{ "date": "2026-10-28", "llama_cpp_commit": "f5222f25c18da062ab2b8a2dbbfbf4feae398566", "hardware": "Framework Desktop, AMD Ryzen AI Max+ 395, Radeon 8060S, 128 GB shared memory", "backend": "Vulkan, KHR cooperative matrices", "driver": "Mesa RADV 26.2.5-arch1.1", "vulkan_reported_memory_mib": 114233, "strict_96_gb_cap_enforced": false, "command": "llama-bench -m MODEL.gguf -p 512 -n 128 -t 6 -ngl 999 -r 5 -o json -v", "notes": [ "Separate pp512 and tg128 tests, not generation after a 512-token prompt.", "Throughput only, no quality evaluation.", "File sizes are GGUF bytes including all shards, not peak memory usage.", "Raw llama-bench model_filename fields omitted to avoid publishing local paths." ], "models": [ { "id": "Qwen3.8-27B-Q8_0", "repository": "unsloth/Qwen3.8-27B-GGUF", "revision": "4ca720788d1e01f1bff70c033e0d0028fd02e502", "gguf_bytes": 29047086048, "status": "OK", "measurements": { "pp512": { "mean_tokens_per_second": 476.889172, "stddev_tokens_per_second": 1.254454, "samples_tokens_per_second": [ 478.024, 478.001, 477.096, 475.062, 476.264 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "qwen35 27B Q8_0", "model_size": 29036089344, "model_n_params": 27320697856, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 512, "n_gen": 0, "n_depth": 0, "test_time": "2026-10-28T13:44:42Z", "avg_ns": 1073630661, "stddev_ns": 2827869, "avg_ts": 476.889172, "stddev_ts": 1.254454, "samples_ns": [ 1071076657, 1071128027, 1073160177, 1077753775, 1075034673 ], "samples_ts": [478.024, 478.001, 477.096, 475.062, 476.264] } }, "tg128": { "mean_tokens_per_second": 7.50314, "stddev_tokens_per_second": 0.002014, "samples_tokens_per_second": [ 7.50359, 7.50188, 7.50334, 7.50078, 7.50611 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "qwen35 27B Q8_0", "model_size": 29036089344, "model_n_params": 27320697856, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 0, "n_gen": 128, "n_depth": 0, "test_time": "2026-10-28T13:44:49Z", "avg_ns": 17059525315, "stddev_ns": 4583550, "avg_ts": 7.50314, "stddev_ts": 0.002014, "samples_ns": [ 17058506331, 17062400446, 17059072201, 17064882059, 17052765542 ], "samples_ts": [7.50359, 7.50188, 7.50334, 7.50078, 7.50611] } } } }, { "id": "Qwen3.8-27B-UD-Q4_K_M", "repository": "unsloth/Qwen3.8-27B-GGUF", "revision": "4ca720788d1e01f1bff70c033e0d0028fd02e502", "gguf_bytes": 16464440224, "status": "OK", "measurements": { "pp512": { "mean_tokens_per_second": 413.768702, "stddev_tokens_per_second": 3.057555, "samples_tokens_per_second": [ 409.699, 414.444, 418.206, 412.982, 413.512 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "qwen35 27B Q4_K - Medium", "model_size": 16453443584, "model_n_params": 27320697856, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 512, "n_gen": 0, "n_depth": 0, "test_time": "2026-10-28T13:46:20Z", "avg_ns": 1237460293, "stddev_ns": 9133490, "avg_ts": 413.768702, "stddev_ts": 3.057555, "samples_ns": [ 1249697481, 1235391179, 1224275832, 1239763282, 1238173695 ], "samples_ts": [409.699, 414.444, 418.206, 412.982, 413.512] } }, "tg128": { "mean_tokens_per_second": 12.816562, "stddev_tokens_per_second": 0.002302, "samples_tokens_per_second": [ 12.8199, 12.8166, 12.816, 12.8134, 12.8168 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "qwen35 27B Q4_K - Medium", "model_size": 16453443584, "model_n_params": 27320697856, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 0, "n_gen": 128, "n_depth": 0, "test_time": "2026-10-28T13:46:27Z", "avg_ns": 9987078264, "stddev_ns": 1793987, "avg_ts": 12.816562, "stddev_ts": 0.002302, "samples_ns": [ 9984486015, 9987033628, 9987491652, 9989514580, 9986865445 ], "samples_ts": [12.8199, 12.8166, 12.816, 12.8134, 12.8168] } } } }, { "id": "Qwen3.8-27B-UD-Q6_K", "repository": "unsloth/Qwen3.8-27B-GGUF", "revision": "4ca720788d1e01f1bff70c033e0d0028fd02e502", "gguf_bytes": 21983677344, "status": "OK", "measurements": { "pp512": { "mean_tokens_per_second": 401.292572, "stddev_tokens_per_second": 0.863539, "samples_tokens_per_second": [ 401.336, 400.206, 402.221, 400.667, 402.032 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "qwen35 27B Q6_K", "model_size": 21972680704, "model_n_params": 27320697856, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 512, "n_gen": 0, "n_depth": 0, "test_time": "2026-10-28T13:47:24Z", "avg_ns": 1275881821, "stddev_ns": 2746308, "avg_ts": 401.292572, "stddev_ts": 0.863539, "samples_ns": [ 1275738310, 1279340962, 1272931598, 1277868451, 1273529784 ], "samples_ts": [401.336, 400.206, 402.221, 400.667, 402.032] } }, "tg128": { "mean_tokens_per_second": 9.711583, "stddev_tokens_per_second": 0.002002, "samples_tokens_per_second": [ 9.71285, 9.71003, 9.70895, 9.71369, 9.71239 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "qwen35 27B Q6_K", "model_size": 21972680704, "model_n_params": 27320697856, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 0, "n_gen": 128, "n_depth": 0, "test_time": "2026-10-28T13:47:31Z", "avg_ns": 13180138209, "stddev_ns": 2717992, "avg_ts": 9.711583, "stddev_ts": 0.002002, "samples_ns": [ 13178415186, 13182247311, 13183707046, 13177280641, 13179040862 ], "samples_ts": [9.71285, 9.71003, 9.70895, 9.71369, 9.71239] } } } }, { "id": "Mistral-Small-4-119B-2603-UD-Q4_K_M", "repository": "unsloth/Mistral-Small-4-119B-2603-GGUF", "revision": "bd93c721735aa32c035c0f19e738cb3371fd56ff", "gguf_bytes": 73763180544, "status": "OK", "measurements": { "pp512": { "mean_tokens_per_second": 675.068877, "stddev_tokens_per_second": 2.995384, "samples_tokens_per_second": [ 672.752, 680.107, 675.3, 672.989, 674.196 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "mistral4 ?B Q4_K - Medium", "model_size": 73755267072, "model_n_params": 118972826624, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 512, "n_gen": 0, "n_depth": 0, "test_time": "2026-10-28T13:48:57Z", "avg_ns": 758453021, "stddev_ns": 3350836, "avg_ts": 675.068877, "stddev_ts": 2.995384, "samples_ns": [ 761052550, 752822930, 758181076, 760785519, 759423031 ], "samples_ts": [672.752, 680.107, 675.3, 672.989, 674.196] } }, "tg128": { "mean_tokens_per_second": 32.71998, "stddev_tokens_per_second": 0.019632, "samples_tokens_per_second": [ 32.7055, 32.7324, 32.7474, 32.7143, 32.7003 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "mistral4 ?B Q4_K - Medium", "model_size": 73755267072, "model_n_params": 118972826624, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 0, "n_gen": 128, "n_depth": 0, "test_time": "2026-10-28T13:49:02Z", "avg_ns": 3911983898, "stddev_ns": 2347515, "avg_ts": 32.71998, "stddev_ts": 0.019632, "samples_ns": [ 3913717284, 3910493852, 3908701341, 3912664105, 3914342910 ], "samples_ts": [32.7055, 32.7324, 32.7474, 32.7143, 32.7003] } } } }, { "id": "Mistral-Small-4-119B-2603-UD-Q5_K_M", "repository": "unsloth/Mistral-Small-4-119B-2603-GGUF", "revision": "bd93c721735aa32c035c0f19e738cb3371fd56ff", "gguf_bytes": 89160470528, "status": "OK", "measurements": { "pp512": { "mean_tokens_per_second": 611.022833, "stddev_tokens_per_second": 3.467403, "samples_tokens_per_second": [ 606.513, 615.047, 610.118, 609.508, 613.929 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "mistral4 ?B Q5_K - Medium", "model_size": 89152557056, "model_n_params": 118972826624, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 512, "n_gen": 0, "n_depth": 0, "test_time": "2026-10-28T13:49:47Z", "avg_ns": 837960819, "stddev_ns": 4756393, "avg_ts": 611.022833, "stddev_ts": 3.467403, "samples_ns": [ 844170321, 832457120, 839182164, 840021663, 833972828 ], "samples_ts": [606.513, 615.047, 610.118, 609.508, 613.929] } }, "tg128": { "mean_tokens_per_second": 29.832304, "stddev_tokens_per_second": 0.018096, "samples_tokens_per_second": [ 29.8329, 29.8181, 29.8136, 29.8593, 29.8376 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "mistral4 ?B Q5_K - Medium", "model_size": 89152557056, "model_n_params": 118972826624, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 0, "n_gen": 128, "n_depth": 0, "test_time": "2026-10-28T13:49:52Z", "avg_ns": 4290652046, "stddev_ns": 2603231, "avg_ts": 29.832304, "stddev_ts": 0.018096, "samples_ns": [ 4290564746, 4292690731, 4293345818, 4286768617, 4289890321 ], "samples_ts": [29.8329, 29.8181, 29.8136, 29.8593, 29.8376] } } } }, { "id": "gemma-4-31B-it-Q4_K_M", "repository": "unsloth/gemma-4-31B-it-GGUF", "revision": "c1ac76e99d5513b141e8adde7288b85c3f9c32ec", "gguf_bytes": 18323733440, "status": "OK", "measurements": { "pp512": { "mean_tokens_per_second": 359.994936, "stddev_tokens_per_second": 1.082351, "samples_tokens_per_second": [ 359.189, 361.733, 359.417, 360.374, 359.262 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "gemma4 31B Q4_K - Medium", "model_size": 18307897584, "model_n_params": 30697345596, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 512, "n_gen": 0, "n_depth": 0, "test_time": "2026-10-28T13:50:19Z", "avg_ns": 1422252487, "stddev_ns": 4265601, "avg_ts": 359.994936, "stddev_ts": 1.082351, "samples_ns": [ 1425433654, 1415408210, 1424531070, 1420746542, 1425142960 ], "samples_ts": [359.189, 361.733, 359.417, 360.374, 359.262] } }, "tg128": { "mean_tokens_per_second": 11.268428, "stddev_tokens_per_second": 0.00155, "samples_tokens_per_second": [ 11.2704, 11.2694, 11.2668, 11.2687, 11.2669 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "gemma4 31B Q4_K - Medium", "model_size": 18307897584, "model_n_params": 30697345596, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 0, "n_gen": 128, "n_depth": 0, "test_time": "2026-10-28T13:50:29Z", "avg_ns": 11359170831, "stddev_ns": 1564548, "avg_ts": 11.268428, "stddev_ts": 0.00155, "samples_ns": [ 11357202126, 11358237917, 11360762047, 11358922745, 11360729321 ], "samples_ts": [11.2704, 11.2694, 11.2668, 11.2687, 11.2669] } } } }, { "id": "gemma-4-31B-it-Q6_K", "repository": "unsloth/gemma-4-31B-it-GGUF", "revision": "c1ac76e99d5513b141e8adde7288b85c3f9c32ec", "gguf_bytes": 25201486784, "status": "OK", "measurements": { "pp512": { "mean_tokens_per_second": 347.981942, "stddev_tokens_per_second": 0.394342, "samples_tokens_per_second": [ 348.583, 348.163, 347.608, 347.826, 347.73 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "gemma4 31B Q6_K", "model_size": 25185650928, "model_n_params": 30697345596, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 512, "n_gen": 0, "n_depth": 0, "test_time": "2026-10-28T13:51:33Z", "avg_ns": 1471342228, "stddev_ns": 1666640, "avg_ts": 347.981942, "stddev_ts": 0.394342, "samples_ns": [ 1468802516, 1470576705, 1472923645, 1472001576, 1472406700 ], "samples_ts": [348.583, 348.163, 347.608, 347.826, 347.73] } }, "tg128": { "mean_tokens_per_second": 8.103701, "stddev_tokens_per_second": 0.001396, "samples_tokens_per_second": [ 8.10537, 8.10277, 8.10412, 8.10441, 8.10184 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "gemma4 31B Q6_K", "model_size": 25185650928, "model_n_params": 30697345596, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 0, "n_gen": 128, "n_depth": 0, "test_time": "2026-10-28T13:51:42Z", "avg_ns": 15795251809, "stddev_ns": 2720661, "avg_ts": 8.103701, "stddev_ts": 0.001396, "samples_ns": [ 15791999249, 15797061971, 15794438401, 15793875096, 15798884328 ], "samples_ts": [8.10537, 8.10277, 8.10412, 8.10441, 8.10184] } } } }, { "id": "gemma-4-31B-it-Q8_0", "repository": "unsloth/gemma-4-31B-it-GGUF", "revision": "c1ac76e99d5513b141e8adde7288b85c3f9c32ec", "gguf_bytes": 32635677632, "status": "OK", "measurements": { "pp512": { "mean_tokens_per_second": 412.117261, "stddev_tokens_per_second": 0.581744, "samples_tokens_per_second": [ 412.791, 412.362, 412.422, 411.5, 411.512 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "gemma4 31B Q8_0", "model_size": 32619841776, "model_n_params": 30697345596, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 512, "n_gen": 0, "n_depth": 0, "test_time": "2026-10-28T13:53:11Z", "avg_ns": 1242366831, "stddev_ns": 1754775, "avg_ts": 412.117261, "stddev_ts": 0.581744, "samples_ns": [ 1240338101, 1241626567, 1241448317, 1244229772, 1244191402 ], "samples_ts": [412.791, 412.362, 412.422, 411.5, 411.512] } }, "tg128": { "mean_tokens_per_second": 6.462613, "stddev_tokens_per_second": 0.001815, "samples_tokens_per_second": [ 6.46055, 6.46085, 6.46442, 6.46311, 6.46413 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "gemma4 31B Q8_0", "model_size": 32619841776, "model_n_params": 30697345596, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 0, "n_gen": 128, "n_depth": 0, "test_time": "2026-10-28T13:53:18Z", "avg_ns": 19806231521, "stddev_ns": 5563238, "avg_ts": 6.462613, "stddev_ts": 0.001815, "samples_ns": [ 19812560755, 19811626791, 19800706901, 19804695171, 19801567987 ], "samples_ts": [6.46055, 6.46085, 6.46442, 6.46311, 6.46413] } } } }, { "id": "GLM-5.3-Flash-UD-IQ1_S", "repository": "unsloth/GLM-5.3-Flash-GGUF", "revision": "a38483c8cd5df544f53d70fb281afe97369d5ab6", "gguf_bytes": 93087245088, "status": "OK", "measurements": { "pp512": { "mean_tokens_per_second": 196.594939, "stddev_tokens_per_second": 0.785736, "samples_tokens_per_second": [ 196.306, 197.856, 195.718, 196.653, 196.442 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "glm5-next 320B.A18B IQ1_S - 1.5625 bpw", "model_size": 93077724536, "model_n_params": 320759404382, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 512, "n_gen": 0, "n_depth": 0, "test_time": "2026-10-28T13:55:22Z", "avg_ns": 2604372871, "stddev_ns": 10381511, "avg_ts": 196.594939, "stddev_ts": 0.785736, "samples_ns": [ 2608167692, 2587742721, 2616012503, 2603569110, 2606372332 ], "samples_ts": [196.306, 197.856, 195.718, 196.653, 196.442] } }, "tg128": { "mean_tokens_per_second": 19.016641, "stddev_tokens_per_second": 0.020365, "samples_tokens_per_second": [ 19.0436, 19.0116, 18.9919, 19.0059, 19.0301 ], "raw_llama_bench": { "build_commit": "f5222f25c", "build_number": 11962, "cpu_info": "AMD RYZEN AI MAX+ 395 w/ Radeon 8060S", "gpu_info": "AMD Radeon 8060S Graphics (RADV STRIX_HALO)", "backends": "Vulkan", "model_type": "glm5-next 320B.A18B IQ1_S - 1.5625 bpw", "model_size": 93077724536, "model_n_params": 320759404382, "n_batch": 2048, "n_ubatch": 512, "n_threads": 6, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": -1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 0, "n_gen": 128, "n_depth": 0, "test_time": "2026-10-28T13:55:38Z", "avg_ns": 6730953025, "stddev_ns": 7207520, "avg_ts": 19.016641, "stddev_ts": 0.020365, "samples_ns": [ 6721410025, 6732714290, 6739699825, 6734748924, 6726192064 ], "samples_ts": [19.0436, 19.0116, 18.9919, 19.0059, 19.0301] } } } } ] }