Llama 2 70B HF — Hardware Requirements & GPU Compatibility
Chat
Llama 2 70B HF is a 69.0B-parameter open language model from Meta in the Llama 2 family. At BF16 it needs about 151.75 GB of VRAM — see which GPUs and Macs can run it below.
Llama 2 70B HF runs on a 192 GB+ Mac or a datacenter GPU
Best local pick → Mac Pro M2 Ultra (192 GB) · BF16 · 156.4 GB · ~4 tok/s
Macs from 192 GB (BF16 · ~4 tok/s)
Llama 2 Community
Not set up for it?
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Llama 2 70B HF
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
vllm serve meta-llama/Llama-2-70b-hf --max-model-len 16384 --tensor-parallel-size 8vLLM repo with its own quantisation
Download: Weights on Hugging Face
Other ways
each opens step-by-stepRun in cloud
Fits on B200 (192 GB) (35 GB headroom) · BF16
- Generation speed
- ~34 tok/s
- generation speed
- Cost per 1M output tokens
- $48.43
- per 1M output tokens
How Much VRAM Does Llama 2 70B HF Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | Runs on | |
|---|---|---|---|---|
| 151.8 GB | 161.8 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Runs on B200 192 GB |
Selected BF16
Brain floating point 16156.4GBat 16K context
- Weights
- 150.8 GB
- KV cache
- 5.4 GB
- Overhead
- 0.3 GB
Server-class: ~156.4 GB at 16K context · one B200 192 GB
Which Hardware Can Run Llama 2 70B HF?
BF16 · 151.8 GBLlama 2 70B HF (BF16) requires 151.8 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 198+ GB is recommended. No consumer GPU has enough memory. 4 devices with unified memory can run Llama 2 70B HF, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Pro M2 Ultra (192 GB).
192+ GB
fits · 35.6 GB free · up to 32K ctx
6 GPUs · 1 Mac class · 2 device families
B200~34Simulate output on B200B300~34Simulate output on B300Instinct MI350X~32Simulate output on Instinct MI350XM3 Ultra256–512 GB~4Simulate output on Mac Studio (M3 Ultra, 256GB)8x H100 SXM5~115Simulate output on NVIDIA DGX H1008x A100 SXM4~70Simulate output on NVIDIA DGX A100 640GB
By VRAM tier · typical GPUs and Macs at BF16
Consumer GPUs (up to 32 GB) and Macs up to 128 GB: too small for this model (needs ≥ 156.4 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →
How good is Llama 2 70B HF?
llmrun Score blends public benchmark boards into one number; 100 = the average open model.
Not enough data yet for an llmrun Score.
A Score needs results on enough independent boards. How the Composite Score is built
Benchmarks
The 6 boards that weigh most in the llmrun Score, of 11 with a result for this model; badges show who ran the test.
- MMLU-Pro
TIGER-LabIndependent
Independent37.5%
#109/145 open
- MMLU
Epoch AI · Jul 2023Lab-reported
Lab-reported69.9%
#28/85 open
- BIG-Bench Hard
Epoch AI · Jul 2023Lab-reported
Lab-reported64.9%
#10/38 open
- GSM8K
Epoch AI · Jul 2023Lab-reported
Lab-reported69.6%
#18/62 open
- HellaSwag
Epoch AI · Jul 2023Lab-reported
Lab-reported85.3%
#6/47 open
- ARC Challenge
Epoch AI · Jul 2023Lab-reported
Lab-reported78.3%
#15/56 open
Alternatives to Llama 2 70B HF
Llama 2 family · 10 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Qwen2.5 VL 72B InstructAlibaba | — | 49.7 GB | —needs > 24 GB | same size |
| Llama 3.1 Tulu 3 70BAllen AI | — | 48 GB | —needs > 24 GB | same size |
| Qwen2 VL 72B InstructAlibaba | — | 49.7 GB | —needs > 24 GB | same size |
| Qwen1.5 72B ChatAlibaba | — | 86.6 GB | —needs > 24 GB | same size |
| StableBeluga2Stability AI | — | 23.5 GB | ~29 | same size |
| Apertus 70B 2509swiss-ai | — | 48 GB | —needs > 24 GB | same size |
Specifications
- Publisher
- Meta
- Family
- Llama 2
- Parameters
- 69.0B
- Release Date
- 2023-07-11
- License
- Llama 2 Community
Frequently Asked Questions
How much VRAM does Llama 2 70B HF need?
Llama 2 70B HF requires 151.8 GB of VRAM at BF16.
VRAM = Weights + KV Cache + Overhead
Weights = 69.0B × 16 bits ÷ 8 = 138 GB
KV Cache + Overhead ≈ 13.8 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
VRAM usage by quantization
Can NVIDIA GeForce RTX 5090 run Llama 2 70B HF?
No — Llama 2 70B HF requires at least 151.8 GB at BF16, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.
Can I run Llama 2 70B HF on a Mac?
Yes — Mac Pro M2 Ultra (192 GB) and 3 other Macs can run Llama 2 70B HF. Apple Silicon uses unified memory, so the model shares RAM with the system. At BF16 you need at least 151.8 GB of usable unified memory (RAM minus macOS overhead).
Can I run Llama 2 70B HF locally?
Yes — Llama 2 70B HF can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Llama 2 70B HF?
At BF16, Llama 2 70B HF can reach ~32 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: NVIDIA B200 → 8000 ÷ 151.8 × 0.65 = ~34 tok/s
Estimated speed at BF16 (151.8 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Llama 2 70B HF?
At BF16, the download is about 137.95 GB.
Which GPUs can run Llama 2 70B HF?
No single consumer GPU has enough VRAM to run Llama 2 70B HF at BF16 (151.8 GB). Multi-GPU or professional hardware is required.
Which devices can run Llama 2 70B HF?
4 devices with unified memory can run Llama 2 70B HF at BF16 (151.8 GB), including Mac Pro M2 Ultra (192 GB), Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio M2 Ultra (192 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.