All WU's completed & validated thus far on my GTX980 with beta app versions 8.44, 8.45 and 8.46. I'm running windows 8.1 and nvidia drivers v. 344.16.

http://www.gpugrid.net/results.php?hostid=142719

All the beta units are finishing valid. Though, the output files are rather large, 44 Megabytes.

52-MJHARVEY_TEST4000-0-10-RND4601_3
Workunit 10123299
Created 29 Sep 2014 | 12:46:43 UTC
Sent 29 Sep 2014 | 19:41:27 UTC
Received 29 Sep 2014 | 22:41:52 UTC
Server state Over
Outcome Success
Client state Done
Exit status 0 (0x0)
Computer ID 127986
Report deadline 4 Oct 2014 | 19:41:27 UTC
Run time 6,363.92
CPU time 6,033.14
Validate state Valid
Credit 1,500.00
Application version ACEMD beta version v8.46 (cuda65)
Stderr output
<core_client_version>7.2.42</core_client_version>
<![CDATA[
<stderr_txt>
# GPU [GeForce GTX 690] Platform [Windows] Rev [3212] VERSION [65]
# SWAN Device 1 :
# Name : GeForce GTX 690
# ECC : Disabled
# Global mem : 2048MB
# Capability : 3.0
# PCI ID : 0000:04:00.0
# Device clock : 1019MHz
# Memory clock : 3004MHz
# Memory width : 256bit
# Driver version : r343_98 : 34411
# GPU 0 : 63C
# GPU 1 : 73C
# GPU 2 : 74C
# GPU 3 : 74C
# GPU 0 : 64C
# GPU 0 : 65C
# GPU 0 : 66C
# GPU 0 : 67C
# GPU 0 : 68C
# GPU 0 : 69C
# GPU 0 : 70C
# GPU 0 : 71C
# Time per step (avg over 2500000 steps): 2.549 ms
# Approximate elapsed time for entire WU: 6371.977 s
# PERFORMANCE: 23558 Natoms 2.549 ns/day 0.000 ms/step 0.000 us/step/atom
18:24:40 (5228): called boinc_finish

</stderr_txt>
]]>

What's the meaning of ns/day performance? Number is same as time (ms) per step.

23558 Natoms 4.726 ns/day-GTX650Ti
23558 Natoms 2.549 ns/day-GTX690
23558 Natoms 1.633 ns/day-GTX980

It completed OK on the GTX 650 Ti, but seems to be causing problems on some higher-end cards. But their versions of ACEMD probably have more changes than the one I got (8.46).
http://www.gpugrid.net/workunit.php?wuid=10123336

I will be trying my GTX 660 Ti next on the same machine to see what happens.

I think the errors on the higher-end cards where caused by to old drivers Jim. I had a lot errors on my 780Ti's yesterday, but when I updated to the latest driver, they run smooth as usual again.
The beta did okay on my 660, so your 660Ti will do great as well.

---

Greetings from TJ

TJ,

Thanks, that is probably it. My GTX 660 Ti did finish fine; I will be trying a couple of GTX 750 Ti's now just for fun.

Just enabled Test Apps for my GTX 680 and GTX 780Ti cards. I'll check back in a while to see how they're doing.

Edit: I saw that TJ recommended updating to the latest drivers. Is this the latest Beta or WHQL driver? I'm currently running 344.11. Thanks.

Edit: I saw that TJ recommended updating to the latest drivers. Is this the latest Beta or WHQL driver? I'm currently running 344.11. Thanks.

344.11 works fine on my GTX 650 Ti and 660 Ti on the test apps. I am running it on my GTX 750 Ti also, but haven't picked up the new apps yet

Thanks, Jim1348. I'll stick with 344.11 for now, then.

Thanks, Jim1348. I'll stick with 344.11 for now, then.

What other options are there? :)

Just enabled Test Apps for my GTX 680 and GTX 780Ti cards. I'll check back in a while to see how they're doing.

Edit: I saw that TJ recommended updating to the latest drivers. Is this the latest Beta or WHQL driver? I'm currently running 344.11. Thanks.

Hello Matt, yes I am running 344.11 the latest WHQL driver. But to be clear it was recommended by Matt from the project.
The older driver I was using, was a bit faster on Win7 as the WDDM was introduced with Vista and can not be switched of, but that is besides the scope of this thread.

---

Greetings from TJ

If I've got things right, the 65 apps shouldn't be sent any driver older than 343.00. The exception to that will be the Linux app, when that finally exists. That will give the WU out to any client that reports CUDA 6.5 capability, as only our patched client reports the driver version.

Matt

If I've got things right, the 65 apps shouldn't be sent any driver older than 343.00. The exception to that will be the Linux app, when that finally exists. That will give the WU out to any client that reports CUDA 6.5 capability, as only our patched client reports the driver version.

Matt

Well Matt with driver 331 on my 780Ti's win7 where a bit faster but then I got cuda65 tasks and errored out. With your advice I updated the driver and no more errors (yesterday one, but that was another reason).
But if you have made changes yesterday or today, then you are probably right.

---

Greetings from TJ

I think it's safe to promote the CUDA6.5 application to the long queue.

Not just yet...

If I've got things right, the 65 apps shouldn't be sent any driver older than 343.00. The exception to that will be the Linux app, when that finally exists. That will give the WU out to any client that reports CUDA 6.5 capability, as only our patched client reports the driver version.

Matt

boinc 7.4.22 (development version) now reports the driver version:

Starting BOINC client version 7.4.22 for x86_64-pc-linux-gnu
CUDA: NVIDIA GPU 0: GeForce GTX 780 Ti (driver version 343.22, CUDA version 6.5, compute capability 3.5, 3072MB, 2814MB available, 5345 GFLOPS peak)

Shows up here too:

http://www.gpugrid.net/show_host_detail.php?hostid=183991

MJH:

I've been processing Beta tasks, and although nearly all are successful for me on the 8.46 app, I did have a failure last night. This is on a completely-stable Windows 8.1 Update 1 x64 machine, on one of my GTX 660 Ti GPUs, using 344.11 driver.

Any ideas?

http://www.gpugrid.net/result.php?resultid=13154266

Name 79-MJHARVEY_TEST4001-2-10-RND8149_0
Workunit 10126844
Created 30 Sep 2014 | 18:55:59 UTC
Sent 1 Oct 2014 | 3:25:13 UTC
Received 1 Oct 2014 | 4:35:24 UTC
Server state Over
Outcome Computation error
Client state Compute error
Exit status -97 (0xffffffffffffff9f) Unknown error number
Computer ID 153764
Report deadline 6 Oct 2014 | 3:25:13 UTC
Run time 1,431.05
CPU time 384.63
Validate state Invalid
Credit 0.00
Application version ACEMD beta version v8.46 (cuda65)
Stderr output

<core_client_version>7.4.22</core_client_version>
<![CDATA[
<message>
(unknown error) - exit code -97 (0xffffff9f)
</message>
<stderr_txt>
# GPU [GeForce GTX 660 Ti] Platform [Windows] Rev [3212] VERSION [65]
# SWAN Device 2 :
# Name : GeForce GTX 660 Ti
# ECC : Disabled
# Global mem : 3072MB
# Capability : 3.0
# PCI ID : 0000:08:00.0
# Device clock : 1045MHz
# Memory clock : 3004MHz
# Memory width : 192bit
# Driver version : r343_98 : 34411
# GPU 0 : 69C
# GPU 1 : 64C
# GPU 2 : 69C
# GPU 1 : 65C
# GPU 1 : 66C
# GPU 1 : 67C
# GPU 0 : 70C
# The simulation has become unstable. Terminating to avoid lock-up (1)
# Attempting restart (step 5505000)
# GPU [GeForce GTX 660 Ti] Platform [Windows] Rev [3212] VERSION [65]
# SWAN Device 2 :
# Name : GeForce GTX 660 Ti
# ECC : Disabled
# Global mem : 3072MB
# Capability : 3.0
# PCI ID : 0000:08:00.0
# Device clock : 1045MHz
# Memory clock : 3004MHz
# Memory width : 192bit
# Driver version : r343_98 : 34411
# The simulation has become unstable. Terminating to avoid lock-up (1)

</stderr_txt>
]]>

Question: would a 4.2CUDA long task running on one card slow down CUDA 6.5 short or Beta tasks running on other or vise versa? I just noticed a CUDA 4.2 Noelia Long task running, with 6.5 Beta and Short. Runtime for Long task is more than normal. It takes ~40Hr to complete, but at ~40Hr the 4.2 task is 80%.

Maybe, if the processes are competing for CPU.

Matt

Any idea why my task failed, 3 posts up?

Any idea why my task failed, 3 posts up?

Have you checked event viewer to locate any occurrences at the time task failed? Any kernel failures ? Or database instances? If you have automatic windows updates or auto Maintenance enabled- this can trigger random failures for other processes. (or sometimes fault any heavy usage process) Also, a security "audit" can trigger background task (GPUGRID) failures.