Welcome to MilkyWay@home

2 x 4870s or 4870X2

Message boards : Number crunching : 2 x 4870s or 4870X2
Message board moderation

To post messages, you must log in.

Previous · 1 · 2

AuthorMessage
Profile The Gas Giant
Avatar

Send message
Joined: 24 Dec 07
Posts: 1947
Credit: 240,884,648
RAC: 0
Message 15305 - Posted: 14 Mar 2009, 6:54:24 UTC

And they are running 0.19e....
ID: 15305 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile UL1

Send message
Joined: 13 Nov 07
Posts: 11
Credit: 916,096,899
RAC: 0
Message 15386 - Posted: 14 Mar 2009, 19:16:23 UTC

I'm sorry, but I have to correct you on this: I'm running 'only' three single 4870s, no 4870X2s. I once had an 4870X2, but I wasn't able to get the GPUs to run at more than 50% load each and so it turned out to be slower than a single 4870. I also tried running four single 4870s on one board, but this only looked fine on paper...in reality something slowed this configuration down and therefore it was less productive than using three cards...
ID: 15386 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Brickhead
Avatar

Send message
Joined: 20 Mar 08
Posts: 108
Credit: 2,607,924,860
RAC: 0
Message 15389 - Posted: 14 Mar 2009, 19:38:20 UTC - in response to Message 15386.  

I'm sorry, but I have to correct you on this: I'm running 'only' three single 4870s, no 4870X2s. I once had an 4870X2, but I wasn't able to get the GPUs to run at more than 50% load each and so it turned out to be slower than a single 4870. I also tried running four single 4870s on one board, but this only looked fine on paper...in reality something slowed this configuration down and therefore it was less productive than using three cards...

So, up to three single-GPU cards in one computer works fine, without the "50% load" issue from the X2? This makes me wonder, are those three cards in a crossfire configuration or just three separate cards?
ID: 15389 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile UL1

Send message
Joined: 13 Nov 07
Posts: 11
Credit: 916,096,899
RAC: 0
Message 15390 - Posted: 14 Mar 2009, 19:53:59 UTC - in response to Message 15389.  

Three seperate cards. They don't run at 100%, but somewhere between 70 - 90%, so if you're for the maximum total output you should run them completely single...
I think CrossFireX wouldn't work...and if, I guess you would have the same issue as with the X2, but this time at 33% each...
ID: 15390 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile The Gas Giant
Avatar

Send message
Joined: 24 Dec 07
Posts: 1947
Credit: 240,884,648
RAC: 0
Message 15394 - Posted: 14 Mar 2009, 20:45:38 UTC

I get enough heat with only 1 4850. How do you handle 3 * 4870's? Love to see a photo of inside your machine.
ID: 15394 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile UL1

Send message
Joined: 13 Nov 07
Posts: 11
Credit: 916,096,899
RAC: 0
Message 15400 - Posted: 14 Mar 2009, 21:50:49 UTC - in response to Message 15394.  

I'll try to do some photos of both of the rigs running three 4870s by tomorrow...
ID: 15400 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 15401 - Posted: 14 Mar 2009, 21:51:36 UTC - in response to Message 15390.  
Last modified: 14 Mar 2009, 22:02:38 UTC

Three seperate cards. They don't run at 100%, but somewhere between 70 - 90%, so if you're for the maximum total output you should run them completely single...
I think CrossFireX wouldn't work...and if, I guess you would have the same issue as with the X2, but this time at 33% each...

I guess that is a driver issue or something in the CAL libraries. Officially the X2 cards are not supported by ATI's Stream SDK 1.3 I've used up to now.

But the Stream SDK 1.4 (released yesterday) added official support for X2 cards. They have also put some more functionality into Brook, so it should be now possible to use multiple GPUs and to have a low CPU load without modifying the official brook runtime library (brook.dll). I added those features to the 1.3 version in a quick and dirty manner. While that hack works quite well with individual cards, there are obviously some qirks with crossfire and X2 configurations. Maybe it will work only with the upcoming Cat 9.3 driver, because that will come with the CAL libraries 1.4.xx which may be needed for that functionality. While Cat 9.2 is already compatible with the 1.4 SDK (opposed to Cat 8.12 and 9.1), it has only CAL 1.3.186, so some things may not work like intended.

I guess the next version (0.19f) will be based on the 1.4 SDK and will require Cat 9.2 or newer. If Travis agrees, I would also like to report the GPU time as WU time, not the CPU time. Otherwise, with the CPU load as low as it is with 0.19d and up, the task duration correction factor is easily falling below 0.02 which may cause some work fetch problems. Furthermore the GPU time is a much more representative value than the CPU time for the actual duration of the WU.
ID: 15401 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
John Clark

Send message
Joined: 4 Oct 08
Posts: 1734
Credit: 64,228,409
RAC: 0
Message 15410 - Posted: 14 Mar 2009, 23:19:49 UTC
Last modified: 14 Mar 2009, 23:23:15 UTC

Is GPU client 0.19e available to the masses yet?

If so, where is a copy located please, Cluster P?

Currently making productive use of 0.19d.
ID: 15410 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 15417 - Posted: 14 Mar 2009, 23:51:06 UTC - in response to Message 15410.  
Last modified: 14 Mar 2009, 23:53:40 UTC

Is GPU client 0.19e available to the masses yet?

If so, where is a copy located please, Cluster P?

I post the links in our forum first to have a smaller testing group in the beginning. That way a version can be pulled much easier, if something goes wrong (like with 0.19c). But some are using it already for 2 days and I've seen no severe complaints yet (quite the opposite ;). So I guess it is now okay to use it as the new recommended version. You can find it here.
Be sure to read the readme file!

Changelog:
0.19e: 12th Mar 2009
- new scheduling system with dynamic granularity of GPU tasks for a better responding system
- new command line option (f) to tune the granularity of the GPU tasks
- extended diagnostic output to stderr.txt like GPU time (visible under task details)

The default options used without any editing of the app_info.xml should enable a good responding system and still a high throughput. Especially systems with HD3800 series cards will see a major improvement of the system response, but also a slightly larger hit of the crunch times (probably smaller than 5%, nothing to really worry about) compared to 4800 cards. Decreasing the f value the system behaviour will get worse (but throughput may increase slightly). One should not set f below 1.
ID: 15417 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Thamir Ghaslan

Send message
Joined: 31 Mar 08
Posts: 61
Credit: 18,325,284
RAC: 0
Message 15419 - Posted: 15 Mar 2009, 0:03:53 UTC - in response to Message 15401.  

Three seperate cards. They don't run at 100%, but somewhere between 70 - 90%, so if you're for the maximum total output you should run them completely single...
I think CrossFireX wouldn't work...and if, I guess you would have the same issue as with the X2, but this time at 33% each...

I guess that is a driver issue or something in the CAL libraries. Officially the X2 cards are not supported by ATI's Stream SDK 1.3 I've used up to now.

But the Stream SDK 1.4 (released yesterday) added official support for X2 cards. They have also put some more functionality into Brook, so it should be now possible to use multiple GPUs and to have a low CPU load without modifying the official brook runtime library (brook.dll). I added those features to the 1.3 version in a quick and dirty manner. While that hack works quite well with individual cards, there are obviously some qirks with crossfire and X2 configurations. Maybe it will work only with the upcoming Cat 9.3 driver, because that will come with the CAL libraries 1.4.xx which may be needed for that functionality. While Cat 9.2 is already compatible with the 1.4 SDK (opposed to Cat 8.12 and 9.1), it has only CAL 1.3.186, so some things may not work like intended.

I guess the next version (0.19f) will be based on the 1.4 SDK and will require Cat 9.2 or newer. If Travis agrees, I would also like to report the GPU time as WU time, not the CPU time. Otherwise, with the CPU load as low as it is with 0.19d and up, the task duration correction factor is easily falling below 0.02 which may cause some work fetch problems. Furthermore the GPU time is a much more representative value than the CPU time for the actual duration of the WU.



OIE! after reading this thread you guys made me suspicious!

I just double checked gpu-z catalyst center readings and I'm getting 90% load on both cards if under full load.

If I lower the value to say one task only one card will get the task.

Win 7, catalyst 9.1 for vista, under compatibility.

I'm not sure if this has anything to do with it, but when win 7 was installing, it installed some serious sounding drivers which names escapes me for the pci-e.

I'd be interested to know if any one under vista is using both x2 cards, considering I jumped ships from xp to 7!

ID: 15419 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 15420 - Posted: 15 Mar 2009, 0:14:12 UTC - in response to Message 15419.  

I think CrossFireX wouldn't work...and if, I guess you would have the same issue as with the X2, but this time at 33% each...

I guess that is a driver issue or something in the CAL libraries. Officially the X2 cards are not supported by ATI's Stream SDK 1.3 I've used up to now.

OIE! after reading this thread you guys made me suspicious! I just double checked gpu-z catalyst center readings and I'm getting 90% load on both cards if under full load. If I lower the value to say one task only one card will get the task.

Win 7, catalyst 9.1 for vista, under compatibility.

I'm not sure if this has anything to do with it, but when win 7 was installing, it installed some serious sounding drivers which names escapes me for the pci-e.

I'd be interested to know if any one under vista is using both x2 cards, considering I jumped ships from xp to 7!

So under Win7 both GPUs of a X2 card are used with Cat 9.1? Can anyone confirm this?
ID: 15420 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Thamir Ghaslan

Send message
Joined: 31 Mar 08
Posts: 61
Credit: 18,325,284
RAC: 0
Message 15426 - Posted: 15 Mar 2009, 0:45:53 UTC - in response to Message 15420.  

I think CrossFireX wouldn't work...and if, I guess you would have the same issue as with the X2, but this time at 33% each...

I guess that is a driver issue or something in the CAL libraries. Officially the X2 cards are not supported by ATI's Stream SDK 1.3 I've used up to now.

OIE! after reading this thread you guys made me suspicious! I just double checked gpu-z catalyst center readings and I'm getting 90% load on both cards if under full load. If I lower the value to say one task only one card will get the task.

Win 7, catalyst 9.1 for vista, under compatibility.

I'm not sure if this has anything to do with it, but when win 7 was installing, it installed some serious sounding drivers which names escapes me for the pci-e.

I'd be interested to know if any one under vista is using both x2 cards, considering I jumped ships from xp to 7!

So under Win7 both GPUs of a X2 card are used with Cat 9.1? Can anyone confirm this?


Also, not sure if hooking up both displays will force both cards to run!

I've got one on monitor and another on TV!

PROOF:

http://i46.photobucket.com/albums/f129/hotthamir/readings.jpg
ID: 15426 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
John Clark

Send message
Joined: 4 Oct 08
Posts: 1734
Credit: 64,228,409
RAC: 0
Message 15427 - Posted: 15 Mar 2009, 0:48:42 UTC - in response to Message 15417.  
Last modified: 15 Mar 2009, 0:49:18 UTC

If so, where is a copy located please, Cluster P?

I post the links in our forum first to have a smaller testing group in the beginning. That way a version can be pulled much easier, some are using it already for 2 days and I've seen no severe complaints yet (quite the opposite ;). So I guess it is now okay to use it as the new recommended version.

You can find it here.
Be sure to read the readme file!


Cluster Physic

Many thanks for access to the SSE2 version, under Win32, of the GPU client.

Watching a few WUs go through I think this is slightly faster than 0.19d (or that is the impression), and the system's graphical response has returned almost to normal.

Thank you for these impressive MW GPU clients, and all the work you and your team testers are carrying out. It is much appreciated.

John
ID: 15427 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 15432 - Posted: 15 Mar 2009, 1:10:42 UTC - in response to Message 15427.  

Watching a few WUs go through I think this is slightly faster than 0.19d (or that is the impression), and the system's graphical response has returned almost to normal.

Thank you for these impressive MW GPU clients, and all the work you and your team testers are carrying out. It is much appreciated.

You're welcome.

But 0.19e isn't faster. It should be even a bit slower because the partitioning of the GPU tasks costs a bit performance (maybe I can tune that a bit in the future). But you would probably gain a bit, if you set the wait factor to 1.1 (instead of 1.2). You could also delete it completely. If the GPU frequency is detected right, the CPU load shouldn't increase (but GPU load may rise slightly).
ID: 15432 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Brickhead
Avatar

Send message
Joined: 20 Mar 08
Posts: 108
Credit: 2,607,924,860
RAC: 0
Message 15435 - Posted: 15 Mar 2009, 1:30:41 UTC - in response to Message 15420.  

So under Win7 both GPUs of a X2 card are used with Cat 9.1? Can anyone confirm this?

I'm running XP64 (not Win7), but I had to check anyway. As expected, no difference between one display, two cloned, and two expanded.

I noticed a small issue with 0.19e though:

4870X2, 0.19e, cat9.2*, works fine. (And for some reason I see ~60% load now, not ~50%.)
4870/1G, 0.19e, cat8.12, works fine.
4870/1G, 0.19d, cat9.2*, works fine.
4870/1G, 0.19e, cat9.2*, no go. Consistently 0 GPU load following a VPU recover.

(* atical*.dll copied to amdcal*.dll)
ID: 15435 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Cluster Physik

Send message
Joined: 26 Jul 08
Posts: 627
Credit: 94,940,203
RAC: 0
Message 15438 - Posted: 15 Mar 2009, 1:55:33 UTC - in response to Message 15435.  
Last modified: 15 Mar 2009, 1:57:24 UTC

4870X2, 0.19e, cat9.2*, works fine. (And for some reason I see ~60% load now, not ~50%.)

Some progress! Looks like we to have wait for Cat 9.3 to get 70% ;)

4870/1G, 0.19e, cat9.2*, no go. Consistently 0 GPU load following a VPU recover.

Strange. The top host at MW (belonging UL1, SETI.Germany) runs 0.19e on WinXP64 with Cat9.2 and even 3 HD4870 (albeit only the 512MB version overclocked to 810MHz, he had even 4 cards running in that box for a day). He reported no real problems.
ID: 15438 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Thamir Ghaslan

Send message
Joined: 31 Mar 08
Posts: 61
Credit: 18,325,284
RAC: 0
Message 15502 - Posted: 15 Mar 2009, 17:46:00 UTC - in response to Message 15438.  

4870X2, 0.19e, cat9.2*, works fine. (And for some reason I see ~60% load now, not ~50%.)

Some progress! Looks like we to have wait for Cat 9.3 to get 70% ;)

4870/1G, 0.19e, cat9.2*, no go. Consistently 0 GPU load following a VPU recover.

Strange. The top host at MW (belonging UL1, SETI.Germany) runs 0.19e on WinXP64 with Cat9.2 and even 3 HD4870 (albeit only the 512MB version overclocked to 810MHz, he had even 4 cards running in that box for a day). He reported no real problems.


Sorry for the lousy youtube downsizing quality:

http://www.youtube.com/watch?v=8COaVZWyqWc&feature=channel_page

The original file is 50 MB capturing a screen res of 1280x1024, it took youtube 12 hours to process it!

Two instances of GPU-Z read a fluctuation of 70-95% on both cards.

Catlyst reads 80-99% on both cards.
ID: 15502 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
MJD1964

Send message
Joined: 24 Nov 07
Posts: 9
Credit: 102,125,541
RAC: 0
Message 16032 - Posted: 19 Mar 2009, 14:04:06 UTC

I have a couple boards that have extra PCIe slots. What do I need to do to get this to run on both cards at the same time.

No crossfire right?
Do I need to have a monitor pluged into the second card or use a VGA dummy?


Mike...
ID: 16032 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
MJD1964

Send message
Joined: 24 Nov 07
Posts: 9
Credit: 102,125,541
RAC: 0
Message 16347 - Posted: 21 Mar 2009, 0:37:26 UTC
Last modified: 21 Mar 2009, 0:42:41 UTC

I figured it out thanks....

I put a couple of HD4830's together..
ID: 16347 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Previous · 1 · 2

Message boards : Number crunching : 2 x 4870s or 4870X2

©2024 Astroinformatics Group