Llama 2 70B Chat on NVIDIA GH200 Grace Hopper (96 GB GPU)

Can the NVIDIA GH200 Grace Hopper (96 GB GPU) run Llama 2 70B Chat?

Yes, it runs great

Runs great. At Q6_K it needs about 63.3 GB of 96.0 GB usable, with an estimated speed of 49.4 tok/s.

Best quant: Q6_K
Generation speed: 49.4 tok/s
Prompt processing: 1,397 tok/s
First token (1K prompt): 0.7 s
Largest safe context: 4K

Best app to run it on the NVIDIA GH200 Grace Hopper (96 GB GPU)

On Linux

vLLM

Inference server · runs on vLLM

The full model fits in GPU memory at 8-bit, so it can batch many requests: best for serving a team or an app.

vllm serve meta-llama/Llama-2-70b-chat-hf

Ollama One command with Ollama’s own build of this model.
LM Studio Point-and-click: search, download and chat in one window.

On Windows

Ollama

Command line + local API · runs on llama.cpp (ggml), MLX (Apple silicon, preview)

One command with Ollama’s own build of this model.
Serves an OpenAI-compatible API that other apps and coding tools can use.

ollama run llama2:70b-chat-q6_K

LM Studio Point-and-click: search, download and chat in one window.
llama.cpp Usually the first to support new model architectures.

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 Llama 2 70B Chat

Recommended for the NVIDIA GH200 Grace Hopper (96 GB GPU): Q6_K. Every other quantization has its own download link in the table below.

Quickest · Ollama

One command downloads a ready-to-run version and starts a chat.

Get Ollama at ollama.com/download (opens in a new tab) · model page (opens in a new tab)

ollama run llama2:70b-chat-q6_K

Quantized files · GGUF

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

huggingface.co/TheBloke/Llama-2-70B-Chat-GGUF (opens in a new tab)

ollama run hf.co/TheBloke/Llama-2-70B-Chat-GGUF

Original weights · Hugging Face

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

huggingface.co/meta-llama/Llama-2-70b-chat-hf (opens in a new tab)

Free Hugging Face account needed: request access on the model page, and Meta approves it (usually within minutes to a few hours).

Official repo · GitHub

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

github.com/meta-llama/llama (opens in a new tab)

Meta’s original Llama 2 repo (now archived).

git clone https://github.com/meta-llama/llama

Links checked 2 days ago against Hugging Face, Ollama and GitHub.

How it’s rated

Quality 35Basic quality (D) · 35 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking.: How capable the model itself is at Q6_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 great: How well it runs on the NVIDIA GH200 Grace Hopper (96 GB GPU): 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.	16.0 / 45
Memory fitFull marks when everything sits in GPU memory; fewer when part spills to system RAM.	25.0 / 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	71.0 / 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 NVIDIA GH200 Grace Hopper (96 GB GPU)

Quant	Memory needed	Available	Speed	Runs	Quality	Download
`Q2_K`	34.3 GB	96.0 GB	117 tok/s	Runs great	25Basic quality (D) · 25 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking.	GGUF (opens in a new tab)
`Q3_K_M`	40.5 GB	96.0 GB	92.0 tok/s	Runs great	30Basic quality (D) · 30 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking.	GGUF (opens in a new tab)
`Q4_K_M`	48.8 GB	96.0 GB	71.2 tok/s	Runs great	34Basic quality (D) · 34 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking.	GGUF (opens in a new tab)
`Q5_K_M`	55.7 GB	96.0 GB	58.8 tok/s	Runs great	34Basic quality (D) · 34 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking.	GGUF (opens in a new tab)
`Q6_K`	63.3 GB	96.0 GB	49.4 tok/s	Runs great	35Basic quality (D) · 35 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 71/100 (quality, fit, speed and memory use) sets the ranking.
`Q8_0`	79.9 GB	96.0 GB	36.2 tok/s	Runs great	36Basic quality (D) · 36 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 70/100 (quality, fit, speed and memory use) sets the ranking.
`F16`	145 GB	96.0 GB	—	Won't run	36Basic quality (D) · 36 of 100: how capable the model is at F16, whatever it runs on. Match score 4/100 (quality, fit, speed and memory use) sets the ranking.	Original weights (opens in a new tab)

Memory figures assume 4K of context. Longer conversations or documents grow the KV cache.

Other hardware that runs it

34Basic quality (D) · 34 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. NVIDIA RTX PRO 6000 Blackwell Q4_K_M · 31.9 tok/s · Runs great
25Basic quality (D) · 25 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. NVIDIA GeForce RTX 5090 Q2_K · 52.4 tok/s · Runs well (tight fit)
30Basic quality (D) · 30 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. Apple M4 Max (64 GB) Q3_K_M · 10.8 tok/s · Runs well
25Basic quality (D) · 25 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. Apple M1 Max (64 GB) Q2_K · 10.1 tok/s · Runs well
34Basic quality (D) · 34 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 57/100 (quality, fit, speed and memory use) sets the ranking. Apple M2 Ultra (192 GB) Q4_K_M · 12.3 tok/s · Runs well
30Basic quality (D) · 30 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. Apple M4 Max (128 GB) Q3_K_M · 10.8 tok/s · Runs well

Quick answers

How much memory does Llama 2 70B Chat need on the NVIDIA GH200 Grace Hopper (96GB GPU)?

At Q6_K it needs about 63.3 GB. The NVIDIA GH200 Grace Hopper (96GB GPU) has 96.0 GB usable.

How fast is Llama 2 70B Chat on the NVIDIA GH200 Grace Hopper (96GB GPU)?

About 49.4 tok/s at Q6_K, an estimate from the hardware’s memory bandwidth and the size of the model.

Which quantization of Llama 2 70B Chat should I use on the NVIDIA GH200 Grace Hopper (96GB GPU)?

Q6_K. Of the versions that fit, it is the one that runs best; when two run equally well, the check prefers the higher-rated one.

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.