https://amdai.net/ Lab Notes posts, dated updates, and official amdai-bench/run@1 cards. Independent consumer notes. Not affiliated with AMD. en-US https://amdai.net/results/local-desktop-20261012-2358-35b-q6k-1gpu-256k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2358-35b-q6k-1gpu-256k-llamacpp-f16 Tue, 13 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q6_K · single · 262144 context · 1 GPU · decode 37.78 tok/s · prefill 471.6 tok/s · peak VRAM 29.3 GB · peak system RAM 43.69 GB https://amdai.net/results/local-desktop-20261013-0005-35b-q6k-2gpu-128k-llamacpp-f16 https://amdai.net/results/local-desktop-20261013-0005-35b-q6k-2gpu-128k-llamacpp-f16 Tue, 13 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q6_K · split · 131072 context · 2 GPU · decode 58.04 tok/s · prefill 1059.7 tok/s · peak VRAM 34.21 GB · peak system RAM 41.41 GB https://amdai.net/results/local-desktop-20261013-0005-35b-q6k-2gpu-32k-llamacpp-f16 https://amdai.net/results/local-desktop-20261013-0005-35b-q6k-2gpu-32k-llamacpp-f16 Tue, 13 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q6_K · split · 32768 context · 2 GPU · decode 75.43 tok/s · prefill 1890.7 tok/s · peak VRAM 30.66 GB · peak system RAM 38.98 GB https://amdai.net/results/local-desktop-20261013-0005-35b-q6k-2gpu-4k-llamacpp-f16 https://amdai.net/results/local-desktop-20261013-0005-35b-q6k-2gpu-4k-llamacpp-f16 Tue, 13 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q6_K · split · 4096 context · 2 GPU · decode 83.5 tok/s · prefill 2572.1 tok/s · peak VRAM 29.64 GB · peak system RAM 38.22 GB https://amdai.net/results/local-desktop-20261013-0007-35b-q6k-2gpu-256k-llamacpp-f16 https://amdai.net/results/local-desktop-20261013-0007-35b-q6k-2gpu-256k-llamacpp-f16 Tue, 13 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q6_K · split · 262144 context · 2 GPU · decode 42.2 tok/s · prefill 666.2 tok/s · peak VRAM 38.96 GB · peak system RAM 45.15 GB https://amdai.net/results/local-desktop-20261012-2311-35b-udq5kxl-1gpu-4k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2311-35b-udq5kxl-1gpu-4k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B UD-Q5_K_XL · single · 4096 context · 1 GPU · decode 102.39 tok/s · prefill 2761.6 tok/s · peak VRAM 25.97 GB · peak system RAM 37.32 GB https://amdai.net/results/local-desktop-20261012-2312-35b-udq5kxl-1gpu-128k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2312-35b-udq5kxl-1gpu-128k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B UD-Q5_K_XL · single · 131072 context · 1 GPU · decode 66.43 tok/s · prefill 942.8 tok/s · peak VRAM 28.82 GB · peak system RAM 40.18 GB https://amdai.net/results/local-desktop-20261012-2312-35b-udq5kxl-1gpu-32k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2312-35b-udq5kxl-1gpu-32k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B UD-Q5_K_XL · single · 32768 context · 1 GPU · decode 90.52 tok/s · prefill 1937.9 tok/s · peak VRAM 26.7 GB · peak system RAM 37.83 GB https://amdai.net/results/local-desktop-20261012-2314-35b-udq5kxl-1gpu-256k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2314-35b-udq5kxl-1gpu-256k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B UD-Q5_K_XL · single · 262144 context · 1 GPU · decode 44.4 tok/s · prefill 519.2 tok/s · peak VRAM 29.34 GB · peak system RAM 42.9 GB https://amdai.net/results/local-desktop-20261012-2319-35b-udq5kxl-2gpu-32k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2319-35b-udq5kxl-2gpu-32k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B UD-Q5_K_XL · split · 32768 context · 2 GPU · decode 74.8 tok/s · prefill 1916 tok/s · peak VRAM 27.41 GB · peak system RAM 38.16 GB https://amdai.net/results/local-desktop-20261012-2319-35b-udq5kxl-2gpu-4k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2319-35b-udq5kxl-2gpu-4k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B UD-Q5_K_XL · split · 4096 context · 2 GPU · decode 82.78 tok/s · prefill 2637.3 tok/s · peak VRAM 26.39 GB · peak system RAM 37.6 GB https://amdai.net/results/local-desktop-20261012-2320-35b-udq5kxl-2gpu-128k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2320-35b-udq5kxl-2gpu-128k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B UD-Q5_K_XL · split · 131072 context · 2 GPU · decode 57.94 tok/s · prefill 1071.1 tok/s · peak VRAM 30.94 GB · peak system RAM 41.02 GB https://amdai.net/results/local-desktop-20261012-2321-35b-udq5kxl-2gpu-256k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2321-35b-udq5kxl-2gpu-256k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B UD-Q5_K_XL · split · 262144 context · 2 GPU · decode 42.03 tok/s · prefill 671.7 tok/s · peak VRAM 35.74 GB · peak system RAM 44.71 GB https://amdai.net/results/local-desktop-20261012-2326-35b-q6k-1gpu-4k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2326-35b-q6k-1gpu-4k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q6_K · single · 4096 context · 1 GPU · decode 82.62 tok/s · prefill 2621.7 tok/s · peak VRAM 29.32 GB · peak system RAM 40.91 GB https://amdai.net/results/local-desktop-20261012-2327-35b-q5km-1gpu-32k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2327-35b-q5km-1gpu-32k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q5_K_M · single · 32768 context · 1 GPU · decode 96.21 tok/s · prefill 1902.6 tok/s · peak VRAM 25.39 GB · peak system RAM 36.46 GB https://amdai.net/results/local-desktop-20261012-2327-35b-q5km-1gpu-4k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2327-35b-q5km-1gpu-4k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q5_K_M · single · 4096 context · 1 GPU · decode 109.73 tok/s · prefill 2710.8 tok/s · peak VRAM 24.68 GB · peak system RAM 35.83 GB https://amdai.net/results/local-desktop-20261012-2328-35b-q5km-1gpu-128k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2328-35b-q5km-1gpu-128k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q5_K_M · single · 131072 context · 1 GPU · decode 69.64 tok/s · prefill 940.3 tok/s · peak VRAM 27.43 GB · peak system RAM 38.65 GB https://amdai.net/results/local-desktop-20261012-2329-35b-q5km-1gpu-256k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2329-35b-q5km-1gpu-256k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q5_K_M · single · 262144 context · 1 GPU · decode 40.54 tok/s · prefill 541.7 tok/s · peak VRAM 29.15 GB · peak system RAM 42.79 GB https://amdai.net/results/local-desktop-20261012-2334-35b-q5km-2gpu-4k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2334-35b-q5km-2gpu-4k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q5_K_M · split · 4096 context · 2 GPU · decode 88.19 tok/s · prefill 2647.2 tok/s · peak VRAM 25.13 GB · peak system RAM 37.17 GB https://amdai.net/results/local-desktop-20261012-2335-35b-q5km-2gpu-128k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2335-35b-q5km-2gpu-128k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q5_K_M · split · 131072 context · 2 GPU · decode 59.86 tok/s · prefill 1067.3 tok/s · peak VRAM 29.76 GB · peak system RAM 40.73 GB https://amdai.net/results/local-desktop-20261012-2335-35b-q5km-2gpu-32k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2335-35b-q5km-2gpu-32k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q5_K_M · split · 32768 context · 2 GPU · decode 78.91 tok/s · prefill 1920.9 tok/s · peak VRAM 26.13 GB · peak system RAM 37.8 GB https://amdai.net/results/local-desktop-20261012-2337-35b-q5km-2gpu-256k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2337-35b-q5km-2gpu-256k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q5_K_M · split · 262144 context · 2 GPU · decode 42.33 tok/s · prefill 664.6 tok/s · peak VRAM 34.54 GB · peak system RAM 46.21 GB https://amdai.net/results/local-desktop-20261012-2341-35b-q4km-1gpu-4k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2341-35b-q4km-1gpu-4k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q4_K_M · single · 4096 context · 1 GPU · decode 117.69 tok/s · prefill 2733.9 tok/s · peak VRAM 21.43 GB · peak system RAM 33.74 GB https://amdai.net/results/local-desktop-20261012-2342-35b-q4km-1gpu-128k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2342-35b-q4km-1gpu-128k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q4_K_M · single · 131072 context · 1 GPU · decode 72.48 tok/s · prefill 942.1 tok/s · peak VRAM 24.27 GB · peak system RAM 36.5 GB https://amdai.net/results/local-desktop-20261012-2342-35b-q4km-1gpu-32k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2342-35b-q4km-1gpu-32k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q4_K_M · single · 32768 context · 1 GPU · decode 101.88 tok/s · prefill 1902.1 tok/s · peak VRAM 22.17 GB · peak system RAM 34.58 GB https://amdai.net/results/local-desktop-20261012-2344-35b-q4km-1gpu-256k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2344-35b-q4km-1gpu-256k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q4_K_M · single · 262144 context · 1 GPU · decode 47.77 tok/s · prefill 558.5 tok/s · peak VRAM 27.03 GB · peak system RAM 39.05 GB https://amdai.net/results/local-desktop-20261012-2349-35b-q4km-2gpu-128k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2349-35b-q4km-2gpu-128k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q4_K_M · split · 131072 context · 2 GPU · decode 62.12 tok/s · prefill 1078.4 tok/s · peak VRAM 26.26 GB · peak system RAM 38.04 GB https://amdai.net/results/local-desktop-20261012-2349-35b-q4km-2gpu-32k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2349-35b-q4km-2gpu-32k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q4_K_M · split · 32768 context · 2 GPU · decode 82.57 tok/s · prefill 1933.1 tok/s · peak VRAM 22.72 GB · peak system RAM 34.39 GB https://amdai.net/results/local-desktop-20261012-2349-35b-q4km-2gpu-4k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2349-35b-q4km-2gpu-4k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q4_K_M · split · 4096 context · 2 GPU · decode 93.28 tok/s · prefill 2660.4 tok/s · peak VRAM 21.69 GB · peak system RAM 33.74 GB https://amdai.net/results/local-desktop-20261012-2351-35b-q4km-2gpu-256k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2351-35b-q4km-2gpu-256k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q4_K_M · split · 262144 context · 2 GPU · decode 42.88 tok/s · prefill 673 tok/s · peak VRAM 31.02 GB · peak system RAM 40.84 GB https://amdai.net/results/local-desktop-20261012-2355-35b-q6k-1gpu-4k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2355-35b-q6k-1gpu-4k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q6_K · single · 4096 context · 1 GPU · decode 103.23 tok/s · prefill 2623.6 tok/s · peak VRAM 29.26 GB · peak system RAM 40.15 GB https://amdai.net/results/local-desktop-20261012-2356-35b-q6k-1gpu-128k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2356-35b-q6k-1gpu-128k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q6_K · single · 131072 context · 1 GPU · decode 54.51 tok/s · prefill 800.3 tok/s · peak VRAM 29.25 GB · peak system RAM 45.43 GB https://amdai.net/results/local-desktop-20261012-2356-35b-q6k-1gpu-32k-llamacpp-f16 https://amdai.net/results/local-desktop-20261012-2356-35b-q6k-1gpu-32k-llamacpp-f16 Mon, 12 Oct 2026 12:00:00 GMT amdai-bench/run@1 · Qwen3.6-35B-A3B Q6_K · single · 32768 context · 1 GPU · decode 71.19 tok/s · prefill 1683 tok/s · peak VRAM 29.17 GB · peak system RAM 44.01 GB https://amdai.net/updates#2026-10-12-consumer-stacks https://amdai.net/updates#2026-10-12-consumer-stacks Mon, 12 Oct 2026 12:00:00 GMT North star is now the two stacks from our test rig. llama.cpp Vulkan daily driver; Strata for Flash; /calculator and /stacks are first-class pages. https://amdai.net/updates#2026-10-12-overnight-sweep https://amdai.net/updates#2026-10-12-overnight-sweep Mon, 12 Oct 2026 12:00:00 GMT 33 official amdai-bench/run@1 cards from local desktop (AMDai) for Qwen3.6-35B-A3B on one R9700 and both cards. Measured, pending re-verify after the DDR5-5200 RAM fix. https://amdai.net/updates#2026-10-12-rocm-10-1 https://amdai.net/updates#2026-10-12-rocm-10-1 Mon, 12 Oct 2026 12:00:00 GMT 10.1.0 lists R9700 as gfx1201 and Radeon Ubuntu 26.04.1 / 24.04.5 / 22.04.5. Published ROCm CLI is v0.2.0, not a 1.0.0 tag. New pages for search, errors, backends, quants, RSS. https://amdai.net/updates#2026-09-22-vllm-0-30 https://amdai.net/updates#2026-09-22-vllm-0-30 Tue, 22 Sep 2026 12:00:00 GMT Dated snapshot. AMD-validated vLLM 0.27 / ROCm 10.0.0 unchanged. Community image moved to vllm/vllm-openai-rocm:v0.30.0. Named Qwen defaults from that week were later withdrawn. https://amdai.net/updates#2026-09-21-weekly-refresh https://amdai.net/updates#2026-09-21-weekly-refresh Mon, 21 Sep 2026 12:00:00 GMT Dated snapshot. ROCm 10.0.0 unchanged for gfx1201. Reliability serve is AMD-aligned vLLM 0.27. vLLM 0.29 ROCm Docker remains the community cutting-edge pin (no newer stable that week). Named Qwen defaults from that week were later withdrawn. R9700 training KIs documented. https://amdai.net/blog/living-status-not-gospel https://amdai.net/blog/living-status-not-gospel Sun, 20 Sep 2026 12:00:00 GMT Why this site tracks ROCm and vLLM as dated snapshots, and how to refresh the YAML without redesigning pages. https://amdai.net/blog/vulkan-is-an-on-ramp https://amdai.net/blog/vulkan-is-an-on-ramp Fri, 18 Sep 2026 12:00:00 GMT Why llama.cpp on RADV still belongs in a ROCm-oriented home lab, especially on RDNA 4. https://amdai.net/blog/rocm-versions-are-a-matching-problem https://amdai.net/blog/rocm-versions-are-a-matching-problem Tue, 15 Sep 2026 12:00:00 GMT A changelog-style note on why “install the latest of everything” is the least reliable way to get gfx1201 working. https://amdai.net/blog/the-name-is-radeon-ai-pro-r9700 https://amdai.net/blog/the-name-is-radeon-ai-pro-r9700 Sat, 12 Sep 2026 12:00:00 GMT Why forums say “R9700 Pro,” why that is not an RX 9700, and which string to put in a bug report. https://amdai.net/updates#2026-09-09-vllm-0-29 https://amdai.net/updates#2026-09-09-vllm-0-29 Wed, 09 Sep 2026 12:00:00 GMT Upstream vLLM 0.29 ships ROCm wheels and Docker. Extra index like wheels.vllm.ai/rocm/0.29.0/rocm723; image vllm/vllm-openai-rocm:v0.29.0. Plain pip install vllm is CUDA. AMD 10.0 docs still documented 0.27. https://amdai.net/updates#2026-09-02-rocm-7-14-1 https://amdai.net/updates#2026-09-02-rocm-7-14-1 Wed, 02 Sep 2026 12:00:00 GMT 2 Sep 2026 patch on the 7.14 TheRock line. Still a valid pin if your wheel/image was built against it. Not a reason to install Ubuntu archive 7.1. https://amdai.net/updates#2026-08-26-rocm-10 https://amdai.net/updates#2026-08-26-rocm-10 Wed, 26 Aug 2026 12:00:00 GMT Production TheRock line. R9700 / gfx1201 first-class. amdrocm-* packages, one gfx family, scoped /opt/rocm prefix. AMD’s notes pair PyTorch 2.13 and vLLM 0.27 — that is the supported combo, not a forever pin.