Question 1

Can NVIDIA H20 96GB run DeepSeek LLM 67B?

Accepted Answer

Yes, NVIDIA H20 96GB can run DeepSeek LLM 67B with a B grade (Runs well). Expected decode speed: 70.5 tok/s.

Question 2

How much VRAM does DeepSeek LLM 67B need?

Accepted Answer

DeepSeek LLM 67B (67B parameters) requires approximately 63.3 GB of memory with Q4_K_M quantization.

Question 3

What is the best quantization for DeepSeek LLM 67B?

Accepted Answer

The recommended quantization for DeepSeek LLM 67B is Q4_K_M, which balances quality and memory efficiency.

Workload	Grade	Fit	Decode	TTFT	Context
Agentic Coding	B	Runs well	70.5 tok/s	3996 ms	4K
Chat	C	Runs well	70.5 tok/s	1499 ms	4K
Coding	B	Runs well	70.5 tok/s	2747 ms	4K
RAG	B	Runs well	70.5 tok/s	4995 ms	4K
Reasoning	B	Runs well	70.5 tok/s	3247 ms	4K

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	26.1 GB	Low	D35
Q3_K_S	3	32.8 GB	Low	D36
NVFP4	4

Can NVIDIA H20 96GB run DeepSeek LLM 67B?

Memory breakdown