WebGPU Micro-Benchmark Result
Per-operation GPU timings for the profiled device. Benchmark details available in the webgpu-bench open source project.
Device
Recorded 10/15/2026, 8:18:01 AM
WebGPU device
Vendorapple
Device—
Architecturemetal-3
Description—
WebGL device
VendorGoogle Inc. (Apple)
RendererANGLE (Apple, ANGLE Metal Renderer: Apple M5, Unspecified Version)
Browser
NameChrome
Version155.0.0.0
Operating system
NameMac OS
Version10.15.7
Benchmark Results
Throughput from this run, in bytes or operations per second. Higher is faster.
| Benchmark | Throughput |
|---|---|
atomic direct | 12.1 Gupdate/s |
atomic sharded | 10.4 Gupdate/s |
atomic workgroup | 12.7 Gupdate/s |
branch coherent | 2.45 TFLOP/s |
branch divergent | 1.55 TFLOP/s |
branch none | 3.61 TFLOP/s |
branch uniform | 2.49 TFLOP/s |
branch vec4 if | 492 Gchoice/s |
branch vec4 select | 465 Gchoice/s |
f32<->f16 convert | 3.05 TOP/s |
fp16 div | 1.53 TFLOP/s |
fp16 ln | 1.52 TFLOP/s |
fp16 mat4 FMA | 3.14 TFLOP/s |
fp16 matvec FMA | 3.52 TFLOP/s |
fp16 pow | 967 GFLOP/s |
fp16 rsqrt | 1.91 TFLOP/s |
fp16 scalar FMA | 4.17 TFLOP/s |
fp16 sin/cos | 296 GFLOP/s |
fp16 sqrt | 1.46 TFLOP/s |
fp16 vec4 FMA | 6.56 TFLOP/s |
fp32 clamp | 1.87 TOP/s |
fp32 div | 1.86 TFLOP/s |
fp32 fma() builtin | 3.66 TFLOP/s |
fp32 ln | 1.84 TFLOP/s |
fp32 mat4 FMA | 2.29 TFLOP/s |
fp32 matvec FMA | 2.28 TFLOP/s |
fp32 min/max | 3.71 TOP/s |
fp32 pow | 953 GFLOP/s |
fp32 rsqrt | 1.86 TFLOP/s |
fp32 scalar FMA | 3.68 TFLOP/s |
fp32 select | 1.43 TOP/s |
fp32 sin/cos | 247 GFLOP/s |
fp32 sqrt | 1.88 TFLOP/s |
fp32 vec4 FMA | 3.69 TFLOP/s |
i32 div | 247 GOP/s |
i32 mat4 multiply-add | 1.9 TOP/s |
i32 matvec multiply-add | 1.78 TOP/s |
i32 scalar multiply-add | 1.95 TOP/s |
i32 vec4 multiply-add | 1.93 TOP/s |
i32<->f32 convert | 934 GOP/s |
int8 dp4a | 1.26 TOP/s |
int8 dp4a matvec | 1.81 TOP/s |
layout aos | 3.58 Gparticle/s |
layout aosoa | 7.41 Gparticle/s |
layout soa | 7.76 Gparticle/s |
read dependent chain | 9.27 Mhop/s |
read gather 16kb | 1.42 TB/s |
read gather 4mb | 74.6 GB/s |
read gather 64mb | 23.4 GB/s |
read linear | 110 GB/s |
reduction serial | 5.15 Gelement/s |
reduction subgroup | 24.9 Gelement/s |
reduction workgroup | 9.67 Gelement/s |
texture interp (built-in) | 608 GB/s |
texture interp (manual) | 812 GB/s |
tile direct | 24.8 Goutput/s |
tile shared | 21.4 Goutput/s |
tile shared padded | 18.9 Goutput/s |
u32 byte pack/unpack | 689 GOP/s |
u32 countOneBits | 932 GOP/s |
u32 firstLeadingBit | 457 GOP/s |
u32 reverseBits | 931 GOP/s |
u32 variable shift | 907 GOP/s |
uint8 dp4a | 1.65 TOP/s |
workgroup 128 | 10.8 Gelement/s |
workgroup 256 | 10.8 Gelement/s |
workgroup 64 | 10.9 Gelement/s |
write linear | 119 GB/s |
write scatter 16kb | 16.2 GB/s |
write scatter 4mb | 47.4 GB/s |
write scatter 64mb | 17.5 GB/s |