Can NVIDIA GeForce RTX 4090 run Llama 3.3 70B Instruct Awq?

Yes, at IQ2_S (23.0 GB) or lower. Higher quantizations like IQ2_M (24.8 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.

What's the best quantization for Llama 3.3 70B Instruct Awq?

For Llama 3.3 70B Instruct Awq, Q4_K_M (43.3 GB) offers the best balance of quality and VRAM usage. Q4_K_L (44.2 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 20.4 GB.

Can I run Llama 3.3 70B Instruct Awq on a Mac?

Llama 3.3 70B Instruct Awq requires at least 20.4 GB at IQ2_XXS, which exceeds the unified memory of most consumer Macs. You would need a Mac Studio or Mac Pro with a high-memory configuration.

Can I run Llama 3.3 70B Instruct Awq locally?

Yes — Llama 3.3 70B Instruct Awq can run locally on consumer hardware. At Q4_K_M quantization it needs 43.3 GB of VRAM. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Llama 3.3 70B Instruct Awq?

At Q4_K_M, Llama 3.3 70B Instruct Awq can reach ~67 tok/s on AMD Instinct MI300X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

What's the download size of Llama 3.3 70B Instruct Awq?

At Q4_K_M, the download is about 42.33 GB. The full-precision Q8_0 version is 70.55 GB. The smallest option (IQ2_XXS) is 19.40 GB.

Casper Hansen·Llama 3·LlamaForCausalLM

Llama 3.3 70B Instruct Awq — Hardware Requirements & GPU Compatibility

Chat

This is an AWQ-quantized version of Meta's Llama 3.3 70B Instruct, repackaged by Casper Hansen. Llama 3.3 70B Instruct is one of the most capable open-weight models available, delivering performance competitive with much larger models across reasoning, coding, math, and multilingual tasks. Casper Hansen's AWQ (Activation-aware Weight Quantization) conversion reduces memory requirements while preserving model quality, making this 70.6-billion-parameter model more accessible for local deployment. AWQ quantization is designed for GPU inference and works with frameworks like vLLM and AutoAWQ. Running this model still requires substantial VRAM, but the quantization brings it within reach of high-end consumer or professional multi-GPU setups.

849.3K downloads 41 likesDec 2024131K context

Based on Llama 3.1 70B

Specifications

Publisher: Casper Hansen
Family: Llama 3
Parameters: 70.6B
Architecture: LlamaForCausalLM
Context Length: 131,072 tokens
Vocabulary Size: 128,256
Release Date: 2024-12-06
License: llama3.3

Get Started

HuggingFace

casperhansen/llama-3.3-70b-instruct-awq

How Much VRAM Does Llama 3.3 70B Instruct Awq Need?

Select a quantization to see compatible GPUs below.

Quantization	Bits	VRAM	+ Context	File Size	Quality
IQ2_XXS	2.20	20.4 GB	62.6 GB	19.40 GB	Importance-weighted 2-bit, extreme compression — significant quality loss
IQ2_XS	2.40	22.1 GB	64.4 GB	21.17 GB	Importance-weighted 2-bit, extra small
IQ2_S	2.50	23.0 GB	65.3 GB	22.05 GB	Importance-weighted 2-bit, small
IQ2_M	2.70	24.8 GB	67.1 GB	23.81 GB	Importance-weighted 2-bit, medium
IQ3_XXS	3.10	28.3 GB	70.6 GB	27.34 GB	Importance-weighted 3-bit
IQ3_XS	3.30	30.1 GB	72.3 GB	29.10 GB	Importance-weighted 3-bit, extra small
Q2_K	3.40	31.0 GB	73.2 GB	29.99 GB	2-bit quantization with K-quant improvements
Q3_K_S	3.50	31.8 GB	74.1 GB	30.87 GB	3-bit small quantization
IQ3_M	3.60	32.7 GB	75 GB	31.75 GB	Importance-weighted 3-bit, medium
Q3_K_M	3.90	35.4 GB	77.6 GB	34.39 GB	3-bit medium quantization
Q4_0	4.00	36.3 GB	78.5 GB	35.28 GB	4-bit legacy quantization
Q3_K_L	4.10	37.1 GB	79.4 GB	36.16 GB	3-bit large quantization
IQ4_XS	4.30	38.9 GB	81.2 GB	37.92 GB	Importance-weighted 4-bit, compact
Q4_1	4.50	40.7 GB	82.9 GB	39.69 GB	4-bit legacy quantization with offset
Q4_K_S	4.50	40.7 GB	82.9 GB	39.69 GB	4-bit small quantization
IQ4_NL	4.50	40.7 GB	82.9 GB	39.69 GB	Importance-weighted 4-bit, non-linear
Q4_K_M	4.80	43.3 GB	85.6 GB	42.33 GB	4-bit medium quantization — most popular sweet spot
Q4_K_L	4.90	44.2 GB	86.5 GB	43.21 GB	4-bit large quantization
Q5_K_S	5.50	49.5 GB	91.8 GB	48.51 GB	5-bit small quantization
Q5_K_M	5.70	51.2 GB	93.5 GB	50.27 GB	5-bit medium quantization — good quality/size tradeoff
Q5_K_L	5.80	52.1 GB	94.4 GB	51.15 GB	5-bit large quantization
Q6_K	6.60	59.2 GB	101.5 GB	58.21 GB	6-bit quantization, very good quality
Q8_0	8.00	71.5 GB	113.8 GB	70.55 GB	8-bit quantization, near-lossless

Which GPUs Can Run Llama 3.3 70B Instruct Awq?

Q4_K_M · 43.3 GB

Show professional

Llama 3.3 70B Instruct Awq (Q4_K_M) requires 43.3 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 57+ GB is recommended. Using the full 131K context window can add up to 42.3 GB, bringing total usage to 85.6 GB. No single GPU has enough memory — multi-GPU or cluster setups are needed.

Which Devices Can Run Llama 3.3 70B Instruct Awq?

Q4_K_M · 43.3 GB

11 devices with unified memory can run Llama 3.3 70B Instruct Awq, including NVIDIA DGX H100, NVIDIA DGX A100 640GB, Mac Studio M4 Max (64 GB).

Runs great

— Plenty of headroom

NVIDIA DGX H100~402 tok/s NVIDIA DGX A100 640GB~245 tok/s Mac Pro M2 Ultra (192 GB)~12 tok/s Mac Studio M2 Ultra (192 GB)~12 tok/s Mac Studio M4 Max (128 GB)~8 tok/s

Decent

— Enough memory, may be tight

Mac Studio M4 Max (64 GB)~8 tok/s MacBook Pro 16" M4 Max (48 GB)~8 tok/s MacBook Pro 16" M4 Max (64 GB)~8 tok/s MacBook Pro 16" M3 Max (48 GB)~6 tok/s Mac Mini M4 Pro (48 GB)~4 tok/s NVIDIA Jetson AGX Orin 64GB~3 tok/s

Related Models

Frequently Asked Questions

How much VRAM does Llama 3.3 70B Instruct Awq need?: Llama 3.3 70B Instruct Awq requires 43.3 GB of VRAM at Q4_K_M, or 71.5 GB at Q8_0. Full 131K context adds up to 42.3 GB (85.6 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 70.6B × 4.8 bits ÷ 8 = 42.3 GB
KV Cache + Overhead ≈ 1 GB (at 2K context + ~0.3 GB framework)
KV Cache + Overhead ≈ 43.3 GB (at full 131K context)
VRAM usage by quantization
Q4_K_M
43.3 GB
Q4_K_M + full context
85.6 GB
Learn more about VRAM estimation →
Can NVIDIA GeForce RTX 4090 run Llama 3.3 70B Instruct Awq?: Yes, at IQ2_S (23.0 GB) or lower. Higher quantizations like IQ2_M (24.8 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
What's the best quantization for Llama 3.3 70B Instruct Awq?: For Llama 3.3 70B Instruct Awq, Q4_K_M (43.3 GB) offers the best balance of quality and VRAM usage. Q4_K_L (44.2 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 20.4 GB.
VRAM requirement by quantization
IQ2_XXS
20.4 GB~53%
Q2_K
31.0 GB~75%
Q3_K_L
37.1 GB~86%
Q4_K_M ★
43.3 GB~89%
Q4_K_L
44.2 GB~90%
Q8_0
71.5 GB~99%
★ Recommended — best balance of quality and VRAM usage.
Learn more about quantization →
Can I run Llama 3.3 70B Instruct Awq on a Mac?: Llama 3.3 70B Instruct Awq requires at least 20.4 GB at IQ2_XXS, which exceeds the unified memory of most consumer Macs. You would need a Mac Studio or Mac Pro with a high-memory configuration.
Can I run Llama 3.3 70B Instruct Awq locally?: Yes — Llama 3.3 70B Instruct Awq can run locally on consumer hardware. At Q4_K_M quantization it needs 43.3 GB of VRAM. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Llama 3.3 70B Instruct Awq?: At Q4_K_M, Llama 3.3 70B Instruct Awq can reach ~67 tok/s on AMD Instinct MI300X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: AMD Instinct MI300X → 5300 ÷ 43.3 × 0.55 = ~67 tok/s
Estimated speed at Q4_K_M (43.3 GB)
AMD Instinct MI300X
~67 tok/s
NVIDIA H100 SXM
~50 tok/s
AMD Instinct MI250X
~42 tok/s
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
Learn more about tok/s estimation →
What's the download size of Llama 3.3 70B Instruct Awq?: At Q4_K_M, the download is about 42.33 GB. The full-precision Q8_0 version is 70.55 GB. The smallest option (IQ2_XXS) is 19.40 GB.