Posts by FalconFly

1) Message boards : Number crunching : Can't get work units (Message 62432) Posted 30 Sep 2014 by FalconFly Post: MilkyWay apparently runs off obsolete GPU capability lists AND had some initial issues with Hawaii GPUs processing tasks correctly. The following AMD GPUs currently cannot get work from MW, as its Server fails to properly recognize them as suitable computing platforms : Bonaire GPU (HD7790) Oland GPU (R7 240/250) Hawaii GPU (R9 290/290X) Tonga GPU (R9 285) Needless to say, all of those GPUs process MW data just like any others. I verified that by loading up a MW cache with a supported GPU, then simply changed the Video Card. The "unsupported" GPU crunched away without any issues and properly reported all finished tasks - of course it never received new work. I think the latest AMD GPU that the MilkyWay server recognizes is the Tahiti GPU (HD7970/R9 280/280X). Anyway, since MW has some deeper rooted Validation issues with the GPUs that do run, maybe it all makes sense. They'd likely need to fix that before taking more GPUs into the mix.
2) Message boards : Number crunching : (Modified Fit) v 1.34 (opencl_ati_101) Large amount with Validate error or Completed, can't validate (Message 62418) Posted 29 Sep 2014 by FalconFly Post: Hm, my failure rate on the ModFit 1.34 is about 5% and half of those seem to be WorkUnits that contain parameter sets that can't achieve valid results regardless of which host processes them... The others are indeed actual validate errors for reasons unknown... (based on running HD7750, HD7850 and GT 720M)
3) Message boards : Number crunching : Updated GPU Requirements (Currently not supporting GPU tasks) (Message 62357) Posted 17 Sep 2014 by FalconFly Post: I've replaced the HD7790 (Bonaire) against a HD7850 (Pitcairn) for a test, using the already installed Catalyst 14.4 Driver - and as expected MilkyWay immediately recognized the GPU and distributed work. The MilkyWay BOINC Server really needs at least updated GPU identification tables.
4) Message boards : Number crunching : Updated GPU Requirements (Currently not supporting GPU tasks) (Message 62350) Posted 16 Sep 2014 by FalconFly Post: From the looks of it - it would seem so. But not having seen an official response to the problem, I don't even know if the staff is working on it or not. If they don't, they'll gradually lose more and more GPU power, as all modern AMD cards will fail to get work assigned.
5) Message boards : Number crunching : Updated GPU Requirements (Currently not supporting GPU tasks) (Message 62334) Posted 13 Sep 2014 by FalconFly Post: I'm using the latest Catalyst 14.4 WHQL Driver. To rule out any Driver issue, I completely un- and re-installed the entire Driver package to make sure (on my Host with HD7790 Bonaire GPU). It didn't make any difference however. My System with the R7 240 Oland GPU is an absolutely fresh Windows install and also uses the Cat 14.4 WHQL. The only difference I can spot in the BOINC host listing is that the GPUs not getting any Work are listed without their Series/ID. Example (getting no work, Win7 64bit w/ latest Cat 14.4 WHQL) : AMD Bonaire (1024MB) OpenCL: 1.02 AMD Oland (2048MB) OpenCL: 1.02 (getting work, Vista 64bit w/ latest Cat 13.12 WHQL ) AMD AMD Radeon HD 7700 series (Capeverde) (1024MB) driver: 1.4.1848 OpenCL: 1.02 Other projects list these Hosts with their GPUs in exactly the same manner but have no issues distributing work to them, so technically it shouldn't make a difference. Hence my suspicion Milkyway relies on other tables/means of GPU identification than them (which I think might be the cause of the issues). After all, it clearly misidentifies the Oland GPU as an old HD38xx series (RV6xx) and thus claims lack of OpenCL support.
6) Message boards : News : New Version of Separation Modified Fit (1.32) (Message 62326) Posted 12 Sep 2014 by FalconFly Post: Got the same : ATI based Host Milkyway@Home 9/12/2014 7:07:33 PM Started download of milkyway_separation__modified_fit_1.32_windows_x86_64__opencl_amd_ati.exe Milkyway@Home 9/12/2014 7:07:34 PM Giving up on download of milkyway_separation__modified_fit_1.32_windows_x86_64__opencl_amd_ati.exe: permanent HTTP error NVidia based host Milkyway@Home 9/12/2014 6:51:02 PM Started download of milkyway_separation__modified_fit_1.32_windows_x86_64__opencl_nvidia.exe Milkyway@Home 9/12/2014 6:51:03 PM Giving up on download of milkyway_separation__modified_fit_1.32_windows_x86_64__opencl_nvidia.exe: permanent HTTP error
7) Message boards : Number crunching : R9 290 (Message 62311) Posted 11 Sep 2014 by FalconFly Post: I may throw the HD7790 (Bonaire) and R7 240 (Oland) GPUs in there as well - not getting a single WorkUnit (Catalyst 14.4 Driver). All other projects work just fine - except Milkyway. Something really needs to be done about the Server GPU detection/identification ASAP. Me thinks it's running off some rather very old lists that do not contain most modern (post Pitcairn) AMD GPUs.
8) Message boards : Number crunching : Very high temps out of two of the VRM's (Message 62288) Posted 9 Sep 2014 by FalconFly Post: Yes, that's very good cooling. Given that, seems weird those VRMs get so hot. Any chance some of the small resistor(s) got damaged or missing on that card's PCB? That would be the only reason I could imagine, as video cards often still run fine with damaged or missing resistors - with some components getting out-of-specs Voltage or Ampere as a result. That can create uncommon/random hotspots on other components. Dry/dead capacitors (typically not visible on modern hardcap variants) could also be a factor - after all those have a limited lifespan depending on operating temperature and time of exposure. That could cause similar effects IMHO but as opposed to damaged resistors, it's very tough to detect (let alone fix).
9) Message boards : Number crunching : Updated GPU Requirements (Currently not supporting GPU tasks) (Message 62282) Posted 8 Sep 2014 by FalconFly Post: I just attached with a R7 240 (Oland) GPU and I also get the BOINC statement Message from server: ATI GPU R600 (R38xx) does not support OpenCL Of course it does (OpenCL 1.2) and it also has Double Precision capability. IMHO this points to a detection/identification issue on the server side, as the Oland GPU is not an R600 (R38xx) (?)
10) Message boards : Number crunching : Very high temps out of two of the VRM's (Message 62281) Posted 8 Sep 2014 by FalconFly Post: How is the case cooling for that card ? The Video card PCB components (especially VRMs) need a good airflow to cool them, for crunching purposes an excellent airflow is mandatory to ensure longlivety of the hardware.
11) Message boards : Number crunching : Milkyway@Home 1.02 (opencl_AMD_ati) all erroring out (Message 62278) Posted 8 Sep 2014 by FalconFly Post: I've tried it again and this time even suspended all other Projects, rendering the entire System idle with only MW being an active Project. BOINC Event log keeps saying "Requesting new Tasks for ATI" - "Scheduler request complete, got 0 Tasks". It really seems that HD7790 can't get any work from MW, no matter what I do (???) I begin to wonder if that's some of the old (long forgotten) BOINC Long Term Debt issues... I know very old Versions of BOINC tended to do that when they internally accumulated out-of-whack debt figures running multiple projects. At least back in the old days, the calculation methods for the figure were unstable and created a BOINC-disabling effect under certain conditions.
12) Message boards : Number crunching : Milkyway@Home 1.02 (opencl_AMD_ati) all erroring out (Message 62275) Posted 8 Sep 2014 by FalconFly Post: Yep, another Host is chugging along just fine (NVidia GPU). Looking at the Event Log, the AMD GPU Host was requesting GPU tasks and the reponse always was "Got 0 new tasks, Scheduler request complete". For now it's busy with another project but I'll try again to get it some work tomorrow. But if that doesn't work, I'm a bit at a loss there. I already toggled the option back & forth but it made no difference. Also, all Applications are allowed to run without restrictions. So far, the host got Collatz and Einstein tasks without issues. Only as I wanted to move to MilkyWay, I noted it not getting any work for the card...
13) Message boards : Number crunching : Milkyway@Home 1.02 (opencl_AMD_ati) all erroring out (Message 62272) Posted 7 Sep 2014 by FalconFly Post: Hm, rejoined MW just recently and right now I have a different issue. My HD7790 doesn't get any WorkUnits at all - has the work distribution for ATI OpenCL been halted maybe ?