My AMD x4 9650 win xp 32 w/5870 runs avg. 70 sec. MW .57 ati4 wu. While running AQUA D-W FP (4 cores) All runs fine. My i7 950 w7 64 w/5870 runs avg 80 sec, same wu as above. Running AQUA D-W ROQS (8 cores) the problem I guess is roqs took all the core time and had to set the affinity in task mgr to 6 cores so the gpu cud be feed. I am waiting to run a D-W FP wu on the i7 and see what happends, not changing the affinity. I have a roqs waiting to crunch on the AMD and will see if I have the same probl there too. So even though the cpu count is .05, it takes more than that. Seeing there is no problem w/AMD FP running w MY gpu. The i7 wu wud sit there and run even tho it seamed to be finished, suspend/resus did not do anything, 1 wu was invalid untill i figuerd out the above core count in task mgr.

---

Updated Catalyst 10.4 to Catalyst 11.4. Now running without errors.

CPU load is 12.5% per task, in other words one CPU core on my 8 core computer. So 2 MilkyWay tasks on HD 5970 are using 25% of CPU resources or 2 of 8 cores.

Running at 99% GPU usage, runtime is about 3 seconds longer than previous 0.23 application on 267.2 credits tasks.

Haven't tried running other CPU projects yet, just listening to GPU vacuum cleaner music again. With current full core CPU usage, would need to leave at least 2 CPU cores free to avoid GPU processing slowdown.

Currently GPU is running a few degrees hotter and drawing 50.5A on VDDC which is slightly more than previously. This may drop a bit when I start to process CPU applications again.

First 2 - 3 wu worked as before, around 90 sec \ wu to complete, GPU load=90%
After that a wu takes 230-270 seconds to complete and the GPU load is only 50%

Q6600@3560MHz ATI HD 5870@970, all on water-cooling
WindowsVista 64 & AMD 11:2


EDIT:

It seams that the slow down was caused by the CPU load, the GPU load is on 97-99% when I only run milkyway@home. Is there a way I can lock 1 cpu core to milkyway?

Computer now processing WU correctly
i7 920
boinc 6.10.58
GPU Radeon 4850

HOWEVER I was originally running
Anonymous Platform (ATI_GPU) version 2.0 by Gypsel installed Dec 2009
These always ran seemlessly

On reading the post about recent events I was expecting failures and to have to do some readjusting. To my surprise overnight ( Australia) the machine downloaded WU and appeared to be working seemlessly using the old 2.0 app. WU were reported as being sucessful on both boincview and boinc manager. I was puzzled and checked the website and discovered all the successful completions were reported as invalid.

**I reset the project... no change

**I detatched the project and re-attached and the computer reported new settings . However didnt send GPU work as ATI drivers were not useable

**I downloaded the latest Catalyst drivers and installed. After a restart
I now have the 0.57 app downloaded and reporting success

Timing for the old app/new app ~ 210s (old anonymous 2.0) 300s (new 0.57)

---

OWN every thing I need
EARN.. enough to live !!!
WANT a solar array on the roof so I can run a BOINC farm( DREAM on!!)
NO wife
NO kids
NO troubles

Seems everything fine here. No broblems Catalyst 10.4P March 23 driver.

<core_client_version>6.12.22</core_client_version>
<![CDATA[
<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_CAYMAN
Revision:              1
CAL Version:           1.4.1353
Compute shader:        CAL_TRUE
Engine clock:          880 Mhz
Memory clock:          1375 Mhz
GPU RAM:               2048
Wavefront size:        64
Double precision:      CAL_TRUE
Compute shader:        CAL_TRUE
Number SIMD:           24
Number shader engines: 2
Pitch alignment:       256
Surface alignment:     4096
Max size 2D:           { 16384, 16384 }

Estimated iteration time 100.952296 ms. Dividing into 4 chunks
Integration range: { nu_steps = 640, mu_steps = 1600, r_steps = 1400 }
Using { 1, 4 } chunk(s) of size { 1400, 400 }
Integration time = 67.577636 s, average per iteration = 105.590057 ms
Integral 0 time = 68.841878 s
Estimated iteration time 25.238074 ms. Dividing into 1 chunks
Integration range: { nu_steps = 640, mu_steps = 400, r_steps = 1400 }
Using { 1, 1 } chunk(s) of size { 1400, 400 }
Integration time = 16.850805 s, average per iteration = 26.329383 ms
Integral 1 time = 17.182738 s
Likelihood time = 5.379981 s
<background_integral> 0.000307098551866 </background_integral>
<stream_integral>  92.501700702443955  1736.205884928765300  205.779079529715720 </stream_integral>
<background_likelihood> -3.062120835651964 </background_likelihood>
<stream_only_likelihood>  -5.882912009451230  -4.498048808511908  -3.984207272741599 </stream_only_likelihood>
<search_likelihood> -2.826061513869698 </search_likelihood>
08:37:16 (444): called boinc_finish

</stderr_txt>
]]>

---

It seams that my problem with low GPU load is fixed. I set the CPU 'use at most' to 75% in the BOINC preferences and after that both GPU & CPU load is on 97 - 100%

Good..after watching for an hour, my AMD has no problm running an D-W ROQS cpu wu and a MY gpu wu. The "only use % cpu" setting in boinc dose not work though on the ROQS on the i7 rig, still have to set affinity to 6 cores.. No big woop.

---

good morning from germany

ok after i got several computation errors, I reattached the
project and it worked. BUT the ATI app used round about 15%
of my quadcore while 4 CPU WUs also was on computing. I guess
this shouldnt be so ...

also i still have the feeling the CPUs WUs are able to
burn my CPU. I don't know what it is. If other programs
like prime95 or other boinc projects use it, which also use
100% of my CPU, my cpu temp always is about 50°.
With milkyway CPU WUs its flipping from 55°-70° (max cpu temp:61°)

sry but i cant believe i m the only one with this problem,
if all other apps and programs are running fine on my system.

CAL ATI Radeon HD 4700/4800 (RV740/RV770) (512MB) driver: 1.4.515

worked fine for months with the 0.23 app.

Edit: they fail within the first 7 sec's

OK, been watching a bit longer now and my observations are:

1) CPU usage is slightly higher than before (higher temps and %)
2) GPU usage is not as efficient as before (lower temps and irregular more spikey waveforms in Afterburner))
3) it's taking longer to process the wu's
(was 49-50secs is now 55-60 for 159.86)
(was 78-80secs is now 89-94 for 267.20)
(was 65-66secs is now 72-73 for 213.76)

Nothing was changed on my system except the new app that it is running

We are crunching again!!!!

Updated Catalyst 10.4 to Catalyst 11.4. Now running without errors.

Where did you find 11.4 ? The amd-site has 11.3 available for download!

Alexander

GIYF!!

http://support.amd.com/us/kbarticles/Pages/Catalyst114earlypreview.aspx

11.4 preview

Also I have the same problem with the gpu computation speed:

old app(ati 0.23 - ati13 ?!):
~150s
~210s
~255s

new app(ati14):
~195s
~260s
~330s

Gainward HD 4870 GS 1024MB
Catalyst 11.3
Boinc 60.10.60

And the 17_3s_fix wu's have 100% failed on all mine

Using cal 1.4.900 and I won't update because it's been running fine for a long time

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

WUs all error out with Catalyst 10.7 XP32. Since AMD broke "classic" OpenGL in 10.8 on and can't be bothered to repair it, updating is not an option on this particular machine.

---

17 920
xp32

was using 10.6 catalyst had to update to 11.3 to get ati .57 app to work

@ cncguru all "17_3s_fix" failed on my rig too

---

OWN every thing I need
EARN.. enough to live !!!
WANT a solar array on the roof so I can run a BOINC farm( DREAM on!!)
NO wife
NO kids
NO troubles

Yep, a upgrade to 11.3 seems to fix it on xp 32.......

xp32
was using 10.6 catalyst had to update to 11.3 to get ati .57 app to work

Cat 11.2, Win 7 64, Boinc 6.10.58, HD5850, Pent D 3.2, GPU load 75%

de_separation_10_3s_fix10_1_94064_1302420332_4
142 secs for 159 credit on an HD5850

de_separation_10_3s_fix20_1_123288_1302425697_0
194 secs for 267 credit

de_separation_13_3s_free_1_118753_1302425037
errored out on 4 machines

Cat 11.3, Win7-64, HD6950, C2Q 2.88 GHz.

The good: no errors.

The bad: running 4 CPU projects the GPU starves. Compared to 0.23 GPU utilization is down from 99% to 75%. Crunching time increases accordingly (this is what many have been reporting as a slow-down). A further slow down comes from the fact that the CPU gets only 50% of one core (in my case) and therefore the CPU part of the calculations take longer, leading to longer idle times for the GPU.

The hot-fix: reserve one CPU core. GPU utilization goes up to 99% again and crunching speed is almost restored.

The ugly: if we can't use the anonymous platform any more we can't configure the app in the way we want it via command line arguments. Specifically:

- I used to run 2 WUs at once to hide the idle GPU time between WUs (when the CPU calculations are done). Since my card needs lass than 60s for a 160 credit WU (including CPU time), there is a lot of this overhead, so overall throughput (and satisfaction) is a few % lower.

- I used "b-1" for the polling behaviour with 0.23. This allowed for optimal GPU performance without loosing a full cpu core -> increased CPU throughput (and satisfaction ;)

- I used f60 instead of the default f30. This makes the Win 7 GUI, light games (Civilization, Plants vs. Zombies etc.) and videos noticeably more smooth. With the default 30 Hz refrseh setting things are choppy because 30 Hz are not achieved, the real result is ~20 Hz. At f60 the throughput is increased again because I don't have to switch MW off for watching video or light games.

How to solve this? Well, get the CPU-utilization sorted out. And could we use 0.57 with an app_info and configure it the same way we used to with 0.23?

MrS

---

Scanning for our furry friends since Jan 2002

Cat 11.3, Win7-64, 3xHD5850, intel 970 4.00 GHz (water cooled)

All 0.57 validating ok

The cpu ussage issue is a right bitch and i have tried reserving cores for the GPU and limiting CPU apps to 8 instead of 12 threads but no joy

The Gpu's are only utizing 65% - 75% per unit and dopping my rac to barely 1,000,000 / day With a power bill in excess of $200/week i can't justify
this performance

I'm afraid that i can't continue with milkyway so i'll be moving all 8 GPU's to collatz to spare my cpu's till Travis and the guys do something about this issue

---