WebGPU Micro-Benchmark Result
Per-operation GPU timings for the profiled device. Benchmark details available in the webgpu-bench open source project.
Device
Recorded 11/6/2026, 3:41:31 PM
WebGPU device
Vendorapple
Device—
Architecture—
DescriptionMetal driver on macOS Version 26.6.2 (Build 25G83)
WebGL device
Vendor—
Renderer—
Browser
Namewebgpu-bench
Version0.10.0
Operating system
NameMac OS
Version25.6.0
Benchmark Results
Throughput from this run, in bytes or operations per second. Higher is faster.
| Benchmark | Throughput |
|---|---|
atomic direct | 7.65 Gupdate/s |
atomic sharded | 6.79 Gupdate/s |
atomic workgroup | 7.49 Gupdate/s |
branch coherent | 9.19 TFLOP/s |
branch divergent | 5.92 TFLOP/s |
branch none | 13.5 TFLOP/s |
branch uniform | 9.57 TFLOP/s |
branch vec4 if | 1.53 Tchoice/s |
branch vec4 select | 1.46 Tchoice/s |
f32<->f16 convert | 12.7 TOP/s |
fp16 div | 5.85 TFLOP/s |
fp16 ln | 5.85 TFLOP/s |
fp16 mat4 FMA | 12.1 TFLOP/s |
fp16 matvec FMA | 13.6 TFLOP/s |
fp16 pow | 3.69 TFLOP/s |
fp16 rsqrt | 7.3 TFLOP/s |
fp16 scalar FMA | 14.7 TFLOP/s |
fp16 sin/cos | 1.14 TFLOP/s |
fp16 sqrt | 5.59 TFLOP/s |
fp16 vec4 FMA | 25.5 TFLOP/s |
fp32 clamp | 6.84 TOP/s |
fp32 div | 7.1 TFLOP/s |
fp32 fma() builtin | 13.5 TFLOP/s |
fp32 ln | 7.02 TFLOP/s |
fp32 mat4 FMA | 8.17 TFLOP/s |
fp32 matvec FMA | 8.03 TFLOP/s |
fp32 min/max | 13.7 TOP/s |
fp32 pow | 3.66 TFLOP/s |
fp32 rsqrt | 7.03 TFLOP/s |
fp32 scalar FMA | 13.5 TFLOP/s |
fp32 select | 5.47 TOP/s |
fp32 sin/cos | 952 GFLOP/s |
fp32 sqrt | 7.19 TFLOP/s |
fp32 vec4 FMA | 13.6 TFLOP/s |
i32 div | 946 GOP/s |
i32 mat4 multiply-add | 6.98 TOP/s |
i32 matvec multiply-add | 6.63 TOP/s |
i32 scalar multiply-add | 7.37 TOP/s |
i32 vec4 multiply-add | 7.33 TOP/s |
i32<->f32 convert | 3.88 TOP/s |
int8 dp4a | 4.44 TOP/s |
int8 dp4a matvec | 6.68 TOP/s |
layout aos | 20.1 Gparticle/s |
layout aosoa | 29.1 Gparticle/s |
layout soa | 27 Gparticle/s |
read dependent chain | 6.85 Mhop/s |
read gather 16kb | 4.35 TB/s |
read gather 4mb | 642 GB/s |
read gather 64mb | 204 GB/s |
read linear | 671 GB/s |
reduction serial | 4.13 Gelement/s |
reduction subgroup | 32.3 Gelement/s |
reduction workgroup | 17.2 Gelement/s |
texture interp (built-in) | 2.08 TB/s |
texture interp (manual) | 3.15 TB/s |
tile direct | 34.9 Goutput/s |
tile shared | 31 Goutput/s |
tile shared padded | 27.7 Goutput/s |
u32 byte pack/unpack | 2.64 TOP/s |
u32 countOneBits | 3.56 TOP/s |
u32 firstLeadingBit | 1.67 TOP/s |
u32 reverseBits | 3.56 TOP/s |
u32 variable shift | 3.49 TOP/s |
uint8 dp4a | 6.31 TOP/s |
workgroup 128 | 22.1 Gelement/s |
workgroup 256 | 21.4 Gelement/s |
workgroup 64 | 21.5 Gelement/s |
write linear | 850 GB/s |
write scatter 16kb | 10.6 GB/s |
write scatter 4mb | 420 GB/s |
write scatter 64mb | 166 GB/s |