julian/amd-strix-halo-toolboxes
1
0
Fork 0
Code Issues Pull Requests 2 Actions Packages Projects Releases Wiki Activity

updated gpt-oss benchmakrs to test rocm7 performance patch

Browse Source
This commit is contained in:
Donato Capitella
2026-02-04 17:46:43 +00:00
parent 785f27b100
commit d97efb0cb9
9 changed files with 122 additions and 118 deletions
Download Patch File Download Diff File Expand all files Collapse all files
benchmark/results/gpt-oss-20b-mxfp4__rocm7-nightlies__fa1__longctx32768.log
+5 -5
View File
@@ -1,8 +1,8 @@
ggml_cuda_init: found 1 ROCm devices:
Device 0: Radeon 8060S Graphics, gfx1151 (0x1151), VMM: no, Wave Size: 32
| model | size | params | backend | ngl | n_ubatch | fa | mmap | test | t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | ---: | --------------: | -------------------: |
| gpt-oss 20B MXFP4 MoE | 11.27 GiB | 20.91 B | ROCm | 99 | 2048 | 1 | 0 | pp2048 @ d32768 | 374.10 ± 0.00 |
| gpt-oss 20B MXFP4 MoE | 11.27 GiB | 20.91 B | ROCm | 99 | 2048 | 1 | 0 | tg32 @ d32768 | 51.98 ± 0.00 |
| model | size | params | backend | ngl | n_ubatch | fa | test | t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | --------------: | -------------------: |
| gpt-oss 20B MXFP4 MoE | 11.27 GiB | 20.91 B | ROCm | 99 | 2048 | 1 | pp2048 @ d32768 | 557.68 ± 0.88 |
| gpt-oss 20B MXFP4 MoE | 11.27 GiB | 20.91 B | ROCm | 99 | 2048 | 1 | tg32 @ d32768 | 57.80 ± 0.01 |
build: 9c142e3a2 (7670)
build: e0c93af2a (7938)