The only option I can think of is the option, used in combination with the anonymous platform. This mechanism has two purposes: to provide fine-grained control of the coprocessors recognized by BOINC (NVIDIA, AMD, and Intel GPUs), and to let you use coprocessors not recognized by BOINC. Client configuration.
The option said earlier to magically detect GPUs that weren't detected before that time. Well, I can only state that the option is only used after BOINC's done all the detecting.
So either it's a coincidence, or there was a longer wait the second time that allowed all drivers to have loaded and then for BOINC to be able to detect all the GPUs (Linux needs this). Or another driver was in the way, and the user cleaned something out, or reinstalled a driver, or uninstalled another driver.
We had for a while that people who installed the Intel GPU driver found that this would break the OpenCL detection of the Nvidia or AMD GPU already in the system. We also still have that when you change videocard brand but do not completely clean out the drivers for your old videocard, that this will play havoc with the detection of the new card.
It's hardware. It's software. It's drivers. It's finicky. It's a different experience for all around.
I'am using and Radeon HD 6950 with ZX87 and I7
Here is what Boinc tell me at start:
Sun 29 Sep 2013 07:29:24 PM CEST | | No config file found - using defaults
Sun 29 Sep 2013 07:29:24 PM CEST | | Starting BOINC client version 7.0.36 for x86_64-pc-linux-gnu
Sun 29 Sep 2013 07:29:24 PM CEST | | log flags: file_xfer, sched_ops, task
Sun 29 Sep 2013 07:29:24 PM CEST | | Libraries: libcurl/7.28.1 OpenSSL/1.0.1e zlib/1.2.7 libidn/1.26 libssh2/1.4.3
Sun 29 Sep 2013 07:29:24 PM CEST | | Running as a daemon
Sun 29 Sep 2013 07:29:24 PM CEST | | Data directory: /var/lib/boinc
Sun 29 Sep 2013 07:29:24 PM CEST | | Processor: 8 GenuineIntel Intel(R) Core(TM) i7-4770 CPU @ 3.40GHz [Family 6 Model 60 Stepping 3]
Sun 29 Sep 2013 07:29:24 PM CEST | | Processor: 8.00 MB cache
Sun 29 Sep 2013 07:29:24 PM CEST | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf eagerfpu pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 fma cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm ida arat epb xsaveopt pln pts dtherm tpr_shadow vnmi flexpriority ept vpid fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm
Sun 29 Sep 2013 07:29:24 PM CEST | | OS: Linux: 3.8.13.4-desktop-1.mga3
Sun 29 Sep 2013 07:29:24 PM CEST | | Memory: 15.62 GB physical, 19.91 GB virtual
Sun 29 Sep 2013 07:29:24 PM CEST | | Disk: 11.69 GB total, 4.79 GB free
Sun 29 Sep 2013 07:29:24 PM CEST | | Local time is UTC +2 hours
Sun 29 Sep 2013 07:29:24 PM CEST | | Couldn't get Device IDs for platform #0: error -1
Sun 29 Sep 2013 07:29:24 PM CEST | | No usable GPUs found
Sun 29 Sep 2013 07:29:24 PM CEST | | A new version of BOINC is available. (7.0.65) Download
What did you have to do to fix your problem?
As of 29 Sep 2013 20:04:59 UTC, the HD6900 was being detected, and has received work from both BRP Search (Arecibo, GPU) and BRP Search (Perseus Arm Survey), but none returned yet.
Edit: His scheduler log does show another problem manifesting, looks as if he has his disk space preference set too low:
Quote:
2013-09-29 20:04:59.7963 [PID=20104] [send] [HOST#8771811] [WU#175321654 p2030.20130518.G34.43+00.56.N.b6s0g0.00000_3344] WU is infeasible: Not enough disk space
2013-09-29 20:04:59.8237 [PID=20104] [mixed] sending locality work second
2013-09-29 20:05:00.4193 [PID=20104] [version] Checking plan class 'X64'
2013-09-29 20:05:00.4193 [PID=20104] [version] plan class ok
2013-09-29 20:05:00.4193 [PID=20104] [version] Checking plan class 'SSE2'
2013-09-29 20:05:00.4194 [PID=20104] [version] plan class ok
2013-09-29 20:05:00.4194 [PID=20104] [version] Best version of app einstein_S6CasA is ID 473 (3.59 GFLOPS)
2013-09-29 20:05:00.4195 [PID=20104] [send] stopping work search - insufficient disk space
2013-09-29 20:05:00.4195 [PID=20104] [send] stopping work search - insufficient disk space
2013-09-29 20:05:00.4195 [PID=20104] [send] stopping work search - insufficient disk space
2013-09-29 20:05:00.4195 [PID=20104] [send] stopping work search - insufficient disk space
As of 29 Sep 2013 20:04:59 UTC, the HD6900 was being detected, and has received work from both BRP Search (Arecibo, GPU) and BRP Search (Perseus Arm Survey), but none returned yet.
hmm...he did it without having to upgrade to the latest and greatest BOINC (v7.0.65). my guess is that he simply reinstalled BOINC, this time as a standard app and not as a service/daemon...but i would love for him to stop by the forums again and confirm it.
As of 29 Sep 2013 20:04:59 UTC, the HD6900 was being detected, and has received work from both BRP Search (Arecibo, GPU) and BRP Search (Perseus Arm Survey), but none returned yet.
hmm...he did it without having to upgrade to the latest and greatest BOINC (v7.0.65). my guess is that he simply reinstalled BOINC, this time as a standard app and not as a service/daemon...but i would love for him to stop by the forums again and confirm it.
I doubt it, from his Data directory location he did a repository installation, this comes already setup, no running an installer.
I doubt it, from his Data directory location he did a repository installation, this comes already setup, no running an installer.
Claggy
yes, but keep in mind you're looking at a snip-it of event log entries from a previous BOINC installation (one that apparently didn't detect his HD 69xx GPU). for all we know he reinstalled BOINC, and the "data directory" line in the BOINC event log now reads something like this: "D:/ProgramData/BOINC"
...granted, i know that's what it would say for a BOINC installation not installed as a service on a Windows host. i can't speak for a Linux platform, but i would imagine the "data directory" line in the BOINC event log would change accordingly if he uninstalled the BOINC service and installed it like a true application. hopefully he'll see this conversation and shed more light on the situation. i too an eager to know exactly what fixed it for him.
Some Linux distributions (Fedora, Ubuntu, Debian, Gentoo, possibly others) have BOINC packages that you can install using your your distro's package manager. Compared to using the Berkeley Installer, this has several advantages:
- The resulting BOINC installation runs applications under an unprivileged account, and is therefore more secure.
- The BOINC binaries are dynamically linked, therefore they require less memory than the binaries in the Berkeley Installer.
- The package manager checks for dependencies and installs any additional libraries required to run BOINC on your Linux distro.
- BOINC is installed as a daemon (BOINC runs automatically at boot time even if no user is logged in).
- BOINC updates can be automated if your Linux distro has automated package update capability (most popular distros do).
As far as I know, it is possible to run BOINC as a daemon under Linux and have the GPU be detected. You just have to wait long enough for the system to boot and load all the X-server drivers, before starting BOINC.
I just scrolled through this thread but things are pretty dated so I'm hoping I can get some current help.
I'm converting a machine over to BOINC. It has an ATI HD7850 and a HD 7950 card and I'm running Linux Mint 17 (based on Ubuntu).
At this point, BOINC cpu WU's are running but it doesn't see the GPU's. I made the change to cc_config.xml. I have NOT installed any proprietary drivers. I looked around for Catalyst 12 since they were mentioned earlier but can't find it anymore although I do have Catalyst 13 and 14 in my archives along with the SDK.
Should I install one of these or can I get what I need from the repository? What would those files be? Also, what about installing the SDK?
Thanks for any help you can provide. All my experience is with Nvidia.
Is Boinc Manager saying "no usable GPUs found"? If it is then this is most likely a driver issue. If not then make sure that "Use GPU while computer is in use" is checked in Boinc Preferences of BOINC manager. You don't need the config file or the SDK. I am running AMD and do not employ either of these and I am running Ubuntu 14 with distro drivers.
[edit] Is the machine you are referencing 989. If so I believe this would be a driver related problem. Using Mint's software package manager see if you have fglrx-amdcccle package installed. This is the package I installed not sure what it would be on Mint. If you do try a restart of the boinc-client "sudo service boinc-client restart". Last try a reboot of the system.
I'm converting a machine over to BOINC. It has an ATI HD7850 and a HD 7950 card and I'm running Linux Mint 17 (based on Ubuntu).
I'm running Einstein GPU tasks on Linux (PCLinuxOS 64bit) and HD7850s are a very sweet card to use from an output per watt perspective. I've not personally tried HD79xx cards but I know they produce extremely well but at the expense of much higher power consumption. If you're looking at credit per watt, you may get the best results if the two cards could be inserted in two separate hosts. Because of vey high PCIe bus and memory bandwidth requirements, you would need an appropriately high spec (and expensive) motherboard to get the most out of two cards in the one host.
Quote:
At this point, BOINC cpu WU's are running but it doesn't see the GPU's. I made the change to cc_config.xml. I have NOT installed any proprietary drivers.
I've zero experience with either Mint or Ubuntu but you do need proprietary drivers if you want to crunch. AMD's driver is called 'fglrx' and also you need the OpenCL libs. Someone will probably come along and mention the packages you need for Mint. Each distro seems to have their own particular way of handling the proprietary stuff. For PCLinuxOS, the driver and the libs are in separate packages and installing the driver package doesn't automatically pull in the OpenCL libs (because most people installing the driver don't actually want to crunch - they just want to play games - poor misguided lot!!). BOINC will not be able to properly detect and use your GPUs if you don't have both the proprietary driver and the OpenCL libs installed.
Quote:
I looked around for Catalyst 12 since they were mentioned earlier but can't find it anymore although I do have Catalyst 13 and 14 in my archives along with the SDK.
You do get some variation with different driver versions but in the main, performance seems to be getting better with more recent versions. I'd be trying whatever is the most recent in your repos. I don't believe you need the SDK. You will probably find that a recent Catalyst (or whatever it's called in your repos) package contains both the fglrx driver and the OpenCL libs. You shouldn't need to install anything extra from AMD.
If you want to see an example of what a HD7850 is capable of under Linux, take a look at this host I built recently. It uses a basic H81 Asrock board with a non-K overclocking BIOS from Asrock so that I can overclock the anniversary edition G3258 Pentium dual core from 3.2GHz to around 3.9GHz.
The HD7850 runs 4 concurrent BRP5 GPU tasks whilst one of the CPU cores runs FGRP4 CPU tasks. Using app_config.xml to modify the project supplied CPU and GPU resource requirements, only one of the CPU cores (rather than both) is reserved for GPU support duties. The CPU is good enough for this not to impact in any noticeable way on the GPU crunch times. In fact, the level of CPU overclocking doesn't seem to have any significant effect on the GPU crunch times, but it certainly does on the CPU crunch time.
Just before the recent outage scrambled things, this very basic machine had a RAC of just under 80K. I upgraded an existing (retired) machine about a month ago and reused the case, PSU and disk. I replaced the board, CPU, GPU and RAM for a total cost of just over $AU300 - around $US 250 or so. You will notice on the details link that the machine started crunching in 2005 and this is now the second reincarnation :-). I seem to be able to get around 5 years from any particular reincarnation :-).
I just scrolled through this thread but things are pretty dated so I'm hoping I can get some current help.
I'm converting a machine over to BOINC. It has an ATI HD7850 and a HD 7950 card and I'm running Linux Mint 17 (based on Ubuntu).
At this point, BOINC cpu WU's are running but it doesn't see the GPU's. I made the change to cc_config.xml. I have NOT installed any proprietary drivers. I looked around for Catalyst 12 since they were mentioned earlier but can't find it anymore although I do have Catalyst 13 and 14 in my archives along with the SDK.
Should I install one of these or can I get what I need from the repository? What would those files be? Also, what about installing the SDK?
Thanks for any help you can provide. All my experience is with Nvidia.
Regards,
Steve
Looks like your AMD GPU has been identified and that you are up and running. Congrats
The only option I can think
)
The only option I can think of is the option, used in combination with the anonymous platform. This mechanism has two purposes: to provide fine-grained control of the coprocessors recognized by BOINC (NVIDIA, AMD, and Intel GPUs), and to let you use coprocessors not recognized by BOINC.
Client configuration.
The option said earlier to magically detect GPUs that weren't detected before that time. Well, I can only state that the option is only used after BOINC's done all the detecting.
So either it's a coincidence, or there was a longer wait the second time that allowed all drivers to have loaded and then for BOINC to be able to detect all the GPUs (Linux needs this). Or another driver was in the way, and the user cleaned something out, or reinstalled a driver, or uninstalled another driver.
We had for a while that people who installed the Intel GPU driver found that this would break the OpenCL detection of the Nvidia or AMD GPU already in the system. We also still have that when you change videocard brand but do not completely clean out the drivers for your old videocard, that this will play havoc with the detection of the new card.
It's hardware. It's software. It's drivers. It's finicky. It's a different experience for all around.
RE: Hello World! No GPU
)
What did you have to do to fix your problem?
As of 29 Sep 2013 20:04:59 UTC, the HD6900 was being detected, and has received work from both BRP Search (Arecibo, GPU) and BRP Search (Perseus Arm Survey), but none returned yet.
Computers belonging to xmal
Edit: His scheduler log does show another problem manifesting, looks as if he has his disk space preference set too low:
Claggy
RE: What did you have to do
)
hmm...he did it without having to upgrade to the latest and greatest BOINC (v7.0.65). my guess is that he simply reinstalled BOINC, this time as a standard app and not as a service/daemon...but i would love for him to stop by the forums again and confirm it.
RE: RE: What did you have
)
I doubt it, from his Data directory location he did a repository installation, this comes already setup, no running an installer.
Claggy
RE: I doubt it, from his
)
yes, but keep in mind you're looking at a snip-it of event log entries from a previous BOINC installation (one that apparently didn't detect his HD 69xx GPU). for all we know he reinstalled BOINC, and the "data directory" line in the BOINC event log now reads something like this: "D:/ProgramData/BOINC"
...granted, i know that's what it would say for a BOINC installation not installed as a service on a Windows host. i can't speak for a Linux platform, but i would imagine the "data directory" line in the BOINC event log would change accordingly if he uninstalled the BOINC service and installed it like a true application. hopefully he'll see this conversation and shed more light on the situation. i too an eager to know exactly what fixed it for him.
RE: I doubt it, from his
)
From http://boinc.berkeley.edu/wiki/Installing_BOINC#Installing_BOINC_as_a_package:
As far as I know, it is possible to run BOINC as a daemon under Linux and have the GPU be detected. You just have to wait long enough for the system to boot and load all the X-server drivers, before starting BOINC.
I just scrolled through this
)
I just scrolled through this thread but things are pretty dated so I'm hoping I can get some current help.
I'm converting a machine over to BOINC. It has an ATI HD7850 and a HD 7950 card and I'm running Linux Mint 17 (based on Ubuntu).
At this point, BOINC cpu WU's are running but it doesn't see the GPU's. I made the change to cc_config.xml. I have NOT installed any proprietary drivers. I looked around for Catalyst 12 since they were mentioned earlier but can't find it anymore although I do have Catalyst 13 and 14 in my archives along with the SDK.
Should I install one of these or can I get what I need from the repository? What would those files be? Also, what about installing the SDK?
Thanks for any help you can provide. All my experience is with Nvidia.
Regards,
Steve
Is Boinc Manager saying "no
)
Is Boinc Manager saying "no usable GPUs found"? If it is then this is most likely a driver issue. If not then make sure that "Use GPU while computer is in use" is checked in Boinc Preferences of BOINC manager. You don't need the config file or the SDK. I am running AMD and do not employ either of these and I am running Ubuntu 14 with distro drivers.
[edit] Is the machine you are referencing 989. If so I believe this would be a driver related problem. Using Mint's software package manager see if you have fglrx-amdcccle package installed. This is the package I installed not sure what it would be on Mint. If you do try a restart of the boinc-client "sudo service boinc-client restart". Last try a reboot of the system.
RE: I'm converting a
)
I'm running Einstein GPU tasks on Linux (PCLinuxOS 64bit) and HD7850s are a very sweet card to use from an output per watt perspective. I've not personally tried HD79xx cards but I know they produce extremely well but at the expense of much higher power consumption. If you're looking at credit per watt, you may get the best results if the two cards could be inserted in two separate hosts. Because of vey high PCIe bus and memory bandwidth requirements, you would need an appropriately high spec (and expensive) motherboard to get the most out of two cards in the one host.
I've zero experience with either Mint or Ubuntu but you do need proprietary drivers if you want to crunch. AMD's driver is called 'fglrx' and also you need the OpenCL libs. Someone will probably come along and mention the packages you need for Mint. Each distro seems to have their own particular way of handling the proprietary stuff. For PCLinuxOS, the driver and the libs are in separate packages and installing the driver package doesn't automatically pull in the OpenCL libs (because most people installing the driver don't actually want to crunch - they just want to play games - poor misguided lot!!). BOINC will not be able to properly detect and use your GPUs if you don't have both the proprietary driver and the OpenCL libs installed.
You do get some variation with different driver versions but in the main, performance seems to be getting better with more recent versions. I'd be trying whatever is the most recent in your repos. I don't believe you need the SDK. You will probably find that a recent Catalyst (or whatever it's called in your repos) package contains both the fglrx driver and the OpenCL libs. You shouldn't need to install anything extra from AMD.
If you want to see an example of what a HD7850 is capable of under Linux, take a look at this host I built recently. It uses a basic H81 Asrock board with a non-K overclocking BIOS from Asrock so that I can overclock the anniversary edition G3258 Pentium dual core from 3.2GHz to around 3.9GHz.
The HD7850 runs 4 concurrent BRP5 GPU tasks whilst one of the CPU cores runs FGRP4 CPU tasks. Using app_config.xml to modify the project supplied CPU and GPU resource requirements, only one of the CPU cores (rather than both) is reserved for GPU support duties. The CPU is good enough for this not to impact in any noticeable way on the GPU crunch times. In fact, the level of CPU overclocking doesn't seem to have any significant effect on the GPU crunch times, but it certainly does on the CPU crunch time.
Just before the recent outage scrambled things, this very basic machine had a RAC of just under 80K. I upgraded an existing (retired) machine about a month ago and reused the case, PSU and disk. I replaced the board, CPU, GPU and RAM for a total cost of just over $AU300 - around $US 250 or so. You will notice on the details link that the machine started crunching in 2005 and this is now the second reincarnation :-). I seem to be able to get around 5 years from any particular reincarnation :-).
Cheers,
Gary.
RE: I just scrolled through
)
Looks like your AMD GPU has been identified and that you are up and running. Congrats