Message boards :
Number crunching :
GPUs: AMD vs nVidia vs The Rest of The World
Message board moderation
Previous · 1 · 2
Author | Message |
---|---|
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
Actually nVidia has even no proper OpenCL 1.0 support. He's pulled across "rev 2058" from Beta testing - that contains the application code which Raistmer was able to copy across from Einstein's sources. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Actually nVidia has even no proper OpenCL 1.0 support. Looks pretty impressive. It makes those blanking blanked tasks really stand out... |
juan BFP Send message Joined: 16 Mar 07 Posts: 9786 Credit: 572,710,851 RAC: 3,799 |
Actually nVidia has even no proper OpenCL 1.0 support. All my hosts are running the rev 2058 from Beta, it has a very low CPU usage (good for me, most of my CPU´s are I5) that allow me to run 2 AP WU at a time on the multiple GPU hosts. But be aware, there still have a bug on this version but that bug is very rare and apears randomly (hard to track). And has a wierd GPU usage on my I7 host. |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Actually nVidia has even no proper OpenCL 1.0 support. Sleep() & wait for event loops will be used in some places That host uses -use_sleep option that never worked in prev builds just because bug I mentioned earlier. And sorry, Richard, this has nothing to do with Einstein's sources though it was close in time with Intel GPU findings that followed from those. EDIT:thanks to Einstein@home project we now have further improved oclFFT library that works much more precise on IntelGPUs than original one (and whole idea of too bad native_sin/cos precision on Intel GPU is inspired by conversations with Oliver) also, comparison between SETI and Einstein's sourses reveal some difference in synchronisation usage (namely, they use direct clFinish() call instead of other possible methods). Extention of this approach to quite ridiculous degree (to call clFinish after EACH OpenCL call) allowed to decrease CPU usage on IntelGPUs considerably (need to note that Einstein's code doesn't use clFinish in that way, but looks like very such (and definitely not the one I could call "normal") way of usage provides best results on Intel GPUs). On nVidia GPUs such excessive clFinish calls do nothing but increase total runtime and CPU overhead (as one could expect from general considerations). SETI apps news We're not gonna fight them. We're gonna transcend them. |
ML1 Send message Joined: 25 Nov 01 Posts: 20283 Credit: 7,508,002 RAC: 20 |
...Perhaps a more everyday readable summary is eloquantly given in this comments post about the nVidia approach: Is this why nVidia are emboldened or driven to try throwing their weight around to attempt to dictate terms?... Top500: Red dragon ... graphical power (See the accelerators graph. Note the recent trend...) IT is what we allow it to be... Martin See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
William Send message Joined: 14 Feb 13 Posts: 2037 Credit: 17,689,662 RAC: 0 |
Please keep the discussion civil and technical or it's going to end up in Politics. There IS a difference between opinion and slander. Competition is one thing, and most of us are biased one way or the other for all sorts of reasons. I don't like the undertone of some of the posts. If you want to be nasty, you can do so in Politics. A person who won't read has no advantage over one who can't read. (Mark Twain) |
ML1 Send message Joined: 25 Nov 01 Posts: 20283 Credit: 7,508,002 RAC: 20 |
Hahaha, yeah I can see how it might be tough for Intel and AMD to come to terms with that they both build inferior compilers to the open source LLVM project, and want nVidia's help while having shoved them out of the x86 market completely. There seems to be too much of that game between the main players that helps noone. You could argue that the latest little game is just another in-kind (negative) step as is always done between such manufacturers. However, in this instance, it is for rather a high profile and widely used and important target... (Thanks for good comment in the various other posts.) For my own personal view: nVidia nicely gained a good lead some time ago with their bold move into what at that time was a whole new GPGPU architecture that would work well enough for both industrial and consumer use. There's various graphs showing how that has been very successful although more recently the competition is starting to gain more market share. I've been nVidia since the early days of their CUDA, all to good gain. Looking around for alternatives, the AMD APUs look good and interesting for their homogenous memory access, but all at a small scale for onboard integrated graphics. I'm guessing such as the Intel Phi, and the Epiphany/Parallela systems are far too new/niche for the time being. Which leaves the AMD (ATI?) Radeon GPUs as a comparable alternative?... For the moment, I'm waiting to see which way things go for the latest controversy. (Shame we seem to always create in effect monopolies!) Happy clean fast crunchin', Martin (All just humble personal opinion as ever and always.) See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.