Message boards :
Graphics cards (GPUs) :
Mixing Cards
Message board moderation
| Author | Message |
|---|---|
|
Send message Joined: 11 Dec 08 Posts: 26 Credit: 648,944,294 RAC: 479 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Hi, I have 2 computers with more than one card. 1st has 9800GX2 and GTX260 2nd has 9800GX2 and 8800GT I was advised this is a really bad idea? |
|
Send message Joined: 25 Oct 08 Posts: 42 Credit: 42,812,268 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
I have an host with 8400, GTX260 and 8800 GT. It works fine with driver 185.2, boinc 6.6.2. and I run 4 wus in same time (2 GPU, and 2 others) With video driver more recent, all my wus go in errors. |
Paul D. BuckSend message Joined: 9 Jun 08 Posts: 1050 Credit: 37,321,185 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Hi, Well, not really a REALLY bad idea ... just that it may not be the best configuration. Sadly we are still in early days so that work is not controlled as to which GPU it would run on. Or to put it another way, that BOINC would be smart enough to start a task on the 8800 and note that it is running too slow to meet the deadline and move it to the 9800 card. Or to use the 9800 to run GPU Grid and the 8800 to run SaH (assuming you were hooked up to both projects). We have asked for this feature, but the developers have been silent as to their intent. The most significant issue is the differences in speed between the cards where the GTX260 is about 4 times faster than the 9800 card... though if I understand the part number that gives you 3 tasks in flight in both systems. Were money no object I would suggest another GTX 260 and pair those and the 9800 cards in the same system and drop the 8800 ... but that is if money is no object ... |
|
Send message Joined: 11 Dec 08 Posts: 26 Credit: 648,944,294 RAC: 479 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Thanks for the comments, I'll see how i goes and if I get problems. I do have 3 tasks on the go at any one time but if the GTX's are 4x faster then, 2x4 is bigger than 3x1. The GX2's are hot and noisy as well. I can trade in the GX2 for another GTX + money, than won't be too hard on the wallet, my 3rd system has 9800GX2 as well. |
|
Send message Joined: 17 Aug 08 Posts: 2705 Credit: 1,311,122,549 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
I was advised this is a really bad idea? Why would it? I can't see a problem with these configs. As has been said earlier in this thread folding@home has problems with mixed setups, but so far BOINC doesn't seem to be affected by this. Maybe who ever gave you this advice was having folding@home in mind? And the GTX 260 is not 4 times faster. Sure, it's more energy efficient, but not that much more. Your 9800GX2: 1512 MHz * 128 shaders * 3 Ops/clock = 580 MFlops per chip. Your GTX 260: 1350 MHz * 216 shader4 * 3 Ops/clock = 875 MFlops As recently determined a theoretical MFlop from a GT200 does about 41% more work than one from a G9x chip, so the final numbers: 2 x 580 = 1160 MFlops for the GX2 and 1233 MFlops for the GTX260. That's about a 6% advantage.. impressive for a "one versus two" match, but probably not worth starting to swap / buy / sell cards. Except if you can get really good deals on the GX2s. MrS Scanning for our furry friends since Jan 2002 |
Paul D. BuckSend message Joined: 9 Jun 08 Posts: 1050 Credit: 37,321,185 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
I was comparing my 9800GT with my GTX280 and GTX295 cards. Where the latter are about 3-4 times faster in my experience. At the moment, BOINC under 6.6.24 does not have a problem with mixed configurations. 6.6.24 and later have a "best" detector and any card that does not match the best is not used. 6.6.25 has a configuration switch that allows you to ignore this test and use all cards. But, that can be removed at any time. I have tried off and on for a couple months to engage the developers in a discussion on the implications of mixed systems to no avail. I think as we move forward this is going to be a significant part of the landscape that they ignore at peril. For example is is very common for people to slowly upgrade their systems with new GPUs, and many systems come with additional slots that the old card can slip into. So, why would I toss a perfectly good card when I add a new faster one? I mean, when I upgrade and buy a new computer to replace the old Dell dual Xeons I have I will likely try to get a fast GPU but the Dells both have a GPU also. So, if I get a system with 3 or 4 PCI-e slots am I to toss the older cards that are perfectly good? Or move them to the new system to keep the computing capability until I can buy 4 GTX295 cards ... |
|
Send message Joined: 17 Aug 08 Posts: 2705 Credit: 1,311,122,549 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
I was comparing my 9800GT with my GTX280 and GTX295 cards. Where the latter are about 3-4 times faster in my experience. Without doing the exact numbers that matches: your 9800GT is a little slower than a single GX2, where you have 2 chips, and your GTX 280 is faster than a GTX 260 Core 216 so you should end up just shy of a factor of 3. The 295 is again faster and should be good for that factor of 4. MrS Scanning for our furry friends since Jan 2002 |
©2025 Universitat Pompeu Fabra