What can AMD Instinct MI300A (APU) run locally?

124 models fit comfortably on AMD Instinct MI300A (APU) with 32 GB system RAM. 2 more run with tradeoffs. Top pick: Hermes 3 Llama 3.1 8B.

What's the recommended runner for this hardware?

llama.cpp (CPU only). This is the runner that gives best tokens-per-second on AMD Instinct MI300A (APU) based on windows support and the GPU backend.

What can AMD Instinct MI300A (APU) run for agents?

Build: AMD Instinct MI300A (APU) + — + 32 GB RAM (windows)

Memory: 128 GB VRAM + 32 GB system RAM

Runner: llama.cpp (CPU only)

Any Chat Coding Agents Reasoning Vision Long context Creative

Runs comfortably
124 models

Ranked by fit for agents use case + predicted speed. Click a row for VRAM breakdown.

#1Hermes 3 Llama 3.1 8B

hermes

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 21.6 GBHeadroom: 106.4 GB

ollama run hermes3:8b

tok/s

Weights

8.50 GB

KV cache

4.00 GB

Activations

8.62 GB

Runtime

0.50 GB

Model details →Run-on benchmark page →

#2Dolphin 3.0 Mistral 24B

24B

dolphin

Commercial OK

Quant: Q4_K_MContext: 8,192VRAM: 35.9 GBHeadroom: 92.1 GB

ollama run dolphin-mistral:24b

tok/s

Weights

14.49 GB

KV cache

12.00 GB

Activations

8.92 GB

Runtime

0.50 GB

Model details →Run-on benchmark page →

#3Qwen 2.5 7B Instruct

qwen

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 17.0 GBHeadroom: 111.0 GB

ollama run qwen2.5:7b

tok/s

Weights

7.44 GB

KV cache

0.47 GB

Activations

8.56 GB

Runtime

0.50 GB

Model details →Run-on benchmark page →

#4Qwen 2.5 14B Instruct

14B

qwen

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 31.3 GBHeadroom: 96.7 GB

ollama run qwen2.5:14b

tok/s

Weights

14.88 GB

KV cache

7.00 GB

Activations

8.94 GB

Runtime

0.50 GB

Model details →Run-on benchmark page →

#5Llama 3.1 Nemotron Nano 8B

llama

Commercial OK

Quant: Q4_K_MContext: 8,192VRAM: 17.8 GBHeadroom: 110.2 GB

140

tok/s

Weights

4.83 GB

KV cache

4.00 GB

Activations

8.43 GB

Runtime

0.50 GB

Model details →Run-on benchmark page →

#6Mistral 7B Instruct v0.3

mistral

Commercial OK

Quant: Q5_K_MContext: 8,192VRAM: 17.2 GBHeadroom: 110.8 GB

ollama run mistral:7b

140

tok/s

Weights

4.81 GB

KV cache

3.50 GB

Activations

8.43 GB

Runtime

0.50 GB

Model details →Run-on benchmark page →

#7Qwen 3 30B-A3B

30B

qwen

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 57.2 GBHeadroom: 70.8 GB

ollama run qwen3:30b

tok/s

Weights

31.88 GB

KV cache

15.00 GB

Activations

9.79 GB

Runtime

0.50 GB

Model details →Run-on benchmark page →

#8Qwen 2.5 Coder 32B Instruct

32B

qwen

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 46.5 GBHeadroom: 81.5 GB

ollama run qwen2.5-coder:32b

tok/s

Weights

34.00 GB

KV cache

2.15 GB

Activations

9.89 GB

Runtime

0.50 GB

Model details →Run-on benchmark page →

#9Qwen 3 32B

32B

qwen

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 60.4 GBHeadroom: 67.6 GB

ollama run qwen3:32b

tok/s

Weights

34.00 GB

KV cache

16.00 GB

Activations

9.89 GB

Runtime

0.50 GB

Model details →Run-on benchmark page →

#10Qwen 2.5 32B Instruct

32B

qwen

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 60.4 GBHeadroom: 67.6 GB

ollama run qwen2.5:32b

tok/s

Weights

34.00 GB

KV cache

16.00 GB

Activations

9.89 GB

Runtime

0.50 GB

Model details →Run-on benchmark page →

#11Nemotron 3 Nano (30B-A3B)

30B

other

Commercial OK

Quant: Q8_0Context: 8,192VRAM: 57.2 GBHeadroom: 70.8 GB

ollama run nemotron3:nano

tok/s

Weights

31.88 GB

KV cache

15.00 GB

Activations

9.79 GB

Runtime

0.50 GB

Model details →Run-on benchmark page →

#12Gemma 4 26B MoE

26B

gemma

Commercial OK

Quant: Q4_K_MContext: 8,192VRAM: 38.2 GBHeadroom: 89.8 GB

ollama run gemma4:26b-moe

tok/s

Weights

15.70 GB

KV cache

13.00 GB

Activations

8.98 GB

Runtime

0.50 GB

Model details →Run-on benchmark page →

Runs with tradeoffs
2 models

Tight VRAM, partial CPU offload, or context-limited.

Command R+ 104B

104B

command-r

Quant: Q4_K_MContext: 8,192VRAM: 126.6 GBHeadroom: 1.4 GB

• Tight VRAM fit — only 1.4 GB headroom left for context growth

ollama run command-r-plus:104b

tok/s

Weights

62.79 GB

KV cache

52.00 GB

Activations

11.33 GB

Runtime

0.50 GB

Model details →Run-on benchmark page →

Command R+ (Aug 2024)

104B

command-r

Quant: AWQ-INT4Context: 2,048VRAM: 124.7 GBHeadroom: 3.3 GB

• Tight VRAM fit — only 3.3 GB headroom left for context growth

tok/s

Weights

104.00 GB

KV cache

13.00 GB

Activations

7.25 GB

Runtime

0.50 GB

Model details →Run-on benchmark page →

What if you upgraded?

Hypothetical scenarios. We re-ran the compatibility engine for each.

+32 GB system RAM

~$80–150

Doubles your CPU-offload working set. Helps when models don't quite fit in VRAM.

Unlocks: 36 new comfortable, 4 new tradeoff

• Gemma 3 1B
• Llama 3.2 1B Instruct
• Gemma 4 E2B (Effective 2B)
• Llama 3.2 3B Instruct

Shop this upgrade↗

Upgrade to AMD Instinct MI300X

see current pricing

192 GB VRAM (vs your 128 GB) plus a bandwidth jump from ~? GB/s to ~5325 GB/s.

Unlocks: 44 new comfortable

• Gemma 3 1B
• Llama 3.2 1B Instruct
• Gemma 4 E2B (Effective 2B)
• Llama 3.2 3B Instruct

Shop this upgrade↗

Add a second AMD Instinct MI300A (APU)

see current pricing

Tensor parallelism splits the model across both cards, effectively doubling VRAM. Bandwidth doesn't double — runs ~1.5× the single-card speed in practice.

Unlocks: 47 new comfortable

• Gemma 3 1B
• Llama 3.2 1B Instruct
• Gemma 4 E2B (Effective 2B)
• Llama 3.2 3B Instruct

Shop this upgrade↗

Some links above are affiliate links. We may earn a commission at no extra cost to you. How we make money.

Won't run
top 5 popular models

Need more memory than you have. Shown for orientation.

DeepSeek V4 Pro (1.6T MoE)

1600B

deepseek

Commercial OK