Best AI Models for AMD Instinct MI350X (288.0GB)
Datacenter CDNA4 accelerator — cloud/server only. Listed for comparison.
With 288 GB of memory, this is a high-end configuration for local AI. You can comfortably run most open-source LLMs including large 70B parameter models at good quantization levels, making it one of the best setups for serious local AI work.
At this memory tier, nearly every popular open-source model is within reach. You can run Llama 3 70B at Q4_K_M or even Q5_K_M quantization with room to spare, handle coding assistants like DeepSeek Coder 33B at high quality, and easily run any 7B–30B model at full or near-full precision. Context windows remain generous even with larger models, so multi-turn conversations and long-document processing work smoothly.
Runs Well
- 70B models (Llama 3 70B, Qwen 72B) at Q4–Q5
- 30B models at Q6–Q8 quality
- 7B–14B models at full FP16 precision
- Vision models (LLaVA, CogVLM) without compromise
Challenging
- Mixture-of-experts models like Mixtral 8x22B at higher quants
- 120B+ models still require lower quantizations
What LLMs Can AMD Instinct MI350X Run?
166 models · 133 excellent · 17 good
Showing compatibility for AMD Instinct MI350X
| Model | Quant | VRAM | Speed | Context | Status | Grade |
|---|---|---|---|---|---|---|
Q4_K_M·66.1 t/s tok/s·131K ctx·RUNS GREAT | Q4_K_M | 72.7 GB | 66.1 t/s | 131K | RUNS GREAT | S85 |
Q4_K_M·64.3 t/s tok/s·262K ctx·RUNS WELL | Q4_K_M | 74.7 GB | 64.3 t/s | 262K | RUNS WELL | A84 |
Q4_K_M·56.4 t/s tok/s·66K ctx·RUNS WELL | Q4_K_M | 85.1 GB | 56.4 t/s | 66K | RUNS WELL | A82 |
Q4_K_M·66.9 t/s tok/s·RUNS GREAT | Q4_K_M | 71.7 GB | 66.9 t/s | — | RUNS GREAT | S85 |
Q4_K_M·56.4 t/s tok/s·66K ctx·RUNS WELL | Q4_K_M | 85.1 GB | 56.4 t/s | 66K | RUNS WELL | A82 |
Q4_K_M·72.0 t/s tok/s·131K ctx·RUNS GREAT | Q4_K_M | 66.7 GB | 72.0 t/s | 131K | RUNS GREAT | S86 |
Q4_K_M·18.7 t/s tok/s·1049K ctx·DECENT | Q4_K_M | 256.7 GB | 18.7 t/s | 1049K | DECENT | B46 |
IQ2_M·18.6 t/s tok/s·1049K ctx·BARELY RUNS | IQ2_M | 258.5 GB | 18.6 t/s | 1049K | BARELY RUNS | C44 |
Q4_K_M·103.1 t/s tok/s·131K ctx·RUNS GREAT | Q4_K_M | 46.6 GB | 103.1 t/s | 131K | RUNS GREAT | S92 |
Q4_K_M·226.1 t/s tok/s·262K ctx·RUNS GREAT | Q4_K_M | 21.2 GB | 226.1 t/s | 262K | RUNS GREAT | S100 |
BF16·66.1 t/s tok/s·262K ctx·RUNS GREAT | BF16 | 72.6 GB | 66.1 t/s | 262K | RUNS GREAT | S85 |
Q4_K_M·103.1 t/s tok/s·131K ctx·RUNS GREAT | Q4_K_M | 46.6 GB | 103.1 t/s | 131K | RUNS GREAT | S92 |
Q4_K_M·107.6 t/s tok/s·33K ctx·RUNS GREAT | Q4_K_M | 44.6 GB | 107.6 t/s | 33K | RUNS GREAT | S92 |
Q4_K_M·97.6 t/s tok/s·262K ctx·RUNS GREAT | Q4_K_M | 49.2 GB | 97.6 t/s | 262K | RUNS GREAT | S91 |
Q4_K_M·236.6 t/s tok/s·41K ctx·RUNS GREAT | Q4_K_M | 20.3 GB | 236.6 t/s | 41K | RUNS GREAT | S100 |
Q4_K_M·289.7 t/s tok/s·262K ctx·RUNS GREAT | Q4_K_M | 16.6 GB | 289.7 t/s | 262K | RUNS GREAT | S100 |
AMD Instinct MI350X Specifications
- Brand
- AMD
- Architecture
- CDNA 4
- VRAM
- 288.0 GB HBM3e
- Memory Bandwidth
- 8000.0 GB/s
- Stream Processors
- 16,384
- FP16 Performance
- 2306.90 TFLOPS
- TDP
- 1000W
- Release Date
- 2025-06-12
Get Started
Prompt Processing
Estimated for MiMo V2.5, the compute-bound phase that reads your prompt before the first reply token appears.
Short chat
627 ms
512 tok prompt
Long chat
5.0 s
4,096 tok prompt
Document / codebase
40.1 s
32,768 tok prompt
Prefill is compute-bound and a different number from the decode tok/s shown elsewhere on this page — how prompt processing works →
Efficiency & Value
Based on MiMo V2.5 at ~25.7 tok/s decode.
Tokens per watt
Higher is better.
Performance figures are estimates calibrated as of 2026-07-30 — see calibration basis →
Frequently Asked Questions
- Can AMD Instinct MI350X run GLM 4.5?
Yes, the AMD Instinct MI350X with 288 GB can run GLM 4.5, GLM 4.6, Llama 3.2 90B Vision Instruct, and 1764 other models. 1654 models run at excellent quality, and 74 at good quality. Check the compatibility table above for the full list with VRAM usage and estimated speed.
- Is AMD Instinct MI350X good for AI?
The AMD Instinct MI350X has 288 GB of HBM3e, making it excellent for running local AI models. It supports 1728 models at good quality or better. With 8000.0 GB/s memory bandwidth, it delivers fast token generation speeds. This is an enthusiast-grade GPU that handles most popular open-source LLMs.
- How many parameters can AMD Instinct MI350X handle?
With 288 GB, the AMD Instinct MI350X supports models from 3B to 70B+ parameters depending on quantization level. At Q4_K_M (the recommended sweet spot), you can fit roughly 480B parameters. This means 7B models at high quality (Q6/Q8) or 30B+ models at Q4.
- What quantization should I use on AMD Instinct MI350X?
For the best balance of quality and speed on the AMD Instinct MI350X, start with Q4_K_M — it preserves ~85% of the original model quality while keeping VRAM usage reasonable. With 24+ GB, you have the headroom to run 7B models at Q5_K_M or even Q6_K for noticeably better output quality. For larger 30B models, Q4_K_M remains the sweet spot.
- How fast is AMD Instinct MI350X for AI inference?
With 8000.0 GB/s memory bandwidth, the AMD Instinct MI350X achieves approximately 1067 tokens/sec on a 7B model at Q4_K_M — that's very fast, well above conversational speed. A 14B model runs at ~533 tok/s. Token generation speed scales inversely with model size — smaller models are significantly faster.
tok/s = (8000 GB/s ÷ model GB) × efficiency
Smaller models = faster inference. Memory bandwidth is the main bottleneck for token generation speed.
Estimated speed on AMD Instinct MI350X
~22 tok/s~22 tok/s~25 tok/s~27 tok/sReal-world results typically within ±20%. Speed depends on quantization kernel, batch size, and software stack.
- What's the best model for AMD Instinct MI350X?
The top-rated models for the AMD Instinct MI350X are GLM 4.5, GLM 4.6, Llama 3.2 90B Vision Instruct. The best choice depends on your use case: coding assistants benefit from code-tuned models, while general chat works well with instruction-tuned models like Llama or Qwen.
- What power supply and cooling does AMD Instinct MI350X need?
The AMD Instinct MI350X has a TDP of 1000 W. A good rule of thumb is to provide at least double the GPU's TDP to cover the rest of the system — that means a 1600 W PSU or larger. At this power level, a high-airflow case matters: aim for at least two front intake fans and one rear exhaust, with tidy cabling so hot air isn't trapped around the card. LLM inference sustains full GPU load continuously — longer and more consistently than most gaming workloads — so also make sure your CPU cooler can keep up under combined load.
- Anything to watch out for with AMD Instinct MI350X?
Datacenter CDNA4 accelerator — cloud/server only. Listed for comparison.