Llama 3.2 1B Instruct: how much VRAM does it need?

Llama 3.2 1B Instruct has 1.24B parameters in 16 layers. It uses grouped-query attention: 32 query heads share 8 key-value heads, so its KV cache is 4× smaller than full multi-head attention would need. It takes up to 131,072 tokens of context; one user at that length needs 4.00 GiB of KV cache at FP16 on top of the weights. Its repository is gated on Hugging Face, so LLMScale ships its architecture in a bundled database.

BF16 · 8k context · one user

3.63 GiB

2.30 GiB of weights, 256 MiB of KV cache and 1.07 GiB of vLLM's overhead. At INT4 the same workload needs 2.29 GiB.

16 layers
hidden 2048
32/8 heads
head_dim 64
128k max context

Llama 3.2 1B Instruct memory by precision and context

Precision	8k context	32k context	128k context
BF16	3.63 GiBKV cache 256 MiB	4.38 GiBKV cache 1.00 GiB	7.38 GiBKV cache 4.00 GiB
FP8	2.72 GiBKV cache 256 MiB	3.47 GiBKV cache 1.00 GiB	6.47 GiBKV cache 4.00 GiB
INT4	2.29 GiBKV cache 256 MiB	3.04 GiBKV cache 1.00 GiB	6.04 GiBKV cache 4.00 GiB

One user, FP16 KV cache and vLLM's overhead on one GPU at its defaults for a card under 70 GB. Quantizing shrinks the weights, not the KV cache, so the saving narrows as the context grows.

Which single GPUs hold it

BF16 (3.63 GiB): RTX 3060, NVIDIA T4, RTX 4060 Ti 16GB, RTX 4070 Ti Super
FP8 (2.72 GiB): RTX 3060, NVIDIA T4, RTX 4060 Ti 16GB, RTX 4070 Ti Super
INT4 (2.29 GiB): RTX 3060, NVIDIA T4, RTX 4060 Ti 16GB, RTX 4070 Ti Super

At 8k context for one user, the smallest cards first. A discrete card is budgeted at vLLM's default 92% of its memory, a unified-memory device at the share its GPU can address, and each carries vLLM's overhead at its own defaults.

How many users fit

GPU	BF16	FP8	INT4
RTX 4090	73	77	78
RTX 5090	101	105	107
RTX 6000 Ada	161	164	166
H100 SXM 80GB	276	279	281
RTX PRO 6000 Blackwell	334	338	339
AMD Ryzen AI Max+ 395 (128GB)	338	341	343
H200 141GB	499	502	504
B200 180GB	638	641	643

Users at 8k context each, on one card. ✗: the weights leave no room for even one.

Fine-tuning Llama 3.2 1B Instruct

Method	Memory	Smallest GPU
Full fine-tune	19.54 GiB	NVIDIA L4
LoRA (rank 16, attention)	3.47 GiB	RTX 3060
QLoRA (rank 16, attention)	1.76 GiB	RTX 3060

Sequence length 2,048, batch size 1, gradient checkpointing on, Adam. Full fine-tuning keeps 16 bytes per parameter before activations; QLoRA stores the frozen base in 4-bit NF4.

Where the numbers come from

Weights are parameters × bytes per parameter. The KV cache is 2 × layers × key-value heads × head dimension × bytes × context × users. For Llama 3.2 1B Instruct that is 32.0 KiB per token at FP16. These are capacity-planning figures: the serving engine — vLLM, TGI, llama.cpp — and its paged-attention efficiency still move the real number.

More on KV cache size and on GQA, MQA and MLA.