Task01 Максим Синицын HSE

[c0ldzy17]

Локальный вывод

Found 1 GPUs in 0.654228 sec (OpenCL: 0.066856 sec, Vulkan: 0.58732 sec)
Available devices:
  Device #0: API: OpenCL+Vulkan. GPU. Apple M3 Pro. Free memory: 27648/27648 Mb.
Using device #0: API: OpenCL+Vulkan. GPU. Apple M3 Pro. Free memory: 27648/27648 Mb.
Using OpenCL API...
matrixes size: 16384x8192 = 3 * 512 MB

Running BAD matrix kernel...
Kernels compilation done in 0.10887 seconds
BAD kernel times, s10 values (min=0.012569 10%=0.012577 median=0.012618 90%=0.164671 max=0.164671)
BAD kernel median bandwidth, gb/s118.878

Running GOOD matrix kernel...
Kernels compilation done in 0.003939 seconds
GOOD kernel times, s10 values (min=0.012919 10%=0.012939 median=0.013008 90%=0.046022 max=0.046022)
GOOD kernel median bandwidth, gb/s115.314

P.S. Как я понял, на MacBook M3 Pro разницы особо нет из-за умной работы с памятью уже на уровне чипа

Вывод Github CI

Logs for PR #1045 (2026-02-25T17:53:30.986255+00:00):

=== СТАТУС: Успешно выполнены программы: main_aplusb_matrix ===
=== main_aplusb_matrix stdout (exit code: -11 (segfault после выполнения)) ===
Found 1 GPUs in 8.58803 sec (CUDA: 0.118098 sec, OpenCL: 0.718217 sec, Vulkan: 7.75166 sec)
Available devices:
Device #0: API: CUDA+OpenCL+Vulkan. GPU. Tesla T4 (CUDA 12020). Free memory: 14822/14930 Mb.
Using device #0: API: CUDA+OpenCL+Vulkan. GPU. Tesla T4 (CUDA 12020). Free memory: 14822/14930 Mb.
Using OpenCL API...
matrixes size: 16384x8192 = 3 * 512 MB
Running BAD matrix kernel...
Kernels compilation done in 3.47536 seconds
BAD kernel times, s10 values (min=0.02029 10%=0.020293 median=0.020397 90%=3.496 max=3.496)
BAD kernel median bandwidth, gb/s73.5402
Running GOOD matrix kernel...
Kernels compilation done in 0.084948 seconds
GOOD kernel times, s10 values (min=0.006437 10%=0.006437 median=0.006439 90%=0.091471 max=0.091471)
GOOD kernel median bandwidth, gb/s232.955

[GPUcourseBOT]

✅ Результаты тестирования PR #1045

Логи тестирования (нажмите чтобы развернуть)

=== СТАТУС: Успешно выполнены программы: main_aplusb_matrix ===
=== main_aplusb_matrix stdout (exit code: -11 (segfault после выполнения)) ===
Found 1 GPUs in 8.58803 sec (CUDA: 0.118098 sec, OpenCL: 0.718217 sec, Vulkan: 7.75166 sec)
Available devices:
Device #0: API: CUDA+OpenCL+Vulkan. GPU. Tesla T4 (CUDA 12020). Free memory: 14822/14930 Mb.
Using device #0: API: CUDA+OpenCL+Vulkan. GPU. Tesla T4 (CUDA 12020). Free memory: 14822/14930 Mb.
Using OpenCL API...
matrixes size: 16384x8192 = 3 * 512 MB
Running BAD matrix kernel...
Kernels compilation done in 3.47536 seconds
BAD kernel times, s10 values (min=0.02029 10%=0.020293 median=0.020397 90%=3.496 max=3.496)
BAD kernel median bandwidth, gb/s73.5402
Running GOOD matrix kernel...
Kernels compilation done in 0.084948 seconds
GOOD kernel times, s10 values (min=0.006437 10%=0.006437 median=0.006439 90%=0.091471 max=0.091471)
GOOD kernel median bandwidth, gb/s232.955

Посмотреть полные логи