Message boards :
Number crunching :
I've Built a Couple OSX CUDA Apps...
Message board moderation
Previous · 1 . . . 23 · 24 · 25 · 26 · 27 · 28 · 29 . . . 58 · Next
Author | Message |
---|---|
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I'm afraid you're going to have to be a little more detailed. What should I do with the lines and where should I do it. I could post the App in the thread at C.A. if you'd like to try it yourself. The Apps are here, OSX v8 apps - dev The 7.5 App compiled in Mavericks was used over the last day on main. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
In AnalyzeFuncs.cpp, Between the succeeding cudaAcc_free(); call, and Boinc_finish(). This looks like it'll require source level and build system debugging, also a bad idea to have broken builds in circulation. Steps to replicate from scratch are this, using older OSX ? - Build boinc libs from current boinc master under XCode - paste code from Petri over the existingapp client folder files - set osx min version flag and build via command line ? (using Berkeley's configure/Make?) Did you see what happens with my flat Makefile (in mac_build) instead of their's ? (still have to run configure first, not completely uncoupled from sah_config.h yet) "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Did you see what happens with my flat Makefile (in mac_build) instead of their's ? (still have to run configure first, not completely uncoupled from sah_config.h yet) I still don't know how to use that method of compiling. If you'd like to post the instructions in that restricted thread at C.A. I'll give it a try. I searched the last used AnalyzeFuncs.cpp for Boinc_finish and received Zero hits. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Did you see what happens with my flat Makefile (in mac_build) instead of their's ? (still have to run configure first, not completely uncoupled from sah_config.h yet) Try cudaAcc_free in that file, (which is where the succeeding things are done), after that it returns control to worker.cpp, where the final boinc_finish() is called. Since it's failing (for whatever reasons) between those two points (perhaps on some pthreads termination of the worker), so far it's in line with either build system issues or boincapi. Trick is to isolate whether it's the client, timer thread, or worker threat initiating the insta-death weirdness. The pipe errors do suggest boinc_diagnostics and shared memory changes may be involved. As we're getting into advanced multithreading concepts, I will probably have to attempt to replicate the conditions. Mavericks the appropriate variant ? ML ? Yosemite would do ? "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
I still don't know how to use that method of compiling. Thought I put a readme in the mac_build folder, checking. [Edit] mac_build_README in mac_build folder contains: Prototype lean makefile (proof of concept only) "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Mavericks the appropriate variant ? ML ? Yosemite would do ? I really couldn't tell much difference between the 7.5 versions, the one compiled in Mavericks completed a few more tasks before giving the Crash after finish. The 6.5 version compiled in ML, using the Exact same code, doesn't Crash. The 7.0 version gives the same Crash as the other two 7 versions, I just added it 'cause it was there. Three different version 7 Apps all with the same Crash After Finish. I saw the README, so, do I need to move the Makefile somewhere else? What do I do with the old Makefiles? Do I leave the BOINC libs in the boinc-master folder with the same SETI_BOINC_DIR=" /Users/Tom/boinc" line? After running the path lines; export PATH=/Developer/NVIDIA/CUDA-7.5/bin:$PATH export DYLD_LIBRARY_PATH=/Developer/NVIDIA/CUDA-7.5/lib:$DYLD_LIBRARY_PATH export PATH=/Developer/NVIDIA/CUDA-7.5/bin/nvcc:$PATH export PATH=/usr/include:$PATH export DYLD_LIBRARY_PATH=/usr/include:$DYLD_LIBRARY_PATH I cd to /Users/Tom/sah_v7_opt/Xbranch, run _autosetup, then run the custom configure line, then edit the makefiles in Xbranch and Xbranch/client and finally run make. What do I do differently? |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Mavericks the appropriate variant ? ML ? Yosemite would do ? Same except replace then edit the makefiles in Xbranch and Xbranch/client and finally run make.with run make while in mac_build directory. Point was to bypass the unmaintained/broken Makefiles, and start clean. If that chokes, then diagnosis and fixes are much easier (single file written by hand by myself, as opposed to 5 broken autogenerated files) "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Nice, FedEx delivers on Saturday! Seems the Second GTX 950 Broke BOINC. Now the thing thinks the 950s are pre-Fermi cards and have disabled OpenCL. It isn't happy about CUDA either but apparently CUDA Does Work, for now. It Removed the OpenCL: 1.2 from the description, and I don't have any APs to test. Wed May 18 15:26:40 2016 | | Starting BOINC client version 7.6.32 for x86_64-apple-darwin Stupid program... Oh, and after Weeks of trying to work around the BOINC "Crash After Finish", the thing Still crashes AFTER the Results are written. BOINC says; Wed May 18 16:20:39 2016 | SETI@home | [sched_op] Reason: Unrecoverable error for task blc2_2bit_guppi_57451_27739_HIP69824_0027.28248.0.17.26.246.vlar_2 |
Chris Adamek Send message Joined: 15 May 99 Posts: 251 Credit: 434,772,072 RAC: 236 |
You could try the MB openCL app if you are in a hurry to try it. Doesn't look like there will be any AP's for a few days. I'm just glad I found an app that doesn't barf constantly on my mac. lol No gaussian problems since I went back to 3347. The midrange AR's sometimes take an oddly long amount of time to finish (I think its when they are running at the same time as VLARs on a given GPU) but I imagine I can fiddle with the settings to get that better. VLARs though are pretty fast. Effective completion time with that app is usually under 10 minutes each running 3 at a time. Chris |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Well, I don't think I need to bother with another App right now. So, BOINC is Also Broke in El Capitan 10.11.4...and it Does see the cards as GTX 950s; Thu May 19 08:06:37 2016 | | Starting BOINC client version 7.6.22 for x86_64-apple-darwin Why is the OpenCL section listing TWO GPU 1s? The card listings are Wrong as well. The 950s are in Slots 1 & 2, the 750Ti is in Slot 4. What's up with that??? |
Chris Adamek Send message Joined: 15 May 99 Posts: 251 Credit: 434,772,072 RAC: 236 |
It is surely something to do with the web driver/cuda drivers. I've not had any trouble with boinc manager in El Capitan with my 3 card nvidia Mac Pro. Had you tried El Capitan before you put a second 950 in? In Windows I often have to reinstall the nvidia drivers when I swap cards around to get them to be seen correctly, did you try reinstalling the web drivers & cuda yet? Chris |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Seems the Second GTX 950 Broke BOINC. Both OSes worked the same with just One GTX 950. It would Appear the Second GTX 950 has caused BOINC to think they are Both 9500 GTs even though they are listed as GTX 950s, hence the statement meant for pre-Fermi GPUs. Why it doesn't consider the GTX 750Ti before disabling OpenCL is anyone's guess as is why it is listing Two GPU 1s. Remember I told you it appears OpenCL is Broken in BOINC with Mavericks and above? The GTX 9xx series Doesn't work with the Apple supplied drivers, You have to use the Webdrivers. The cards work Fine with CUDA. |
Urs Echternacht Send message Joined: 15 May 99 Posts: 692 Credit: 135,197,781 RAC: 211 |
Both OSes worked the same with just One GTX 950.Idea : On win or linux it helped sometimes to switch GPU slots. _\|/_ U r s |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
On the Older Mac Pros Slots 1 & 2 are PCIe2 16x while slots 3 & 4 are PCIe1 4x. Also, you can't mount a double height card in Slot 4 without altering the faceplate. I'm not interested in altering another faceplate. Seems to me CUDA sees the cards correctly. The problem is with OpenCL detection. Not only is BOINC incapable of determining there is only One GPU 1, it's also incapable of reading what it printed for Compute Capability. If it would realize that CC 5.2 is Not a pre-Fermi card it wouldn't print that driver 6.5 warning. It seems the warning is coming from the OpenCL detection since it's listing GPU 1 & 2. Things improved a bit after rebooting back into Yosemite, now it appears it realizes the machine really does have OpenCL. At least the OpenCL: 1.2 is back and it's not complaining about the GPU not having OpenCL. Thu May 19 17:58:49 2016 | | Starting BOINC client version 7.6.32 for x86_64-apple-darwin Right now it's still working. |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14679 Credit: 200,643,578 RAC: 874 |
Is there a Mac/Darwin version of clinfo, which could perform the OpenCL card enumeration independently of BOINC? |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Is there a Mac/Darwin version of clinfo, which could perform the OpenCL card enumeration independently of BOINC? app itself does basic enumeration being started standalone. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
There are versions around you can compile yourself; Last login: Thu May 19 11:18:11 on console TomsMacPro:~ Tom$ cd /Users/Tom/MacCLinfo TomsMacPro:MacCLinfo Tom$ ./clinfo Found 1 platform(s). platform[0x7fff0000]: profile: FULL_PROFILE platform[0x7fff0000]: version: OpenCL 1.2 (Nov 12 2015 15:39:11) platform[0x7fff0000]: name: Apple platform[0x7fff0000]: vendor: Apple platform[0x7fff0000]: extensions: cl_APPLE_SetMemObjectDestructor cl_APPLE_ContextLoggingFunctions cl_APPLE_clut cl_APPLE_query_kernel_names cl_APPLE_gl_sharing cl_khr_gl_event platform[0x7fff0000]: Found 4 device(s). device[0xffffffff]: NAME: Intel(R) Xeon(R) CPU E5472 @ 3.00GHz device[0xffffffff]: VENDOR: Intel device[0xffffffff]: PROFILE: FULL_PROFILE device[0xffffffff]: VERSION: OpenCL 1.2 device[0xffffffff]: EXTENSIONS: cl_APPLE_SetMemObjectDestructor cl_APPLE_ContextLoggingFunctions cl_APPLE_clut cl_APPLE_query_kernel_names cl_APPLE_gl_sharing cl_khr_gl_event cl_khr_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_byte_addressable_store cl_khr_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_3d_image_writes cl_khr_image2d_from_buffer cl_APPLE_fp64_basic_ops cl_APPLE_fixed_alpha_channel_orders cl_APPLE_biased_fixed_point_image_formats cl_APPLE_command_queue_priority device[0xffffffff]: DRIVER_VERSION: 1.1 device[0xffffffff]: Type: CPU device[0xffffffff]: EXECUTION_CAPABILITIES: Kernel Native device[0xffffffff]: GLOBAL_MEM_CACHE_TYPE: Read-Write (2) device[0xffffffff]: CL_DEVICE_LOCAL_MEM_TYPE: Global (2) device[0xffffffff]: SINGLE_FP_CONFIG: 0xbf device[0xffffffff]: QUEUE_PROPERTIES: 0x2 device[0xffffffff]: VENDOR_ID: 4294967295 device[0xffffffff]: MAX_COMPUTE_UNITS: 8 device[0xffffffff]: MAX_WORK_ITEM_DIMENSIONS: 3 device[0xffffffff]: MAX_WORK_GROUP_SIZE: 1024 device[0xffffffff]: PREFERRED_VECTOR_WIDTH_CHAR: 16 device[0xffffffff]: PREFERRED_VECTOR_WIDTH_SHORT: 8 device[0xffffffff]: PREFERRED_VECTOR_WIDTH_INT: 4 device[0xffffffff]: PREFERRED_VECTOR_WIDTH_LONG: 2 device[0xffffffff]: PREFERRED_VECTOR_WIDTH_FLOAT: 4 device[0xffffffff]: PREFERRED_VECTOR_WIDTH_DOUBLE: 2 device[0xffffffff]: MAX_CLOCK_FREQUENCY: 3000 device[0xffffffff]: ADDRESS_BITS: 64 device[0xffffffff]: MAX_MEM_ALLOC_SIZE: 1610612736 device[0xffffffff]: IMAGE_SUPPORT: 1 device[0xffffffff]: MAX_READ_IMAGE_ARGS: 128 device[0xffffffff]: MAX_WRITE_IMAGE_ARGS: 8 device[0xffffffff]: IMAGE2D_MAX_WIDTH: 8192 device[0xffffffff]: IMAGE2D_MAX_HEIGHT: 8192 device[0xffffffff]: IMAGE3D_MAX_WIDTH: 2048 device[0xffffffff]: IMAGE3D_MAX_HEIGHT: 2048 device[0xffffffff]: IMAGE3D_MAX_DEPTH: 2048 device[0xffffffff]: MAX_SAMPLERS: 16 device[0xffffffff]: MAX_PARAMETER_SIZE: 4096 device[0xffffffff]: MEM_BASE_ADDR_ALIGN: 1024 device[0xffffffff]: MIN_DATA_TYPE_ALIGN_SIZE: 128 device[0xffffffff]: GLOBAL_MEM_CACHELINE_SIZE: 6291456 device[0xffffffff]: GLOBAL_MEM_CACHE_SIZE: 64 device[0xffffffff]: GLOBAL_MEM_SIZE: 6442450944 device[0xffffffff]: MAX_CONSTANT_BUFFER_SIZE: 65536 device[0xffffffff]: MAX_CONSTANT_ARGS: 8 device[0xffffffff]: LOCAL_MEM_SIZE: 32768 device[0xffffffff]: ERROR_CORRECTION_SUPPORT: 0 device[0xffffffff]: PROFILING_TIMER_RESOLUTION: 1 device[0xffffffff]: ENDIAN_LITTLE: 1 device[0xffffffff]: AVAILABLE: 1 device[0xffffffff]: COMPILER_AVAILABLE: 1 device[0x1022700]: NAME: GeForce GTX 750 Ti device[0x1022700]: VENDOR: NVIDIA device[0x1022700]: PROFILE: FULL_PROFILE device[0x1022700]: VERSION: OpenCL 1.2 device[0x1022700]: EXTENSIONS: cl_APPLE_SetMemObjectDestructor cl_APPLE_ContextLoggingFunctions cl_APPLE_clut cl_APPLE_query_kernel_names cl_APPLE_gl_sharing cl_khr_gl_event cl_khr_byte_addressable_store cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_APPLE_fp64_basic_ops cl_khr_fp64 cl_khr_3d_image_writes cl_khr_depth_images cl_khr_gl_depth_images cl_khr_gl_msaa_sharing cl_khr_image2d_from_buffer cl_APPLE_ycbcr_422 cl_APPLE_rgb_422 device[0x1022700]: DRIVER_VERSION: 10.5.2 346.02.03f06 device[0x1022700]: Type: GPU device[0x1022700]: EXECUTION_CAPABILITIES: Kernel device[0x1022700]: GLOBAL_MEM_CACHE_TYPE: None (0) device[0x1022700]: CL_DEVICE_LOCAL_MEM_TYPE: Local (1) device[0x1022700]: SINGLE_FP_CONFIG: 0x9f device[0x1022700]: QUEUE_PROPERTIES: 0x2 device[0x1022700]: VENDOR_ID: 16918272 device[0x1022700]: MAX_COMPUTE_UNITS: 5 device[0x1022700]: MAX_WORK_ITEM_DIMENSIONS: 3 device[0x1022700]: MAX_WORK_GROUP_SIZE: 1024 device[0x1022700]: PREFERRED_VECTOR_WIDTH_CHAR: 1 device[0x1022700]: PREFERRED_VECTOR_WIDTH_SHORT: 1 device[0x1022700]: PREFERRED_VECTOR_WIDTH_INT: 1 device[0x1022700]: PREFERRED_VECTOR_WIDTH_LONG: 1 device[0x1022700]: PREFERRED_VECTOR_WIDTH_FLOAT: 1 device[0x1022700]: PREFERRED_VECTOR_WIDTH_DOUBLE: 1 device[0x1022700]: MAX_CLOCK_FREQUENCY: 1110 device[0x1022700]: ADDRESS_BITS: 32 device[0x1022700]: MAX_MEM_ALLOC_SIZE: 536870912 device[0x1022700]: IMAGE_SUPPORT: 1 device[0x1022700]: MAX_READ_IMAGE_ARGS: 256 device[0x1022700]: MAX_WRITE_IMAGE_ARGS: 16 device[0x1022700]: IMAGE2D_MAX_WIDTH: 16384 device[0x1022700]: IMAGE2D_MAX_HEIGHT: 16384 device[0x1022700]: IMAGE3D_MAX_WIDTH: 2048 device[0x1022700]: IMAGE3D_MAX_HEIGHT: 2048 device[0x1022700]: IMAGE3D_MAX_DEPTH: 2048 device[0x1022700]: MAX_SAMPLERS: 32 device[0x1022700]: MAX_PARAMETER_SIZE: 4352 device[0x1022700]: MEM_BASE_ADDR_ALIGN: 1024 device[0x1022700]: MIN_DATA_TYPE_ALIGN_SIZE: 128 device[0x1022700]: GLOBAL_MEM_CACHELINE_SIZE: 0 device[0x1022700]: GLOBAL_MEM_CACHE_SIZE: 0 device[0x1022700]: GLOBAL_MEM_SIZE: 2147483648 device[0x1022700]: MAX_CONSTANT_BUFFER_SIZE: 65536 device[0x1022700]: MAX_CONSTANT_ARGS: 9 device[0x1022700]: LOCAL_MEM_SIZE: 49152 device[0x1022700]: ERROR_CORRECTION_SUPPORT: 0 device[0x1022700]: PROFILING_TIMER_RESOLUTION: 1000 device[0x1022700]: ENDIAN_LITTLE: 1 device[0x1022700]: AVAILABLE: 1 device[0x1022700]: COMPILER_AVAILABLE: 1 device[0x2022700]: NAME: Graphics Device device[0x2022700]: VENDOR: NVIDIA device[0x2022700]: PROFILE: FULL_PROFILE device[0x2022700]: VERSION: OpenCL 1.2 device[0x2022700]: EXTENSIONS: cl_APPLE_SetMemObjectDestructor cl_APPLE_ContextLoggingFunctions cl_APPLE_clut cl_APPLE_query_kernel_names cl_APPLE_gl_sharing cl_khr_gl_event cl_khr_byte_addressable_store cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_APPLE_fp64_basic_ops cl_khr_fp64 cl_khr_3d_image_writes cl_khr_depth_images cl_khr_gl_depth_images cl_khr_gl_msaa_sharing cl_khr_image2d_from_buffer cl_APPLE_ycbcr_422 cl_APPLE_rgb_422 device[0x2022700]: DRIVER_VERSION: 10.5.2 346.02.03f06 device[0x2022700]: Type: GPU device[0x2022700]: EXECUTION_CAPABILITIES: Kernel device[0x2022700]: GLOBAL_MEM_CACHE_TYPE: None (0) device[0x2022700]: CL_DEVICE_LOCAL_MEM_TYPE: Local (1) device[0x2022700]: SINGLE_FP_CONFIG: 0x9f device[0x2022700]: QUEUE_PROPERTIES: 0x2 device[0x2022700]: VENDOR_ID: 33695488 device[0x2022700]: MAX_COMPUTE_UNITS: 6 device[0x2022700]: MAX_WORK_ITEM_DIMENSIONS: 3 device[0x2022700]: MAX_WORK_GROUP_SIZE: 1024 device[0x2022700]: PREFERRED_VECTOR_WIDTH_CHAR: 1 device[0x2022700]: PREFERRED_VECTOR_WIDTH_SHORT: 1 device[0x2022700]: PREFERRED_VECTOR_WIDTH_INT: 1 device[0x2022700]: PREFERRED_VECTOR_WIDTH_LONG: 1 device[0x2022700]: PREFERRED_VECTOR_WIDTH_FLOAT: 1 device[0x2022700]: PREFERRED_VECTOR_WIDTH_DOUBLE: 1 device[0x2022700]: MAX_CLOCK_FREQUENCY: 1316 device[0x2022700]: ADDRESS_BITS: 32 device[0x2022700]: MAX_MEM_ALLOC_SIZE: 536870912 device[0x2022700]: IMAGE_SUPPORT: 1 device[0x2022700]: MAX_READ_IMAGE_ARGS: 256 device[0x2022700]: MAX_WRITE_IMAGE_ARGS: 16 device[0x2022700]: IMAGE2D_MAX_WIDTH: 16384 device[0x2022700]: IMAGE2D_MAX_HEIGHT: 16384 device[0x2022700]: IMAGE3D_MAX_WIDTH: 2048 device[0x2022700]: IMAGE3D_MAX_HEIGHT: 2048 device[0x2022700]: IMAGE3D_MAX_DEPTH: 2048 device[0x2022700]: MAX_SAMPLERS: 32 device[0x2022700]: MAX_PARAMETER_SIZE: 4352 device[0x2022700]: MEM_BASE_ADDR_ALIGN: 1024 device[0x2022700]: MIN_DATA_TYPE_ALIGN_SIZE: 128 device[0x2022700]: GLOBAL_MEM_CACHELINE_SIZE: 0 device[0x2022700]: GLOBAL_MEM_CACHE_SIZE: 0 device[0x2022700]: GLOBAL_MEM_SIZE: 2147483648 device[0x2022700]: MAX_CONSTANT_BUFFER_SIZE: 65536 device[0x2022700]: MAX_CONSTANT_ARGS: 9 device[0x2022700]: LOCAL_MEM_SIZE: 49152 device[0x2022700]: ERROR_CORRECTION_SUPPORT: 0 device[0x2022700]: PROFILING_TIMER_RESOLUTION: 1000 device[0x2022700]: ENDIAN_LITTLE: 1 device[0x2022700]: AVAILABLE: 1 device[0x2022700]: COMPILER_AVAILABLE: 1 device[0x3022700]: NAME: Graphics Device device[0x3022700]: VENDOR: NVIDIA device[0x3022700]: PROFILE: FULL_PROFILE device[0x3022700]: VERSION: OpenCL 1.2 device[0x3022700]: EXTENSIONS: cl_APPLE_SetMemObjectDestructor cl_APPLE_ContextLoggingFunctions cl_APPLE_clut cl_APPLE_query_kernel_names cl_APPLE_gl_sharing cl_khr_gl_event cl_khr_byte_addressable_store cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_APPLE_fp64_basic_ops cl_khr_fp64 cl_khr_3d_image_writes cl_khr_depth_images cl_khr_gl_depth_images cl_khr_gl_msaa_sharing cl_khr_image2d_from_buffer cl_APPLE_ycbcr_422 cl_APPLE_rgb_422 device[0x3022700]: DRIVER_VERSION: 10.5.2 346.02.03f06 device[0x3022700]: Type: GPU device[0x3022700]: EXECUTION_CAPABILITIES: Kernel device[0x3022700]: GLOBAL_MEM_CACHE_TYPE: None (0) device[0x3022700]: CL_DEVICE_LOCAL_MEM_TYPE: Local (1) device[0x3022700]: SINGLE_FP_CONFIG: 0x9f device[0x3022700]: QUEUE_PROPERTIES: 0x2 device[0x3022700]: VENDOR_ID: 50472704 device[0x3022700]: MAX_COMPUTE_UNITS: 6 device[0x3022700]: MAX_WORK_ITEM_DIMENSIONS: 3 device[0x3022700]: MAX_WORK_GROUP_SIZE: 1024 device[0x3022700]: PREFERRED_VECTOR_WIDTH_CHAR: 1 device[0x3022700]: PREFERRED_VECTOR_WIDTH_SHORT: 1 device[0x3022700]: PREFERRED_VECTOR_WIDTH_INT: 1 device[0x3022700]: PREFERRED_VECTOR_WIDTH_LONG: 1 device[0x3022700]: PREFERRED_VECTOR_WIDTH_FLOAT: 1 device[0x3022700]: PREFERRED_VECTOR_WIDTH_DOUBLE: 1 device[0x3022700]: MAX_CLOCK_FREQUENCY: 1316 device[0x3022700]: ADDRESS_BITS: 32 device[0x3022700]: MAX_MEM_ALLOC_SIZE: 536870912 device[0x3022700]: IMAGE_SUPPORT: 1 device[0x3022700]: MAX_READ_IMAGE_ARGS: 256 device[0x3022700]: MAX_WRITE_IMAGE_ARGS: 16 device[0x3022700]: IMAGE2D_MAX_WIDTH: 16384 device[0x3022700]: IMAGE2D_MAX_HEIGHT: 16384 device[0x3022700]: IMAGE3D_MAX_WIDTH: 2048 device[0x3022700]: IMAGE3D_MAX_HEIGHT: 2048 device[0x3022700]: IMAGE3D_MAX_DEPTH: 2048 device[0x3022700]: MAX_SAMPLERS: 32 device[0x3022700]: MAX_PARAMETER_SIZE: 4352 device[0x3022700]: MEM_BASE_ADDR_ALIGN: 1024 device[0x3022700]: MIN_DATA_TYPE_ALIGN_SIZE: 128 device[0x3022700]: GLOBAL_MEM_CACHELINE_SIZE: 0 device[0x3022700]: GLOBAL_MEM_CACHE_SIZE: 0 device[0x3022700]: GLOBAL_MEM_SIZE: 2147483648 device[0x3022700]: MAX_CONSTANT_BUFFER_SIZE: 65536 device[0x3022700]: MAX_CONSTANT_ARGS: 9 device[0x3022700]: LOCAL_MEM_SIZE: 49152 device[0x3022700]: ERROR_CORRECTION_SUPPORT: 0 device[0x3022700]: PROFILING_TIMER_RESOLUTION: 1000 device[0x3022700]: ENDIAN_LITTLE: 1 device[0x3022700]: AVAILABLE: 1 device[0x3022700]: COMPILER_AVAILABLE: 1 TomsMacPro:MacCLinfo Tom$ |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Missing names and compute capabilities imply a driver issue there. Probably the compute capability fields are zero and omitted, which would trigger the behaviour you see in the Boinc client. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Except the Compute Capabilities are listed by BOINC and in El Capitan the cards are identified as GTX 950. It didn't seem to make a difference. Thu May 19 08:06:37 2016 | | Starting BOINC client version 7.6.22 for x86_64-apple-darwin |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Could be they're matching up Cuda-OpenCL Devices by name or similar parameter then (would have to look at the current detection code later). Only problem with that is the only guaranteed way to get the same device in the different enumerations, is by PCIe Bus+Slot# . What happens if you do just the 2 x 950s ? "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.