Found all I need to know.

Need a better AMD-card.

Well, we haven't clearly identified a 4xxx specific problem so far. We're still investigating the overall numerical accuracy/correctness of our OpenCL apps in general though. Given AMD's driver quality that's no trivial, straight-forward task unfortunately.

Please be patient with us.

Oliver

Upon startup BOINC will detect if the GPU is OpenCL capable, which driver it's using and how much memory is on the GPU and what is available. Would you mind exiting BOINC and restarting it, then post those lines about your videocards?

I mean these ones:
18/01/2012 19:52:15 | | ATI GPU 0: AMD Radeon HD 6800 series (Barts) (CAL version 1.4.1646, 2048MB, 2015MB available, 2400 GFLOPS peak)
18/01/2012 19:52:15 | | OpenCL: ATI GPU 0: Barts (driver version CAL 1.4.1646 (VM), device version OpenCL 1.1 AMD-APP (831.4), 2048MB, 2015MB available)

-5850(1gb), 4850(515MB)

Well, I suppose 515 MB is right around what our app currently uses (we set the requirement to 490 MB). Which of these cards feeds your desktop? If any, is it multi-headed? Anyway, the app doesn't start since BOINC determines that the GPU in question doesn't have enough free/available memory.

-Only SETI on both GPU's

No, I wanted to know whether you ran any other GPU app while you tried to run our app.


Oliver

-5850(1gb), 4850(515MB)

Well, I suppose 515 MB is right around what our app currently uses (we set the requirement to 490 MB). Which of these cards feeds your desktop? If any, is it multi-headed? Anyway, the app doesn't start since BOINC determines that the GPU in question doesn't have enough free/available memory.

-Only SETI on both GPU's

No, I wanted to know whether you ran any other GPU app while you tried to run our app.


Oliver

-SETI app OpenCl MB version .390 for HD5xxx ATI cards
(I have completed tasks on them both earliar).

You can't use the 5xxx app on a HD4850.

From Lunatics Windows Installer v0.39 release notes:

ATI MB apps - upgraded to r390

We offer the application in two different variants, depending on card:
plain - for most HD4xxx owners and those with HD5xxx and HD6xxx cards who
suffer from driver incompatibilities.
HD5 - for most if not all owners of HD5xxx GPUs and HD6xxx GPUs.

The ATI MB application will not work on ATI cards with workgroup size 128 (e.g. HD43xx). To check workgroupsize run CLinfo, downloadable (at present) from {http://boinc.berkeley.edu/dl/clinfo.zip}
If you have problems running the HD5 flavour (anything from reduced performance to BSOD), you should run the plain variant.

The recommended apps (r521 for Astropulse, r390 for MultiBeam) need OpenCL support from the ATI driver - for recent driver versions, look for a driver *with* APP or OpenCL listed. For compatible drivers see http://support.amd.com/us/gpudownload/Pages/index.aspx
For cards, see hhttp://ttp://developer.amd.com/gpu/AMDAPPSDK/pages/DriverCompatibility.aspx
On most systems GPU-Z (see above) will tell you if your setup supports OpenCL.
Minimum required driver: Something that is OpenCL capable. You can check with GPU-Z (see above).
Catalyst 10.7b with SDK
Catalyst 10.10 Accelerated Parallel Processing (APP) Technology Edition
The following drivers are known to be incompatible with the app:
11.10 and 11.11.
On some drivers the app shows increased CPU usage.

For cards without OpenCL support (from HD2xxxx on) use the 'hybrid' r453 Astropulse application. NB this is a combined CPU/GPU application and will only use the GPU for some of the calculations, while doing the rest on a CPU core.
Minimum driver version Catalyst 9.x.

NB using this installer will give you arbitrary default cmdline parameter entries.
If you are already running ATI applications: Make a note of what numbers you are using to be able to restore to those values PRIOR to running the installer. This also applies to setting the <count> variable according to your -instances_per_device variable for multithreading.

The Hybrid AP application uses about 80% of a CPU core and does only about 20% of the calculations on the GPU. Parameters have been choosen to reflect this usage. However present BOINC can not reserve '0.8' of a CPU core and therefore reserves none. This means your CPUs will be overcommitted and runtimes and DCF will suffer. You do increase the overall output though.

I run SETI today and I tried to run on both my ATI cards 5850 + 4850

and I get (schedular wait) after 25seconds of runtime on the 4850 card

I need more details:
- You forced BOINC to use all GPUs via cc_config.xml?
- So our app works in your 5850?
- What's the total RAM of both GPUs?
- What SETI app? CAL or OpenCL?
- Did you run SETI and E@H at the same time?

Oliver

Ok, we could finally track down and fix the problem. BOINC 7.0.11 contains the fix.

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

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, 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.

T4T already runs a Virtual Box wrapper application, but even though they are a Beta project they do not want to go test all the adjustments and bug fixes that the BOINC developers added to the Vboxwrapper app (which begs the question why they run as a beta project then, but heck ;-)), to Virtual Box (in coop with Oracle devs) and to the BOINC client. Hence the CERNVM/Vboxwrapper Test Project set up and administrated by the BOINC developers, where we test all that.

There's no validator running over there, so there's no credits even though we do real work through the hypervisor. The project will be deleted when we're done testing.

Eventually, when we're done testing the crap out of all that, hopefully T4T will take over the vboxwrapper app and run it at their project.

I am using 7.0.7 because it's needed for the VBoxwrapper test project.

My other two OpenCL tasks went a lot quicker, 4,800 and 5,600 seconds.
But then this one didn't validate against the CUDA32. Unknown at this time if this is because of me or him.

My Radeon HD6850 just finished its first task, it got validated against a CUDA32. How refreshing.

As I wrote earlier, the computations for the BRP app have to be rearranged/resized to fit the hardware of the 4xxx series cards (you'll see some messages indicting this in the log output). Either this resizing process has a bug or it's causing a loss in precision. So there is still some hope that this can be made to work. Something the devs have to look into after the holiday season.

HB

Dear Oliver,
I have the same problem as others here.
All tasks are finished as error while computing with error -5 and 2021.

I have W XP 32bit, boinc 7.0.2, catalyst 11.12 (all in one package-with openCL 1.1 and SDK 2.5) on C2D with 2x6GB ram and HD5870 with more than 950 MB free memory.

As you ordered here I stopped other projects on GPU and I was watching GPU-Z during the Albert WU processing.
Usage of GPU before start is 0%, after the unit is computed it still doesn´t work on GPU (0%). Task crushes after about 30 seconds.

More is probably here:http://albert.phys.uwm.edu/results.php?userid=333329

This is in BM protokol (for example):
14.12.2011 13:53:28 | Albert@Home | Starting task p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1 using einsteinbinary_BRP4 version 119 (atiOpenCL)
14.12.2011 13:53:46 | Albert@Home | Computation for task p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1 finished
14.12.2011 13:53:46 | Albert@Home | Output file p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1_0 for task p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1 absent
14.12.2011 13:53:46 | Albert@Home | Output file p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1_1 for task p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1 absent
14.12.2011 13:53:46 | Albert@Home | Output file p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1_2 for task p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1 absent
14.12.2011 13:53:46 | Albert@Home | Output file p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1_3 for task p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1 absent
14.12.2011 13:53:46 | Albert@Home | Output file p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1_4 for task p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1 absent
14.12.2011 13:53:46 | Albert@Home | Output file p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1_5 for task p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1 absent
14.12.2011 13:53:46 | Albert@Home | Output file p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1_6 for task p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1 absent
14.12.2011 13:53:46 | Albert@Home | Output file p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1_7 for task p2030.20100913.G44.54-00.26.S.b3s0g0.00000_848_1 absent


Is anything alse I can do for Albert app to make it running?

Doktor z CNT

Sorry, not up to us. I'm not sure whether the BOINC devs can do anything about it since this might even be an AMD driver issue.

Oliver

Of course it's up to you. Apparently the server back-end version that you use doesn't store the ATI/CAL driver version, but it is sent to you.

From my 7.0.2 sched_request_albert.phys.uwm.edu.xml file:

<coproc_ati>
<count>1</count>
<name>ATI Radeon HD 4700/4800 (RV740/RV770)</name>
<available_ram>1040187392.000000</available_ram>
<have_cal>1</have_cal>
<have_opencl>1</have_opencl>
<req_secs>0.000000</req_secs>
<req_instances>0.000000</req_instances>
<estimated_delay>0.000000</estimated_delay>
<peak_flops>2000000000000.000000</peak_flops>
<CALVersion>1.4.1417</CALVersion>
<target>5</target>
<localRAM>1024</localRAM>
<uncachedRemoteRAM>2047</uncachedRemoteRAM>
<cachedRemoteRAM>2047</cachedRemoteRAM>
<engineClock>625</engineClock>
<memoryClock>950</memoryClock>
<wavefrontSize>64</wavefrontSize>
<numberOfSIMD>10</numberOfSIMD>
<doublePrecision>1</doublePrecision>
<pitch_alignment>256</pitch_alignment>
<surface_alignment>4096</surface_alignment>
<maxResource1DWidth>8192</maxResource1DWidth>
<maxResource2DWidth>8192</maxResource2DWidth>
<maxResource2DHeight>8192</maxResource2DHeight>
<atirt_detected/>
<coproc_opencl>
<name>ATI RV770</name>
<vendor>Advanced Micro Devices, Inc.</vendor>
<vendor_id>4098</vendor_id>
<available>1</available>
<half_fp_config>0</half_fp_config>
<single_fp_config>62</single_fp_config>
<double_fp_config>63</double_fp_config>
<endian_little>1</endian_little>
<execution_capabilities>1</execution_capabilities>
<extensions>cl_amd_fp64 cl_khr_gl_sharing cl_amd_device_attribute_query cl_khr_d3d10_sharing </extensions>
<global_mem_size>1073741824</global_mem_size>
<local_mem_size>16384</local_mem_size>
<max_clock_frequency>625</max_clock_frequency>
<max_compute_units>10</max_compute_units>
<opencl_platform_version>OpenCL 1.1 AMD-APP-SDK-v2.5 (775.2)</opencl_platform_version>
<opencl_device_version>OpenCL 1.0 AMD-APP-SDK-v2.5 (775.2)</opencl_device_version>
<opencl_driver_version>CAL 1.4.1417</opencl_driver_version>
</coproc_opencl>
</coproc_ati>

You can even use the OpenCL information.
Then with the CAL version we can figure out which Catalysts they are. E.g. CAL 1.4.1417 is Catalysts 11.6