There is new MultiBeam build added to testing - OpenCL one for Intel GPUs.

Please report any found issues in this thread.

Well, my offline bench reference tasks are ready for a pre-release validation test now - but I see we didn't need to wait for that?

Well, my offline bench reference tasks are ready for a pre-release validation test now - but I see we didn't need to wait for that?

If you find smth wrong in offline validation we will retract app. With only 2 hosts for alpha we could not do too much.

Well, my offline bench reference tasks are ready for a pre-release validation test now - but I see we didn't need to wait for that?

If you find smth wrong in offline validation we will retract app. With only 2 hosts for alpha we could not do too much.

Now that we have some agreed observations for AP, I'll set the bench running overnight. If it finishes before my train in the morning, I'll try and post it - if not, at least it'll be ready to upload when I get back late on Saturday evening. Bench run will have to be at 100% CPU usage, so as not to interfere with the ongoing AP timing run.

Bench test has completed and been reported. Live running has commenced - successfully, if rather slowly.

As Werinbert has found out in the Astropulse area, Intel GPU crunching requires that there is a monitor loading (monitor, dummy plug, or secondary video cable) connected to the Intel video output. How quickly we all forget these things!

interesting error case:
http://setiweb.ssl.berkeley.edu/beta/result.php?resultid=15247583

both app and BOINC failed to get info from device though BOINc detected OpenCL 1.1 GPU on host.
worth to report to BOINc team: BOINC 7.2.5

worth to report to BOINc team: BOINC 7.2.5

I'd say not. That's an absurdly old version to be trying to run intel_gpu apps under - there's no way the BOINC team will be dragged that far back.

He was already about six versions out of date when he set that machine up!

7.2.5 ??? I running 7.0.64 on my host and no issues.
Are you sure ?

Looks like BOINC changes versions like socks - better less but better, as one man said ;)
If 7.0.64 can operate intel GPU why 7.2.5 should not ???

[if we would go from one set of features to more rich set of features new versions would be good. But mostly we go from one set of bugs to another unknown one, and in this case better known bug than unknown one :P. It's true for BOINC API as well as for BOINC client binaries]

If 7.0.64 can operate intel GPU why 7.2.5 should not ???

Because 7.2.5 is a very early development version of the Boinc 7.2.x line, where they introduced detecting GPUs via a child process.

Claggy

I did have an inconclusive with another Intel GPU running an older driver, looks like precision problem with the older 1.1 drivers:

http://setiweb.ssl.berkeley.edu/beta/workunit.php?wuid=5822771

Check out the differences in Red between my result and my intel GPU wingman (I started at the bottom and have only gone up as far as the Best spike)

Eric might want to restrict work to intel GPUs running OpenCL 1.2 drivers.

My Result:

OpenCL version by Raistmer, r2061

Number of OpenCL platforms: 2


OpenCL Platform Name: Intel(R) OpenCL
Number of devices: 1
Max compute units: 16
Max work group size: 512
Max clock frequency: 350Mhz
Max memory allocation: 425721856
Cache type: Read/Write
Cache line size: 64
Cache size: 2097152
Global memory size: 1702887424
Constant buffer size: 65536
Max number of constant args: 8
Local memory type: Scratchpad
Local memory size: 65536
Queue properties:
Out-of-Order: No
Name: Intel(R) HD Graphics 4000
Vendor: Intel(R) Corporation
Driver version: 9.18.10.3257
Version: OpenCL 1.2
Extensions: cl_khr_icd 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_gl_sharing cl_khr_d3d10_sharing cl_intel_dx9_media_sharing cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_gl_event cl_khr_gl_msaa_sharing cl_khr_depth_images cl_khr_gl_depth_images cl_khr_dx9_media_sharing cl_khr_d3d11_sharing cl_khr_image2d_from_buffer



Work Unit Info:
...............
Credit multiplier is : 2.85
WU true angle range is : 0.429362
Used GPU device parameters are:
Number of compute units: 16
Single buffer allocation size: 64MB
max WG size: 512
period_iterations_num=20
Spike: peak=24.84762, time=73.82, d_freq=1418789733.65, chirp=-0.83461, fft_len=128k
Spike: peak=24.81097, time=73.82, d_freq=1418789733.66, chirp=-0.83553, fft_len=128k
Spike: peak=24.61703, time=6.711, d_freq=1418790071.32, chirp=-9.1576, fft_len=128k
Spike: peak=24.60221, time=6.711, d_freq=1418790071.31, chirp=-9.1585, fft_len=128k
Pulse: peak=0.5884006, time=31.26, period=0.07578, d_freq=1418788287.11, score=1.017, chirp=-10.159, fft_len=64
Gaussian: peak=3.264745, mean=0.5102067, ChiSq=1.397039, time=12.58, d_freq=1418787335.69,
score=2.879469, null_hyp=2.392513, chirp=-68.975, fft_len=16k
Gaussian: peak=3.367662, mean=0.5108123, ChiSq=1.26971, time=14.26, d_freq=1418787219.97,
score=2.433692, null_hyp=2.298166, chirp=-68.975, fft_len=16k

Best spike: peak=24.84762, time=73.82, d_freq=1418789733.65, chirp=-0.83461, fft_len=128k
Best autocorr: peak=16.28008, time=6.711, delay=6.3508, d_freq=1418789176.24, chirp=16.949, fft_len=128k
Best gaussian: peak=3.264745, mean=0.5102067, ChiSq=1.397039, time=12.58, d_freq=1418787335.69,
score=2.879469, null_hyp=2.392513, chirp=-68.975, fft_len=16k
Best pulse: peak=0.5884006, time=31.26, period=0.07578, d_freq=1418788287.11, score=1.017, chirp=-10.159, fft_len=64
Best triplet: peak=0, time=-2.121e+011, period=0, d_freq=0, chirp=0, fft_len=0


Flopcounter: 21426740938298.609000

Spike count: 4
Autocorr count: 0
Pulse count: 1
Triplet count: 0
Gaussian count: 2
Wallclock time elapsed since last restart: 7443.8 seconds

My intel GPU Wingman's result:

OpenCL version by Raistmer, r2061

Number of OpenCL platforms: 1


OpenCL Platform Name: Intel(R) OpenCL
Number of devices: 1
Max compute units: 6
Max work group size: 256
Max clock frequency: 350Mhz
Max memory allocation: 425721856
Cache type: Read/Write
Cache line size: 64
Cache size: 2097152
Global memory size: 1702887424
Constant buffer size: 65536
Max number of constant args: 8
Local memory type: Scratchpad
Local memory size: 65536
Queue properties:
Out-of-Order: No
Name: Intel(R) HD Graphics 2500
Vendor: Intel(R) Corporation
Driver version: 8.15.10.2712
Version: OpenCL 1.1
Extensions: cl_khr_icd 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_gl_sharing cl_khr_d3d10_sharing cl_intel_dx9_media_sharing cl_khr_3d_image_writes cl_khr_byte_addressable_store


Work Unit Info:
...............
Credit multiplier is : 2.85
WU true angle range is : 0.429362
Used GPU device parameters are:
Number of compute units: 6
Single buffer allocation size: 64MB
max WG size: 256
period_iterations_num=20
Spike: peak=24.84763, time=73.82, d_freq=1418789733.65, chirp=-0.83461, fft_len=128k
Spike: peak=24.81097, time=73.82, d_freq=1418789733.66, chirp=-0.83553, fft_len=128k
Autocorr: peak=25083.42, time=33.55, delay=3.9507, d_freq=1418788818.71, chirp=-7.2656, fft_len=128k
Spike: peak=24.61703, time=6.711, d_freq=1418790071.32, chirp=-9.1576, fft_len=128k
Spike: peak=24.60222, time=6.711, d_freq=1418790071.31, chirp=-9.1585, fft_len=128k
Pulse: peak=0.5884005, time=31.26, period=0.07578, d_freq=1418788287.11, score=1.017, chirp=-10.159, fft_len=64
GPU device synched
Termination request detected or computations are finished. GPU device synched, exiting...
Running on device number: 0
Priority of worker thread raised successfully
Priority of process adjusted successfully, below normal priority class used
OpenCL platform detected: Intel(R) Corporation
BOINC assigns device 0
Info: BOINC provided OpenCL device ID used

Build features: SETI7 Non-graphics OpenCL USE_OPENCL_INTEL OCL_SYNCHED OCL_CHIRP3 FFTW AMD specific USE_SSE x86
CPUID: Intel(R) Core(TM) i3-3220 CPU @ 3.30GHz

Cache: L1=64K L2=256K

CPU features: FPU TSC PAE CMPXCHG8B APIC SYSENTER MTRR CMOV/CCMP MMX FXSAVE/FXRSTOR SSE SSE2 HT SSE3 SSSE3 SSE4.1 SSE4.2 AVX
OpenCL-kernels filename : MultiBeam_Kernels_r2061.cl
ar=0.429362 NumCfft=195439 NumGauss= 1100839256 NumPulse= 118452272639 NumTriplet= 16596055687168

Currently allocated 145 MB for GPU buffers
In v_BaseLineSmooth: NumDataPoints=1048576, BoxCarLength=8192, NumPointsInChunk=32768
Restarted at 36.70 percent.
Used GPU device parameters are:
Number of compute units: 6
Single buffer allocation size: 64MB
max WG size: 256
period_iterations_num=20
Gaussian: peak=3.264746, mean=0.5102066, ChiSq=1.39704, time=12.58, d_freq=1418787335.69,
score=2.879473, null_hyp=2.392514, chirp=-68.975, fft_len=16k
Gaussian: peak=3.367662, mean=0.5108122, ChiSq=1.26971, time=14.26, d_freq=1418787219.97,
score=2.433709, null_hyp=2.298167, chirp=-68.975, fft_len=16k

Best spike: peak=24.84763, time=73.82, d_freq=1418789733.65, chirp=-0.83461, fft_len=128k
Best autocorr: peak=25083.42, time=33.56, delay=3.9507, d_freq=1418788818.71, chirp=-7.2656, fft_len=128k
Best gaussian: peak=3.264746, mean=0.5102066, ChiSq=1.39704, time=12.58, d_freq=1418787335.69,
score=2.879473, null_hyp=2.392514, chirp=-68.975, fft_len=16k
Best pulse: peak=0.5884005, time=31.27, period=0.07578, d_freq=1418788287.11, score=1.017, chirp=-10.159, fft_len=64
Best triplet: peak=0, time=-2.121e+011, period=0, d_freq=0, chirp=0, fft_len=0


Flopcounter: 21426653906490.609000

Spike count: 4
Autocorr count: 1
Pulse count: 1
Triplet count: 0
Gaussian count: 2
Wallclock time elapsed since last restart: 3647.3 seconds

My Cuda wingman:

Spike count: 4
Autocorr count: 0
Pulse count: 1
Triplet count: 0
Gaussian count: 2

Claggy

I did have an inconclusive with another Intel GPU running an older driver, looks like precision problem with the older 1.1 drivers:

http://setiweb.ssl.berkeley.edu/beta/workunit.php?wuid=5822771

Check out the differences in Red between my result and my intel GPU wingman (I started at the bottom and have only gone up as far as the Best spike)

Eric might want to restrict work to intel GPUs running OpenCL 1.2 drivers.
...
My intel GPU Wingman's result:

...
Best autocorr: peak=25083.42
...

Claggy

I don't think that can be classed as a precision problem. It must be a corrupted buffer or something similar, having shown only once. If that amount of autocorrelation actually were found in the data it would show up at many sequential chirp rates and cause a result_overflow. It takes something like a pure noiseless repeating pattern to get that much correlation.

My guess would be hardware operating at the verge of instability, though a vulnerabilty in Intel OpenCL 1.1 drivers is also possible.

                                                                  Joe

Another inconclusive, this time on my Nexus 7 between the v7.23 (armv6-vfp) Arm app and v7.03 (opencl_intel_gpu_sah) app on my wingman's GPU, the Autocorr counts being different, and the intel GPU is running OpenCL 1.1 drivers again:

http://setiweb.ssl.berkeley.edu/beta/workunit.php?wuid=5833620

The Second wingman also has an Intel GPU, hasn't completed the Wu yet, and is running OpenCL 1.2 drivers.

Claggy

The Second wingman also has an Intel GPU, hasn't completed the Wu yet, and is running OpenCL 1.2 drivers.

I take that Back, that host is no longer showing an Intel GPU, only a HD6900 running OpenCL 1.2 drivers, and hasn't completed any work since August,
any work is either getting cancelled because the other wingmen's Wu's are in, or gets cancelled at deadline, so we'll probably have to wait until 26 Dec before a replacement is sent.

Claggy

Few days ago my host completed its 200th valid Intel OpenCL MB7 result.
No invalids (besides "can't validate") and no computation errors (besides "cancelled by server").

So, I think we have stable enough and precise enough app for release.
Please confirm this conclusion from own experience.
I stop beta testing for Intel MB7 on own host switching completely to Intel OpenCL AP testing.

I use HD2500 under x64 Win7 with driver:

Driver version: 9.18.10.3257
Version: OpenCL 1.2

Host 7137742 has been running r2061 for MB (and r2058 for AP) - the low-CPU usage variants - since the last week of November, and is currently displaying 703 and 52 consecutive valid tasks, respectively. I'd agree that the app is fit for wider stock Beta testing.

Host 7137742 has been running r2061 for MB (and r2058 for AP) - the low-CPU usage variants - since the last week of November, and is currently displaying 703 and 52 consecutive valid tasks, respectively. I'd agree that the app is fit for wider stock Beta testing.

?
App already on beta testing, did you miss it? speak about release on main here.

Host 7137742 has been running r2061 for MB (and r2058 for AP) - the low-CPU usage variants - since the last week of November, and is currently displaying 703 and 52 consecutive valid tasks, respectively. I'd agree that the app is fit for wider stock Beta testing.

?
App already on beta testing, did you miss it? speak about release on main here.

Sorry, brain still asleep after the holidays. Just in case Eric is in the same state, might be wise to be explicit about which app version(s) you would like to see deployed, and where.

App passed beta, of course, on main.
There was only one MB7 Intel GPU app as I can recall, you just test it as anonymous platform perhaps.

What prevented deployment on this week outage ?

On my side, since I got many task errors on Jan, 27 of the same type as 15904216, I got no WU anymore for this App.

---

WU's on first