Radeon 8065S
19 records across 3 models (quants: MXFP4-Q5K mixed (4.29 bpw), Q5K-IQ3S mixed (3.84 bpw), UD-IQ4_XS).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Radeon 8065S | GLM-5.3-Flash llama-bench -fa 1 -b 2048 -ub 2048, 3 runs; ROCmFPX main (GLM-5.3 support), Mesa RADV | Q5K-IQ3S mixed (3.84 bpw) | llama.cpp (Vulkan) | 17.4 | – | – | – | – | 2026-10-02 | community | – | source |
| Radeon 8065S | GLM-5.3-Flash llama-bench -fa 1 -b 2048 -ub 2048, 3 runs; ROCmFPX main (GLM-5.3 support), Mesa RADV | Q5K-IQ3S mixed (3.84 bpw) | llama.cpp (Vulkan) | 15.0 | – | – | – | 32768 | 2026-10-02 | community | – | source |
| Radeon 8065S | GLM-5.3-Flash llama-bench -fa 1 -b 2048 -ub 2048, 3 runs; ROCmFPX main (GLM-5.3 support), Mesa RADV | Q5K-IQ3S mixed (3.84 bpw) | llama.cpp (Vulkan) | – | 256.4 | – | – | – | 2026-10-02 | community | – | source |
| Radeon 8065S | GLM-5.3-Flash llama-bench -fa 1 -b 2048 -ub 2048, 3 runs; ROCmFPX main (GLM-5.3 support), Mesa RADV | Q5K-IQ3S mixed (3.84 bpw) | llama.cpp (Vulkan) | – | 95.9 | – | – | 32768 | 2026-10-02 | community | – | source |
| Radeon 8065S | MiMo-V2.6-Flash-RL llama-bench -fa 1 -b 2048 -ub 2048, 3 runs; ROCmFPX main (fused gate/up expert tensors, from PR #33), Mesa RADV | MXFP4-Q5K mixed (4.29 bpw) | llama.cpp (Vulkan) | 23.6 | – | – | – | tg128 | 2026-10-02 | community | – | source |
| Radeon 8065S | MiMo-V2.6-Flash-RL llama-bench -fa 1 -b 2048 -ub 2048, 3 runs; ROCmFPX main (fused gate/up expert tensors, from PR #33), Mesa RADV | MXFP4-Q5K mixed (4.29 bpw) | llama.cpp (Vulkan) | 22.1 | – | – | – | 32768 | 2026-10-02 | community | – | source |
| Radeon 8065S | MiMo-V2.6-Flash-RL llama-bench -fa 1 -b 2048 -ub 2048, 3 runs; ROCmFPX main (fused gate/up expert tensors, from PR #33), Mesa RADV | MXFP4-Q5K mixed (4.29 bpw) | llama.cpp (Vulkan) | – | 484.9 | – | – | – | 2026-10-02 | community | – | source |
| Radeon 8065S | MiMo-V2.6-Flash-RL llama-bench -fa 1 -b 2048 -ub 2048, 3 runs; ROCmFPX main (fused gate/up expert tensors, from PR #33), Mesa RADV | MXFP4-Q5K mixed (4.29 bpw) | llama.cpp (Vulkan) | – | 258.7 | – | – | 32768 | 2026-10-02 | community | – | source |
| Radeon 8065S | MiniMax-M2.7 | UD-IQ4_XS | llama.cpp (Vulkan) | – | 54.6 | – | – | – | 2026-10 | community | – | source |
| Radeon 8065S | MiniMax-M2.7 | UD-IQ4_XS | llama.cpp (Vulkan) | – | 42.6 | – | – | 32768 | 2026-10 | community | – | source |
| Radeon 8065S | MiniMax-M2.7 | UD-IQ4_XS | llama.cpp (Vulkan) | – | 78.9 | – | – | – | 2026-10 | community | – | source |
| Radeon 8065S | MiniMax-M2.7 | UD-IQ4_XS | llama.cpp (Vulkan) | – | 55.7 | – | – | 32768 | 2026-10 | community | – | source |
| Radeon 8065S | MiniMax-M2.7 | UD-IQ4_XS | llama.cpp (Vulkan) | – | 124.0 | – | – | – | 2026-10 | community | – | source |
| Radeon 8065S | MiniMax-M2.7 | UD-IQ4_XS | llama.cpp (Vulkan) | – | 77.5 | – | – | 32768 | 2026-10 | community | – | source |
| Radeon 8065S | MiniMax-M2.7 | UD-IQ4_XS | llama.cpp (Vulkan) | – | 51.5 | – | – | 32768 | 2026-10 | community | – | source |
| Radeon 8065S | MiniMax-M2.7 | UD-IQ4_XS | llama.cpp (Vulkan) | – | 309.3 | – | – | – | 2026-10 | community | – | source |
| Radeon 8065S | MiniMax-M2.7 | UD-IQ4_XS | llama.cpp (Vulkan) | – | 13.3 | – | – | 32768 | 2026-10 | community | – | source |
| Radeon 8065S | MiniMax-M2.7 | UD-IQ4_XS | llama.cpp (ROCm) | – | 246.4 | – | – | – | 2026-10 | community | – | source |
| Radeon 8065S | MiniMax-M2.7 | UD-IQ4_XS | llama.cpp (ROCm) | – | 151.0 | – | – | 32768 | 2026-10 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085