I see you picked up 2x gamma ray tasks for the ARC, will check back later to see how they ran.
I pauded the pending 1.70 Binary Radio Pulsar Search (Arecibo) (opencl-intel_gpu-newer) to run the 1.22 Gamma-ray pulsar binary search #1 on GPUs (FGRPopencl-intel_gpu) earlier
I see you picked up 2x gamma ray tasks for the ARC, will check back later to see how they ran.
I paused the pending 1.70 Binary Radio Pulsar Search (Arecibo) (opencl-intel_gpu-newer) to run the 1.22 Gamma-ray pulsar binary search #1 on GPUs (FGRPopencl-intel_gpu) earlier
OpenCL compiling FAILED! : -11 . Error message: 1:10:30: error: unknown type name 'double2'; did you mean 'double'? __kernel void test( __global double2 *vec) { ^~~~~~~ double 1:10:30: error: use of type 'double' requires cl_khr_fp64 extension to be enabled OpenCL device has no FP64 support</p>
<p>
I wonder if that last FP64 section (the final 10%) fell back to the CPU for processing and that's why it took so long. or it uses some other technique to kinda sorta do FP64 work really slowly using multiple clock cycles on the FP32 cores.
do you recall how long it took to get from 0% to 89%?
but this seems to confirm what some earlier reports said, no hardware FP64 support on these cards.
yeah that's not too bad for the FP32 part. not blazing fast, but decent for the power draw. that last 10% really kills it though for gamma ray.
it's unfortunate, because it would actually be pretty productive for gamma ray if it had any hw FP64 capability to get that last 10% done in a reasonable time.
and the credit reward for BRP4 is so low at just 62cr.
I see you picked up 2x gamma
)
I see you picked up 2x gamma ray tasks for the ARC, will check back later to see how they ran.
_________________________________________________________________________
Ian&Steve C. wrote: I see
)
I pauded the pending 1.70 Binary Radio Pulsar Search (Arecibo) (opencl-intel_gpu-newer) to run the 1.22 Gamma-ray pulsar binary search #1 on GPUs (FGRPopencl-intel_gpu) earlier
Ian&Steve C. wrote:I see
)
I paused the pending 1.70 Binary Radio Pulsar Search (Arecibo) (opencl-intel_gpu-newer) to run the 1.22 Gamma-ray pulsar binary search #1 on GPUs (FGRPopencl-intel_gpu) earlier
Edit:
Einstein@Home 1.22 Gamma-ray pulsar binary search #1 on GPUs (FGRPopencl-intel_gpu) LATeah3012L12221022_836.0_0_0.0_2286342_1 00:29:50 (00:29:41) 99,49 70.03 MB 89,998 00:10:43 13d,22:31:18 1C + 1INT Running 48,0 °C 0 A750
This one has been "stuck" at 89.998% for about 15mn.... but has completed
https://einsteinathome.org/task/1375878092
OpenCL compiling FAILED! : -11 . Error message: 1:10:30: error: unknown type name 'double2'; did you mean 'double'? __kernel void test( __global double2 *vec) { ^~~~~~~ double 1:10:30: error: use of type 'double' requires cl_khr_fp64 extension to be enabled OpenCL device has no FP64 support</p> <p>Well, it has been validated
)
Well, it has been validated without FP64.
https://einsteinathome.org/task/1375878092
https://einsteinathome.org/task/1375878090
Granted credit:3,465
TASK ID WORKUNIT
)
Application
I wonder if that last FP64
)
I wonder if that last FP64 section (the final 10%) fell back to the CPU for processing and that's why it took so long. or it uses some other technique to kinda sorta do FP64 work really slowly using multiple clock cycles on the FP32 cores.
do you recall how long it took to get from 0% to 89%?
but this seems to confirm what some earlier reports said, no hardware FP64 support on these cards.
_________________________________________________________________________
Ian&Steve C. wrote:do you
)
seems to be about 7mn (I just managed to capture it on a currently running one)
Einstein@Home 1.22 Gamma-ray pulsar binary search #1 on GPUs (FGRPopencl-intel_gpu) LATeah3012L12221022_820.0_0_0.0_1667151_2 00:07:05 (00:06:58) 98,34 133.97 MB 89,998 00:07:07 13d,22:19:33 1C + 1INT Running 53,2 °C 0 A750
edit:
28mn total
Einstein@Home 1.22 Gamma-ray pulsar binary search #1 on GPUs (FGRPopencl-intel_gpu) LATeah3012L12221022_820.0_0_0.0_1667151_2 00:28:20 (00:28:10) 99,42 100,000 - 13d,21:57:40 1C + 1INT Ready to report A750
Can you try running two tasks
)
Can you try running two tasks in parallel. Maybe it is scaling good.
yeah that's not too bad for
)
yeah that's not too bad for the FP32 part. not blazing fast, but decent for the power draw. that last 10% really kills it though for gamma ray.
it's unfortunate, because it would actually be pretty productive for gamma ray if it had any hw FP64 capability to get that last 10% done in a reasonable time.
and the credit reward for BRP4 is so low at just 62cr.
looking at about 100,000 ppd running BRP4
and about 165,000 ppd running FGRPB1G
_________________________________________________________________________
Landjunge wrote: Can you try
)
thats a good idea
_________________________________________________________________________