Good to hear, a Teammate of mine posted in our Teams Forum about error's with the 6.62 with 64 Bit Windows right on top of a Post I made about Updating their Video Drivers to 181.22, I'll let him figure it out on his own if he can't read whats Posted ... ;)

I am not sure that I am really seeing a decrease in productivity. Looking at 14 tasks run on my computer W2 which have been run on my GTX 295 and GTX 280 ... all the tasks seem to be of the same type.

GPU    Total TIme   Step (ms)  version   Task ID
295       32,796        43.72         6.62      259000
295       17,102        34.2           6.61      258856
280       15,996        31               6.61      258699
295       25,243        33.65         6.61      258509
295       17,277        34.55         6.61      257990
280       16,063        32.127      6.61       257975
295       17,339        34.67         6.61       257861
280       17,128        34.258      6.62       257485
295       18,088        36.177      6.62       257364
295       17,624        35.249      6.62       257038
295       17,493        34.982      6.62       257007
295       18,396        36.793      6.62       256746
295       18,401        36.803      6.62       256602
280       17,120        34.24        6.62       256039

As I look at these numbers, granted not a HUGE sample, but, 50 / 50 roughly there is only a slight increase in step size and a minor increase in run time ...

I make it 18,170 sec average (5 hours) and 33.36 ms for the 6 each 6.61 tasks in the list...
and 19,630 sec average (5 hours 27 Minutes) and 36.52 ms for the 8 each 6.62 tasks ...

which is a change of roughly 8-9% ...

The one other point that I would raise is that there is sufficient variation in the runtimes and time steps that the actual LONG TERM values may even be lower ... but the tedium of gathering the data means that I am not sure that there is sufficient reason to dig deeper.

Also not clear is the impact on the actual daily average credit. I have only been doing GPU Grid for a little over a month and the numbers on the stat sites don't seem to align with what I am seeing in my account, though I have not tried to do my own data capture... worse, I have been moving cards around and increasing the number I have which also makes my account a poor candidate to establish a baseline this way ...

If the change impacts credit grant I would expect my daily current average of about 43.5K per day would drop to about 40K ...

Another caveat is that I have two different classes of card doing the processing ... anyway, make of it what you like ... I am seeing a very small increase in GPU time and a huge decrease in CPU usage ... I am not sure for the credit hounds if the one balances the other ... but for those of us that are supporting multiple projects I still think that this is a positive development ...

{edit}

I was getting about 4.8 tasks done per day per core with a total of 14.4 total per day (6.61)
With the 6.62 version and an increase of 30 min per task my calculations indicate this drops to 12.9 total per day or a reduction of 1.5 tasks per day ...

Meaning if I exchange the 280 for another GTX 295 I will still see a significant increase in processing... but that is just me I suppose ....

After 6.61 I was about to hang up on GPUGRID but with the 6.62 I'm back on board full force.
6.62 is definitely on the right track.
Keep up the good work!

---

Proud Member of Team [H]ard|OCP and the [H]ard DC Commandos

Excellent results here on the first 6.62 WU. Has made my system much more responsive for the additional Boinc clients I run.

System:XPPro SP2 x86, Boinc 6.4.2 (running 4+1) 9600GSO 575/1540/850 Driver 177.84

Results: WU M13486-SH2_US_3-8-40-SH2_US_31000000_0
6.61: CPU Time - 61773.92, Elapsed Time - 73545.438, MS/Step - 98.061
6.62: CPU Time - 1247.219, Elapsed Time - 75388.047, MS/Step - 100.517

Old CPU efficiency (as reported via Boincview) with 6.61 for 4x CPU clients averaged 67%

CPU efficiency with 6.62 for 4x CPU clients back to their normal 98%

Overall 6.62 has provided a 4.6% increase in CPU efficiency for a 2.4% decrease in GPU production. Great work on this client. I have 1 more 6.62 currently running with 2 more in queue. I'll followup if there is any notable change.

Sorry, have not the same WU name.

First 6.62-WU with WinXP/32, nVidia 181.22: 241966 - 2478 credits
- GTX 260/192 @666/1512/1150
- CPU time: 366.7031 (3+1 configuration)
- Time per step: 43.210 ms
- Approximate elapsed time for entire WU: 21605.234 s

"Old" 6.61-WU with WinXP/32, nVidia 181.20 (think so): 243068 - 2478 credits
- GTX 260/192 @666/1512/1150
- CPU time: 9720.516 (3+1 configuration)
- Time per step: 35.068 ms
- Approximate elapsed time for entire WU: 17534.094 s

Next two 6.62-WUs at the same computer (WindowsXP/32, 181.22)
239626 - 2478 credits:
- GTX 260/192 @666/1512/1150
- CPU time: 366.4062
- Time per step: 36.637 ms
- Approximate elapsed time for entire WU: 18318.438 s

258100 - 2478 credits:
- GTX 260/192 @666/1512/1150
- CPU time: 545.2812 (switched back meanwhile to 4+1 configuration using manager 6.5.0)
- Time per step: 38.747 ms
- Approximate elapsed time for entire WU: 19373.393 s

v6.62 is actually a bit faster here:

My last v6.61 WU:

CPU time	47005.82

# Using CUDA device 0
# Device 0: "GeForce 9600 GSO"
#  Clock rate: 1674000 kilohertz
#  Total amount of global memory:                 402325504 bytes
#  Number of multiprocessors:                     12
#  Number of cores:                               96
# Time per step: 	110.087 ms
# Approximate elapsed time for entire WU:  	55043.295 s

Validate state	Valid
Claimed credit	2478.98611111111
Granted credit	2478.98611111111
application version	6.61


My first v6.62 WU:

CPU time	546.0781

# Using CUDA device 0
# Device 0: "GeForce 9600 GSO"
#  Clock rate: 1674000 kilohertz
#  Total amount of global memory:                 402325504 bytes
#  Number of multiprocessors:                     12
#  Number of cores:                               96
# Time per step: 	108.343 ms
# Approximate elapsed time for entire WU:  	54171.699 s

Validate state	Valid
Claimed credit	2478.98611111111
Granted credit	2478.98611111111
application version	6.62

CPU time was cut from 47005 to 546, what an improvement!

I confirm 6.62 working fine on XP 32-bit with 8800GT. CPU usage usually below 2% on a Athlon64x2 4400+. I'm happy!

No edit button?
Anyway, just wanted to add: Great work, GDF! Thank you very much!

No edit button?
Anyway, just wanted to add: Great work, GDF! Thank you very much!

You can only edit a post for up to an hour after posting... then no more edits ... have to make a new post ...

But the sentiment is correct ... great work ...

my daily number went down by about the amount I would have expected. Tomorrows will be lower as I did an experiment with SETI@Home Beta with CUDA and have had a new experience with the world of bugs ...

The good news is that I ordered another GTX 295 this morning ... should be here mid week next and i can get my number back up ...

Here my values for the 6.62 on Vista 64:
For my GTX280 the 6.62 needs 4 to 5% CPU-Power of one core on a Phenom 9950 BE (2,6 GHz), for my GTX260² is one core needed with 2 to 3% on a Phenom II 940 (3,0 GHz) and for my 8800GT needs one core of a Phenom 9850 BE 2 to 3% CPU power.

The Linux version 6.59 needs on my Intel Q9550 (3,2 GHz) for a GTX260 4 bis 7% of a core, the Windows version is now better than the Linux version here.

---

Finished one on on 6.62 XP64 bit, 181.22 Nvidia driver. All good.

Finished one on on 6.62 XP64 bit, 181.22 Nvidia driver. All good.

It appears that there never was a problem with v5.56 in XP64 either, just a bug in the NVidia drivers.

v6.62 is actually a bit faster here:

I think I'm finding the same thing. I don't see things slowing down. I compared several of my 6.62 "appx elapsed times" and see they dropped about 1500-2000 seconds.

v6.62 is actually a bit faster here:

I think I'm finding the same thing. I don't see things slowing down. I compared several of my 6.62 "appx elapsed times" and see they dropped about 1500-2000 seconds.

The problem is that there is variability in the total execution times of the tasks apparently depending on the complexity of the model or other factors which may be the interaction of the tasks with the system and OS ...

This is why I looked at 14 tasks and calculated an average between the 6.61 and 6.62 versions. To be more fair I should have sampled more, but, I had not done that many 6.62 tasks by that time so I would have had problems coming up with enough tasks done by that application...

Just small info:

Also, let us know if any Win Xp driver works.
gdf

I know that`s pretty late, but good news are always good news I hope. 181.22 works absolutelly perfect for WinXP Pro SP2 and SP3. I`ve already crunched many WU`s with this driver.

About Win x64 I can just write that I don`t see anything better than 178.24. I know this driver is not perfect, but was best for me until now.

---

Have anybody seen a difference of WU-time from 19000s - 30000s on a GTX260 with the same credit rate (4+1)? Normally Iam loosing around 2000s but this is mystic.

Let's see if I have any useful performance data:

6.62: 4 WUs with 2,478.99 credits

69.540 ms | QlY2031-SH2_US_2-12-40-SH2_US_2760000_0
71.756 ms | x29439-SH2_US_2-10-40-SH2_US_21450000_0
69.490 ms | c29521-SH2_US_2-8-40-SH2_US_21630000_0
69.725 ms | JA25492-SH2_US_1-27-40-SH2_US_11660000_0

6.61: also 2,478.99 credit WUs
66.994 ms | Zr16723-SH2_US-25-40-SH2_US1900000_0
67.746 ms | Hf22398-SH2_US_5-5-10-SH2_US_52390000_0
67.812 ms | DXS2851-SH2_US_5-2-10-SH2_US_51300000_0
67.726 ms | HWP4428-SH2_US_5-2-10-SH2_US_51390000_1
71.363 ms | v28865-SH2_US_2-7-40-SH2_US_2120000_0
67.069 ms | Ot10079-SH2_US-17-40-SH2_US340000_3
67.835 ms | aBL4709-SH2_US_2-4-40-SH2_US_21460000_0

All are run in 4+1 mode. I'm seeing fairly consistent times and an average of 70.13 ms/step with 6.62 versus 68.08 ms/step with 6.61. That's a performance loss of 3%, or 2.2% if the 71.756 ms value from 6.62 is omitted (may have been playing Civ 4).

System responsiveness ahs not been terrible before and is not great now, but it's certainly improved. Overall I appreciate the new client.. for me a 3% loss on the GPU is more than made up for by the benefits of the new client.

And an interesting question: which part of the WU name actually tells me if they're the same? Judging from the runtime my WUs were surely similar enough of a comparison.

Edit: still using 178.24 on XP32, no problems.

MrS

---

Scanning for our furry friends since Jan 2002

All are run in 4+1 mode. I'm seeing fairly consistent times and an average of 70.13 ms/step with 6.62 versus 68.08 ms/step with 6.61. That's a performance loss of 3%, or 2.2% if the 71.756 ms value from 6.62 is omitted (may have been playing Civ 4).

System responsiveness ahs not been terrible before and is not great now, but it's certainly improved. Overall I appreciate the new client.. for me a 3% loss on the GPU is more than made up for by the benefits of the new client

My data showed a slightly larger loss of as much as 9% ... yet my daily numbers don't reflect that ... so, I am not sure what the reality is ...

The problem as I see it is that we have slight instability in the run times of the tasks to begin with and that makes it hard to really pin down the "true" performance. Of course the other problem is that you also get the issue where one task or more falls over the boundary so that you have as more tasks done on some particular days than others.

Even worse for my account is that I have been changing the number and class of the GPUs I have on hand so that the numbers keep shifting ... and will change again this week ... :)

Interesting times ...

Compared to previous versions / performance investigations I find these runtimes to be surprisingly consistent. But maybe that's just because I don't have time to actually use my machine over the week.. :p

MrS

---

Scanning for our furry friends since Jan 2002

Another one with doubled time + same credits, I cant explain me why ^^
http://www.ps3grid.net/result.php?resultid=264367