DeepSeek V3 0324 (671B) on AMD Instinct MI350X

Can the AMD Instinct MI350X run DeepSeek V3 0324 (671B)?

Yes, it runs well (offloads to RAM)

Runs well (offloads to RAM). At Q2_K it needs about 289 GB of 288 GB usable, with an estimated speed of 139 tok/s.

Best quant: Q2_K
Generation speed: 139 tok/s
Prompt processing: 2,288 tok/s
First token (1K prompt): 0.4 s
Largest safe context: —

Best app to run it on the AMD Instinct MI350X

On Linux

llama.cpp

Command line + local API · runs on llama.cpp

Splits the model between GPU memory and system RAM, which this setup needs.
Gives you direct control of how many layers sit on the GPU, the best way to squeeze a model that doesn’t quite fit.
Usually the first to support new model architectures.

llama-server -hf unsloth/DeepSeek-V3-0324-GGUF:Q2_K

LM Studio Splits the model between GPU memory and system RAM, which this setup needs.
Lemonade Splits the model between GPU memory and system RAM, which this setup needs.

On Windows

llama.cpp

Command line + local API · runs on llama.cpp

Splits the model between GPU memory and system RAM, which this setup needs.
Gives you direct control of how many layers sit on the GPU, the best way to squeeze a model that doesn’t quite fit.
Usually the first to support new model architectures.

llama-server -hf unsloth/DeepSeek-V3-0324-GGUF:Q2_K

LM Studio Splits the model between GPU memory and system RAM, which this setup needs.
Lemonade Splits the model between GPU memory and system RAM, which this setup needs.

Want an agent? DeepSeek V3 0324 (671B) supports tool use, so you can pair a llama.cpp or MLX LM server with one of these. Apps that run on this computer come first.

Coding agents

Personal assistants

How we choose: the files published for this model, your operating system and GPU, whether it fits fully in memory, and published speed tests (MLX vs llama.cpp on Apple silicon, vLLM vs llama.cpp vs Ollama on NVIDIA). Details.

Download DeepSeek V3 0324 (671B)

Recommended for the AMD Instinct MI350X: Q2_K. Every other quantization has its own download link in the table below.

Quantized files · GGUF

Works in LM Studio, llama.cpp, Jan and Ollama. No sign-in needed.

Download the Q2_K files (opens in a new tab)

huggingface.co/unsloth/DeepSeek-V3-0324-GGUF (opens in a new tab)

ollama run hf.co/unsloth/DeepSeek-V3-0324-GGUF

Original weights · Hugging Face

Full-precision files from DeepSeek, for fine-tuning, vLLM, Transformers.

huggingface.co/deepseek-ai/DeepSeek-V3-0324 (opens in a new tab)

No sign-in needed.

Official repo · GitHub

Reference code, inference scripts and docs from DeepSeek. The weights themselves download from Hugging Face.

github.com/deepseek-ai/DeepSeek-V3 (opens in a new tab)

git clone https://github.com/deepseek-ai/DeepSeek-V3

Links checked 1 day ago against Hugging Face, Ollama and GitHub.

How it’s rated

Quality 58Good quality (B) · 58 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking.: How capable the model itself is at Q2_K, from its editorial quality tier scaled by the quantization or precision. It’s the same on any hardware that runs this variant. Scale: A+ 85+ · A 70+ · B 55+ · C 40+ · D under 40.
Runs Runs welloffloads to RAM: How well it runs on the AMD Instinct MI350X: speed, capped by how well it fits in memory. Runs great: 30+ tokens/s (or near the image, video or decision speed target) · Runs well: 10+ tokens/s · Slow: 3+ tokens/s · Very slow: under 3 · Won’t run: doesn’t fit even with offload. A tight fit or RAM offload caps it at Runs well; heavy offload caps it at Slow.

Match score

To order lists, each result also gets a match score out of 100 from four parts. It favors the most capable model that still runs well.

Model qualityEditorial quality tier of the model, scaled by the quantization or precision used.	26.1 / 45
Memory fitFull marks when everything sits in GPU memory; fewer when part spills to system RAM.	10.0 / 25
SpeedGeneration speed against a 40 tokens/s target for LLMs, or render time against a target for image and video.	15.0 / 15
Memory useRewards using the memory you have, so a bigger, better variant beats a tiny one that leaves most memory idle.	15.0 / 15
Match score	66.1 / 100

The match score orders results; it is not shown as a grade. Quality and Runs are the two ratings in the tables.

Every quantization on the AMD Instinct MI350X

Quant	Memory needed	Available	Speed	Runs	Quality	Download
`Q2_K`	289 GB	288 GB	139 tok/s	Runs welloffloads to RAM	58Good quality (B) · 58 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking.	GGUF · folder (opens in a new tab)
`Q3_K_M`	351 GB	288 GB	—	Won't run	69Good quality (B) · 69 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 9/100 (quality, fit, speed and memory use) sets the ranking.	GGUF · folder (opens in a new tab)
`Q4_K_M`	433 GB	288 GB	—	Won't run	78Great quality (A) · 78 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 10/100 (quality, fit, speed and memory use) sets the ranking.	GGUF · folder (opens in a new tab)
`Q5_K_M`	502 GB	288 GB	—	Won't run	80Great quality (A) · 80 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 10/100 (quality, fit, speed and memory use) sets the ranking.	GGUF · folder (opens in a new tab)
`Q6_K`	577 GB	288 GB	—	Won't run	82Great quality (A) · 82 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 11/100 (quality, fit, speed and memory use) sets the ranking.	GGUF · folder (opens in a new tab)
`Q8_0`	741 GB	288 GB	—	Won't run	83Great quality (A) · 83 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 11/100 (quality, fit, speed and memory use) sets the ranking.	GGUF · folder (opens in a new tab)
`F16`	1,385 GB	288 GB	—	Won't run	83Great quality (A) · 83 of 100: how capable the model is at F16, whatever it runs on. Match score 11/100 (quality, fit, speed and memory use) sets the ranking.	GGUF · folder (opens in a new tab) Original weights (opens in a new tab)

Memory figures assume 8K of context. Longer conversations or documents grow the KV cache.

Other hardware that runs it

69Good quality (B) · 69 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 82/100 (quality, fit, speed and memory use) sets the ranking. Apple M3 Ultra (512 GB) Q3_K_M · 27.5 tok/s · Runs well (tight fit)
69Good quality (B) · 69 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 85/100 (quality, fit, speed and memory use) sets the ranking. Apple M5 Ultra (512 GB) Q3_K_M · 38.8 tok/s · Runs well (tight fit)
78Great quality (A) · 78 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 79/100 (quality, fit, speed and memory use) sets the ranking. AMD EPYC 7002 / 7003 series NAS (512 GB) Q4_K_M · 3.7 tok/s · Slow (tight fit)
58Good quality (B) · 58 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking. Intel Xeon Silver (4100 to 4300 series) NAS (512 GB) Q2_K · 3.0 tok/s · Slow
78Great quality (A) · 78 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. Intel Xeon W-2100 series NAS (512 GB) Q4_K_M · 1.8 tok/s · Very slow (tight fit)
78Great quality (A) · 78 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 77/100 (quality, fit, speed and memory use) sets the ranking. Intel Xeon D-2700 series NAS (512 GB) Q4_K_M · 1.4 tok/s · Very slow (tight fit)

Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.