What people run, by hardware
Pick the kind of machine you have. You'll see the model sizes people report running on it, most reported first, with the speed they typically get. It tells you what works on hardware like yours. It doesn't tell you which model gives the best answers.
To check a specific model on your exact card, use the calculator. For coding, see the coding models by VRAM.
AMD and Intel up to 16 GB· Intel only
Discrete AMD and Intel cards up to 16 GB, on ROCm, Vulkan or SYCL. Memory speed 456 to 560 GB/s across the 2 devices with reports, so a typical figure here spans all of them. 15 single-device reports.
| # | Model | Reports | Gen t/s | PP t/s | With speculation |
|---|---|---|---|---|---|
| 1 | Bonsai 227B PrismML · mostly ternary Gen 45 t/s (3 runs) · PP 1188 t/s (2 runs) | 3 | 45 3 runs | 1188 2 runs | none |
| 2 | Qwen3.827B Alibaba · mostly Q3_K_S Gen 6.3 t/s (2 runs) | 2 | 6.3 2 runs, 2 devices | no PP data | none |
| 3 | Qwen31.7B Alibaba · mostly Q8_0 Gen 33 t/s (1 run) | 1 | 33 1 run | no PP data | none |
| 4 | GLM-49B mostly Q4_0 Gen 21 t/s (1 run) | 1 | 21 1 run | no PP data | none |
| 5 | Qwen3.6 Coder27B · 3B active Alibaba · mostly Q4_K_M Gen 14 t/s (1 run) | 1 | 14 1 run | no PP data | none |
| 6 | Qwen34B Alibaba · mostly Q4_K_M Gen 13 t/s (1 run) · PP 261 t/s (1 run) | 1 | 13 1 run | 261 1 run | none |
| 7 | Bonsai 2 Ternary27B PrismML · mostly 2-bit ternary No generation data | 1 | no generation data | no PP data | none |
| 8 | Gemma 412B Google DeepMind · mostly QAT No generation data · with speculation 20 t/s (1) | 1 | no generation data | no PP data | 20 1 run |
| 9 | Llama 27B mostly Q4_0 No generation data · PP 4454 t/s (1 run) | 1 | no generation data | 4454 1 run | none |
| 10 | Qwen3.5 DeepSeek-V4-Flash9B Alibaba · mostly Q6_K No generation data · with speculation 49 t/s (1) | 1 | no generation data | no PP data | 49 1 run |
| 11 | Qwen3.635B · 3B active Alibaba · mostly IQ2_XXS No generation data | 1 | no generation data | no PP data | none |
| 12 | Qwen3.8 Flash-Next125B · 6B active Alibaba · mostly IQ2_XS No generation data | 1 | no generation data | no PP data | none |
How this list works
Each row is one model size, because a 27B and a 397B from the same family are different choices. Rows are ordered by how many people report running that size on this hardware.
Generation speed is the median of runs on one device, serving one request, without speculative decoding. PP is the median prompt-processing speed from plain runs that reported it, so its run count can be smaller. Speculative decoding (MTP, a draft model, DFlash, n-gram lookup) can make the same model several times faster, so those runs get their own column. A report that doesn't make clear which kind of run it was counts in neither.
Hardware is grouped by how much it holds and how fast its memory is, since memory speed sets how fast a model writes. Click a model for every report behind its row.