LLM GPU requirements

Llama 3.3 70B GPU requirements: VRAM, GPUs and servers

How much GPU memory Llama 3.3 70B needs in BF16, FP8 and 4-bit, how its KV cache grows with context, what fine-tuning takes, and which servers run it, with monthly prices.

Parameters: 70.6B
Architecture: Dense 80 layers
KV heads: 8 × 128 dim
Context: 128k tokens

Open in the VRAM calculator

Cheapest servers that run it

FP8 weights · 8k context · 8 concurrent requests · about 100 GB

NVIDIA H200 In stock

141 GB HBM3e · Single GPU

$3,090/mo
NVIDIA B200 In stock

180 GB HBM3e · Single GPU

$4,590/mo
NVIDIA B300 Limited

288 GB HBM3e · Single GPU

$5,190/mo

Delivered in under 10 minutes after payment.

Inference memory

How much VRAM Llama 3.3 70B needs

Weights take parameters × bytes per parameter. Serving adds the KV cache and about 10% runtime overhead; the totals below assume an 8k context with 8 concurrent requests and a BF16 KV cache.

Llama 3.3 70B memory by precision and the cheapest server that fits
Precision	Weights	Total to serve	Cheapest server that fits
BF16 2 bytes per parameter	141 GB	178 GB	NVIDIA B300 $5,190/mo
FP8 1 byte per parameter	71 GB	100 GB	NVIDIA H200 $3,090/mo
4-bit 0.5 bytes per parameter	35 GB	61 GB	NVIDIA RTX PRO 6000 Blackwell $1,390/mo

GGUF quantizations for llama.cpp and Ollama

Llama 3.3 70B GGUF file sizes and the smallest GPU for one 8k request
Quantization	Weights	Smallest GPU, one 8k request
Q8_0 about 8.5 bits per weight	75 GB	NVIDIA RTX PRO 6000 Blackwell 96 GB · $1,390/mo
Q6_K about 6.56 bits per weight	58 GB	NVIDIA H100 SXM 80 GB · $2,190/mo
Q5_K_M about 5.69 bits per weight	50 GB	NVIDIA H100 SXM 80 GB · $2,190/mo
Q4_K_M about 4.85 bits per weight	43 GB	NVIDIA H100 SXM 80 GB · $2,190/mo

Estimates with about 8% of GPU memory kept free. Quantized weights also keep some layers in higher precision, and frameworks differ in overhead; check the free memory your engine reports at startup. A model split across GPUs with tensor parallelism runs best on NVLink servers (HGX); the 8× RTX PRO 6000 server links its GPUs over PCIe.

GPUs and servers

Which GPUs can run Llama 3.3 70B?

Yes. In FP8, a single NVIDIA H200 (141 GB, $3,090 a month at OverGPU) runs Llama 3.3 70B with an 8k context and 8 concurrent requests; one NVIDIA RTX PRO 6000 Blackwell (96 GB) holds a single 8k request. The table checks every configuration we rent in each precision.

Whether each GPU configuration runs Llama 3.3 70B in BF16, FP8 and 4-bit
Configuration	BF16	FP8	4-bit
NVIDIA GeForce RTX 5090 32 GB · $449/mo	—	—	—
NVIDIA L40S 48 GB · $890/mo	—	—	1 req
NVIDIA H100 SXM 80 GB · $2,190/mo	—	—	Yes
NVIDIA RTX PRO 6000 Blackwell 96 GB · $1,390/mo	—	1 req	Yes
NVIDIA H200 141 GB · $3,090/mo	—	Yes	Yes
NVIDIA B200 180 GB · $4,590/mo	1 req	Yes	Yes
NVIDIA B300 288 GB · $5,190/mo	Yes	Yes	Yes
8× NVIDIA HGX H100 8 × 80 GB · $16,900/mo	Yes	Yes	Yes
8× NVIDIA RTX PRO 6000 Blackwell 8 × 96 GB · $10,490/mo	Yes	Yes	Yes
8× NVIDIA HGX H200 8 × 141 GB · $23,900/mo	Yes	Yes	Yes
8× NVIDIA HGX B200 8 × 180 GB · $34,900/mo	Yes	Yes	Yes
8× NVIDIA HGX B300 8 × 288 GB · $39,900/mo	Yes	Yes	Yes
8× AMD Instinct MI355X 8 × 288 GB · $19,900/mo	Yes	Yes	Yes

Yes: 8 concurrent requests with an 8k context each. 1 req: a single 8k request at a time. —: the weights, KV cache and runtime do not fit. Same formulas as the tables above, with about 8% of GPU memory kept free.

KV cache

Llama 3.3 70B KV cache by context length

Each token in flight stores keys and values for every layer: 2 × 80 layers × 8 KV heads × 128 dimensions × 2 bytes = 320 KB per token in BF16. An FP8 KV cache halves these figures.

KV cache size by context length and number of concurrent requests
Context per request	1 request	8 requests	32 requests
4k tokens	1.3 GB	11 GB	43 GB
32k tokens	11 GB	86 GB	344 GB
128k tokens	43 GB	344 GB	1,374 GB

Fine-tuning

Memory to fine-tune Llama 3.3 70B

Rough totals before tuning batch size and sequence length: QLoRA keeps 4-bit base weights, LoRA keeps BF16 weights frozen, and a full fine-tune with AdamW needs about 16 bytes per parameter plus activations, sharded across GPUs.

Fine-tuning memory by method and the cheapest server that fits
Method	Approximate memory	Cheapest server that fits
QLoRA (4-bit base)	53 GB	NVIDIA RTX PRO 6000 Blackwell $1,390/mo
LoRA (BF16 base)	175 GB	NVIDIA B300 $5,190/mo
Full fine-tune (AdamW, mixed precision)	1,356 GB	8× AMD Instinct MI355X $19,900/mo

Full fine-tuning assumes sharded training (FSDP or ZeRO-3) with activation checkpointing. See the fine-tuning guide for methods and settings.

About Llama 3.3 70B

Llama 3.3 70B is Meta’s 70B dense instruction model: 80 layers, grouped-query attention with 8 KV heads and a 128k-token context window. It is the reference point for 70B-class serving: in BF16 the weights alone take about 141 GB, more than a single H200 can hold once the KV cache is added, while FP8 halves that and fits one H200 with room to spare.

Every figure on this page comes from the model’s published architecture and the same formulas as our VRAM calculator, where you can change the context length, concurrency and KV cache precision.

Model facts

Developer: Meta
License: Llama 3.3 Community License
Type: Dense
Parameters: 70.6B
Layers: 80
Attention: 64 query heads, 8 KV headsgrouped-query attention, head dimension 128
Context window: 128k tokens

FAQ

Llama 3.3 70B GPU FAQ

The short version. The help center has the details, and customers with an active server get answers by ticket, 24/7.

Visit the help center

How much VRAM does Llama 3.3 70B need?

About 141 GB for the weights in BF16, 71 GB in FP8 and 35 GB in 4-bit, plus the KV cache: about 328 MB per 1,000 tokens in flight with a BF16 cache. Serving it in FP8 with an 8k context and 8 concurrent requests takes about 100 GB in total.

Can Llama 3.3 70B run on a single GPU?

Yes. In FP8, a single NVIDIA H200 (141 GB) serves it with an 8k context and 8 concurrent requests. Longer contexts or more users need more memory for the KV cache, so check your case in the VRAM calculator.

What is the cheapest OverGPU server for Llama 3.3 70B?

For FP8 serving with an 8k context and 8 concurrent requests, the NVIDIA H200 at $3,090 per month is the lowest-priced configuration that fits. For other contexts and concurrency, size it in the VRAM calculator.

How much memory do I need to fine-tune Llama 3.3 70B?

Roughly 53 GB with QLoRA, 175 GB with LoRA on BF16 weights, and 1,356 GB for a full fine-tune with AdamW in mixed precision, before tuning batch size and sequence length. Full fine-tunes of large models are sharded across GPUs with FSDP or ZeRO-3.

Deploy a GPU server

Run Llama 3.3 70B on your own GPUs.

Choose a configuration, create your account in the same form and pay the first month from your balance. Your server is delivered in under 10 minutes.

Delivered in under 10 minutes
Cancel any month
Pay with crypto, up to +50% bonus

Configure H200 See all prices