WebGPU Micro-Benchmark Result
Per-operation GPU timings for the profiled device. Benchmark details available in the webgpu-bench open source project.
Device
Recorded 10/10/2026, 10:49:57 AM
WebGPU device
Vendormesa
Device—
Architecturesoftware
Descriptionllvmpipe: Mesa 26.2.2-1~bpo13+0~rpt1 (LLVM 19.1.7)
WebGL device
Vendor—
Renderer—
Browser
Namewebgpu-bench
Version0.10.0
Operating system
NameLinux
Version6.18.50+rpt-rpi-v8
Benchmark Results
Throughput from this run, in bytes or operations per second. Higher is faster.
| Benchmark | Throughput |
|---|---|
atomic direct | 66.7 Mupdate/s |
atomic sharded | 167 Mupdate/s |
atomic workgroup | 111 Mupdate/s |
branch coherent | 9.73 GFLOP/s |
branch divergent | 6.65 GFLOP/s |
branch none | 18.2 GFLOP/s |
branch uniform | 8.64 GFLOP/s |
branch vec4 if | 1.43 Gchoice/s |
branch vec4 select | 1.23 Gchoice/s |
f32<->f16 convert | 1.63 GOP/s |
fp16 div | 8.83 GFLOP/s |
fp16 ln | 800 MFLOP/s |
fp16 mat4 FMA | 11.3 GFLOP/s |
fp16 matvec FMA | 17.4 GFLOP/s |
fp16 pow | 376 MFLOP/s |
fp16 rsqrt | 4.53 GFLOP/s |
fp16 scalar FMA | 20.8 GFLOP/s |
fp16 sin/cos | 415 MFLOP/s |
fp16 sqrt | 7.83 GFLOP/s |
fp16 vec4 FMA | 29 GFLOP/s |
fp32 clamp | 9.37 GOP/s |
fp32 div | 8.33 GFLOP/s |
fp32 fma() builtin | 23.4 GFLOP/s |
fp32 ln | 2.46 GFLOP/s |
fp32 mat4 FMA | 9.44 GFLOP/s |
fp32 matvec FMA | 12.5 GFLOP/s |
fp32 min/max | 16.9 GOP/s |
fp32 pow | 1.14 GFLOP/s |
fp32 rsqrt | 4.37 GFLOP/s |
fp32 scalar FMA | 21.6 GFLOP/s |
fp32 select | 8.91 GOP/s |
fp32 sin/cos | 641 MFLOP/s |
fp32 sqrt | 7.07 GFLOP/s |
fp32 vec4 FMA | 29 GFLOP/s |
i32 div | 1.74 GOP/s |
i32 mat4 multiply-add | 10.3 GOP/s |
i32 matvec multiply-add | 19.9 GOP/s |
i32 scalar multiply-add | 17.1 GOP/s |
i32 vec4 multiply-add | 22.8 GOP/s |
i32<->f32 convert | 11.5 GOP/s |
int8 dp4a | 11.3 GOP/s |
int8 dp4a matvec | 14.7 GOP/s |
layout aos | 76 Mparticle/s |
layout aosoa | 81.6 Mparticle/s |
layout soa | 94.1 Mparticle/s |
read dependent chain | 26.9 Mhop/s |
read gather 16kb | 2.69 GB/s |
read gather 4mb | 703 MB/s |
read gather 64mb | 616 MB/s |
read linear | 923 MB/s |
reduction serial | 325 Melement/s |
reduction subgroup | 82.8 Melement/s |
reduction workgroup | 27 Melement/s |
texture interp (built-in) | 5.31 GB/s |
texture interp (manual) | 3.23 GB/s |
tile direct | 82.8 Moutput/s |
tile shared | 56.9 Moutput/s |
tile shared padded | 58.8 Moutput/s |
u32 byte pack/unpack | 20.1 GOP/s |
u32 countOneBits | 10.7 GOP/s |
u32 firstLeadingBit | 1.35 GOP/s |
u32 reverseBits | 12 GOP/s |
u32 variable shift | 7.61 GOP/s |
uint8 dp4a | 12.2 GOP/s |
workgroup 128 | 18.4 Melement/s |
workgroup 256 | 19.2 Melement/s |
workgroup 64 | 18.8 Melement/s |
write linear | 1.69 GB/s |
write scatter 16kb | 896 MB/s |
write scatter 4mb | 584 MB/s |
write scatter 64mb | 540 MB/s |