| Model |
Size |
Params |
Image |
Backend |
pp512 (t/s) |
tg128 (t/s) |
| tinyllama-1.1b-q4 | 636.18 MiB | 1.10 B | llama-hipblaslt-0 | ROCm | 5820.68 | 212.35 |
| tinyllama-1.1b-q4 | 636.18 MiB | 1.10 B | llama-vulkan | Vulkan | 5642.90 | 258.95 |
| tinyllama-1.1b-q4 | 636.18 MiB | 1.10 B | llama-vulkan-radv-fa | Vulkan | 5602.10 | 257.81 |
| tinyllama-1.1b-q4 | 636.18 MiB | 1.10 B | llama-vulkan-radv | Vulkan | 5597.76 | 257.86 |
| tinyllama-1.1b-q4 | 636.18 MiB | 1.10 B | llama-vulkan-radv-fa-finegrain | Vulkan | 5590.72 | 258.72 |
| tinyllama-1.1b-q4 | 636.18 MiB | 1.10 B | llama-vulkan-radv-ub1024 | Vulkan | 5583.54 | 259.68 |
| tinyllama-1.1b-q4 | 636.18 MiB | 1.10 B | llama-moe | Vulkan | 5546.77 | 257.35 |
| tinyllama-1.1b-q4 | 636.18 MiB | 1.10 B | llama-hipblaslt-1 | ROCm | 4913.05 | 212.77 |
| tinyllama-1.1b-q4 | 636.18 MiB | 1.10 B | llama-hip-rocm72-vram | ROCm | 4867.30 | 209.73 |
| tinyllama-1.1b-q4 | 636.18 MiB | 1.10 B | llama-hip-rocm72-gfx1151 | ROCm | 4765.45 | 184.93 |
| tinyllama-1.1b-q4 | 636.18 MiB | 1.10 B | llama-hip-rocm72-nofa-gfx1151 | ROCm | 4757.30 | 185.08 |
| tinyllama-1.1b-q4 | 636.18 MiB | 1.10 B | llama-hip-rocm72-fa0-gfx1151 | ROCm | 4731.79 | 184.94 |
| tinyllama-1.1b-q4 | 636.18 MiB | 1.10 B | llama-hip-rocm72-mmap-gfx1151 | ROCm | 4726.40 | 184.59 |
| tinyllama-1.1b-q4 | 636.18 MiB | 1.10 B | llama-hip-rocm72-sdma1-gfx1151 | ROCm | 4720.79 | 184.34 |
| tinyllama-1.1b-q4 | 636.18 MiB | 1.10 B | llama-hip-rocwmma-rocm72-gfx1151 | ROCm | 4720.17 | 184.57 |
| LFM2-2.6B | 1.45 GiB | 2.57 B | llama-hipblaslt-0 | ROCm | 3428.04 | 106.14 |
| LFM2-2.6B | 1.45 GiB | 2.57 B | llama-hip-rocm72-vram | ROCm | 3237.39 | 106.43 |
| LFM2-2.6B | 1.45 GiB | 2.57 B | llama-hip-rocm72-gfx1151 | ROCm | 3228.27 | 101.55 |
| LFM2-2.6B | 1.45 GiB | 2.57 B | llama-hipblaslt-1 | ROCm | 3225.97 | 105.95 |
| LFM2-2.6B | 1.45 GiB | 2.57 B | llama-hip-rocm72-nofa-gfx1151 | ROCm | 3213.26 | 101.13 |
| LFM2-2.6B | 1.45 GiB | 2.57 B | llama-hip-rocm72-fa0-gfx1151 | ROCm | 3203.39 | 101.25 |
| LFM2-2.6B | 1.45 GiB | 2.57 B | llama-hip-rocm72-mmap-gfx1151 | ROCm | 3199.57 | 100.67 |
| LFM2-2.6B | 1.45 GiB | 2.57 B | llama-hip-rocwmma-rocm72-gfx1151 | ROCm | 3193.19 | 100.64 |
| LFM2-2.6B | 1.45 GiB | 2.57 B | llama-hip-rocm72-sdma1-gfx1151 | ROCm | 3175.27 | 100.57 |
| LFM2-2.6B | 1.45 GiB | 2.57 B | llama-vulkan-radv-ub1024 | Vulkan | 2620.22 | 118.38 |
| LFM2-2.6B | 1.45 GiB | 2.57 B | llama-vulkan | Vulkan | 2609.16 | 118.47 |
| LFM2-2.6B | 1.45 GiB | 2.57 B | llama-vulkan-radv-fa | Vulkan | 2606.02 | 118.72 |
| LFM2-2.6B | 1.45 GiB | 2.57 B | llama-vulkan-radv | Vulkan | 2593.01 | 118.54 |
| LFM2-2.6B | 1.45 GiB | 2.57 B | llama-vulkan-radv-fa-finegrain | Vulkan | 2580.14 | 118.16 |
| LFM2-2.6B | 1.45 GiB | 2.57 B | llama-moe | Vulkan | 2554.20 | 118.59 |
| llama3.2-3b-q8 | 3.18 GiB | 3.21 B | llama-vulkan-radv-fa | Vulkan | 2141.66 | 60.28 |
| llama3.2-3b-q8 | 3.18 GiB | 3.21 B | llama-vulkan-radv-ub1024 | Vulkan | 2125.40 | 59.39 |
| llama3.2-3b-q8 | 3.18 GiB | 3.21 B | llama-vulkan | Vulkan | 2114.53 | 60.27 |
| llama3.2-3b-q8 | 3.18 GiB | 3.21 B | llama-moe | Vulkan | 2112.83 | 60.19 |
| llama3.2-3b-q8 | 3.18 GiB | 3.21 B | llama-vulkan-radv | Vulkan | 2107.62 | 59.60 |
| llama3.2-3b-q8 | 3.18 GiB | 3.21 B | llama-vulkan-radv-fa-finegrain | Vulkan | 2081.04 | 60.18 |
| llama3.2-3b-q8 | 3.18 GiB | 3.21 B | llama-hipblaslt-0 | ROCm | 1057.93 | 54.18 |
| llama3.2-3b-q8 | 3.18 GiB | 3.21 B | llama-hip-rocm72-vram | ROCm | 1008.53 | 54.29 |
| llama3.2-3b-q8 | 3.18 GiB | 3.21 B | llama-hip-rocm72-nofa-gfx1151 | ROCm | 1003.06 | 52.39 |
| llama3.2-3b-q8 | 3.18 GiB | 3.21 B | llama-hip-rocm72-sdma1-gfx1151 | ROCm | 1002.85 | 52.32 |
| llama3.2-3b-q8 | 3.18 GiB | 3.21 B | llama-hip-rocm72-fa0-gfx1151 | ROCm | 1002.56 | 52.29 |
| llama3.2-3b-q8 | 3.18 GiB | 3.21 B | llama-hipblaslt-1 | ROCm | 998.56 | 54.53 |
| llama3.2-3b-q8 | 3.18 GiB | 3.21 B | llama-hip-rocm72-mmap-gfx1151 | ROCm | 996.14 | 52.38 |
| llama3.2-3b-q8 | 3.18 GiB | 3.21 B | llama-hip-rocm72-gfx1151 | ROCm | 994.97 | 52.56 |
| llama3.2-3b-q8 | 3.18 GiB | 3.21 B | llama-hip-rocwmma-rocm72-gfx1151 | ROCm | 992.68 | 52.25 |
| qwen2.5-coder-7b-q6 | 5.82 GiB | 7.62 B | llama-hipblaslt-0 | ROCm | 1421.37 | 34.55 |
| qwen2.5-coder-7b-q6 | 5.82 GiB | 7.62 B | llama-hip-rocm72-vram | ROCm | 1280.00 | 34.75 |
| qwen2.5-coder-7b-q6 | 5.82 GiB | 7.62 B | llama-hip-rocwmma-rocm72-gfx1151 | ROCm | 1272.16 | 33.74 |
| qwen2.5-coder-7b-q6 | 5.82 GiB | 7.62 B | llama-hip-rocm72-fa0-gfx1151 | ROCm | 1263.39 | 33.68 |
| qwen2.5-coder-7b-q6 | 5.82 GiB | 7.62 B | llama-hipblaslt-1 | ROCm | 1256.41 | 34.72 |
| qwen2.5-coder-7b-q6 | 5.82 GiB | 7.62 B | llama-hip-rocm72-sdma1-gfx1151 | ROCm | 1252.19 | 33.49 |
| qwen2.5-coder-7b-q6 | 5.82 GiB | 7.62 B | llama-hip-rocm72-mmap-gfx1151 | ROCm | 1250.34 | 33.57 |
| qwen2.5-coder-7b-q6 | 5.82 GiB | 7.62 B | llama-hip-rocm72-nofa-gfx1151 | ROCm | 1245.68 | 33.80 |
| qwen2.5-coder-7b-q6 | 5.82 GiB | 7.62 B | llama-hip-rocm72-gfx1151 | ROCm | 1245.38 | 33.79 |
| qwen2.5-coder-7b-q6 | 5.82 GiB | 7.62 B | llama-vulkan-radv-ub1024 | Vulkan | 1001.49 | 35.99 |
| qwen2.5-coder-7b-q6 | 5.82 GiB | 7.62 B | llama-vulkan-radv-fa-finegrain | Vulkan | 997.92 | 36.35 |
| qwen2.5-coder-7b-q6 | 5.82 GiB | 7.62 B | llama-vulkan-radv-fa | Vulkan | 996.57 | 36.38 |
| qwen2.5-coder-7b-q6 | 5.82 GiB | 7.62 B | llama-moe | Vulkan | 983.99 | 36.55 |
| qwen2.5-coder-7b-q6 | 5.82 GiB | 7.62 B | llama-vulkan | Vulkan | 969.68 | 36.26 |
| qwen2.5-coder-7b-q6 | 5.82 GiB | 7.62 B | llama-vulkan-radv | Vulkan | 925.70 | 36.40 |
| Qwen2.5-14B-Instruct | 8.37 GiB | 14.77 B | llama-hipblaslt-0 | ROCm | 675.83 | 22.42 |
| Qwen2.5-14B-Instruct | 8.37 GiB | 14.77 B | llama-hipblaslt-1 | ROCm | 615.91 | 22.37 |
| Qwen2.5-14B-Instruct | 8.37 GiB | 14.77 B | llama-hip-rocm72-sdma1-gfx1151 | ROCm | 614.06 | 21.71 |
| Qwen2.5-14B-Instruct | 8.37 GiB | 14.77 B | llama-hip-rocm72-vram | ROCm | 610.54 | 22.35 |
| Qwen2.5-14B-Instruct | 8.37 GiB | 14.77 B | llama-hip-rocm72-gfx1151 | ROCm | 609.17 | 21.86 |
| Qwen2.5-14B-Instruct | 8.37 GiB | 14.77 B | llama-hip-rocm72-fa0-gfx1151 | ROCm | 608.11 | 21.80 |
| Qwen2.5-14B-Instruct | 8.37 GiB | 14.77 B | llama-hip-rocm72-nofa-gfx1151 | ROCm | 607.53 | 21.83 |
| Qwen2.5-14B-Instruct | 8.37 GiB | 14.77 B | llama-hip-rocm72-mmap-gfx1151 | ROCm | 607.40 | 21.76 |
| Qwen2.5-14B-Instruct | 8.37 GiB | 14.77 B | llama-hip-rocwmma-rocm72-gfx1151 | ROCm | 603.82 | 21.78 |
| Qwen2.5-14B-Instruct | 8.37 GiB | 14.77 B | llama-moe | Vulkan | 543.50 | 23.99 |
| Qwen2.5-14B-Instruct | 8.37 GiB | 14.77 B | llama-vulkan-radv-ub1024 | Vulkan | 535.23 | 24.27 |
| Qwen2.5-14B-Instruct | 8.37 GiB | 14.77 B | llama-vulkan-radv-fa-finegrain | Vulkan | 533.73 | 24.35 |
| Qwen2.5-14B-Instruct | 8.37 GiB | 14.77 B | llama-vulkan-radv | Vulkan | 532.82 | 24.07 |
| Qwen2.5-14B-Instruct | 8.37 GiB | 14.77 B | llama-vulkan-radv-fa | Vulkan | 527.74 | 24.25 |
| Qwen2.5-14B-Instruct | 8.37 GiB | 14.77 B | llama-vulkan | Vulkan | 523.25 | 23.95 |
| openai_gpt-oss-20b (MoE) | 10.86 GiB | 20.91 B | llama-vulkan-radv-ub1024 | Vulkan | 1168.20 | 81.53 |
| openai_gpt-oss-20b (MoE) | 10.86 GiB | 20.91 B | llama-vulkan-radv-fa | Vulkan | 1139.21 | 81.16 |
| openai_gpt-oss-20b (MoE) | 10.86 GiB | 20.91 B | llama-moe | Vulkan | 1126.14 | 81.33 |
| openai_gpt-oss-20b (MoE) | 10.86 GiB | 20.91 B | llama-vulkan-radv | Vulkan | 1105.93 | 81.06 |
| openai_gpt-oss-20b (MoE) | 10.86 GiB | 20.91 B | llama-vulkan-radv-fa-finegrain | Vulkan | 1058.48 | 80.79 |
| openai_gpt-oss-20b (MoE) | 10.86 GiB | 20.91 B | llama-vulkan | Vulkan | 1015.54 | 79.96 |
| openai_gpt-oss-20b (MoE) | 10.86 GiB | 20.91 B | llama-hipblaslt-0 | ROCm | 537.28 | 73.41 |
| openai_gpt-oss-20b (MoE) | 10.86 GiB | 20.91 B | llama-hip-rocm72-vram | ROCm | 521.57 | 73.33 |
| openai_gpt-oss-20b (MoE) | 10.86 GiB | 20.91 B | llama-hipblaslt-1 | ROCm | 520.70 | 73.45 |
| openai_gpt-oss-20b (MoE) | 10.86 GiB | 20.91 B | llama-hip-rocm72-nofa-gfx1151 | ROCm | 520.35 | 69.42 |
| openai_gpt-oss-20b (MoE) | 10.86 GiB | 20.91 B | llama-hip-rocm72-gfx1151 | ROCm | 519.64 | 69.57 |
| openai_gpt-oss-20b (MoE) | 10.86 GiB | 20.91 B | llama-hip-rocm72-sdma1-gfx1151 | ROCm | 518.01 | 69.11 |
| openai_gpt-oss-20b (MoE) | 10.86 GiB | 20.91 B | llama-hip-rocm72-mmap-gfx1151 | ROCm | 515.21 | 69.42 |
| openai_gpt-oss-20b (MoE) | 10.86 GiB | 20.91 B | llama-hip-rocm72-fa0-gfx1151 | ROCm | 509.81 | 68.14 |
| openai_gpt-oss-20b (MoE) | 10.86 GiB | 20.91 B | llama-hip-rocwmma-rocm72-gfx1151 | ROCm | 505.59 | 69.11 |
| GLM-4.7-Flash (MoE) | 16.88 GiB | 29.94 B | llama-vulkan-radv-fa-finegrain | Vulkan | 863.28 | 70.65 |
| GLM-4.7-Flash (MoE) | 16.88 GiB | 29.94 B | llama-vulkan-radv-ub1024 | Vulkan | 861.42 | 70.64 |
| GLM-4.7-Flash (MoE) | 16.88 GiB | 29.94 B | llama-vulkan-radv-fa | Vulkan | 860.89 | 70.71 |
| GLM-4.7-Flash (MoE) | 16.88 GiB | 29.94 B | llama-vulkan-radv | Vulkan | 859.53 | 70.50 |
| GLM-4.7-Flash (MoE) | 16.88 GiB | 29.94 B | llama-moe | Vulkan | 858.79 | 70.64 |
| GLM-4.7-Flash (MoE) | 16.88 GiB | 29.94 B | llama-vulkan | Vulkan | 855.10 | 70.57 |
| GLM-4.7-Flash (MoE) | 16.88 GiB | 29.94 B | llama-hip-rocm72-mmap-gfx1151 | ROCm | 807.11 | 47.73 |
| GLM-4.7-Flash (MoE) | 16.88 GiB | 29.94 B | llama-hip-rocwmma-rocm72-gfx1151 | ROCm | 807.01 | 47.69 |
| GLM-4.7-Flash (MoE) | 16.88 GiB | 29.94 B | llama-hip-rocm72-gfx1151 | ROCm | 804.78 | 47.69 |
| GLM-4.7-Flash (MoE) | 16.88 GiB | 29.94 B | llama-hip-rocm72-nofa-gfx1151 | ROCm | 804.73 | 47.68 |
| GLM-4.7-Flash (MoE) | 16.88 GiB | 29.94 B | llama-hip-rocm72-fa0-gfx1151 | ROCm | 804.58 | 47.78 |
| GLM-4.7-Flash (MoE) | 16.88 GiB | 29.94 B | llama-hip-rocm72-sdma1-gfx1151 | ROCm | 801.99 | 47.76 |
| GLM-4.7-Flash (MoE) | 16.88 GiB | 29.94 B | llama-hip-rocm72-vram | ROCm | 799.82 | 50.43 |
| GLM-4.7-Flash (MoE) | 16.88 GiB | 29.94 B | llama-hipblaslt-1 | ROCm | 796.97 | 50.16 |
| GLM-4.7-Flash (MoE) | 16.88 GiB | 29.94 B | llama-hipblaslt-0 | ROCm | 774.87 | 50.08 |
| Qwen3-30B-A3B (MoE) | 17.28 GiB | 30.53 B | llama-hipblaslt-0 | ROCm | 1096.33 | 68.32 |
| Qwen3-30B-A3B (MoE) | 17.28 GiB | 30.53 B | llama-vulkan-radv-ub1024 | Vulkan | 982.64 | 88.17 |
| Qwen3-30B-A3B (MoE) | 17.28 GiB | 30.53 B | llama-hip-rocm72-gfx1151 | ROCm | 969.18 | 62.29 |
| Qwen3-30B-A3B (MoE) | 17.28 GiB | 30.53 B | llama-hipblaslt-1 | ROCm | 964.86 | 68.38 |
| Qwen3-30B-A3B (MoE) | 17.28 GiB | 30.53 B | llama-hip-rocm72-vram | ROCm | 959.97 | 68.30 |
| Qwen3-30B-A3B (MoE) | 17.28 GiB | 30.53 B | llama-hip-rocm72-nofa-gfx1151 | ROCm | 933.24 | 62.10 |
| Qwen3-30B-A3B (MoE) | 17.28 GiB | 30.53 B | llama-hip-rocm72-fa0-gfx1151 | ROCm | 929.99 | 62.12 |
| Qwen3-30B-A3B (MoE) | 17.28 GiB | 30.53 B | llama-hip-rocm72-sdma1-gfx1151 | ROCm | 927.09 | 62.20 |
| Qwen3-30B-A3B (MoE) | 17.28 GiB | 30.53 B | llama-hip-rocm72-mmap-gfx1151 | ROCm | 926.34 | 62.27 |
| Qwen3-30B-A3B (MoE) | 17.28 GiB | 30.53 B | llama-hip-rocwmma-rocm72-gfx1151 | ROCm | 923.76 | 62.28 |
| Qwen3-30B-A3B (MoE) | 17.28 GiB | 30.53 B | llama-vulkan-radv-fa-finegrain | Vulkan | 895.96 | 84.46 |
| Qwen3-30B-A3B (MoE) | 17.28 GiB | 30.53 B | llama-vulkan-radv | Vulkan | 893.84 | 85.98 |
| Qwen3-30B-A3B (MoE) | 17.28 GiB | 30.53 B | llama-moe | Vulkan | 887.85 | 82.55 |
| Qwen3-30B-A3B (MoE) | 17.28 GiB | 30.53 B | llama-vulkan | Vulkan | 885.82 | 86.27 |
| Qwen3-30B-A3B (MoE) | 17.28 GiB | 30.53 B | llama-vulkan-radv-fa | Vulkan | 882.20 | 84.13 |
| Qwen2.5-32B-Instruct | 18.48 GiB | 32.76 B | llama-hipblaslt-0 | ROCm | 307.07 | 10.47 |
| Qwen2.5-32B-Instruct | 18.48 GiB | 32.76 B | llama-hip-rocm72-vram | ROCm | 289.57 | 10.47 |
| Qwen2.5-32B-Instruct | 18.48 GiB | 32.76 B | llama-hipblaslt-1 | ROCm | 289.41 | 10.47 |
| Qwen2.5-32B-Instruct | 18.48 GiB | 32.76 B | llama-hip-rocm72-sdma1-gfx1151 | ROCm | 288.31 | 10.34 |
| Qwen2.5-32B-Instruct | 18.48 GiB | 32.76 B | llama-hip-rocm72-fa0-gfx1151 | ROCm | 287.94 | 10.29 |
| Qwen2.5-32B-Instruct | 18.48 GiB | 32.76 B | llama-hip-rocm72-nofa-gfx1151 | ROCm | 287.11 | 10.35 |
| Qwen2.5-32B-Instruct | 18.48 GiB | 32.76 B | llama-hip-rocwmma-rocm72-gfx1151 | ROCm | 286.47 | 10.35 |
| Qwen2.5-32B-Instruct | 18.48 GiB | 32.76 B | llama-hip-rocm72-mmap-gfx1151 | ROCm | 286.34 | 10.31 |
| Qwen2.5-32B-Instruct | 18.48 GiB | 32.76 B | llama-hip-rocm72-gfx1151 | ROCm | 285.43 | 10.31 |
| Qwen2.5-32B-Instruct | 18.48 GiB | 32.76 B | llama-vulkan-radv-ub1024 | Vulkan | 233.66 | 11.18 |
| Qwen2.5-32B-Instruct | 18.48 GiB | 32.76 B | llama-vulkan-radv | Vulkan | 230.12 | 11.11 |
| Qwen2.5-32B-Instruct | 18.48 GiB | 32.76 B | llama-moe | Vulkan | 230.07 | 11.10 |
| Qwen2.5-32B-Instruct | 18.48 GiB | 32.76 B | llama-vulkan | Vulkan | 229.63 | 10.98 |
| Qwen2.5-32B-Instruct | 18.48 GiB | 32.76 B | llama-vulkan-radv-fa-finegrain | Vulkan | 228.20 | 11.00 |
| Qwen2.5-32B-Instruct | 18.48 GiB | 32.76 B | llama-vulkan-radv-fa | Vulkan | 228.15 | 11.11 |
| Meta-Llama-3.1-70B | 39.59 GiB | 70.55 B | llama-hipblaslt-0 | ROCm | 127.94 | 4.85 |
| Meta-Llama-3.1-70B | 39.59 GiB | 70.55 B | llama-hip-rocm72-vram | ROCm | 127.53 | 4.86 |
| Meta-Llama-3.1-70B | 39.59 GiB | 70.55 B | llama-hipblaslt-1 | ROCm | 127.28 | 4.86 |
| Meta-Llama-3.1-70B | 39.59 GiB | 70.55 B | llama-hip-rocm72-gfx1151 | ROCm | 125.99 | 4.75 |
| Meta-Llama-3.1-70B | 39.59 GiB | 70.55 B | llama-hip-rocm72-fa0-gfx1151 | ROCm | 123.41 | 4.76 |
| Meta-Llama-3.1-70B | 39.59 GiB | 70.55 B | llama-hip-rocwmma-rocm72-gfx1151 | ROCm | 122.93 | 4.71 |
| Meta-Llama-3.1-70B | 39.59 GiB | 70.55 B | llama-hip-rocm72-sdma1-gfx1151 | ROCm | 122.86 | 4.72 |
| Meta-Llama-3.1-70B | 39.59 GiB | 70.55 B | llama-hip-rocm72-mmap-gfx1151 | ROCm | 122.24 | 4.74 |
| Meta-Llama-3.1-70B | 39.59 GiB | 70.55 B | llama-hip-rocm72-nofa-gfx1151 | ROCm | 121.30 | 4.75 |
| Meta-Llama-3.1-70B | 39.59 GiB | 70.55 B | llama-vulkan-radv-ub1024 | Vulkan | 107.69 | 5.07 |
| Meta-Llama-3.1-70B | 39.59 GiB | 70.55 B | llama-moe | Vulkan | 101.16 | 5.04 |
| Meta-Llama-3.1-70B | 39.59 GiB | 70.55 B | llama-vulkan | Vulkan | 99.52 | 5.03 |
| Meta-Llama-3.1-70B | 39.59 GiB | 70.55 B | llama-vulkan-radv-fa-finegrain | Vulkan | 99.48 | 5.04 |
| Meta-Llama-3.1-70B | 39.59 GiB | 70.55 B | llama-vulkan-radv | Vulkan | 99.33 | 5.04 |
| Meta-Llama-3.1-70B | 39.59 GiB | 70.55 B | llama-vulkan-radv-fa | Vulkan | 98.83 | 5.01 |
| Qwen3-Next-80B-A3B (MoE) | 45.08 GiB | 79.67 B | llama-hipblaslt-0 | ROCm | 615.37 | 39.49 |
| Qwen3-Next-80B-A3B (MoE) | 45.08 GiB | 79.67 B | llama-vulkan | Vulkan | 579.26 | 43.98 |
| Qwen3-Next-80B-A3B (MoE) | 45.08 GiB | 79.67 B | llama-vulkan-radv-fa-finegrain | Vulkan | 577.79 | 43.75 |
| Qwen3-Next-80B-A3B (MoE) | 45.08 GiB | 79.67 B | llama-vulkan-radv-ub1024 | Vulkan | 532.50 | 39.59 |
| Qwen3-Next-80B-A3B (MoE) | 45.08 GiB | 79.67 B | llama-vulkan-radv | Vulkan | 528.70 | 39.66 |
| Qwen3-Next-80B-A3B (MoE) | 45.08 GiB | 79.67 B | llama-moe | Vulkan | 525.85 | 39.50 |
| Qwen3-Next-80B-A3B (MoE) | 45.08 GiB | 79.67 B | llama-vulkan-radv-fa | Vulkan | 524.64 | 43.75 |
| Qwen3-Next-80B-A3B (MoE) | 45.08 GiB | 79.67 B | llama-hip-rocwmma-rocm72-gfx1151 | ROCm | 347.89 | 30.63 |
| Qwen3-Next-80B-A3B (MoE) | 45.08 GiB | 79.67 B | llama-hip-rocm72-sdma1-gfx1151 | ROCm | 347.55 | 30.87 |
| Qwen3-Next-80B-A3B (MoE) | 45.08 GiB | 79.67 B | llama-hip-rocm72-mmap-gfx1151 | ROCm | 346.74 | 30.82 |
| Qwen3-Next-80B-A3B (MoE) | 45.08 GiB | 79.67 B | llama-hip-rocm72-nofa-gfx1151 | ROCm | 344.01 | 30.83 |
| Qwen3-Next-80B-A3B (MoE) | 45.08 GiB | 79.67 B | llama-hip-rocm72-fa0-gfx1151 | ROCm | 343.63 | 30.81 |
| Qwen3-Next-80B-A3B (MoE) | 45.08 GiB | 79.67 B | llama-hip-rocm72-vram | ROCm | 342.22 | 39.81 |
| Qwen3-Next-80B-A3B (MoE) | 45.08 GiB | 79.67 B | llama-hip-rocm72-gfx1151 | ROCm | 340.45 | 30.85 |
| Qwen3-Next-80B-A3B (MoE) | 45.08 GiB | 79.67 B | llama-hipblaslt-1 | ROCm | 339.38 | 39.74 |
| openai_gpt-oss-120b (MoE) | 58.51 GiB | 116.83 B | llama-vulkan-radv-ub1024 | Vulkan | 488.47 | 57.87 |
| openai_gpt-oss-120b (MoE) | 58.51 GiB | 116.83 B | llama-vulkan-radv-fa | Vulkan | 488.01 | 57.84 |
| openai_gpt-oss-120b (MoE) | 58.51 GiB | 116.83 B | llama-vulkan-radv | Vulkan | 487.30 | 57.60 |
| openai_gpt-oss-120b (MoE) | 58.51 GiB | 116.83 B | llama-moe | Vulkan | 479.71 | 57.61 |
| openai_gpt-oss-120b (MoE) | 58.51 GiB | 116.83 B | llama-vulkan-radv-fa-finegrain | Vulkan | 446.55 | 54.29 |
| openai_gpt-oss-120b (MoE) | 58.51 GiB | 116.83 B | llama-vulkan | Vulkan | 443.45 | 52.57 |
| openai_gpt-oss-120b (MoE) | 58.51 GiB | 116.83 B | llama-hipblaslt-0 | ROCm | 180.50 | 51.91 |
| openai_gpt-oss-120b (MoE) | 58.51 GiB | 116.83 B | llama-hipblaslt-1 | ROCm | 177.66 | 52.41 |
| openai_gpt-oss-120b (MoE) | 58.51 GiB | 116.83 B | llama-hip-rocm72-nofa-gfx1151 | ROCm | 177.18 | 49.44 |
| openai_gpt-oss-120b (MoE) | 58.51 GiB | 116.83 B | llama-hip-rocm72-vram | ROCm | 177.16 | 52.18 |
| openai_gpt-oss-120b (MoE) | 58.51 GiB | 116.83 B | llama-hip-rocm72-sdma1-gfx1151 | ROCm | 176.90 | 49.49 |
| openai_gpt-oss-120b (MoE) | 58.51 GiB | 116.83 B | llama-hip-rocm72-mmap-gfx1151 | ROCm | 176.84 | 49.47 |
| openai_gpt-oss-120b (MoE) | 58.51 GiB | 116.83 B | llama-hip-rocwmma-rocm72-gfx1151 | ROCm | 176.57 | 49.53 |
| openai_gpt-oss-120b (MoE) | 58.51 GiB | 116.83 B | llama-hip-rocm72-fa0-gfx1151 | ROCm | 176.57 | 49.55 |
| openai_gpt-oss-120b (MoE) | 58.51 GiB | 116.83 B | llama-hip-rocm72-gfx1151 | ROCm | 172.86 | 49.41 |
| Qwen3-235B-A22B (MoE) | 104.72 GiB | 235.09 B | llama-hipblaslt-0 | ROCm | 172.20 | 15.03 |
| Qwen3-235B-A22B (MoE) | 104.72 GiB | 235.09 B | llama-hip-rocm72-fa0-gfx1151 | ROCm | 166.29 | 14.56 |
| Qwen3-235B-A22B (MoE) | 104.72 GiB | 235.09 B | llama-hip-rocm72-vram | ROCm | 166.14 | 15.11 |
| Qwen3-235B-A22B (MoE) | 104.72 GiB | 235.09 B | llama-hip-rocm72-sdma1-gfx1151 | ROCm | 166.12 | 14.55 |
| Qwen3-235B-A22B (MoE) | 104.72 GiB | 235.09 B | llama-hip-rocm72-nofa-gfx1151 | ROCm | 166.08 | 14.51 |
| Qwen3-235B-A22B (MoE) | 104.72 GiB | 235.09 B | llama-hip-rocm72-gfx1151 | ROCm | 165.98 | 14.30 |
| Qwen3-235B-A22B (MoE) | 104.72 GiB | 235.09 B | llama-hipblaslt-1 | ROCm | 165.63 | 15.11 |
| Qwen3-235B-A22B (MoE) | 104.72 GiB | 235.09 B | llama-hip-rocwmma-rocm72-gfx1151 | ROCm | 164.60 | 14.24 |
| Qwen3-235B-A22B (MoE) | 104.72 GiB | 235.09 B | llama-hip-rocm72-mmap-gfx1151 | ROCm | 157.07 | 14.28 |
| Qwen3-235B-A22B (MoE) | 104.72 GiB | 235.09 B | llama-vulkan-radv-fa-finegrain | Vulkan | 122.11 | 16.07 |
| Qwen3-235B-A22B (MoE) | 104.72 GiB | 235.09 B | llama-vulkan-radv-ub1024 | Vulkan | 121.55 | 16.04 |
| Qwen3-235B-A22B (MoE) | 104.72 GiB | 235.09 B | llama-vulkan-radv-fa | Vulkan | 121.50 | 15.88 |