Models › Llama-3.2-3B-Instruct

Llama-3.2-3B-Instruct: VRAM needs and speed by GPU

At Q4_K_M, Llama-3.2-3B-Instruct needs about 2.9 GB of VRAM. No measured runs yet. Speeds below are estimates. Submit a run.

3.21B parameters · license: llama3.2 · 131,072-token context · GGUF published Sep 25, 2024 · GGUF files on Hugging Face

Memory needed per quantization

QuantFile sizeNeeds (estimate)Smallest card it fits
UD-IQ1_S0.9 GB1.8 GB8 GB cards
UD-IQ1_M1.0 GB1.8 GB8 GB cards
UD-IQ2_XXS1.0 GB1.9 GB8 GB cards
UD-IQ2_M1.3 GB2.1 GB8 GB cards
Q2_K1.4 GB2.2 GB8 GB cards
Q2_K_L1.4 GB2.2 GB8 GB cards
UD-IQ3_XXS1.4 GB2.3 GB8 GB cards
UD-Q2_K_XL1.4 GB2.3 GB8 GB cards
Q3_K_S1.5 GB2.4 GB8 GB cards
Q3_K_M1.7 GB2.6 GB8 GB cards
UD-Q3_K_XL1.7 GB2.6 GB8 GB cards
IQ4_XS1.8 GB2.7 GB8 GB cards
IQ4_NL1.9 GB2.8 GB8 GB cards
Q4_01.9 GB2.8 GB8 GB cards
Q4_K_S1.9 GB2.8 GB8 GB cards
Q4_K_M2.0 GB2.9 GB8 GB cards
UD-Q4_K_XL2.1 GB2.9 GB8 GB cards
Q4_12.1 GB3.0 GB8 GB cards
Q5_K_S2.3 GB3.1 GB8 GB cards
Q5_K_M2.3 GB3.2 GB8 GB cards
UD-Q5_K_XL2.3 GB3.2 GB8 GB cards
Q6_K2.6 GB3.5 GB8 GB cards
UD-Q6_K_XL3.0 GB3.8 GB8 GB cards
Q8_03.4 GB4.3 GB8 GB cards
UD-Q8_K_XL4.2 GB5.1 GB8 GB cards
F166.4 GB7.3 GB8 GB cards
BF166.4 GB7.3 GB8 GB cards

Expected speed on popular hardware

HardwareBest quant that fitsExpected tok/s
NVIDIA RTX 3060Q8_090.0 estimate
NVIDIA RTX 4060 TiQ8_071.5 estimate
NVIDIA RTX 3090Q8_0192 estimate
NVIDIA RTX 4090Q8_0212 estimate
NVIDIA RTX 5090Q8_0336 estimate
AMD Radeon RX 7900 XTXQ8_0213 estimate
Apple M4 Pro (20-core GPU) (64 GB)Q8_067.7 estimate
Apple M3 Max (40-core GPU) (64 GB)Q8_098.8 estimate
AMD Ryzen AI Max+ 395 (Radeon 8060S)Q8_060.8 estimate

Check your own card

Too big for your card?

Run Llama-3.2-3B-Instruct on a rented 24-80 GB GPU by the hour.

Rent a bigger GPU on RunPod

Affiliate link: if you sign up, RigBench may earn a referral credit. It never affects rankings or numbers.