There have many changes, but most on the science, nothing apparent from a user perspective if not that we report the number of shader cores.

gdf

Hope, these won't bomb out with the 'out of memory' error so often (preferrably ever).

Greetings,

---

I still see 6.48, 6.47 listed on the applciation's page.

NO 6.51 or 6.52 are shown.

When will these be avaialble ?

My mistake. Here they are.
gdf

My first 6.52 ran fine in windows 7 x64 with BOINC x64 6.3.21. I did notice in the info returned for the task it said I had 16 multiprocessors. Is this something internal for the GeForce 8800 GTS 512, or is it supposed to be the number of cores in the CPU? It did correctly detect my shader speed.

GPU internal: 16 multiprocessors with 8 shaders each -> 128 shaders.

MrS

---

Scanning for our furry friends since Jan 2002

Well, what did you do with this app?
Since I crunched some 6.52 WUs now I noticed a quite big increase in runtime on my 9600GT, an increase from 24 to 36 hours!
For example these two WUs here: WU and WU.
The approximate runtime therein fits the real runtime very good, indeed they did run on high priority, not only with 0.9 CPU usage and no manual stops between...
All other Wus before with the 6.48 app were quite faster and crunched in a day, like this one.
Did anyone else noticed such a huge runtime increase or am I the only one?

---

Member of BOINC@Heidelberg and ATA!

I haven't seen any change in runtime with the new applications. Are you sure this is since the new app is out and not since CPU < 1 ?

---

pixelicious.at - my little photoblog

"Changes in science"... IMHO, it would be nice to read more about project science in this forum or on gpugrid.net site in common...

Are you sure this is since the new app is out and not since CPU < 1 ?

I am. The change for cpu<1 was on 25.10., that's over 2 weeks ago now. I crunched several WUs since then, all without problems and within one day.
I've just looked with RivaTuner about something odd, but the only thing I notice now is a slight more video memory usage of 124 MB instead of 115 before with the 6.48. That can't be the reason obviously. ;)
I remember one of the elder app versions also did extend my WU crunch time enormously, fortunately it didn't stayed for long.
I don't know the version now anymore, but can it be that it is the same mechanism now again?

---

Member of BOINC@Heidelberg and ATA!

Your change doesn't correspond to version number. Your quicker times were returned on or before November 4th. November 5th and on you have returned both 6.48 and 6.52 units that are longer (time step jumps from about 110 to about 155). I do not have any changes in speed with my 9600GSO over this same period (time steps consistent around 86).

Have you considered machine specific possibilities? I would assume that you aren't using the GPU more heavily (e.g., gaming, CAD, etc.) recently. Perhaps you have developed an overheating issue or something similar?

Regarding to add more on science.

There are the publications for now. We will try to add more readable material as commented movies and extracts of papers, possibly using a forum for it.

gdf

DrNow, sometimes the clock of your 9600GT is reported as 1.2 GHz and sometimes 1.8 GHz. Could it be that it clocks down during the slow WUs, due to some changes in the power saving setting or the driver?

Took another look: both slow WUs are crunched entirely at 1.2 GHz, whereas the fast one has 1 1.2 GHz and 2 1.8 GHz entries.

MrS

---

Scanning for our furry friends since Jan 2002

Your change doesn't correspond to version number. Your quicker times were returned on or before November 4th. November 5th and on you have returned both 6.48 and 6.52 units that are longer (time step jumps from about 110 to about 155).

Well, I'm pretty curious about that, after I did a closer look on the WUs now. I didn't do anything that it could have changed...

Have you considered machine specific possibilities? I would assume that you aren't using the GPU more heavily (e.g., gaming, CAD, etc.) recently. Perhaps you have developed an overheating issue or something similar?

Well, in the last days I have played a little bit more, but when I do so, I turn the GPU-crunching off, because my card is too slow for doing both and becomes very sluggish then.
But that was only about 2 or 3 hours per day in maximum, couldn't be the reason for such big WU-extending.
And I don't think the card was overheated, I check the temp regularly, it's mostly about 70°, and that's deep in the green area for my card.

DrNow, sometimes the clock of your 9600GT is reported as 1.2 GHz and sometimes 1.8 GHz. Could it be that it clocks down during the slow WUs, due to some changes in the power saving setting or the driver?

That's pretty strange. I didn't change anything, I don't know what could have caused that.

Well, I wait until the actual WU has finished (will be somewhat this evening maybe), then I make a project reset to see if it changes something.
I don't know if it's possible but maybe the software from 6.48 was a bit messed up with files from 6.52... ?

---

Member of BOINC@Heidelberg and ATA!

If the GPU speed changes I would guess that it has something to do with power saving settings in the NVIDIA drivers. Which driver version are you using?
I had a simmilar problem with one of the first Linux 64 drivers for my GTX 260...

---

pixelicious.at - my little photoblog

If the GPU speed changes I would guess that it has something to do with power saving settings in the NVIDIA drivers. Which driver version are you using?

I just got home and checked here.
Well, I'm using driver version 177.84, stefan@home.
So far the word "power saving setting" reminded me on something.
So I did a look in the systray nvidia icon and just found out, that the system performance setting there was checked for "quality", not for "best performance"!
Oh crap, I wasn't aware that this was changed, at least not from me, maybe this was the reason for the slowdown?
Drat, I'm not sure about it. I just changed it to best performance and counted the seconds from one percentage progress to the next, but it seems it hasn't changed at all, it's the same time difference like before...

---

Member of BOINC@Heidelberg and ATA!

Damn, after it didn't change, I think I found out now what was the cause for the slowdown.
It weren't the settings per nvidia driver, it was the RivaTuner I installed!
That was at the sixth of november. According to my logs that's exactly the date where the slowdown started.
So I have to deinstall that thing again!
Anyway, I only needed it to check things, after the 6.48 caused me some broken WUs.
Sorry for the mess, people... ;-)

---

Member of BOINC@Heidelberg and ATA!

I will post this here again because its not a BOINC client problem:

Happening with running a GPU WU:
if Iam running GPU and download some WUs in high priority mode because just of a short deadline GPU is going to waiting mode while there is a need/must of working on other WUs (sample: last seen with evolution WUs from yoyo, runtime 13min, 0.3d cache, got 20)

The question is why? Also GPU depends too much of CPU power (0.90), this must be reduced very soon.

So there are 2 existing issues:
- scheduling policy between gpu/cpu
- high cpu load for running gpu

I will post this here again because its not a BOINC client problem:

Happening with running a GPU WU:
if Iam running GPU and download some WUs in high priority mode because just of a short deadline GPU is going to waiting mode while there is a need/must of working on other WUs (sample: last seen with evolution WUs from yoyo, runtime 13min, 0.3d cache, got 20)

The question is why? Also GPU depends too much of CPU power (0.90), this must be reduced very soon.

So there are 2 existing issues:
- scheduling policy between gpu/cpu
- high cpu load for running gpu

The 0.90 does not mean the app uses 0.90 (90%) of the cpu. That number is only for boinc to reserve 0.90 for scheduling purposes. Normally you run a cpu job using 1 cpu (or core). so your max would be 2.00 (Lets assume a dual core for this). If you have 2 cpus this means it will schedlue 2 cpu jobs + 1 cpu/gpu job for a single gpu. If you have 2 gpus it means it will schedule 1 cpu job + 2 cpu/jobs using 1.80 (0.90 x 2) for cpu scheduling. With the gpu's cpu usage is up to 0.99 over your max cpus, so for a dual the max boinc will schedule is 2.99. The numbers shown, in this case, can be any number between 0.01 and 0.99 to cause this. If the projects sets it to 1.00, the task will be treated as any normal cpu task and you max would be a flat 2.00. Read the new rules in the boinc wiki for how scheduling is done using the new clients 6.3.XX and above. I've put that link in the forum somewhere a few times, here it is again http://boinc.berkeley.edu/trac/wiki/GpuSched

Look in windows task manager to see the actual cpu useage. The high load of 10-40% is a windows problem, linux only uses 2%.