| GrueMaster | http://www.phoronix.com/scan.php?page=article&item=phoronix_effimass_cluster | 00:11 |
|---|---|---|
| lilstevie | GrueMaster: wow, 6 pandaboard ES clustered was heaps more efficient than the atom, that is unexpected | 00:42 |
| infinity | lilstevie: It's an Atom 330, that's not a useful or fair comparison. | 00:44 |
| lilstevie | infinity: yeah just realized, old as atom | 00:45 |
| infinity | lilstevie: Oldest Atom core there is, yeah. | 00:45 |
| lilstevie | the z530 atom was also less efficient too though | 00:46 |
| infinity | lilstevie: z530 is the same vintage. | 00:47 |
| infinity | lilstevie: Both are 45mn non-SoC cores from 2008. | 00:47 |
| infinity | lilstevie: A comparison with something from this year would be much more interesting. | 00:47 |
| lilstevie | ah | 00:47 |
| lilstevie | I didn't know where the z530 fit | 00:47 |
| lilstevie | the medfield atom would be interesting | 00:48 |
| infinity | Indeed. | 00:48 |
| lilstevie | also would be interesting to see performance of something like the S4 in the same situation | 00:50 |
| infinity | Well, it would also be interesting to make that an OpenCL cluster, and let the AMD Fusion actually use its GPU. | 00:51 |
| infinity | Perf/W could skew a bit there. | 00:51 |
| infinity | Cause it's powered up, despite it not being used. | 00:51 |
| lilstevie | yeah | 00:52 |
| infinity | (Though this is true of the PVR on the OMAP4 as well, there's no way it can come close to delivering the performance of the on-die Radeon on the Fusion) | 00:52 |
| lilstevie | agreed | 00:52 |
| lilstevie | but then use the tegra powered openCL rig and see how that one goes :p | 00:52 |
| infinity | Has anyone mangled OpenCL for the Tegra yet? | 00:53 |
| lilstevie | that isn't an on-die gpu though | 00:53 |
| infinity | "Tegra does not currently support CUDA or OpenCL. NVIDIA believes in and invests heavily in GPGPU computing so this is an area we're investigating but do not have any announcements for Tegra at this time." | 00:54 |
| lilstevie | yeah | 00:54 |
| lilstevie | not likely to happen | 00:54 |
| lilstevie | :p | 00:54 |
| infinity | Oh, right, I remember researching this before. There was some internal corporate fear that supporting CUDA and OpenCL on the Tegra parts would cut into other business units. | 00:55 |
| infinity | Cause one could ditch the x86/GeForce clusters and build Tegra clusters instead. | 00:55 |
| infinity | To which, I say, "let 'em", cause they'd still hit two different HPC markets. | 00:55 |
| infinity | Given the workload-per-core characteristics would be dissimilar. | 00:56 |
| infinity | But, arguing with management over money issues is never easy. | 00:56 |
| lilstevie | heh | 00:56 |
| infinity | That actually puts AMD in a unique situation. I wonder if they realise this and have tried selling into that market more heavily. | 00:58 |
| infinity | Cause, one would think, bang-for-buck, Fusion would beat Atom, i7, and any ARM part for an OpenCL solution. | 00:59 |
| lilstevie | yeah | 00:59 |
| lilstevie | imagine how killer they could be with tegra if they put their gpu tech to work though, full OpenGL rather than just the embedded standard | 01:00 |
| infinity | Yeah, it could be shiny. | 01:01 |
| lilstevie | I can't see why the ULP GeForce would be ruined with GL | 01:01 |
| infinity | I'd also love to see an AMD ARM part, but I suspect they don't have the spare cash flying around internally to hedge their bets on forking their CPU division. | 01:01 |
| lilstevie | yeah | 01:02 |
| infinity | And no, there's no reason whatsoever that the GeForce on the Tegra couldn't support full OpenGL, it was a conscious decision not to. | 01:02 |
| infinity | Partially because of the CUDA/shader/CL concerns and, of course, because GLES is the mobile standard. | 01:03 |
| lilstevie | yeah, well GLES being the mobile standard isn't so much of a problem | 01:03 |
| lilstevie | you could handle compat in driver really | 01:04 |
| infinity | But, from what I understand, it's just a "normal" GeForce under the hood. | 01:04 |
| infinity | Like, electrically. | 01:04 |
| lilstevie | I suspect you may be right with CUDA/shader/CL | 01:04 |
| lilstevie | yeah | 01:04 |
| lilstevie | thats what I have heard too | 01:04 |
| lilstevie | it will be better if the open driver ever makes it too | 01:05 |
| lilstevie | the nvidia ones are pretty bad | 01:05 |
| === prp^2 is now known as prpplague | ||
| infinity | lilstevie: Pretty bad is an understatement. | 01:14 |
| lilstevie | infinity: well I have a tegra3 device and it is pretty bad, but the earlier drivers were way worse on the trimslice | 01:18 |
| lilstevie | :p | 01:18 |
| prpplague | ho ho hum | 01:28 |
| _william_ | Hi all :) | 12:22 |
| _william_ | i'd like to setup a chrooted environment and i'm looking for packages repositories in order to add software into the chroot. Is there some ARM repo existing ? | 12:23 |
| ogra_ | _william_, http://ports.ubuntu.com/ubuntu-ports/ | 13:51 |
| ogra_ | use that as the mirror in your sources.list | 13:51 |
| === ogra_ is now known as ogra | ||
| === ogra is now known as ogra_ | ||
| === prpplague^2 is now known as prpplague | ||
| === prp^2 is now known as prpplague | ||
| === tr-808_ is now known as tr-808 | ||
| mlankhorst | *sighs at binary drivers* immediately know when they're loaded because you start seeing glitches :\ | 20:04 |
| mahmoh | does /etc/flash-kernel.conf get used any longer with Quantal flash-kernel? | 22:01 |
| GrueMaster | mahmoh: Simple test. Delete it (or change it) and see what flash-kernel does. | 22:36 |
| dannf | mahmoh: no ref to it in the source i can find | 22:49 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!