AMD · CDNA 3
Best AI Models for AMD Instinct MI300X (192 GB)
Runs everything up to 120B MoE well.
192 GB HBM35300 GB/s1307 TFLOPS FP16750 WCDNA 3 2023
| # | Model | Speed | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Inkling SmallThinking Machines Q4_K_M · 162.7 GB · 115 t/s | 115 tok/s | Q4_K_M · 162.7 GB | |
| 3 | DeepSeek V4 FlashDeepSeek Q4_K_M · 175.1 GB · 88.7 t/s | 88.7 tok/s | Q4_K_M · 175.1 GB | |
| 4 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 5 | Q4_K_M · 19.3 GB · 92.1 t/s | 92.1 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Coding Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 2 | DeepSeek V4 FlashDeepSeek Q4_K_M · 175.1 GB · 88.7 t/s | 88.7 tok/s | Q4_K_M · 175.1 GB | |
| 3 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 156 t/s | 156 tok/s | Q4_K_M · 21.7 GB | |
| 4 | Qwen3.6 35B A3BAlibaba Q4_K_M · 22.0 GB · 112 t/s | 112 tok/s | Q4_K_M · 22.0 GB | |
| 5 | Q4_K_M · 19.3 GB · 92.1 t/s | 92.1 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Reasoning Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Inkling SmallThinking Machines Q4_K_M · 162.7 GB · 115 t/s | 115 tok/s | Q4_K_M · 162.7 GB | |
| 3 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 156 t/s | 156 tok/s | Q4_K_M · 21.7 GB | |
| 4 | DeepSeek V4 FlashDeepSeek Q4_K_M · 175.1 GB · 88.7 t/s | 88.7 tok/s | Q4_K_M · 175.1 GB | |
| 5 | Qwen3.8 Flash NextAlibaba Q4_K_M · 109.0 GB · 47.0 t/s | 47.0 tok/s | Q4_K_M · 109.0 GB |
| # | Model | Speed | Math Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Inkling SmallThinking Machines Q4_K_M · 162.7 GB · 115 t/s | 115 tok/s | Q4_K_M · 162.7 GB | |
| 3 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 4 | Qwen3.6 35B A3BAlibaba Q4_K_M · 22.0 GB · 112 t/s | 112 tok/s | Q4_K_M · 22.0 GB | |
| 5 | Q4_K_M · 19.3 GB · 92.1 t/s | 92.1 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Long context Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 3 | MiniMax M2.7MiniMax Q4_K_M · 139.6 GB · 69.0 t/s | 69.0 tok/s | Q4_K_M · 139.6 GB | |
| 4 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 156 t/s | 156 tok/s | Q4_K_M · 21.7 GB | |
| 5 | Q4_K_M · 19.3 GB · 92.1 t/s | 92.1 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed ↓ | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 976 t/s | 976 tok/s | Q4_K_M · 3.4 GB | |
| 2 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB · 500 t/s | 500 tok/s | Q4_K_M · 6.6 GB | |
| 3 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 178 t/s | 178 tok/s | Q4_K_M · 13.1 GB | |
| 4 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 156 t/s | 156 tok/s | Q4_K_M · 21.7 GB | |
| 5 | Q4_K_M · 19.3 GB · 92.1 t/s | 92.1 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Vision Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Inkling SmallThinking Machines Q4_K_M · 162.7 GB · 115 t/s | 115 tok/s | Q4_K_M · 162.7 GB | |
| 3 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 4 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 156 t/s | 156 tok/s | Q4_K_M · 21.7 GB | |
| 5 | JEV 27B VLautotrust Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB |
See all 419 models on the Instinct MI300X →Best quant that fits · speeds estimated on this card
The Instinct MI300X runs everything up to 120B MoE well. CDNA 3 generation, released 2023.
GenerationCDNA 3released 2023
Instinct MI300X strengths and weaknesses
Strengths
- Runs 120B MoE models: GPT OSS 120B at 121 tok/s in 70.8 GB
- Room for long context: Llama 3.3 70B Instruct at Q4 uses 51.3 GB, leaving 140.7 GB free
- Reads long prompts fast: A 4K-token prompt on Qwen3.8 27B starts answering in 791 ms
Weaknesses
- 750 W under load: Plan a 1,500 W PSU; ≈$164 a year at 4 h a day
Output speed depends on memory bandwidth (5300 GB/s).
- 9BQwen3.5 9B · Q4_K_M · 6.6 GB500tok/s
- 14BQwen3 14B · Q4_K_M · 11.9 GB335tok/s
- 27–31BQwen3.8 27B · Q4_K_M · 17.9 GB183tok/s
- 35B MoEQwen3.6 35B A3B · Q4_K_M · 22.0 GB112tok/s
- 70BLlama 3.3 70B Instruct · Q4_K_M · 51.3 GB68.3tok/s
- 120B MoEGPT OSS 120B · Q4_K_M · 70.8 GB121tok/s
Estimated speeds: Q4_K_M, 4K context, one user. How we estimate
All 1318 models on the Instinct MI300X
Featured onlyBest quant that fits, ordered by llmrun Score. Fit grades how well it runs here, S best. How ranking works
| Model | Quant | VRAM | Speed | Context | Score | Fit |
|---|---|---|---|---|---|---|
| Q4_K_M | 0.9 GB | 4676.5 t/s | 41K | S100, Runs great | ||
| Q4_K_M | 7.9 GB | 600.0 t/s | — | S100, Runs great | ||
| Q4_K_M | 85.3 GB | 87.1 t/s | 262K | S89, Runs great | ||
Llama 3.1 Nemotron 70B Instruct HFChatFunctions | Q4_K_M | 48 GB | 73.4 t/s | 131K | S86, Runs great | |
OvisOCR3VisionFunctions | Q4_K_M | 0.9 GB | 3697.7 t/s | 262K | S100, Runs great | |
JoyAI LLM FlashChat | Q4_K_M | 30.6 GB | 117.9 t/s | 131K | S94, Runs great | |
| BF16 | 66.6 GB | 48.7 t/s | 66K | A79, Runs well | ||
Athene v2 ChatChatFunctions | Q4_K_M | 49.3 GB | 71.3 t/s | 33K | S86, Runs great | |
| Q4_K_M | 79.8 GB | 42.6 t/s | 131K | A76, Runs well | ||
Penguin VL 2BChatVision | BF16 | 6.5 GB | 653.0 t/s | 41K | S100, Runs great | |
| Q4_K_M | 7.0 GB | 627.2 t/s | 33K | S100, Runs great | ||
MiniCPM OCR 1.0VisionFunctions | Q4_K_M | 1.5 GB | 2814.2 t/s | 262K | S100, Runs great | |
Qwen3 VL 30B A3B InstructVisionFunctions | Q4_K_M | 19.7 GB | 96.6 t/s | 262K | S90, Runs great | |
Qwen3 Coder NextChatCodeFunctions | Q4_K_M | 48.3 GB | 96.7 t/s | 262K | S90, Runs great | |
A.X 4.0 LightChatFunctions | Q4_K_M | 5.6 GB | 666.7 t/s | 16K | S100, Runs great |
Specs
- Architecture
- CDNA 3
- VRAM
- 192 GB HBM3
- Memory bandwidth
- 5300.0 GB/s
- FP16 (dense)
- 1307.40 TFLOPS
- Stream processors
- 19,456
- TDP
- 750 W
- Released
- 2023-12-06
Instinct MI300X FAQ
What is the largest model the Instinct MI300X can run?
MiMo V2.6 Flash RL (309B) is the largest at 4-bit or better: at Q4_K_M it uses 186.5 GB of 192 GB, and runs at about 80.4 tok/s.
Can the Instinct MI300X run Qwen3.8 27B?
Yes. Qwen3.8 27B fits at Q4_K_M, using 17.9 GB of 192 GB, and it is the best model that runs well on this card.
How many tokens per second does the Instinct MI300X give?
About 183 tok/s on Qwen3.8 27B at Q4_K_M, with the first token after 0.8 s on a 4K-token prompt.
What power supply does the Instinct MI300X need?
Plan a 1500 W PSU (2× the 750 W TDP).
Is 192 GB enough for coding and agents?
Yes for coding: Qwen3.8 27B, the best of the 19 coding models that fit, runs at about 183 tok/s.