Mine is only an Intel P4(HT) at 3.8GHz on Windows XP (32bit). I've been using it lightly off and on all day.

I'm up to 35:52 at 54% done.

I can understand some difference in P4's vs Quad Cores or linux vs windows

But still this is a big difference.

Why would an AMD show so much less useage ?

Even if it were 50% less I might understand, but this is like 99.6% less.

Mine is only an Intel P4(HT) at 3.8GHz on Windows XP (32bit). I've been using it lightly off and on all day.

I'm up to 35:52 at 54% done.

I can understand some difference in P4's vs Quad Cores or linux vs windows

But still this is a big difference.

Why would an AMD show so much less useage ?

Even if it were 50% less I might understand, but this is like 99.6% less.

Well if the 1st one is puzzling you, then the 2nd workunit at 9.53 seconds is going to add even more confusion ;)

---

Down with the Kredit Kops!!!

Application 6.45 under Windows Vista 64 is much slower (about 56.000 Seconds - old one: about 30.000 Seconds), and uses much less GPU.

WU: http://www.ps3grid.net/workunit.php?wuid=44433

only 74 Degrees with Fan Speed of 700 RPM - old client (with one CPU-Core usage) 83 Degrees and 1000 RPM.

Is there a new version coming regarding this issue?

Application 6.45 under Windows Vista 64 is much slower (about 56.000 Seconds - old one: about 30.000 Seconds), and uses much less GPU.

WU: http://www.ps3grid.net/workunit.php?wuid=44433

only 74 Degrees with Fan Speed of 700 RPM - old client (with one CPU-Core usage) 83 Degrees and 1000 RPM.

Is there a new version coming regarding this issue?

This is the one that fixes the CPU things.
As all the work is done on the GPU, The cpu just done other things.
Have a look at the cpu usage now on windows, it uses about 1% of the cpu.

And since all the work is done on the GPU, GPU time is more imprortant.
Have a look at one of your Task ID's and you see that info.

---

Down with the Kredit Kops!!!

I am aware of the fact that the CPU-core is freed now, but i just wanted to point out the fact, that the GPU is not utilized to the max with the new client.

Otherwise, there would be more heat generated by the calculation and load on the GPU core like in the old client - even if the parameters of the simulation have changed, leading to the increase in computation time.

As already stated in my previous post, the GPU time it took to calculate the WU was 56002 seconds. The old client took about 30.000 seconds (GPU time AND CPU time, as one core was used 100% we can assume that the used GPU time is proportional to the used CPU time).

The actual running WU also will be finished after around 56k seconds after looking at the calculation time of the first 10%. GPU utilization still stays low.

I am aware of the fact that the CPU-core is freed now, but i just wanted to point out the fact, that the GPU is not utilized to the max with the new client.

Otherwise, there would be more heat generated by the calculation and load on the GPU core like in the old client - even if the parameters of the simulation have changed, leading to the increase in computation time.

As already stated in my previous post, the GPU time it took to calculate the WU was 56002 seconds. The old client took about 30.000 seconds (GPU time AND CPU time, as one core was used 100% we can assume that the used GPU time is proportional to the used CPU time).

The actual running WU also will be finished after around 56k seconds after looking at the calculation time of the first 10%. GPU utilization still stays low.

In our tests the speed of the application was not affected by the better utilization of the cpu core. Let's run another wu and see. Also, you have far too many compute error on your machine. You should expect zero errors.


gdf

Those compute errors were from a) from using another CUDA dll (trying to speed up the computation) and b) a faulty power supply in my computer. As you can see, all WUs have completed successfully a week ago (old app, working power supply ;-)

Dont get me wrong - 5000 Credits a day additional to CPU-crunched WIs is still a lot, but i don´t think the client makes use of the full capabilites of the GPU anymore.

Well - i´ll give it another chance for a few more days and hope its only an issue of the current workunits.

Edit:
i just checked with a fellow from my team. Same OS (Vista x64) but his Setup using a 9800 GTX has a) MUCH lower CPU Usage and b) MUCH faster computation on the GPU. (about 46-47 ms per Step).

His results:
http://www.ps3grid.net/result.php?resultid=58207
http://www.ps3grid.net/result.php?resultid=57930

What could be the reason? Nothing changed on my system instead of the new client.

I FOUND THE PROBLEM

After Rebooting the machine, windows firewall popped up "acemd_6.45_windows_intelx86__cuda.exe blocked...allow?"

I allowed the process, and voila - GPU Utilization went up, Temperature went up, computation time (estimated) went down to 30.000 - 31.000 seconds.

CPU Utilization also dropped from about 3/4 of an hour per WU to about nothing..

VERY strange!

Someone know about that issue?

Overall I'm impressed with this near zero CPU utilization release! Too many obvious good points.

However, I have one concern.

My scheduler got messed up!

To clarify:

Before this release, I gave ps3grid 25% more resource priority than other projects, and another 25% went to burp.boinc.dk, the rest of the 10s of projects each get less than 3%.

This made ps3grid run non stop 24/7.

Now, the CPU time dropped from 7 hours to 12 minutes, and each time a ps3grid task would finish the scheduler would think that the rest of the ps3grid tasks on queue can wait because it thinks it would take 12 minutes, and not 7 hours to complete.

Even if I further increase resource sharing for ps3grid the scheduler would still lower the priority because of this 12 minutes cpu completion!

Any workarounds?

It seems like I'm way slower with the new app. Previously I had ~44.000s / 12h. My first 6.45 WU came in at 60.000 s / 70 ms, but I was gaming for 1 or 2 (probably 2 :p ) hours, so this may have messed the time up. The 2nd one was 53100s / 62 ms and the machine was almost undisturbed. I checked the wall clock time between start and finish of the WU, it's 14:46h, which agrees very well with the time reported by the app. So this speed decrease is real!
And my room temp dropped ~1°C but the GPU ~5°C, so I think it's safe to say it's used less now.

My config: XP32, SP2, firewall deactivated, 6.3.10, 177.92, 9800GTX+ w/o OC. I rebootet last weekend but should probably do it again, just to be sure. Oh, and CPU utilization was very low for the 2nd result, just 14.9s.

MrS

---

Scanning for our furry friends since Jan 2002

Computation time increased once again, seems to be a problem with the Windows Scheduler.

Although one Processor core (of 4 in my quad) is unused: as early as i start another task on the computer which takes one core (single-threaded-mp3 encoder using only one core for example) PS3Grid Application drops in performance again. Also as earlier visible a significant drop in GPU temperatures. After exiting the encoder, GPU utilization goes up again.

Manual priorization of the process even to "highest" or "real time" doesn´t help.

Heres an image - temperature increases again as the mp3-encoder is closed.

I also tried to increase the priority of the CPU part. However, I couldn't: "access denied".

MrS

---

Scanning for our furry friends since Jan 2002

I also tried to increase the priority of the CPU part. However, I couldn't: "access denied".

MrS

Tonight, we will upload a new windows application that will remove the firewall issue, hoping that this is the cause of the slower application (I doubt it).

gdf

It appears to me 6.45 is slower.

Home #0 took 55,000s before
Result 58254 CPU 3723s GPU 58289s 68ms
Result 58428 CPU 3436s GPU 59270s 69ms

Work #1 took 59,000s-60,000s before
Result 58017 CPU 285s GPU 68463s 80ms
Result 58579 CPU 203s GPU - ERROR - See Below

Work #2 took same as #1
Result 58264 CPU 3456s GPU 61631s 72ms

Work #1 and #2 are both same GPU and only a little slower than Home #0. Not counting the GPU slowdown time, they fall in an understandable range. The CPU time also, except for the one that is 285s (4m) vs 3456s (57m)? These two hosts are nearly identical in hardware and software, one may have a few extra applications.

This is the same question I asked before when another user reported 16s vs my estimated time, actual above is 62m and 57m (an hour).

Why this wild difference in CPU times ?

I would expect work #1 and work #2 to be nearly identical and only a little longer, since they are a little slower, than home #0.

---
Also this drill down into each work unit to get the GPU time is not liked. It is going to take a lot of extra effort, especially with multiple hosts during testing to see what is going on. I think it would be desirable to have the GPU time and maybe the ms step included in the table of result information next to the CPU time. I know this will take a custom hack as it's not part of the BOINC default web layout, but give it some thought.

My other thought too was to have version number in that table, especially during all this testing. So you can know which results are from older apps and easily see differnece in run times vs app version.

I'll add these suggestions on the wish list.

---

Error:
<core_client_version>6.3.10</core_client_version>
<![CDATA[
<message>
Incorrect function. (0x1) - exit code 1 (0x1)
</message>
<stderr_txt>
# Using CUDA device 0
# Device 0: "GeForce 8800 GT"
# Clock rate: 1512000 kilohertz
MDIO ERROR: cannot open file "restart.coor"
Cuda error: Kernel [frc_sum_kernel_dihed] failed in file 'force.cu' in line 568 : unspecified launch failure.

</stderr_txt>
]]>
No credit and no GPU time listed, so I do not know how far along task was when error occured.

It appears to me 6.45 is slower.

Home #0 took 55,000s before
Result 58254 CPU 3723s GPU 58289s 68ms
Result 58428 CPU 3436s GPU 59270s 69ms

Work #1 took 59,000s-60,000s before
Result 58017 CPU 285s GPU 68463s 80ms
Result 58579 CPU 203s GPU - ERROR - See Below

Work #2 took same as #1
Result 58264 CPU 3456s GPU 61631s 72ms

Keith, was the Work #2 machine running anything else, or being used interactively whilst it was crunching?

Matt

For the CPU time difference.
The machine that had the 16 Second time (9 second lowest), all it does is crunch.
So there is no activity on the screen for it to update.

I have noticed when looking at task manager on my main PC. the PS3GRID app sits at 0% CPU load, however when something happens on screen the cpu load increases.

So that will explain why my AMD has such a low CPU time, since I never use the computer to do anything (apart from Boinc), the screen will hardly ever change, resulting in 9 seconds of cpu time, and since my Intel is the PC I use the most, the screen is updated quite a lot (web pages, msn, video, etc) it ends up with the higher CPU time of 44 minutes.

And yes 6.45 is slower.
Looking at my Boincview logfiles, there used to be just under a 12 hour gap between workunits, now it has gone to 14.5 hours between workunits :(

---

Down with the Kredit Kops!!!

epsecially to GDF

The increase in computation time did not result from the firewall, this was a false positive from me - sorry for that. Here is true reason (at least it was reproducable by many others on their computers). Maybe its a issue of windows scheduler (maybe only on vista - don´t know right now if it happened also on xp, will look into it).

http://www.ps3grid.net/forum_thread.php?id=371&nowrap=true#2387

To add my 2 cents, I also encounter slowdown with 6.45 with a 9800 GT on XP32 (PIV HT 3.00):

Previously with 6.43 around 57k sec, example1

Now with 6.45 around 66k seconds:
- example2
- example3

In the meantime I increased slightly the overclocking, thus the real loss for a defined set of frequencies is somewhat higher.

Hope this helps.

McRoger

Edit: tests done with a machine 100% dedicated to GPUGRID

It appears to me 6.45 is slower.

Home #0 took 55,000s before
Result 58254 CPU 3723s GPU 58289s 68ms
Result 58428 CPU 3436s GPU 59270s 69ms

Work #1 took 59,000s-60,000s before
Result 58017 CPU 285s GPU 68463s 80ms
Result 58579 CPU 203s GPU - ERROR - See Below

Work #2 took same as #1
Result 58264 CPU 3456s GPU 61631s 72ms

Keith, was the Work #2 machine running anything else, or being used interactively whilst it was crunching?

Matt

Yes both work #1 and work #2 are used during an 8 hour day by me at the same time, except that work #2 is used slighly less at times and slightly more at times. Basically both get about the same use.

There is something running at home and on work #2 that is common to both that is not on work #1, but at this time I do not know what it is. Those two exhibit the largely slower CPU times.

Except at home the first task was processed with only 1 CPU running BOINC, not both, so at least 41% of the system was idle.

It's hard to figutre out what may be the cause.

The new Windows-application shows since hours the same percentage and the time to complete is rising. 2 hours ago they show 97,294% and a completion time 4:08, now 97,294% and a completion time of 4:25. The shown CPU-time is 2:30:38, the taskmanager shows 3% for the core. Looks like there is no progress. The real total time is running since 8:44 UTC, now it's 17:34 UTC, that's far too long for my GTX280.

---