No OpenCL library found, Mint Linux on AMD CPU/GPU.

Skip Da Shu
Skip Da Shu
Joined: 18 Jan 05
Posts: 151
Credit: 1030200633
RAC: 1021276

Thanx guys. Right now I'm

Thanx guys.

Right now I'm back with the

$ sudo amdgpu-install -y --usecase=opencl --opencl=legacy --accept-eula

and the 580 with the iGPU on.  W/o the rocm BOINC doesn't see the iGPU as a valid device.

It's starting to get warm so my next mission will be to put the Iceberg whirling airplane noise exhaust fan in the case and re-do the BIOS fan curve for it then...

I'll uninstall amdgpu, re-install with rocm, swap the 6600 back in, disable the iGPU in BIOS and see what happens.

The MEMORY error MIGHT be related to the UDMA being allocated to support the iGPU.  I seem to remember seeing the 256MB (now 64MB) looking like it was being assigned to both the iGPU and the GPU.  We'll see.

 

Skip Da Shu
Skip Da Shu
Joined: 18 Jan 05
Posts: 151
Credit: 1030200633
RAC: 1021276

$ sudo amdgpu-install -y

$ sudo amdgpu-install -y --usecase=opencl --opencl=legacy,rocr --accept-eula

 

2023-04-20 15:47:32.4779 (2863) [normal]: Validating SFTs (detectors: H1, L1, ) ... success.
XLAL Error - XLALOpenCLInit (/home/jenkins/workspace/workspace/EaH-GW-OpenCL-Testing/SLAVE/LIBC215/TARGET/linux-x86_64/EinsteinAtHome/source/lalsuite/lalpulsar/lib/GPUUtils/OpenCLUtils.c:379): OpenCL CreateCommandQueue failed with OpenCL error: CL_OUT_OF_HOST_MEMORY

etc/OpenCL/vendors$ dir has these 3 files:
amdocl64_50403_121.icd     has libamdocl64.so

amdocl-orca64.icd                has libamdocl-orca64.so

nvidia.icd                              has libnvidia-opencl.so.1

 

I'm about to give up and send the 6600 back.  I would assume I'd have the same issues with any Navi based card.

 

PS:  mountkidd - why r u running w/ anonymous platform?

mountkidd
mountkidd
Joined: 14 Jun 12
Posts: 176
Credit: 12433192555
RAC: 8103808

Skip Da Shu wrote: ... I'm

Skip Da Shu wrote:

...

I'm about to give up and send the 6600 back.  I would assume I'd have the same issues with any Navi based card.

 

PS:  mountkidd - why r u running w/ anonymous platform?

I see BRP7, GW and GR tasks on your computer with the 6600.  Are you trying to run all of these concurrently and more than one of each at a time? You just might have too much going on for the 8GB card memory.  Run GW only at x1 to see if you can complete tasks successfully - you shouldn't be running out of memory with this configuration.  Then try GR only at x1...  Then try BRP7 only at x1...  If you are wanting to run GR & GW at the same time, you might be able to get away with x2 on GR + x1 on GW.   Running a mix of apps on one host might not be very productive.  With your collection of hosts, you might be more successful dedicating a host to a specific app.

Your 6600 should be ~2x more productive than your 580 with similar power consumption.  I'd keep the 6600 and send the 580 back...  Sorting out the 6600 and your runtime config is the best way forward.

I switched all my AMD cards to NVIDIA and am running the Petri GR app which requires NV/Linux/anonymous platform.  Petri GR & BRP7 apps outperform the standard project apps.

mikey
mikey
Joined: 22 Jan 05
Posts: 12640
Credit: 1839026286
RAC: 5469

Skip Da Shu wrote: $ sudo

Skip Da Shu wrote:

$ sudo amdgpu-install -y --usecase=opencl --opencl=legacy,rocr --accept-eula

 

2023-04-20 15:47:32.4779 (2863) [normal]: Validating SFTs (detectors: H1, L1, ) ... success.
XLAL Error - XLALOpenCLInit (/home/jenkins/workspace/workspace/EaH-GW-OpenCL-Testing/SLAVE/LIBC215/TARGET/linux-x86_64/EinsteinAtHome/source/lalsuite/lalpulsar/lib/GPUUtils/OpenCLUtils.c:379): OpenCL CreateCommandQueue failed with OpenCL error: CL_OUT_OF_HOST_MEMORY

etc/OpenCL/vendors$ dir has these 3 files:
amdocl64_50403_121.icd     has libamdocl64.so

amdocl-orca64.icd                has libamdocl-orca64.so

nvidia.icd                              has libnvidia-opencl.so.1

 

I'm about to give up and send the 6600 back.  I would assume I'd have the same issues with any Navi based card.

How much memory do you have in that pc? I run a 6600 just fine in one of my windows pc's, maybe you can setup a VM and run it that way.

Skip Da Shu
Skip Da Shu
Joined: 18 Jan 05
Posts: 151
Credit: 1030200633
RAC: 1021276

Dang... power glitched and

Dang... power glitched and everything shut down... of course I'd been replying to you with links and all...

Highlights... All of my running RX 580s are Sapphire 4GB cards.  Project web page has 0.5 settings for all 3 apps using every bit of the 4096 on these cards.  No longer have a project config.xml on E@H.

MW runx 3x Sep but uses less vram and more wattage/heat:

<app>
  <name>milkyway</name>
  <gpu_versions>
    <cpu_usage>0.5</cpu_usage>
    <gpu_usage>0.333</gpu_usage>
  </gpu_versions>
 </app>

They only mix when the 'puter is finishing up a batch of one or the other and sometimes during that 'transition' to the next projects batch a card will run an MW or two with the longer running GW.  No problems to date.

Skip

PS:  I've got till mid-May to have the 6600 back if that's the route I take.

PPS: The two single RX 580 card 'puters have 32GB in them.  Nearly identical boxes skip-MS7C95-1 & skip-MS7C95-2

Skip Da Shu
Skip Da Shu
Joined: 18 Jan 05
Posts: 151
Credit: 1030200633
RAC: 1021276

OK, time for another shot at

OK, time for another shot at the RX 6000 under Linux Mint:

1st I shut it down, used systemctl to disable boinc-client so it wouldn't start up.

I swapped the RX 6600 in where the RX 580 was and powered the box back up (iGPU still on and my monitor is on it).

1st a ran amdgpu-uninstall.

2nd $ sudo amdgpu-install -y --usecase=opencl --opencl=rocr --accept-eula

3rd

$ clinfo
Number of platforms                               1
  Platform Name                                   AMD Accelerated Parallel Processing
  Platform Vendor                                 Advanced Micro Devices, Inc.
  Platform Version                                OpenCL 2.1 AMD-APP (3513.0)
  Platform Profile                                FULL_PROFILE
  Platform Extensions                             cl_khr_icd cl_amd_event_callback
  Platform Host timer resolution                  1ns
  Platform Extensions function suffix             AMD

  Platform Name                                   AMD Accelerated Parallel Processing
Number of devices                                 2
  Device Name                                     gfx1032
  Device Vendor                                   Advanced Micro Devices, Inc.
  Device Vendor ID                                0x1002
  Device Version                                  OpenCL 2.0
  Driver Version                                  3513.0 (HSA1.1,LC)
  Device OpenCL C Version                         OpenCL C 2.0
  Device Type                                     GPU
  Device Board Name (AMD)                         AMD Radeon RX 6600

.

. deleted

.

  Device Name                                     gfx90c:xnack-
  Device Vendor                                   Advanced Micro Devices, Inc.
  Device Vendor ID                                0x1002
  Device Version                                  OpenCL 2.0
  Driver Version                                  3513.0 (HSA1.1,LC)
  Device OpenCL C Version                         OpenCL C 2.0
  Device Type                                     GPU
  Device Board Name (AMD)                         (n/a)
  Device Topology (AMD)                           PCI-E, 30:00.0

.

.

.
  Max work item sizes                             1024x1024x1024
  Max work group size                             256
  Preferred work group size (AMD)                 256
  Max work group size (AMD)                       1024

Memory access fault by GPU node-2 (Agent handle: 0x562017d375c0) on address (nil). Reason: Page not present or supervisor privilege.
Aborted (core dumped)

SAVED HERE SO I COULD REBOOT W/O iGPU enabled after seeing the clinfo abort on the iGPU) and new clinfo below w/o iGPU.

$ clinfo
Number of platforms                               1
  Platform Name                                   AMD Accelerated Parallel Processing
  Platform Vendor                                 Advanced Micro Devices, Inc.
  Platform Version                                OpenCL 2.1 AMD-APP (3513.0)
  Platform Profile                                FULL_PROFILE
  Platform Extensions                             cl_khr_icd cl_amd_event_callback
  Platform Host timer resolution                  1ns
  Platform Extensions function suffix             AMD

  Platform Name                                   AMD Accelerated Parallel Processing
Number of devices                                 1
  Device Name                                     gfx1032
  Device Vendor                                   Advanced Micro Devices, Inc.
  Device Vendor ID                                0x1002
  Device Version                                  OpenCL 2.0
  Driver Version                                  3513.0 (HSA1.1,LC)
  Device OpenCL C Version                         OpenCL C 2.0
  Device Type                                     GPU
  Device Board Name (AMD)                         AMD Radeon RX 6600
  Device Topology (AMD)                           PCI-E, 12:00.0
  Device Profile                                  FULL_PROFILE
  Device Available                                Yes
  Compiler Available                              Yes
  Linker Available                                Yes
  Max compute units                               14
  SIMD per compute unit (AMD)                     4
  SIMD width (AMD)                                32
  SIMD instruction width (AMD)                    1
  Max clock frequency                             2750MHz
  Graphics IP (AMD)                               10.3
  Device Partition                                (core)
    Max number of sub-devices                     14
    Supported partition types                     None
    Supported affinity domains                    (n/a)
  Max work item dimensions                        3
  Max work item sizes                             1024x1024x1024
  Max work group size                             256
  Preferred work group size (AMD)                 256
  Max work group size (AMD)                       1024
  Preferred work group size multiple              32
  Wavefront width (AMD)                           32
  Preferred / native vector sizes                 
    char                                                 4 / 4       
    short                                                2 / 2       
    int                                                  1 / 1       
    long                                                 1 / 1       
    half                                                 1 / 1        (cl_khr_fp16)
    float                                                1 / 1       
    double                                               1 / 1        (cl_khr_fp64)
  Half-precision Floating-point support           (cl_khr_fp16)
    Denormals                                     No
    Infinity and NANs                             No
    Round to nearest                              No
    Round to zero                                 No
    Round to infinity                             No
    IEEE754-2008 fused multiply-add               No
    Support is emulated in software               No
  Single-precision Floating-point support         (core)
    Denormals                                     Yes
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 Yes
    Round to infinity                             Yes
    IEEE754-2008 fused multiply-add               Yes
    Support is emulated in software               No
    Correctly-rounded divide and sqrt operations  Yes
  Double-precision Floating-point support         (cl_khr_fp64)
    Denormals                                     Yes
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 Yes
    Round to infinity                             Yes
    IEEE754-2008 fused multiply-add               Yes
    Support is emulated in software               No
  Address bits                                    64, Little-Endian
  Global memory size                              8573157376 (7.984GiB)
  Global free memory (AMD)                        8372224 (7.984GiB)
  Global memory channels (AMD)                    4
  Global memory banks per channel (AMD)           4
  Global memory bank width (AMD)                  256 bytes
  Error Correction support                        No
  Max memory allocation                           7287183768 (6.787GiB)
  Unified memory for Host and Device              No
  Shared Virtual Memory (SVM) capabilities        (core)
    Coarse-grained buffer sharing                 Yes
    Fine-grained buffer sharing                   Yes
    Fine-grained system sharing                   No
    Atomics                                       No
  Minimum alignment for any data type             128 bytes
  Alignment of base address                       1024 bits (128 bytes)
  Preferred alignment for atomics                 
    SVM                                           0 bytes
    Global                                        0 bytes
    Local                                         0 bytes
  Max size for global variable                    7287183768 (6.787GiB)
  Preferred total size of global vars             8573157376 (7.984GiB)
  Global Memory cache type                        Read/Write
  Global Memory cache size                        16384 (16KiB)
  Global Memory cache line size                   64 bytes
  Image support                                   Yes
    Max number of samplers per kernel             29695
    Max size for 1D images from buffer            134217728 pixels
    Max 1D or 2D image array size                 8192 images
    Base address alignment for 2D image buffers   256 bytes
    Pitch alignment for 2D image buffers          256 pixels
    Max 2D image size                             16384x16384 pixels
    Max 3D image size                             16384x16384x8192 pixels
    Max number of read image args                 128
    Max number of write image args                8
    Max number of read/write image args           64
  Max number of pipe args                         16
  Max active pipe reservations                    16
  Max pipe packet size                            2992216472 (2.787GiB)
  Local memory type                               Local
  Local memory size                               65536 (64KiB)
  Local memory syze per CU (AMD)                  65536 (64KiB)
  Local memory banks (AMD)                        32
  Max number of constant args                     8
  Max constant buffer size                        7287183768 (6.787GiB)
  Preferred constant buffer size (AMD)            16384 (16KiB)
  Max size of kernel argument                     1024
  Queue properties (on host)                      
    Out-of-order execution                        No
    Profiling                                     Yes
  Queue properties (on device)                    
    Out-of-order execution                        Yes
    Profiling                                     Yes
    Preferred size                                262144 (256KiB)
    Max size                                      8388608 (8MiB)
  Max queues on device                            1
  Max events on device                            1024
  Prefer user sync for interop                    Yes
  Number of P2P devices (AMD)                     0
  P2P devices (AMD)                               <printDeviceInfo:147: get number of CL_DEVICE_P2P_DEVICES_AMD : error -30>
  Profiling timer resolution                      1ns
  Profiling timer offset since Epoch (AMD)        0ns (Wed Dec 31 18:00:00 1969)
  Execution capabilities                          
    Run OpenCL kernels                            Yes
    Run native kernels                            No
    Thread trace supported (AMD)                  No
    Number of async queues (AMD)                  8
    Max real-time compute queues (AMD)            8
    Max real-time compute units (AMD)             14
  printf() buffer size                            4194304 (4MiB)
  Built-in kernels                                (n/a)
  Device Extensions                               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_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_fp16 cl_khr_gl_sharing cl_amd_device_attribute_query cl_amd_media_ops cl_amd_media_ops2 cl_khr_image2d_from_buffer cl_khr_subgroups cl_khr_depth_images cl_amd_copy_buffer_p2p cl_amd_assembly_program

NULL platform behavior
  clGetPlatformInfo(NULL, CL_PLATFORM_NAME, ...)  No platform
  clGetDeviceIDs(NULL, CL_DEVICE_TYPE_ALL, ...)   No platform
  clCreateContext(NULL, ...) [default]            No platform
  clCreateContext(NULL, ...) [other]              Success [AMD]
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_DEFAULT)  Success (1)
    Platform Name                                 AMD Accelerated Parallel Processing
    Device Name                                   gfx1032
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CPU)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_GPU)  Success (1)
    Platform Name                                 AMD Accelerated Parallel Processing
    Device Name                                   gfx1032
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_ACCELERATOR)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CUSTOM)  No devices found in platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_ALL)  Success (1)
    Platform Name                                 AMD Accelerated Parallel Processing
    Device Name                                   gfx1032

Other than to start BOINC back I'm not sure what else I can do at this point.

1 E@H WU attempted and crashed:  https://einsteinathome.org/task/1458703384

Same error about creating the OpenCL command queue.

boinc_get_opencl_ids returned [0x14f6e10 , 0x7f99e0c6be50] 
Using OpenCL platform provided by: Advanced Micro Devices, Inc.
Using OpenCL device "gfx1032" by: Advanced Micro Devices, Inc.
Max allocation limit: 7287183768
Global mem size: 8573157376
Couldn't create OpenCL command queue (error: -6)!
OpenCL shutdown complete!
initialize_ocl returned error [2013]
OCL context null
OCL queue null
Error generating generic FFT context object [5]
17:11:40 (8674): [CRITICAL]: ERROR: MAIN() returned with error '5'

Radeon-Profile shows NO load on the processor and no utilization of the vram with the main clock bouncing from 0MHz to 800Mhz and back.

Also noticed this on some pending/in que BRP WUs (all the GW and MW WUs bombed out within seconds):

Fri 21 Apr 2023 05:20:47 PM CDT | Einstein@Home | Task M22_1_dns_cfbf00017_segment_8_dms_100_20000_32_2950000_2 postponed for 900 seconds: Not enough free CPU/GPU memory available! Delaying next attempt for at least 15 minutes...

Only thing I can think to try is to go back to BIOS, re-enable the iGPU so I can set a larger UDMA and then disable the iGPU again.

SAVE for reboot.

No change.  Did notice the 6600 is only using 19w (from Radeon-Profile) which has to be idle wattage.

Can anyone bail me out?

Skip

mountkidd
mountkidd
Joined: 14 Jun 12
Posts: 176
Credit: 12433192555
RAC: 8103808

Your recent opencl installs

Your recent opencl installs appear to be ok (installing opencl=rocr,legacy will eliminate a bunch of re-installs) but there is one more thing to check. In your boinc-client.service file set parameter ProtectSystem=off.  There has been a problem with boinc versions newer than 7.18.x and more recent amdgpu drivers.

Disable iGPU for boinc or do it in the bios.  It is recognized as a valid rocm device and you don't want boinc feeding it dGPU work.

Skip Da Shu
Skip Da Shu
Joined: 18 Jan 05
Posts: 151
Credit: 1030200633
RAC: 1021276

mountkidd wrote:Your recent

mountkidd wrote:

Your recent opencl installs appear to be ok (installing opencl=rocr,legacy will eliminate a bunch of re-installs) but there is one more thing to check. In your boinc-client.service file set parameter ProtectSystem=off.  There has been a problem with boinc versions newer than 7.18.x and more recent amdgpu drivers.

Disable iGPU for boinc or do it in the bios.  It is recognized as a valid rocm device and you don't want boinc feeding it dGPU work.

boinc-client.service part is done.  Should be able to swap cards and do driver uninstall / reinstall (currently have opencl=legacy) in a couple hours.

Thanx.

EDIT:  Looks like it's gonna be late tonight or tomorrow b4 I can swap the cards.  OK, It'll be Monday.

Skip Da Shu
Skip Da Shu
Joined: 18 Jan 05
Posts: 151
Credit: 1030200633
RAC: 1021276

RX 6600 again...1st I

RX 6600 again...

1st I used systemctl to stop and disable BOINC so it wouldn't autostart after reboots.

Then I ran $ amdgpu-uninstall and did the dpkg grep for opencl to make sure no mesa lurking about.

Next up was the install for both opencl & rocm

sudo amdgpu-install -y --usecase=opencl --opencl=legacy,rocr --accept-eula

Ran clinfo and saved output if it's needed.  A quick perusal didn't see anything new.

Confirmed boinc-client.service has:

#ProtectSystem=strict
ProtectSystem=off

Guess I'm ready for a reboot and see what BOINC thinks (Radeon-Profile will autostart so I'll see that first).

SAVE FOR REBOOT.

1st thing I noticed was 0 RPM fan (from Radeon-Profile).  Went to 'fixed' fan and set 30% and it comes alive and reads ~850 rpm.  I know this comes from the card but I believe it may come thru amdgpu to get to RP.  Also different sensors shown in PSensors: fan1, amdgpu edge, amdgpu junction and amdgpu mem temp.  The only point is amdgpu is exposing the sensors to these monitoring tools.

Well, I guess I can't put it off any longer:  $ sudo systemctl start boinc-client

Mon 24 Apr 2023 02:57:37 PM CDT |  | Starting BOINC client version 7.20.5 for x86_64-pc-linux-gnu
Mon 24 Apr 2023 02:57:37 PM CDT |  | This a development version of BOINC and may not function properly
Mon 24 Apr 2023 02:57:37 PM CDT |  | log flags: file_xfer, sched_ops, task
Mon 24 Apr 2023 02:57:37 PM CDT |  | Libraries: libcurl/7.68.0 OpenSSL/1.1.1f zlib/1.2.11 brotli/1.0.7 libidn2/2.2.0 libpsl/0.21.0 (+libidn2/2.2.0) libssh/0.9.3/openssl/zlib nghttp2/1.40.0 librtmp/2.3
Mon 24 Apr 2023 02:57:37 PM CDT |  | Data directory: /var/lib/boinc-client
Mon 24 Apr 2023 02:57:37 PM CDT |  | OpenCL: AMD/ATI GPU 0: AMD Radeon RX 6600 (driver version 3513.0 (HSA1.1,LC), device version OpenCL 2.0, 8176MB, 8176MB available, 9856 GFLOPS peak)
Mon 24 Apr 2023 02:57:37 PM CDT |  | libc:  version 2.31
Mon 24 Apr 2023 02:57:37 PM CDT |  | Host name: skip-MS7C95-1

OMG!

It finished a VERY short PrimeGrid WU (sub 1 minute) on the card... will go look at that in a minute... and a GammaRay with a BRP and they HAVEN'T errored for >4 minutes.

BEFORE I absolutely go ecstatic let me go look at that PrimeGird WU that finished and in 10 minutes the GammaRay will be done.

Trying to contain myself :-)

Skip

 

 

Skip Da Shu
Skip Da Shu
Joined: 18 Jan 05
Posts: 151
Credit: 1030200633
RAC: 1021276

From only a couple completed

From only a couple completed tasks and just rough numbers:

GammaRay - 32.5% decrease in times with RX 6600 over the slightly oc'd RX580

BuRP - 45% decrease in times 6600 over 580

 

Other projects:

 

Prime Grid works ...

GFN18 - 6600 has reduced times by 53%.

GFN17 - 6600 reduces run times by 27%.

GFN16 - I don't see any gain on these and possibly a very slight degradation.

GFN15 - the 580 was anywhere from 77-105 seconds, the two 6600 completions where both a hair under 50s..

AP27 - 6600 reduces run times by 58%.

 

MW Separation tasks fail for reasons Ian&Steve C. have already found:


http://man.opencl.org/deprecated.html

OpenCL Features Deprecated in OpenCL 2.0

clCreateCommandQueue

 

Error creating command queue (-6): CL_OUT_OF_HOST_MEMORY

 

OK, 'nuff of this.

A few side notes.  My used Sapphire RX 580 (Pulse variant) had a TDP/PPT cap of 155w (the Nitro's have 160w cap).  I had the core clock pushed to 1420MHz (think it comes at 1366) and the memory @ 1785MHz (vs 1750).  If loaded enough it would sometimes run up close to that 155w PPT cap and start backing the clock down to around 1400MHz.  The Nitro+ versions will hold closer to 1420MHz due to the 160w PPT.

The RX 6600 (an ASUS card, also bought used) has 100w TDP/PPT cap and I thought a 2750MHz clock but it seems to think it's 1615~1620MHz and a mem clock of 875MHz.  Radeon-Profile doesn't seem to be able to move those but it will set a multi-point fan curve and I have as yet to see it get over 59C and right now it's at 55C running 2x  BuRPs.  A significant bit cooler than the RX 580 ran and is only reporting 81~89w against it's 100w cap.  It does however seem to be much warier of that 100w cap as above 81w it seems to draw the clock down from 2620 to 2615MHz.  It might really run 2750MHz at a mid range load (time will tell).  None of which bothers me at this point.

Now for the important stuff:

I owe MOUNTKIDD a HUGE debt of gratitude for all his help with this box and it's AMD cards over the last few weeks.  I don't know how I'll ever repay him.  Many thanx my friend.

Skip

 

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.