⚡ WebGPU Benchmark Results (7.55x speedup)
#110
by
SuperPauly - opened
Chrome browser
| Batch Size | WASM (int8) | WASM (fp16) | WASM (fp32) | WebGPU (fp32) |
| 1 | 1048.20 | 1011.80 | 943.10 | 157.90 |
| 2 | 2092.30 | 2026.00 | 1901.30 | 307.10 |
| 4 | 4162.40 | 4078.50 | 3803.20 | 523.00 |
| 8 | 8257.70 | 8173.60 | 7810.00 | 1067.70 |
| 16 | 16675.60 | 17104.80 | 16076.50 | 2475.10 |
| 32 | 34500.30 | 34857.10 | 32062.60 | 4618.20 |
- Model: Xenova/all-MiniLM-L6-v2
- Tests run: WASM (int8), WASM (fp16), WASM (fp32), WebGPU (fp32)
- Sequence length: 512
- Browser: Mozilla/5.0 (Linux; Android 10; K) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/144.0.0.0 Mobile Safari/537.36
- GPU: vendor=Mali G-715, architecture=undefined, device=Pixel 8 Pro, description=undefined
Microsoft Edge Browser
| Batch Size | WASM (int8) | WebGPU (fp32) |
| 1 | 1046.30 | 183.60 |
| 2 | 2051.40 | 297.40 |
| 4 | 4063.40 | 541.60 |
| 8 | 8180.20 | 1065.90 |
| 16 | 16595.00 | 2227.80 |
| 32 | 33304.00 | 4882.30 |
- Model: Xenova/all-MiniLM-L6-v2
- Tests run: WASM (int8), WebGPU (fp32)
- Sequence length: 512
- Browser: Mozilla/5.0 (Linux; Android 10; K) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/146.0.0.0 Mobile Safari/537.36 EdgA/146.0.0.0
- GPU: vendor=Mali G-715, architecture=undefined, device=Pixel 8 Pro, description=undefined