Hardware guide · October 2026
What GPU do you need to run Llama 3.3 70B Instruct?
Llama 3.3 70B Instruct is a 70.6B dense model. Here is exactly how much memory it needs and which GPUs and Macs run it well.
The short answer
To run Llama 3.3 70B Instruct at Q4_K_M with 8K tokens of context you need about 47GB of VRAM (42.5GB of weights plus KV cache and overhead). The cheapest way in is a 48GB card like the RTX A6000 (~$4,400 used). For the best speed per dollar, get the RTX 6000 Ada (~22 tokens/sec). On a Mac or other unified-memory system, the Ryzen AI Max+ 395 runs it at ~2.6 tokens/sec.
- VRAM at Q4_K_M
- ~47GB
- GPUs that run it
- 12
- Parameters
- 70.6B
- CPU only (est.)
- ~1.1 tok/s
Our recommendations
The best GPUs for Llama 3.3 70B Instruct
Minimum
RTX A600048GB · ~$4,400 used
- Quant
- Q3_K_M
- Speed (est.)
- ~16 tok/s
The smallest card that runs it at Q3_K_M with 8K context.
Best value
RTX 6000 Ada48GB · ~$4,750 used
- Quant
- Q3_K_M
- Speed (est.)
- ~22 tok/s
The most tokens per second per dollar among cards with room to spare.
Multi-GPU
3× RTX 309072GB total · ~$3,150 used
- Quant
- Q4_K_M
- Speed (est.)
- ~15 tok/s
Split across cards with llama.cpp, vLLM, or ExLlama. Needs a board and PSU that can feed every card.
Faster multi-GPU
3× RTX 409072GB total · ~$7,500 used
- Quant
- Q4_K_M
- Speed (est.)
- ~18 tok/s
Split across cards with llama.cpp, vLLM, or ExLlama. Needs a board and PSU that can feed every card.
Mac / unified memory
Ryzen AI Max+ 395128GB unified · ~$3,300
- Quant
- Q8_0
- Speed (est.)
- ~2.6 tok/s
Lots of memory per dollar and silent, but slower than a GPU with the same capacity.
We may earn a commission when you buy through links on this page. It never changes what we recommend. How we make money
Memory
Llama 3.3 70B Instruct VRAM requirements
Memory needed at each quantization level. Q4_K_M is the best balance of size and quality for most people; move up to Q5 or Q6 if you have the room.
| Quantization | Quality | Weights | Total @ 8K ctx | Total @ 32K ctx |
|---|---|---|---|---|
| Q2_K | Rough | 26 GB | 30 GB | 38 GB |
| Q3_K_M | Usable | 34 GB | 39 GB | 47 GB |
| Q4_K_M | RecommendedDefault | 43 GB | 47 GB | 55 GB |
| Q5_K_M | Very good | 50 GB | 55 GB | 63 GB |
| Q6_K | Near-lossless | 58 GB | 63 GB | 71 GB |
| Q8_0 | Practically lossless | 75 GB | 81 GB | 89 GB |
| BF16 | Full precision | 141 GB | 149 GB | 157 GB |
Context
How context length changes the memory you need
Every token in the context window is stored in the KV cache. At Q4_K_M, with a 16-bit KV cache, Llama 3.3 70B Instruct needs:
| Context | KV cache | Total memory | With 8-bit KV |
|---|---|---|---|
| 4K tokens | 1.3 GB | 45.7 GB | 45.1 GB |
| 8K tokens | 2.7 GB | 47.1 GB | 45.7 GB |
| 16K tokens | 5.4 GB | 49.8 GB | 47.1 GB |
| 32K tokens | 10.7 GB | 55.1 GB | 49.8 GB |
| 64K tokens | 21.5 GB | 65.9 GB | 55.1 GB |
| 128K tokens | 42.9 GB | 87.3 GB | 65.9 GB |
2 x 80 layers x 8 KV heads x 128 dims x 2 bytes = 327,680 bytes (320 KiB) per token: about 2.7GB at 8K context and 43GB at the full 128K.
Compatibility
Every GPU that runs Llama 3.3 70B Instruct
Each GPU at the highest-quality quantization that fits with 8K context, fastest first. Speeds are single-user estimates.
| GPU | Memory | Quant | Speed | Fit | Price |
|---|---|---|---|---|---|
| RTX 6000 Ada | 48GB | Q3_K_M | ~22 tok/s | Runs well | ~$4,750 used |
| RTX PRO 6000 | 96GB | Q8_0 | ~19 tok/s | Runs well | ~$16,000 |
| RTX PRO 5000 72GB | 72GB | Q6_K | ~18 tok/s | Runs well | ~$9,100 |
| RTX A6000 | 48GB | Q3_K_M | ~16 tok/s | Runs well | ~$4,400 used |
| M5 Ultra 512GB | 512GB unified | Q8_0 | ~8.9 tok/s | Runs well | — |
| M5 Ultra 256GB | 256GB unified | Q8_0 | ~8.9 tok/s | Runs well | ~$9,499 |
| M3 Ultra 512GB | 512GB unified | Q8_0 | ~6.0 tok/s | Runs well | $9,499 |
| M5 Max 128GB | 128GB unified | Q8_0 | ~5.3 tok/s | Runs well | ~$5,100 |
| M4 Max 128GB | 128GB unified | Q8_0 | ~4.7 tok/s | Runs well | $3,499 |
| Ryzen AI Max+ PRO 495 | 192GB unified | Q8_0 | ~2.7 tok/s | Runs well | ~$6,800 |
| DGX Spark | 128GB unified | Q8_0 | ~2.7 tok/s | Runs well | ~$6,950 |
| Ryzen AI Max+ 395 | 128GB unified | Q8_0 | ~2.6 tok/s | Runs well | ~$3,300 |
With CPU offload (partial offload, assuming 64GB of system RAM): RTX 5090 (~2.7 tok/s), V100 32GB (~2.5 tok/s), MI50 32GB (~2.5 tok/s), R9700 (~2.4 tok/s), Arc Pro B70 (~2.3 tok/s), RTX 4090 (~1.8 tok/s), RTX 3090 (~1.8 tok/s), RX 7900 XTX (~1.8 tok/s) and 16 more.
Tips
Getting the most out of Llama 3.3 70B Instruct
- Start at Q4_K_M. It keeps nearly all of the quality at about a third of the 16-bit size. Step up to Q5_K_M or Q6_K only if it still fits with the context you need.
- Budget memory for context. The KV cache grows with every token. Quantizing it to 8-bit roughly halves that cost with little quality loss.
- Avoid heavy CPU offload. For a dense model, every token reads every weight. Once a large share of the model sits in system RAM, speed falls off a cliff. A second GPU usually beats offloading.
- Pick the right runtime. Ollama and LM Studio are easiest. llama.cpp gives the most control. vLLM or SGLang are best when several people or agents share one machine.
Want to try different quantizations, context lengths, or GPU counts? Use our VRAM calculator, or read the Llama 3.3 70B Instruct overview for setup commands and benchmarks.
Frequently asked questions
How much VRAM do I need to run Llama 3.3 70B Instruct?
About 47GB at Q4_K_M with an 8K context window. At Q8_0 (near-lossless) it needs about 81GB, and at Q3_K_M about 39GB with some quality loss. Longer contexts add 0.34GB per 1K tokens with a 16-bit KV cache.
Can I run Llama 3.3 70B Instruct on a 24GB GPU?
Only with partial CPU offload. On a 24GB card like the RTX 3090 or RTX 4090, about 55% of the weights would sit in system RAM and generation drops to about 1.8 tokens/sec with CPU offload.
Can I run Llama 3.3 70B Instruct on a Mac?
Yes. With 128GB of unified memory or more (for example the Ryzen AI Max+ 395), macOS can give the GPU enough memory to run it at Q8_0, at about ~2.6 tokens/sec. MLX builds are usually fastest on Apple silicon.
Can I run Llama 3.3 70B Instruct on CPU only, without a GPU?
Yes, if you have at least 64GB of system RAM. On a desktop with dual-channel DDR5, expect roughly 1.1 tokens/sec. A GPU is strongly recommended for interactive use.