I believe it can be changed with an environment variable. Have you updated the driver before you started noticing these changes? Or updated your BOINC client version?
Thanks, armstrdj |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Would not be surprised if this varies when more than 1 HCC-GPU is run concurrent on a host.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hello armstrd and SekeRob
----------------------------------------References: armstrd [Jan 7, 2013 3:56:26 PM] post SekeRob [Jan 7, 2013 5:46:30 PM] post Thanks for responding. The Global Memory size I would think would be static. The Max allocation size is a percentage of the Global size and I believe it can be changed with an environment variable. I'm seeing the opposite in my wingWUs. The GlobalMemorySize I'm seeing there changes values more often than what I've seen for MAX_MEM_ALLOC_SIZE. That's the pattern I'm seeing even in my HD7770 where values for the GlobalMemorySize parameter changes more frequently comparatively than the value for the MAX_MEM_ALLOC_SIZE parameter.Have you updated the driver before you started noticing these changes? Or updated your BOINC client version? Perhaps it's the way I've installed BOINC_v7.0.42. My orig BOINC_v7.0.27 is from the Ubuntu repo. I didn't want to have BOINC_v7.0.42 replace my tried-and-tested-solid BOINC_v7.0.27. But I want to have app_config customization. So, I installed BOINC in a different directory and like a self-contained and isolated executable. Maybe that was what changed how the memory values are seen/deployed, and therefore how the HD7770 is used. However, and as before, I continue to have my done HCC_v7.08-GPU-WUs evaluated as valid from using either BOINC_v7.0.27, or BOINC_v7.0.42.As for the GPU-drivers for my HD7770, there are only two named drivers I use under Ubu12.10: the 'fglrx' and the 'fglrx-updates' driver for my HD7770. I'll do a test and see if the choice of a driver changes the picture, although I doubt if it's a good idea that what is as substantial as a memory is allowed to be held hostage 'depending on the driver'. Another thing is this: my Ubu-machine's mainboard has an onboard videoController. Is this affecting the values we see? Any ideas about which is supposed to be better: turn on or off that mainboard videoController? Would not be surprised if this varies when more than 1 HCC-GPU is run concurrent on a host. Of all the Radeon GPU-cards, a Tahiti is the GPU-card to see for high count of concurrent HCC_v7.08-GPU-WUs. The Tahiti's of my wingWUs almost-always have static values for the two mentioned CL_DEVICE parameters. It would seem therefore that a high count of concurrent HCC_v7.08-GPU-WUs does not affect a Tahiti but Pitcaim, Capeverde and perhaps all other non-Tahiti cards are expected to see the values for their card change? Hmm... I protest!  ; ; andzgridPost#794 ; ; 4 edit instances > all for spell-check [Edit 4 times, last edit by Former Member at Jan 8, 2013 12:07:47 AM] |
||
|
|
|