I'm happy to be able to crunch Einstein tasks on ATI at all! That's a great leap ahead already! All good on the developers so far!

Thanks, that's exactly our focus right now. I mean, we not only had to implement the OpenCL version of our algorithm but also update the server backend as well as the client requirements since we are the first project deploying a "real" (fully BOINC compliant) OpenCL app on all three major platforms (Win, Mac, Lin). This is not straight-forward, in particular when AMD's drivers come into play... Again, we first aim at supporting AMD GPUs at all, then we'll start optimizing the app. And hey, our test rigs show a performance drop of 20% only, compared to our CUDA app - not that bad of the first release (but you can't directly compare those results anyway)...

Best,
Oliver

Please, look at this WU.
http://albert.phys.uwm.edu/workunit.php?wuid=12832
What's wrong? Why Ati open'cl app made invalid result?

ATI driver 11.8

I ran that one with Catalysts 11.7; also wonder why we didn't match.

---

Jord.

BOINC FAQ Service

They say most of your brain shuts down in cryo-sleep. All but the primitive side, the animal side. No wonder I'm still awake.

iMac OS X 10.7.2/7.0.2 BOINC Client

all ATI WUs failing with "Error while computing" status.

Sample Stderr Output:

http://albert.phys.uwm.edu/result.php?resultid=57408

http://albert.phys.uwm.edu/result.php?resultid=51158

--
Jon

---

As the error messages indicate, your GPU doesn't have enough free memory. Sorry but we can't reliably check this in advance right now...

Oliver

That's the way I read the error as well but didn't want to assume -- also wanted to make sure it was reported in case there was an error in the Work Unit assignment algorithms that should have been able to map work to memory configuration of a specific GPU.

I'll stop processing work and monitor updates for any enhancements that will provide a better match between work units and GPU configuration.

Thanks for the feedback and explanation.

--
Jon

---

Same Codebase different platform / Compiler .. ?

Linux: Intel(R) Xeon(R) CPU X5472 @ 3.00GHz [2] AMD ATI Radeon HD 5700 series (Juniper) (1024MB) driver: 1.4.1646 runtime 45000 - 61000 sec

Windows:Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz [2] AMD ATI Radeon HD 5700 series (Juniper) (1024MB) driver: 1.4.1607 runtime 7100 - 8800 sec
hmm... .
I would expect ~ 8000sec for juniper radeon

The successors (cypress, cayman) looking good.

After several failed attempts to get BOINC 7.0.2 (DEV) working under Ubuntu 11.04 and 11.10, I have returned an Albert OpenCL result under Win7-64.
The GPGPU was a HD 5770, latest AMD driver version available 11.12, BOINC client is development version 7.0.2
The standard 500 coblestones was awarded.
Run time was 30,744.65 seconds
CPU time was 1,304.54 seconds
So I make that just over 9 points per 10 minute period.
Barely better than a CPU but using bucket-loads more power.
The completion time is not quite 1 quarter of one of CPU-core's completion time on the same machine.
But the CPU has 6 cores and so can still beat the snot out of the OpenCL performance on the HD 5770.
Not really worth it then. Well, not with the current AMD Catalyst driver anyway.

Once I get BOINC 7.0.2 working under Linux_AMD64 I should be able to test the different Catalyst drivers. But until then, this is the best I can do.

---

NEW 12.1 drivers fixed the HD6990 "high CPU usage" problem... work great!

http://support.amd.com/us/kbarticles/Pages/Catalyst121Previewdriver.aspx

9-)

I saw an intriguing trend in the ATI HD 5770 WUs.
the first finished at 33k+ seconds
the second finished at 28K+seconds
the third finished at 11k+ seconds
I'm waiting on the fourth to see where this is going.

---

NEW 12.1 drivers fixed the HD6990 "high CPU usage" problem... work great!

http://support.amd.com/us/kbarticles/Pages/Catalyst121Previewdriver.aspx

This is great news, finally! Can others confirm that, too? I'll update our test rigs to check it myself...

Oliver

I haven't had the high CPU usage problem since the 11.11 drivers. I even put in a whole new videocard yesterday, so the drivers for it are installed clean. Using the 11.12 Catalysts here. Your app is using 1% CPU cycles every so many seconds.

---

Jord.

BOINC FAQ Service

They say most of your brain shuts down in cryo-sleep. All but the primitive side, the animal side. No wonder I'm still awake.

I am running 7.0.7, 12.1, XP64, C2Q, 5870, and the task is still consuming a about 40% of a CPU core.

Edit: Some more info. When I reserve a core, GPU load is ~60%. When I do not reserve a core, GPU load drops to ~0%.

Edit 2: Same behavior with 7.0.7, 12.1, Win7 64, i7 920, dual 5870s.

---

Dublin, California
Team: SETI.USA

Ok, there's a very weird thing going on here.
When I run work on my HD6850, and do not touch BOINC, don't suspend it or worse, exit it and later restart it, then most all tasks will run and finish in between 5,000 and 6,400 seconds.

Do I suspend BOINC and resume it, or exit it and restart it, then work will become very slow and sluggish, running for multiple hours, up to 20K seconds.

Before the GPU load will be sustained at around 70%, all the way through.
After the GPU load is sporadic, coming and going but mostly zero. Like this:
Date , GPU Core Clock [MHz] , GPU Memory Clock [MHz] , GPU Temperature [°C] , Fan Speed (%) [%] , Fan Speed (RPM) [RPM] , GPU Load [%] , GPU Temp. #1 [°C] , GPU Temp. #2 [°C] , GPU Temp. #3 [°C] , Memory Usage (Dedicated) [MB] , Memory Usage (Dynamic) [MB] , VDDC [V]
2012-01-08 02:17:58 , 625.0 , 1000.0 , 51.0 , 33 , 1131 , 0 , 52.0 , 53.0 , 53.5 , 495 , 71 , 0.000 ,

2012-01-08 02:17:59 , 500.0 , 700.0 , 51.0 , 33 , 1133 , 23 , 51.5 , 52.0 , 52.5 , 495 , 71 , 0.000 ,

2012-01-08 02:18:00 , 625.0 , 1000.0 , 51.0 , 33 , 1133 , 0 , 51.5 , 52.0 , 52.5 , 495 , 71 , 0.000 ,

2012-01-08 02:18:01 , 625.0 , 1000.0 , 51.0 , 33 , 1130 , 0 , 51.5 , 52.5 , 53.0 , 495 , 71 , 0.000 ,

2012-01-08 02:18:11 , 500.0 , 700.0 , 51.0 , 33 , 1139 , 10 , 51.0 , 51.0 , 52.0 , 495 , 71 , 0.000 ,

2012-01-08 02:18:12 , 625.0 , 1000.0 , 51.0 , 33 , 1133 , 0 , 51.5 , 52.0 , 53.0 , 495 , 71 , 0.000 ,

.... all zero GPU load here ...

2012-01-08 02:19:04 , 625.0 , 1000.0 , 52.0 , 33 , 1130 , 26 , 52.0 , 53.0 , 53.5 , 495 , 70 , 0.000 ,

2012-01-08 02:19:05 , 625.0 , 1000.0 , 52.0 , 33 , 1131 , 27 , 52.0 , 52.0 , 53.0 , 495 , 71 , 0.000 ,

2012-01-08 02:19:06 , 625.0 , 1000.0 , 52.0 , 33 , 1132 , 25 , 52.0 , 52.5 , 53.0 , 495 , 70 , 0.000 ,

2012-01-08 02:19:07 , 625.0 , 1000.0 , 51.0 , 33 , 1129 , 26 , 51.5 , 52.0 , 52.5 , 495 , 71 , 0.000 ,

2012-01-08 02:19:08 , 625.0 , 1000.0 , 51.0 , 33 , 1129 , 25 , 51.5 , 51.5 , 52.5 , 495 , 71 , 0.000 ,

2012-01-08 02:19:09 , 625.0 , 1000.0 , 52.0 , 33 , 1131 , 23 , 51.5 , 52.5 , 53.0 , 495 , 71 , 0.000 ,

2012-01-08 02:19:10 , 500.0 , 700.0 , 51.0 , 33 , 1135 , 0 , 51.0 , 51.0 , 51.5 , 495 , 71 , 0.000 ,

2012-01-08 02:19:11 , 625.0 , 1000.0 , 51.0 , 33 , 1130 , 0 , 51.5 , 52.0 , 52.5 , 495 , 71 , 0.000 ,

2012-01-08 02:19:12 , 625.0 , 1000.0 , 52.0 , 33 , 1128 , 0 , 52.0 , 52.5 , 53.0 , 495 , 71 , 0.000 ,

Then when I do a full exit & restart of BOINC, it'll go into full overdrive again to catch up.
Back to the sustained ~70%.

Date , GPU Core Clock [MHz] , GPU Memory Clock [MHz] , GPU Temperature [°C] , Fan Speed (%) [%] , Fan Speed (RPM) [RPM] , GPU Load [%] , GPU Temp. #1 [°C] , GPU Temp. #2 [°C] , GPU Temp. #3 [°C] , Memory Usage (Dedicated) [MB] , Memory Usage (Dynamic) [MB] , VDDC [V]

2012-01-08 02:25:42 , 625.0 , 1000.0 , 51.0 , 33 , 1127 , 68 , 51.5 , 52.5 , 53.0 , 495 , 70 , 0.000 ,

2012-01-08 02:25:42 , 625.0 , 1000.0 , 51.0 , 33 , 1126 , 69 , 51.5 , 53.0 , 53.5 , 495 , 70 , 0.000 ,

2012-01-08 02:25:43 , 625.0 , 1000.0 , 52.0 , 33 , 1125 , 70 , 51.5 , 53.0 , 53.5 , 495 , 70 , 0.000 ,

2012-01-08 02:25:44 , 625.0 , 1000.0 , 51.0 , 33 , 1125 , 72 , 52.0 , 52.5 , 53.0 , 495 , 71 , 0.000 ,

2012-01-08 02:25:46 , 625.0 , 1000.0 , 52.0 , 33 , 1126 , 73 , 52.0 , 53.0 , 54.0 , 495 , 103 , 0.000 ,

2012-01-08 02:25:46 , 625.0 , 1000.0 , 52.0 , 33 , 1127 , 74 , 52.0 , 53.0 , 53.5 , 495 , 71 , 0.000 ,

2012-01-08 02:25:47 , 625.0 , 1000.0 , 51.0 , 33 , 1125 , 76 , 52.0 , 53.0 , 53.5 , 495 , 71 , 0.000 ,

2012-01-08 02:25:49 , 625.0 , 1000.0 , 52.0 , 33 , 1123 , 76 , 52.0 , 53.5 , 54.0 , 495 , 71 , 0.000 ,

---

Jord.

BOINC FAQ Service

They say most of your brain shuts down in cryo-sleep. All but the primitive side, the animal side. No wonder I'm still awake.

Ok, there's a very weird thing going on here.
When I run work on my HD6850, and do not touch BOINC, don't suspend it or worse, exit it and later restart it, then most all tasks will run and finish in between 5,000 and 6,400 seconds.

Do I suspend BOINC and resume it, or exit it and restart it, then work will become very slow and sluggish, running for multiple hours, up to 20K seconds.

I tried suspending and then resumed....
I don't get the slow and sluggish thing...
I run boinc 7.0.3

I also have marked "Leave task in memory while suspended".
So tasks can resume without being read to memory when resumed.

---

Ok! Got my first valid WU on a ATI GPU. Yaaaaa! Soon there might be a second.

Ageless. Plz keep your beak shut, dont want any comment from you.

---

Ok! Second one validated both against NVIDA. Here is what i have done. Went to AMD site and got their under development OpenCL and APP SDK installed those two and now i dont know where my Catlyst Control Center has gone to? Any way now getting valid units at Albert and Poem.

Ok! Second one validated both against NVIDA. Here is what i have done. Went to AMD site and got their under development OpenCL and APP SDK installed those two and now i dont know where my Catlyst Control Center has gone to? Any way now getting valid units at Albert and Poem.

I use 11.12 Catalyst suite and I have a working Control Center and the APP for OpenCl that is included with the driver.
I don't need to install the separate SDK to run POEM and Albert or SETI OpenCl.

---

Ok! Second one validated both against NVIDA. Here is what i have done. Went to AMD site and got their under development OpenCL and APP SDK installed those two and now i dont know where my Catlyst Control Center has gone to? Any way now getting valid units at Albert and Poem.

Nice, the friends at SETI also noticed corrupt results seemingly caused by certain driver/SDK combinations. Maybe AMD finally fixed/improved their software. I'll try to build a new set of apps using the latest SDK for the latest drivers ASAP.

Best,
Oliver

Ok! Second one validated both against NVIDA. Here is what i have done. Went to AMD site and got their under development OpenCL and APP SDK installed those two and now i dont know where my Catlyst Control Center has gone to? Any way now getting valid units at Albert and Poem.

Nice, the friends at SETI also noticed corrupt results seemingly caused by certain driver/SDK combinations. Maybe AMD finally fixed/improved their software. I'll try to build a new set of apps using the latest SDK for the latest drivers ASAP.

Best,
Oliver

Any chance that the new apps will work on 4xxx ATI cards??

---