I received the same errors on the new 6.48 Application

The output of Stoffelstasks shows the "incorrect function" error, whereas you get "<message> - exit code -1073741819 (0xc0000005)</message>]]>". Don't know what that means though.

MrS

It may have something to do with both Stoffel & my card being older ones, his being a 9800 & mine a 8800GT. I have 2 other Box's with GTX 260's both the 216 Shader variety type that are Processing the 6.48's just fine in under 7 hr's time ... :)

My point is that you both are getting different errors, not the same - despite of what the BOINC manager output says (it can only report the symptom, not the cause). So while you could be correct that the new client may have a problem with older hardware, I expect that this would be a special problem and not a general one. Otherwise we should be seeing many more user complaints.

MrS

---

Scanning for our furry friends since Jan 2002

There is no big difference between my GTX260 and GTX280.

GTX280 Vista64 Clock rate: 1404000 khz, Time per step: 28.564 ms, WU: 24279.624 s
GTX260 XPPro64 Clock rate: 1242000 khz, Time per step: 28.580 ms, WU: 24292.703 s

Both running under Boincmanager 6.3.14 the application 6.48.

The GTX280 has 240 shader and is running with 1404000 khz, the GTX260 has 192 shader is only running with 1242000 khz. The difference in time per WU should be bigger.

The GTX280 times should be ~10-15% faster -- it seems they are a little bit to slow. The speed of the 260 is ok (if it´s a stock GTX260² with 216 shaders)


Cu KyleFL

The GTX280 times should be ~10-15% faster -- it seems they are a little bit to slow.

It was only a single WU anyway, wasn't it?

MrS

---

Scanning for our furry friends since Jan 2002

It was only a single WU anyway, wasn't it?

MrS

Yeas. Meanwhile I have 3 crunched on the GTX280, here the values:

Clock rate: 1296000 kilohertz # Time per step: 28.124 ms WU: 23905.296 s
Clock rate: 1404000 kilohertz # Time per step: 28.564 ms WU: 24279.624 s
Clock rate: 1296000 kilohertz # Time per step: 27.046 ms WU: 22989.391 s

The clockrate in the second WU is manual set to a higher value, also the memory and the shader. Looks like the WUs are very different in the length.

---

a 280 normally clocked is around 24ms on linux. I don't know why your 280 is slower. Next applications will print also the number of shaders. All the wus have the same length, although this is going to change soon.

gdf

Unfortunately 6.45 is a rather significant step backwards for me:

App 6.45: 78.143ms, minutes usage of CPU, run with realtime priority
App 6.48: 87.283ms, ~40% of 1 core, run with standard low priority

As it is now 6.48 runs MORE SLOWLY on my Opteron 165/8800GT while USING MORE CPU time than 6.45. I'm running the next WU using realtime priority to see what difference that makes, which appears to be using up ~80% of 1 core.

I have two 8800GT's, with only one 6.48 result in so far. Given the lack of a large enough sample, I can't make any conclusions yet. However, it appears this WU completed about 13% faster than 6.45's using the same drivers. The CPU usage did increase from 1-2% to about 19% on that core (Q6600, so using about 75% of that core instead of only 5%).

I'll certainly take this increase in performance, if it holds up over the long run. Good job!

I received the same errors on the new 6.48 Application

The output of Stoffelstasks shows the "incorrect function" error, whereas you get "<message> - exit code -1073741819 (0xc0000005)</message>]]>". Don't know what that means though.

MrS

It may have something to do with both Stoffel & my card being older ones, his being a 9800 & mine a 8800GT. I have 2 other Box's with GTX 260's both the 216 Shader variety type that are Processing the 6.48's just fine in under 7 hr's time ... :)

I did a Reboot & Project Reset early this morning on the Box giving the Error's and running a 8800GT OC. I was finally able to get a couple of PS3 Wu's a few Minutes ago and 1 of the Wu's did start up & run 20 or so seconds before the other Project (Poem) took back over and started running it's Wu's in Priority Mode (roll's eyes) so I'll have to wait until the PS3 Wu decides to start back up again ...

A general remark for everyone: be careful with statements like "the new client is xx% faster/slower" based on a single result. On my machine I've seen times between 70 and 52 ms/step with application 6.45 (average ~62 ms), so I'd say the time can vary significantly.

MrS

---

Scanning for our furry friends since Jan 2002

My first 6.48-WU resulted in an error, and I'm pretty sure it was almost finished, but I don't know exactly, it was in the night time as it broke.
I'm a bit disappointed about the increased cpu-usage of this version, it's way too much again.
I will watch that a bit more now, I hope the next WU finishes normally.

---

Member of BOINC@Heidelberg and ATA!

My first 6.48 WU is being crunched right now, so far no problems (9800GTX+, 177.13, 6.3.10, XP SP2) and cpu usage is a fairly constant 16-17% of 25% (quad core). BUT I'm seeing an increase in GPU temperature by ~2°C, which hopefully means that the higher CPU usage buys me better GPU utilization, which would really be worth it.

MrS

---

Scanning for our furry friends since Jan 2002

I've now nine of the 6.48 WUs crunched on my GTX280 and GTX260 cards and so far found no problems. The performance is much better than with the 6.45 and the 6.43, only 6:20h instead over 8 hours calculation time. The CPU utilization is higher (~15% of the power of one Quad CPU, ~60% of the power of one core), but I don't use a ncpus=5 entry, so this is not important for me. The temperature of the GPU is a little bit higher, but not so high as with the 6.43.

---

Actual the first 6.48 WU is running on my 8800GT. There is the CPU utilization much higher then on my GTX260 or GTX280 cards. It's also on a Phenom Quad CPU (9850 BE). The workload is about 20% of the CPU, that's 80% of the core. It's a higher load as I thought.

---

A general remark for everyone: be careful with statements like "the new client is xx% faster/slower" based on a single result. On my machine I've seen times between 70 and 52 ms/step with application 6.45 (average ~62 ms), so I'd say the time can vary significantly.

MrS

The time varies because it is measured in ELAPSED time, from start to finish including pauses/slowdowns by other task/programs (GPU waits for CPU to poll it) and boinc itself, if it moves the task to waiting to run and the task is left in memory, the elapsed time keeps counting. May happen in other cases, that is the only one I know for sure as I observed it happen and recorded GPU time reflected 5 extra hours in the time and a poorer ms/step.

You cannot compare 6.45 shown GPU elapsed times to 6.43 and before which was CPU time, not necessarily reflecting the true GPU time used.

---

Now that being said, I've only compared two results on one Windows x86 host for 6.48 on 8800GT-640Mhz stock speed. I see a large increase in CPU time from about 1 hour to 12 hours, CPU usage from 2-4% to 38-40%. GPU elapsed time decreased some, about 2000s or 33min at most from most previous 6.45 tasks, so now they run 15.75hours elapsed GPU time. The 6.45 were mostly in the range of 68ms at the low end, the two 6.48's are 66.55 and 66.6ms. This is running the task at low priority as determined by boinc, not using ncpus+1. NVIDIA driver 178.13, and buggy client 6.3.14. I did just upgrade the driver to 178.24 after those two ran.

WELL DONE!

Boinc 6.3.14 running 5 tasks, with ps3 6.45 and the latest Nvidia .24 drivers used to average 40 - 45 ms time per step on a stock GTX 280.

http://www.ps3grid.net/result.php?resultid=85965
http://www.ps3grid.net/result.php?resultid=86350

Last 2 results gave a 29 ms and 27 ms crunch on the new 6.48 app.

Still running 5 tasks considering I've noticed that the timings are minimal if running 4 or 5 tasks.

Bottom line, after much tweaking with drivers & boinc releases, I blame the old near zero cpu app for the slow time per steps, the older app gave me lower steps, and this app is giving even lowest.

Hardware:
- QX9770 4GHz
- 3x 280GTX 650MHz

Software:
- XP32
- 178.24
- 6.3.14
- 6.48

Over 30 WU done successfuly with 6.48.

Compare to old app:
- Ca. 10% faster
- Ca. 30% more CPU usage
- Until now absolutely stable - no any problems.

Really god job!


Btw, with my 3x280 I would like to ask for more than 10WU/daily quota. 12 is minimum for me :)

---

Hi!
After reading about relatively low completition times here, I just had to abort my first WU, as it stuck at 57.822% (IIRC) with 24+h runtime. I must admit, that I had several system crashes during computation of said WU, however unrelated to WU itself. My second WU is at 60+% with ~7h elapsed. All are 6.48, computer is Q9450 stock speed, 2G 1600MHz DDR3 @ 1600MHz, XFX 8800GT XXX (factory overclocked 670/1950).

Greetings,

An update to my previous comparison between 6.45 and 6.48. This is from my last post:

App 6.45: 78.143ms, minutes usage of CPU, run with realtime priority
App 6.48: 87.283ms, ~40% of 1 core, run with standard low priority

I've since been running with realtime priority again, and got 69.xxxms for 4 WU's run on 8800GT's in WinXP. The CPU usage has gone up to 80-100% of 1 core when run at realtime priority.

AiDec,

I took a look at your results and I see two things: most 6.45 WUs needed 33 - 37 ms/step, a few 32 or 31 and some over 40 ms. With 6.48 most WUs are between 24 and 27 ms/step. How did you arrive at 10%?

And your GPU clock is changing a lot in the recent WUs. Did you take this into account when looking at the results?

MrS

---

Scanning for our furry friends since Jan 2002