What can the Instinct MI100 run?
With 32 GB, 8 of 12 popular open models fit. The best pick is Qwen3 32B at about 34 tok/s, faster than you can read.
ollama run qwen3:32bFirst time? Install Ollama, then paste that into a terminal. Is this your machine? Check it on the homepage.
Every model the Instinct MI100 can run
- Qwen3 4BQuick everyday chat and summaries~270 tok/sRuns great
qwen3:4b - Qwen3 8BA solid all-rounder for chat and writing~130 tok/sRuns great
qwen3:8b - Gemma 3 12BReads images as well as text~83 tok/sRuns great
gemma3:12b - Qwen3 14BNoticeably smarter at reasoning and code~73 tok/sRuns great
qwen3:14b - gpt-oss 20BOpenAI's open model. Fast, strong reasoning~220 tok/sRuns great
gpt-oss:20b - Gemma 3 27BGoogle's best open model for its size~40 tok/sRuns great
gemma3:27b - Qwen3 30B-A3B30B smarts at small-model speed~220 tok/sRuns great
qwen3:30b - Qwen3 32BStrong coding and long answers~34 tok/sRuns great
qwen3:32b - Llama 3.3 70BClose to the big cloud modelsneeds 45 GBToo big
llama3.3:70b - gpt-oss 120BOpenAI's largest open modelneeds 67 GBToo big
gpt-oss:120b - Qwen3 235B-A22BAlibaba's flagship open model, for multi-GPU rigsneeds 144 GBToo big
qwen3:235b - DeepSeek R1 671BThe full DeepSeek R1, for the biggest rigsneeds 406 GBToo big
deepseek-r1:671b
Can the Instinct MI100 run it?
Can the Instinct MI100 run Qwen3 30B-A3B?
Yes. Qwen3 30B-A3B needs about 21 GB and the Instinct MI100 has 32 GB of VRAM. Expect about 220 tok/s, faster than you can read.
Can the Instinct MI100 run Qwen3 32B?
Yes. Qwen3 32B needs about 22 GB and the Instinct MI100 has 32 GB of VRAM. Expect about 34 tok/s, faster than you can read.
Can the Instinct MI100 run Llama 3.3 70B?
No. Llama 3.3 70B needs about 45 GB and the Instinct MI100 has at most 32 GB of VRAM. Ollama can still start it by putting part of the model in system RAM, but it runs several times slower. The most capable model it runs well is Qwen3 32B, at about 34 tok/s.
Real runs on the Instinct MI100
Loading measured runs…
Generating a token reads the model's active weights from memory once, so speed is roughly memory bandwidth (1229 GB/s) divided by the active weight size, with efficiency calibrated on our own rack. Sizes are Ollama's default 4-bit downloads plus about 1.2 GB for context. Real results are usually within about a third. Measured runs are on the leaderboard.
The Ryzen AI Max+ 395 mini PC, 128 GB runs gpt-oss 120B at about 31 tok/s, from $3,400 used.