The new application for cuda3.1 is now out, please report here.

gdf

Got one 6.11 task.
It's running at 91-93% GPU usage, and completed 50% in 52 minutes. (GTX480 stock clocking, SWAN_SYNC=0)
It's very promising.

Got one 6.11 task.
It's running at 91-93% GPU usage, and completed 50% in 52 minutes. (GTX480 stock clocking, SWAN_SYNC=0)
It's very promising.

It's completed successfully, and much faster (6236 sec) than 6.05 (8200 sec), that's approx. 23% faster.
# Time per step (avg over 1300000 steps): 4.794 ms
# Approximate elapsed time for entire WU: 6232.813 s
GPU load was 90-93%, and the Memory Controller Load was 22%, both of them is higher than it used to be with the 6.05 client (66% and 14%).

Faster depends on what way you work it out!

For your two TONI tasks I make it 23% less time, which is 30% faster

2696626 1716504 20 Jul 2010 9:12:35 UTC 20 Jul 2010 11:04:52 UTC Completed and validated 6,235.95 6,196.56 4,727.93 7,091.89 ACEMD2: GPU molecular dynamics v6.11 (cuda31)

2696083 1716202 20 Jul 2010 6:30:00 UTC 20 Jul 2010 8:51:33 UTC Completed and validated 8,135.92 7,649.42 4,727.93 7,091.89 ACEMD2: GPU molecular dynamics v6.05 (cuda30)

I'm also running a 3.1 task on a GTX470. Its been running for a while and looks slightly faster (but not as much as the TONI tasks); should finish in 7420sec total. It is a different type of WU (468-KASHIF_HIVPR_auto_spawn_2_90_ba1-28-100-RND7028_0). A similar 3.0 task on my card took 8208sec. So the 3.1 KASHIF_HIVPR tasks are about 10% faster than the 3.0 tasks.

GPU says 96% usage, temp 76degC, fan at 83%, clocked to 704MHZ.

Faster depends on what way you work it out!

For your two TONI tasks I make it 23% less time, which is 30% faster

You're right.

Well my first 6.11 task finished in 7402.484sec (as predicted about 10% faster than a similar 6.05 task). I have picked up another 6.11 WU, this time its an IBUCH :)

I see a few people with GTX460 cards are running these 6.11 based tasks, but no results back yet.

My second 6.11 WU was a TONI too, and the speed was the same fast.
I'm very pleased with this performance.
Now I got two new TONIs, but I really like to have a GCY.

Does anyone finish a 6.11 WU under Win7?
I know that CUDA 3.0 tasks finish faster under driver 197.75 than under 257.xx. So I am hesitating to change drivers again. I'd be happy to read about your experience with CUDA 3.1 under Win7.

My card finish v6.11 under Win7
http://www.gpugrid.net/result.php?resultid=2696901 ACEMD2: GPU molecular dynamics v6.11 (cuda31)
http://www.gpugrid.net/result.php?resultid=2692327 ACEMD2: GPU molecular dynamics v6.05 (cuda30)

That is almost 60% faster for the IBUCH tasks under Win7 using the 6.11 app.
It looks like that improvement holds for XP on a GTX470 as well.

roundup, I think it is worth a try; as long as you get plenty of IBUCH tasks, it should be a good improvement. If you get an even spread of tasks (with 10%, 30% and 60% improvement), it should cover the drivers losses (as long as it was not over 33%)!

roundup, I think it is worth a try; [...]

Okay skgiven, you've won ;-). I'll give it a try.

My card finish v6.11 under Win7
http://www.gpugrid.net/result.php?resultid=2696901 ACEMD2: GPU molecular dynamics v6.11 (cuda31)
http://www.gpugrid.net/result.php?resultid=2692327 ACEMD2: GPU molecular dynamics v6.05 (cuda30)

That's about a 40% improvement O_O

Here's the first off my 1GB 460. Slower than I had hoped in Win7. GPU showed about 73% load during the task.

I am going to try some with Swan_Sync=0 now and see what happens.


http://www.gpugrid.net/result.php?resultid=2697154

Vista and Win7 are slow - nothing has changed there; that is 88% slower than a similar 6.05 task I ran on my GTX470.
Remember to free up your CPU; If you run other CPU projects on every CPU core, using swan_sync will make little difference! You have to free up at least one core/thread:
On my i7-920, I leave 2 threads free, and it is significantly faster than with only 1 thread free.

Also, the improvement for GF100 Fermi's was only about 10% for the KASHIF_HIVPR 6.11 WU's (your last run).

The TONI tasks (which you have this time) are about 30% faster on 6.11 than on 6.05, and the IBUCH tasks are about 59% faster. So your task should be a bit faster this time.

This IBUCH_101b_pYEEI task was about 13% faster. We seem to have a large variety of 6.11 tasks and improvements over the 6.05 tasks.

Swan_Sync=0 is showing an 80% GPU load with this KASHIF_HIVPR task compared to the 73% GPU load on the last KASHIF_HIVPR with no Swan_Sync.

So far, so good.
I've got one erratic task, which is failed 3 times already.

I have a nan Invalid task:

Name I529-TONI_KIDln-23-100-RND9840_0
Workunit 1719104
Created 21 Jul 2010 4:49:41 UTC
Sent 21 Jul 2010 5:54:43 UTC
Received 21 Jul 2010 8:44:18 UTC
Server state Over
Outcome Client error
Client state Compute error
Exit status 98 (0x62)
Computer ID 71363
Report deadline 26 Jul 2010 5:54:43 UTC
Run time 2838.5625
CPU time 2835.453
stderr out

<core_client_version>6.10.56</core_client_version>
<![CDATA[
<message>
- exit code 98 (0x62)
</message>
<stderr_txt>
# Using device 0
# There is 1 device supporting CUDA
# Device 0: "GeForce GTX 470"
# Clock rate: 1.41 GHz
# Total amount of global memory: 1341718528 bytes
# Number of multiprocessors: 14
# Number of cores: 112
SWAN: Using synchronization method 0
MDIO ERROR: cannot open file "restart.coor"
ERROR: file deven.cpp line 855: # Energies have become nan

called boinc_finish

</stderr_txt>
]]>

Validate state Invalid
Claimed credit 4727.92939814815
Granted credit 0
application version ACEMD2: GPU molecular dynamics v6.11 (cuda31)

Bikermatt, thats a 9% improvement for that type of task by using swan_sync - Not bad.

This task caused an "acemd2_6.11_windows_intelx86__cuda31.exe has stopped working" error, and I didn't click "OK", I restarted my pc instead. (I'm used to do that every time I see this error message, because if I click "OK" then the task will abort, and the processing time is lost)
After restart, the task is finished well, just like every other task with that error message (and restart).

Yes, I have seen this stopped working error too, and try to do the same thing - a hot restart ASAP usually allows the task to pick up at the last checkpoint. If we had a way to tell a failed task to go back to the last checkpoint it would save quite a few tasks. In Windows there are a lot of false failures, for example in IE, sometimes you get a failure message saying IE has crashed, but if you just click the X in the top right corner, IE works fine. If you click OK Windows closes IE. Unfortunately clicking the X in the case of an acemd task kills it.