perf: A100 baseline for the PixelHistogramCUDA probes
Build on RHEL8 / build (push) Successful in 4m8s
Build on RHEL9 / build (push) Successful in 4m28s
Run tests using data on local RHEL8 / build (push) Failing after 4m52s

Fill kernel-bound at 1000 bins on raw uint16 frames (23 us/frame vs
12 us H2D); H2D-bound at 100 bins.
This commit is contained in:
Ferjaoui committed 2026-10-29 16:47:32 +01:00
1 parent a3c2ac1572
commit bb9b04d7ad
2 files changed
+17

No files matched your search

@@ -0,0 +1,12 @@
{
"timestamp": "2026-10-29 14:57:21",
"host": "gpu105",
"git_rev": "6f67823",
"git_branch": "feature/cuda_pixelhistogram",
"git_dirty": true,
"aare_version": "2026.9.30",
"gpu": "NVIDIA A100-SXM4-80GB",
"driver": "610.57.04",
"nvcc": "Build cuda_12.9.r12.9/compiler.36037853_0",
"python": "3.14.10"
}
@@ -0,0 +1,5 @@
n_frames,window_ms,window_us_per_frame,kernel_n,kernel_sum_ms,kernel_busy_ms,kernel_overlap,kernel_duty_pct,kernel_us_per_frame,H2D_n,H2D_sum_ms,H2D_busy_ms,H2D_overlap,H2D_duty_pct,H2D_us_per_frame,bottleneck,roofline_us_per_frame,roofline_fps,readout_kernel_n,readout_kernel_ms,readout_D2H_ms,readout_D2H_MB,readout_D2H_GBps,label,dtype,n_bins,batch
20000,467.863379,23.39316895,313,466.57334,466.57334,1.0,99.72427014852984,23.328667000000003,312,240.406117,240.406117,1.0,51.38382865396268,12.02030585,kernel,23.328667000000003,42865.71538785306,1,5.820164,24.349518,640.0,26.28388783712269,u16_b1000_bs64,u16,1000,64
20000,483.297376,24.1648688,313,470.523191,470.523191,1.0,97.35686853801582,23.52615955,312,480.106159,480.106159,1.0,99.33969908415145,24.00530795,H2D,24.00530795,41657.45351331767,1,5.820296,24.348635,640.0,26.284841018808653,f32_b1000_bs64,f32,1000,64
20000,242.032464,12.1016232,313,87.348257,87.348257,1.0,36.089479715415365,4.36741285,312,240.548915,240.548915,1.0,99.38704545023349,12.02744575,H2D,12.02744575,83143.17277215738,1,0.43635,2.440375,64.0,26.225477641755877,u16_b100_bs64,u16,100,64
20000,458.435197,22.92175985,40,458.281981,458.281981,1.0,99.9665784824109,22.91409905,39,234.368762,234.368762,1.0,51.1236404913299,11.7184381,kernel,22.91409905,43641.253265857726,1,5.819049,24.363299,640.0,26.26902046393635,u16_b1000_bs512,u16,1000,512
1 n_frames window_ms window_us_per_frame kernel_n kernel_sum_ms kernel_busy_ms kernel_overlap kernel_duty_pct kernel_us_per_frame H2D_n H2D_sum_ms H2D_busy_ms H2D_overlap H2D_duty_pct H2D_us_per_frame bottleneck roofline_us_per_frame roofline_fps readout_kernel_n readout_kernel_ms readout_D2H_ms readout_D2H_MB readout_D2H_GBps label dtype n_bins batch
2 20000 467.863379 23.39316895 313 466.57334 466.57334 1.0 99.72427014852984 23.328667000000003 312 240.406117 240.406117 1.0 51.38382865396268 12.02030585 kernel 23.328667000000003 42865.71538785306 1 5.820164 24.349518 640.0 26.28388783712269 u16_b1000_bs64 u16 1000 64
3 20000 483.297376 24.1648688 313 470.523191 470.523191 1.0 97.35686853801582 23.52615955 312 480.106159 480.106159 1.0 99.33969908415145 24.00530795 H2D 24.00530795 41657.45351331767 1 5.820296 24.348635 640.0 26.284841018808653 f32_b1000_bs64 f32 1000 64
4 20000 242.032464 12.1016232 313 87.348257 87.348257 1.0 36.089479715415365 4.36741285 312 240.548915 240.548915 1.0 99.38704545023349 12.02744575 H2D 12.02744575 83143.17277215738 1 0.43635 2.440375 64.0 26.225477641755877 u16_b100_bs64 u16 100 64
5 20000 458.435197 22.92175985 40 458.281981 458.281981 1.0 99.9665784824109 22.91409905 39 234.368762 234.368762 1.0 51.1236404913299 11.7184381 kernel 22.91409905 43641.253265857726 1 5.819049 24.363299 640.0 26.26902046393635 u16_b1000_bs512 u16 1000 512