We have uploaded the new beta application for windows.
Please report any problem here.

gdf

2 gtx 680 W7

BOINC 7.0.25, Driver= 301.42

Downloaded 4 CUDA 4.0 tasks=failed

Win 7 64bit/ XP sp3, GTX 670, NV driver 301.42, Boinc Mgr. 7.0.25/27/28


GPUgrid keeps to send WUs of ACEMD beta version 6.46(cuda42 OK), and (cuda40 compu. error) to the said computers.


EDIT -- It seems on my computers that Fermi cards work OK with cuda40 and cuda42, but Kelper cards work OK only with cuda42.

---

I've received a couple of Cuda4.0 and Cuda4.2 beta workunints, and they are working fine.
(WinXP 32 bit, GTX 480 @ 800MHz, driver v301.42)
(WinXP 64 bit, GTX 590 @ 725MHz, driver v302.59)
(WinXP 64 bit, GTX 480 @ 800MHz, driver v301.42)
(WinXP 64 bit, GTX 480 @ 800MHz, driver v301.24)
However, the Cuda 4.2 version is much faster than the Cuda 4.0.

I cannot find a single reason while the server gives both applications out to the same computer.

The decision on which application to give is based on driver version and CUDA runtime. So, this should not have changed between one request and another.

Also the difference between cuda4.0 and cuda4.2 seems too much.

gdf

I have done some changes on the server side to see if the application delivery improves.

gdf

I'm still receiving cuda4.0 tasks.
Task 5484191
Task 5484657

Win 7 64bit/ XP sp3, GTX 670, NV driver 301.42, Boinc Mgr. 7.0.25/27/28


GPUgrid keeps to send WUs of ACEMD beta version 6.46(cuda42 OK), and (cuda40 compu. error) to the said computers.


EDIT -- It seems on my computers that Fermi cards work OK with cuda40 and cuda42, but Kelper cards work OK only with cuda42.

Only cuda42 application are received for GTX470+GTX670 and GTX670+GTX580 computers and OK

---

Cuda40 &42 evenly split. XP64,GTX580, 7.0.28, 301.42. All completing OK.

Win7 64bit, GTX 680, Driver 301.42, Boinc 7.0.28
(http://www.gpugrid.net/show_host_detail.php?hostid=125758)

Got 15 WUs (cuda42). All worked fine.

There are two strange things about the cuda4.2 beta client v6.46:
1. My GTX 580 @ 850MHz seems to be faster than your GTX 680. (67 sec vs. 70-71 sec)
2. The beta client seems to be ignoring the swan_sync=0 setting on my hosts.

Cuda 40 = GPU load ~97% CPU load 8% No screen lag, 122secs
Cuda 42 = GPU load ~95% CPU load 12.5% No screen lag, 70 secs

XP64,GTX580, 7.0.28, 301.42.

Just had two cuda40 tasks error. One at 20secs, other at 41secs.
SWAN : FATAL : Cuda driver error 700 in file 'swanlibnv2.cpp' in line 1574.
Assertion failed: a, file swanlibnv2.cpp, line 59

Swan_sync is not enabled on this machine.

Both 680's now receiving 4.2 constantly. So far so good. Runtimes of roughly 60 sec on each one with a 3820@ 4.3.

Looking good guys. Also, no screen lag apparent using W7 with GPU utilization at 91%.

This is with WCG suspended currently. Will be resuming it and see how things go.

EDIT: I've noticed a bug. Whenever one task finishes. The other one TIME freezes. However, the GPU keeps working, and once it un-"freezes", the percent jumps to where it should be. But, the time remains where it last was. This will cause runtimes to vary WIDELY.

EDIT: I've noticed a bug. Whenever one task finishes. The other one TIME freezes. However, the GPU keeps working, and once it un-"freezes", the percent jumps to where it should be. But, the time remains where it last was. This will cause runtimes to vary WIDELY.

Probably a BOINC client bug.

2. The beta client seems to be ignoring the swan_sync=0 setting on my hosts.

The new clients don't honour SWAN_SYNC right now. They automatically set blocking sync (low CPU load).

MJH

2. The beta client seems to be ignoring the swan_sync=0 setting on my hosts.

The new clients don't honour SWAN_SYNC right now. They automatically set blocking sync (low CPU load).

The swan_sync=0 seems to be working with the 6.46 beta client on this host.

Seems there is not lot difference between running a GTX 480 or a GTX 680 on cuda 4.2 , at this time, i think. Will that be improved in future ?

I do not want to rush you :)

Right now it looks like RZs 480 on xp was around 6-7 seconds slower than a 680 on W7. Probably 10 seconds difference if the 680 was on xp. That's pretty big. No? Especially considering the power consumption, which is my favorite factor. :)

Both 680's now receiving 4.2 constantly. So far so good. Runtimes of roughly 60 sec on each one with a 3820@ 4.3.

Would you reveal your computers so that I can see your WUs, please?

MJH

Been looking for that tick for awhile now and still can't find it....

Here's a link to my page though.?

http://www.gpugrid.net/results.php?userid=86780

EDIT: After a bumpy start had about 60 in a row that were all 4.2