GPUs tracked45Price history8 dayssince Sep 3, 2026Price points (24h)6,955 pointsBiggest 24h dropBiggest 24h riseData updated

How to Deploy Llama 3.2 3B Locally

Llama 3.2 3B is a dense model with 3.2B parameters. At Q4 it needs at least 4GB of VRAM, and 8GB is the comfortable amount for an 8K context. The cheapest card that runs it today is the Tesla V100 16GB at about $659, at roughly 227.3 tokens/s.

Llama 3.2 3B details

Launch dateSep 18, 2024
Revision
ArchitectureDense
Ollama tagllama3.2:3b
Min VRAM at Q44 GB · Q4_K_M · 8K
Parameters3.2B
Layers28
Hidden size3,072
KV heads8
Head dim128
Max context128K
VendorMeta AI

How much VRAM does Llama 3.2 3B need for local deployment?

QuantizationWeightsGBTotal at 8KGBTotal at 32KGBTotal at 128KGB
1-bit0.8est.2.65.315.8
2-bit1.2est.3.15.716.2
3-bit1.9GGUF3.4616.5
4-bit2GGUF3.76.316.8
5-bit2.3GGUF46.717.2
8-bit3.4GGUF5.27.818.3
FP166.4GGUF8.210.821.3

Total = weights + KV cache (FP16) + 1GB runtime overhead. Weights marked GGUF are measured file sizes from the repository; the rest are estimated from bytes per parameter. Contexts beyond this model's 128K limit show a dash. Each tier takes whichever file of that bit width the repository has, preferring unsloth dynamic quants, so the exact file name varies by model.

Which GPUs can run Llama 3.2 3B?

AmazonJP
DeviceVRAMGBPriceEst. t/sSuggested context
Tesla V100 16GB16$6594 listings227.332K
Radeon RX 9070 XT16$8848 listings126.832K
GeForce RTX 5060 Ti 16GB16$9249 listings124.532K
GeForce RTX 507012$95610 listings177.932K
Radeon RX 7800 XT16$96610 listings12432K
Tesla V100 32GB32$1,1413 listings227.3128K
Arc Pro B6024$1,1643 listings84.4128K
Radeon RX 7900 XT20$1,2449 listings154.6128K
Radeon RX 6950 XT16$1,4044 listings115.332K
GeForce RTX 5070 Ti16$1,57610 listings226.532K
Radeon RX 7900 XTX24$1,7389 listings181.1128K
GeForce RTX 508016$1,8684 listings239.632K
Arc Pro B7032$2,0743 listings110.1128K
GeForce RTX 408016$2,1878 listings18832K
Radeon AI PRO R970032$2,2446 listings126.8128K
GeForce RTX 309024$2,5209 listings234.8128K
Ryzen AI Max+ 395 128GB128$3,9136 listings54128K
RTX PRO 4000 Blackwell24$3,9565 listings177.9128K
GeForce RTX 409024$4,9059 listings249.2128K
GeForce RTX 509032$7,4899 listings383.5128K
DGX Spark 128GB128$8,2233 listings79128K
GeForce RTX 2080 Ti 22GB22No listings165.1128K
GeForce RTX 5090 D V224No listings311.5128K
Arc Pro B6532No listings110.1128K
GeForce RTX 4080 SUPER 32GB32No listings192.4128K
GeForce RTX 5090 D32No listings383.5128K
Instinct MI10032No listings222.7128K
Instinct MI50 32GB32No listings191.3128K
Radeon PRO W780032No listings115.3128K
RTX PRO 4500 Blackwell32No listings226.5128K
A100 40GB PCIe40No listings346.9128K
CMP 170HX 40GB40No listings347.7128K
GeForce RTX 4090 48GB48No listings249.2128K
Radeon PRO W790048No listings165.3128K
RTX PRO 5000 Blackwell 48GB48No listings311.5128K
CMP 170HX 64GB64No listings336.8128K
Instinct MI21064No listings280.3128K
RTX PRO 5000 Blackwell 72GB72No listings311.5128K
RTX PRO 6000D84No listings349128K
RTX PRO 6000 Blackwell96No listings383.5128K
Mac Studio M4 Max 128GB128No listings119.8128K
Mac Studio M5 Max 128GB128No listings133.1128K
MacBook Pro M4 Max 128GB128No listings119.8128K
MacBook Pro M5 Max 128GB128No listings133.1128K
Mac Studio M3 Ultra 512GB512No listings171.4128K

Fits when weights + KV cache + runtime overhead is at or below usable memory, single card. "Offload" on MoE models means it runs with expert weights in system RAM (only attention layers and KV cache stay in VRAM); that speed assumes 70 GB/s RAM bandwidth. The context column is the largest tier that fits.

Deploy Llama 3.2 3B locally with Ollama, llama.cpp or vLLM

Ollama

Ollama pulls and runs it in one command; the quantization comes from the official tag.

ollama run llama3.2:3b
llama.cpp

llama.cpp pulls the GGUF straight from Hugging Face; -c sets the context length.

llama-server -hf bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M -c 8192
vLLM

vLLM serves the original-precision weights, which needs far more memory than GGUF and usually more than one GPU.

vllm serve meta-llama/Llama-3.2-3B-Instruct

File names follow that week's model-library snapshot; the definitions are in the methodology. Methodology

FAQ

How much VRAM does Llama 3.2 3B need?

At Q4_K_M with an 8K context Llama 3.2 3B needs about 4GB of VRAM; 8GB leaves comfortable headroom.

What is the cheapest GPU that runs Llama 3.2 3B?

The Tesla V100 16GB: 16GB of VRAM, currently about $659, at roughly 227.3 tokens/s.

Can you run Llama 3.2 3B with Ollama?

Yes: ollama run llama3.2:3b.