Models › gemma-3-4b-it

gemma-3-4b-it: VRAM needs and speed by GPU

At Q4_K_M, gemma-3-4b-it needs about 3.4 GB of VRAM. No measured runs yet. Speeds below are estimates. Submit a run.

3.88B parameters · license: gemma · 131,072-token context · GGUF published Mar 12, 2025 · GGUF files on Hugging Face

Memory needed per quantization

QuantFile sizeNeeds (estimate)Smallest card it fits
UD-IQ1_S1.2 GB2.1 GB8 GB cards
UD-IQ1_M1.2 GB2.1 GB8 GB cards
UD-IQ2_XXS1.3 GB2.2 GB8 GB cards
UD-IQ2_M1.6 GB2.5 GB8 GB cards
UD-IQ3_XXS1.7 GB2.6 GB8 GB cards
Q2_K1.7 GB2.6 GB8 GB cards
Q2_K_L1.7 GB2.6 GB8 GB cards
UD-Q2_K_XL1.8 GB2.7 GB8 GB cards
Q3_K_S1.9 GB2.8 GB8 GB cards
Q3_K_M2.1 GB3.0 GB8 GB cards
UD-Q3_K_XL2.2 GB3.1 GB8 GB cards
IQ4_XS2.3 GB3.2 GB8 GB cards
IQ4_NL2.4 GB3.3 GB8 GB cards
Q4_02.4 GB3.3 GB8 GB cards
Q4_K_S2.4 GB3.3 GB8 GB cards
Q4_K_M2.5 GB3.4 GB8 GB cards
UD-Q4_K_XL2.5 GB3.4 GB8 GB cards
Q4_12.6 GB3.5 GB8 GB cards
Q5_K_S2.8 GB3.7 GB8 GB cards
Q5_K_M2.8 GB3.7 GB8 GB cards
UD-Q5_K_XL2.8 GB3.7 GB8 GB cards
Q6_K3.2 GB4.1 GB8 GB cards
UD-Q6_K_XL3.6 GB4.5 GB8 GB cards
Q8_04.1 GB5.0 GB8 GB cards
UD-Q8_K_XL5.2 GB6.1 GB8 GB cards
BF167.8 GB8.7 GB12 GB cards

Expected speed on popular hardware

HardwareBest quant that fitsExpected tok/s
NVIDIA RTX 3060Q8_074.6 estimate
NVIDIA RTX 4060 TiQ8_059.3 estimate
NVIDIA RTX 3090Q8_0159 estimate
NVIDIA RTX 4090Q8_0176 estimate
NVIDIA RTX 5090Q8_0278 estimate
AMD Radeon RX 7900 XTXQ8_0177 estimate
Apple M4 Pro (20-core GPU) (64 GB)Q8_056.0 estimate
Apple M3 Max (40-core GPU) (64 GB)Q8_081.9 estimate
AMD Ryzen AI Max+ 395 (Radeon 8060S)Q8_050.3 estimate

Check your own card

Too big for your card?

Run gemma-3-4b-it 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.