Could people with errors post their driver versions and try updating them?

Is Matt going to release a 64-bit ATI version as well?

That might account for some of the speed difference that people are seeing.

I could but it really shouldn't matter since everything that matters is on the GPU. I don't really believe that it's slower (at least for the 5870) with the cited numbers since they look like the run times of 2 of the different sets of units that have been out recently. Can anyone else confirm that it's slower?

EDIT:[/b]

Are you using the math functions included with cal++ (include\cal\il\math) or something else ?

No, I'm using custom functions.
https://github.com/Milkyway-at-home/milkywayathome_client/blob/master/separation/src/separation_cal_kernelgen.cpp

Could people with errors post their driver versions and try updating them?

I can't update as I am running 11.4 beta currently.
1.4.1353

<core_client_version>6.12.15</core_client_version>
<![CDATA[
<message>
Incorrect function. (0x1) - exit code 1 (0x1)
</message>
<stderr_txt>
<search_application> milkywayathome_client separation 0.57 Windows x86 double CAL++ </search_application>
Found 1 CAL devices
Chose device 0

Device target:         CAL_TARGET_CYPRESS
Revision:              2
CAL Version:           1.4.1353
Compute shader:        CAL_TRUE
Engine clock:          850 Mhz
Memory clock:          1101 Mhz
GPU RAM:               1024
Wavefront size:        64
Double precision:      CAL_TRUE
Compute shader:        CAL_TRUE
Number SIMD:           14
Number shader engines: 2
Pitch alignment:       256
Surface alignment:     4096
Max size 2D:           { 16384, 16384 }

Estimated iteration time 179.169118 ms. Dividing into 5 chunks
Invalid number of chunks: 5
Integral 0 time = 3.590042 s
Failed to calculate integral 0
18:46:00 (4232): called boinc_finish

</stderr_txt>
]]>

---

All WU - error :(

ATI 4890, WinXP x64, driver: ati2dvag 6.14.10.7018

I could but it really shouldn't matter since everything that matters is on the GPU. I don't really believe that it's slower (at least for the 5870) with the cited numbers since they look like the run times of 2 of the different sets of units that have been out recently. Can anyone else confirm that it's slower?

Matt

- the tasks that are earning me ~267 points are taking between 90 and 115 seconds to complete.
- the tasks that are earning me ~213 points are taking between 84 and 90 seconds to complete.
- the tasks that are earning me ~160 points are taking approx. 70 seconds to complete.

tasks don't seem to be crunching any slower from what i can tell. i'm using a 5870 GPU with Catalyst 11.3 drivers on a WinXP Pro 32-bit platform though...

---

GPU time seems normal but varies by ~5 seconds (haven't seen that on the old app).
CPU time seems 2 - 3 times higher than before (varies too)
System is very sluggish due to high Cpu use by system (new too).

Win XP, HD5850, Catalyst 10.10

Yay, just did my first WU in a few days, and these are my observations on the new app so far. Not to be taken as a complaint, just reporting perceived differences on my 5850, AMD 600e, Cat. 10.12 drivers, Win7-64.

It does take longer than the 0.23 opti app I was using. It doesn't load the card as much as 0.23 did either, it ran about 75% utilization on the first one and about 50% on the second one. I had to use command line parameters to slow down 0.23 to 91%. Before the tasks ran in 115 to 160 seconds, while the small sample on the new app ran from about 140 to 270 seconds.

CPU utilization is a different story. Before it was barely noticeable, using about 12-20 secs. per WU, now it utilizes half a core for the duration of the WU, about 58-130 secs.

On Win 7 64-Bit with two 5850 (Catalyst 11.1) my first WU going over 100 %.

Stopped them at more than 700 % and more then 22 min of runtime.

Trying to update the driver to 11.3

CPU-usage is higher (on my fully used Quad-Core) about 8 % each. I notice a high memory usage.

---

I'm running new ati14 app and the Asus driver on a non-reference 6950 Direct CUII card due to ati's driver are not compatible with this card asus has modified the 11.2 drivers to work with this card and it is the only drivers that I can get to work on this card. I have noticed that all of my WU's that are erroring out are the de_seperation_17_3s_fix_4_ ones, two of these that I have run ran for over 20+ minutes and still showed 0.000% complete. (7 out of 9 WU of this run errored within 1 second and the other 2 would show any progress.) I've been completeing WUs in about 80-90 seconds up until now. I am now running de_seperation_13_3s_fix_1_, de_seperation_13_3s_free_1_, and de_seperation_10_3s_free_1_ and all seems normal and they are validating.

New twist. This system does primarily Seti, but after the first few MW WU's processed normally, everyone after that starts immediately in high priority mode. Could be my version of Boinc Manager, could need a reboot, etc. Can't start troubleshooting right now so I'll just set NNT till I have more time.

XP x86, HD4870... some work, most don't- they all ran fine with the old ATI version>

Stderr output

<core_client_version>6.10.58</core_client_version>
<![CDATA[
<message>
Incorrect function. (0x1) - exit code 1 (0x1)
</message>
<stderr_txt>
<search_application> milkywayathome_client separation 0.57 Windows x86 double CAL++ </search_application>
Found 1 CAL devices
Chose device 0

Device target: CAL_TARGET_770
Revision: 2
CAL Version: 1.4.1016
Compute shader: CAL_TRUE
Engine clock: 790 Mhz
Memory clock: 1100 Mhz
GPU RAM: 512
Wavefront size: 64
Double precision: CAL_TRUE
Compute shader: CAL_TRUE
Number SIMD: 10
Number shader engines: 1
Pitch alignment: 256
Surface alignment: 256
Max size 2D: { 8192, 8192 }

Estimated iteration time 194.106013 ms. Dividing into 5 chunks
Invalid number of chunks: 5
Integral 0 time = 1.211306 s
Failed to calculate integral 0
20:56:18 (408): called boinc_finish

</stderr_txt>
]]>

Win7-64, HD4770, ATI 10:12

The de_separation_17 WUs don't run at all, the de_separation_13 run OK, the de_separation_10 WUs run OK. By boosting the priority to normal the WUs run about 10% slower than the equivalent WUs on the old app. If I leave them at the default priority they use only 50% GPU and run SLOW. They're also hogging a whole CPU core.

If you want us to help debug this you really have to leave the WUs in the database longer as they're being purged too quickly.

On Win 7 64-Bit with two 5850 (Catalyst 11.1) my first WU going over 100 %.

Stopped them at more than 700 % and more then 22 min of runtime.

Trying to update the driver to 11.3

CPU-usage is higher (on my fully used Quad-Core) about 8 % each. I notice a high memory usage.

Updated to 11.3, reattached the project. WU run without errors, but a lot slower. 0.57 uses a full core on my system. The computer is fully used by cpu-projects so a full core isn´t available. So the runtime varies from the amount of core which is granted to the app.

GPU-usage is not more than 50 %.

---

I am also seeing large swings in gpu usage on both my XFX 5870's with Driver 11.2. on Win. Vista 64
Drops from 98, 99 to the 20's, 40's back up again. I have not seen this before with MW.

my 5870 GPU usage occasionally dips down to ~85%, but i'd say its at 97% on average. my per task CPU usage is still the normal 5% (0.05 CPU).

---

OK I'm running the older 10.12 driver which has always run most stable for whatever reasons on mine and I have tried them all!

The 17's are not working at all and I am seeing an increase accross the board in wu crunch times of approx 10% plus there seems to be a wider spread between min-max also which also scales with the crunch times so wider for longer

And I also see a wider fluctuation in GPU usage
All in all it doesn't seem as stable or fast as before so far but I am just starting to monitor

wu count is not running away as before

On a 3G Intel quad I'm seeing high CPU loading. While running AQUA my GPU load was only 50%. Turning off the CPU crunching jumped GPU loading to 97% on both GPUs of a 5970. Completion times are much slower than .23 and there is no comparison on CPU loading.

A single 5870 on my I7 (also running AQUA) gets 95% loading while the CPU crunches but here again the completion time is considerably longer than .23.

I consider the CPU loading unacceptable if this is the way it is going to be in the long term.

---

Bryan

I have 2 ATI GPUs (4870 & 3200) and am getting Computation error on all WU. Currently there are 16 WUs

With my 1090T and single 5870:

1) High CPU load -- virtually an entire core is dedicated to the GPU WU
2) GPU load is *maybe* a little less (based on observed UPS power load)
3) Run time is longer than before by about as much as 50%
4) BOINC manager wants to keep 16 in the queue
5) 4 CPU WUs get their own core, 2 others share the remaining core

MarkR

---