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 |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -------: | -: | ---: | --------------: | -------------------: |
| qwen3moe 235B.A22B Q3_K - Medium |  96.99 GiB |   235.09 B | ROCm       |  99 |     2048 |  1 |    0 | pp2048 @ d32768 |         29.54 ± 0.00 |
/opt/llama.cpp/ggml/src/ggml-cuda/ggml-cuda.cu:96: ROCm error
:0:rocdevice.cpp            :3618: 63153010732 us:  Callback: Queue 0x7f3225300000 aborting with error : HSA_STATUS_ERROR_MEMORY_APERTURE_VIOLATION: The agent attempted to access memory beyond the largest legal address. code: 0x29
/usr/local/lib64/libggml-base.so.0(+0x35a5) [0x7f3234ef25a5]
/usr/local/lib64/libggml-base.so.0(ggml_print_backtrace+0x1eb) [0x7f3234ef296b]
/usr/local/lib64/libggml-base.so.0(ggml_abort+0x11f) [0x7f3234ef2aef]
/usr/local/lib64/libggml-hip.so.0(+0x2ca3912) [0x7f3237c52912]
/usr/local/lib64/libggml-hip.so.0(+0x2ca8cde) [0x7f3237c57cde]
/usr/local/lib64/libggml-base.so.0(ggml_backend_sched_synchronize+0x2e) [0x7f3234f09e5e]
/usr/local/lib64/libllama.so.0(_ZN13llama_context11synchronizeEv+0x10) [0x7f3238331630]
/usr/local/bin/llama-bench() [0x408ca6]
/lib64/libc.so.6(+0x35b5) [0x7f32348885b5]
/lib64/libc.so.6(__libc_start_main+0x88) [0x7f3234888668]
/usr/local/bin/llama-bench() [0x409cf5]
✖ ! [rocm-7alpha-rocwmma] Qwen3-235B-A22B-Instruct-2507-UD-Q3_K_XL-00001-of-00003__fa1 __longctx32768 failed (exit 0)