xhystos ~/llama.cpp/build/bin>./llama-bench --hf-repo unsloth/gemma-4-E4B-it-GGUF
ggml_cuda_init: found 1 ROCm devices (Total VRAM: 15609 MiB):
  Device 0: AMD Radeon 860M Graphics, gfx1150 (0x1150), VMM: no, Wave Size: 32, VRAM: 15609 MiB
| model                          |       size |     params | backend    | ngl |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | --------------: | -------------------: |
| gemma4 E4B Q4_K - Medium       |   4.62 GiB |     7.52 B | ROCm       |  99 |           pp512 |       291.48 ± 49.07 |
| gemma4 E4B Q4_K - Medium       |   4.62 GiB |     7.52 B | ROCm       |  99 |           tg128 |          6.65 ± 0.48 |

build: 82764d8f4 (8770)