RunsOnMy ModelsGPUs & MacsBy VRAMAPI pricesVRAM calculatorNew

Can I run gemma 3 12b it locally?

gemma 3 12b it by google has 12B parameters. At 4-bit with an 8K context it needs about 8.5 GB of GPU memory. It fits on 65 of 65 GPUs and Macs we track. See all models that fit in 12 GB of VRAM.

8.5 GBQ4_K_M · 8K context
RTX 3080 10GBsmallest consumer GPU that fits
M1 16GBsmallest Mac that fits
128Kmax context
GQA + sliding windowattention type

Memory needed by quant and context length

GB of GPU memory = weights + KV cache (FP16) + runtime overhead. Weights use real file sizes from unsloth/gemma-3-12b-it-GGUF where available (marked •), otherwise parameter count × bits per weight. How we calculate.

QuantQuality4K ctx8K ctx32K ctx128K ctx
2-bit (Q2_K) •rough5.9 GB6.2 GB7.7 GB13.7 GB
3-bit (Q3_K_M) •usable7.1 GB7.3 GB8.8 GB14.8 GB
4-bit (Q4_K_M) •good, most popular8.3 GB8.5 GB10.0 GB16.0 GB
5-bit (Q5_K_M) •very good9.3 GB9.6 GB11.1 GB17.1 GB
6-bit (Q6_K) •near lossless10.5 GB10.7 GB12.2 GB18.2 GB
8-bit (Q8_0) •effectively lossless13.1 GB13.4 GB14.9 GB20.9 GB
16-bit (F16/BF16) •original weights23.4 GB23.6 GB25.1 GB31.1 GB

Which GPUs and Macs run gemma 3 12b it?

Best quality quant that fits at 8K context. "Tight" = within 10% of usable memory. Speed is a ceiling from memory bandwidth (real speed is typically 50–80% of it).

HardwareMemoryFits?Best quantNeedsMax tok/s (est.)
RTX 3060 12GB12 GBFitsQ5_K_M9.6 GB43details
RTX 4060 Ti 16GB16 GBFitsQ8_013.4 GB23details
RTX 309024 GBFitsQ8_013.4 GB75details
RTX 409024 GBFitsQ8_013.4 GB81details
RTX 509032 GBFitsF1623.6 GB76details
RTX 5070 Ti16 GBFitsQ8_013.4 GB72details
RX 7900 XTX24 GBFitsQ8_013.4 GB77details
Mac M4 Pro 48GB48 GBFitsF1623.6 GB12details
Mac M4 Max 128GB128 GBFitsF1623.6 GB23details
Mac M3 Ultra 512GB512 GBFitsF1623.6 GB35details
DGX Spark128 GBFitsF1623.6 GB12details
RTX 407012 GBFitsQ5_K_M9.6 GB60details
RTX 5060 Ti 16GB16 GBFitsQ8_013.4 GB36details
RTX 40608 GBFitsQ2_K6.2 GB57details
Arc B58012 GBFitsQ5_K_M9.6 GB54details
RTX 3060 Ti8 GBFitsQ2_K6.2 GB94details
RTX 30708 GBFitsQ2_K6.2 GB94details
RTX 3080 10GB10 GBFitsQ4_K_M8.5 GB104details
RTX 3080 12GB12 GBFitsQ5_K_M9.6 GB108details
RTX 3090 Ti24 GBFitsQ8_013.4 GB81details
RTX 4060 Ti 8GB8 GBFitsQ2_K6.2 GB60details
RTX 4070 Super12 GBFitsQ5_K_M9.6 GB60details
RTX 4070 Ti12 GBFitsQ5_K_M9.6 GB60details
RTX 4070 Ti Super16 GBFitsQ8_013.4 GB54details
RTX 408016 GBFitsQ8_013.4 GB57details
RTX 4080 Super16 GBFitsQ8_013.4 GB59details
RTX 50608 GBFitsQ2_K6.2 GB94details
RTX 507012 GBFitsQ5_K_M9.6 GB80details
RTX 508016 GBFitsQ8_013.4 GB77details
2x RTX 30902×24 GBFitsF1623.6 GB40details
2x RTX 40902×24 GBFitsF1623.6 GB43details
2x RTX 50902×32 GBFitsF1623.6 GB76details
RTX A600048 GBFitsF1623.6 GB33details
RTX 6000 Ada48 GBFitsF1623.6 GB41details
RTX PRO 6000 Blackwell96 GBFitsF1623.6 GB76details
L40S48 GBFitsF1623.6 GB37details
A100 40GB40 GBFitsF1623.6 GB66details
A100 80GB80 GBFitsF1623.6 GB87details
H100 PCIe80 GBFitsF1623.6 GB85details
H100 SXM80 GBFitsF1623.6 GB142details
H200141 GBFitsF1623.6 GB204details
RX 76008 GBFitsQ2_K6.2 GB60details
RX 7700 XT12 GBFitsQ5_K_M9.6 GB51details
RX 7800 XT16 GBFitsQ8_013.4 GB50details
RX 7900 GRE16 GBFitsQ8_013.4 GB46details
RX 7900 XT20 GBFitsQ8_013.4 GB64details
RX 907016 GBFitsQ8_013.4 GB51details
RX 9070 XT16 GBFitsQ8_013.4 GB51details
Ryzen AI Max+ 395 128GB128 GBFitsF1623.6 GB11details
Arc A770 16GB16 GBFitsQ8_013.4 GB45details
Mac M1 16GB16 GBFitsQ5_K_M9.6 GB8.1details
Mac M2 24GB24 GBFitsQ8_013.4 GB8.0details
Mac M3 24GB24 GBFitsQ8_013.4 GB8.0details
Mac M4 16GB16 GBFitsQ5_K_M9.6 GB14details
Mac M4 32GB32 GBFitsQ8_013.4 GB9.6details
Mac M1 Pro 32GB32 GBFitsQ8_013.4 GB16details
Mac M3 Pro 36GB36 GBFitsQ8_013.4 GB12details
Mac M4 Pro 24GB24 GBFitsQ8_013.4 GB22details
Mac M4 Pro 64GB64 GBFitsF1623.6 GB12details
Mac M1 Max 64GB64 GBFitsF1623.6 GB17details
Mac M2 Max 96GB96 GBFitsF1623.6 GB17details
Mac M3 Max 128GB128 GBFitsF1623.6 GB17details
Mac M4 Max 64GB64 GBFitsF1623.6 GB23details
Mac M2 Ultra 192GB192 GBFitsF1623.6 GB34details
Mac M3 Ultra 256GB256 GBFitsF1623.6 GB35details

Prefer an API?

Google: Gemma 3 12B is on OpenRouter at $0.05 per 1M input tokens and $0.15 per 1M output tokens. Price details and history →

Questions

How much VRAM does gemma 3 12b it need?

About 8.5 GB at 4-bit (Q4_K_M) with an 8K context: 6.8 GB of weights (real GGUF file size), 0.8 GB of KV cache and 0.9 GB of runtime overhead. At 8-bit it needs about 13.4 GB.

What is the smallest GPU that runs gemma 3 12b it?

At Q4_K_M and 8K context, the smallest single consumer GPU in our table that fits it is the RTX 3080 10GB (10 GB).

Can I run gemma 3 12b it on a Mac?

Yes. At Q4_K_M with 8K context it fits on a Mac M1 16GB or larger, using the default macOS GPU memory limit.

Similar models

Same family or a similar parameter count, with the VRAM each needs at Q4_K_M and 8K context.

gemma 3 4b it

4.3B · 3.5 GB at 4-bit

medgemma 1.5 4b it

4.3B · 3.5 GB at 4-bit

medgemma 4b it

4.3B · 3.5 GB at 4-bit

Llama Guard 4 12B

12B · 9.1 GB at 4-bit

Llama 3.2 11B Vision Instruct

11B · 7.4 GB at 4-bit

Llama 3.2 11B Vision

11B · 8.0 GB at 4-bit

Model facts

Hugging Face repogoogle/gemma-3-12b-it
Parameters12,187,325,040 (12B)
Layers / hidden size48 / 3840
AttentionGQA + sliding window; 16 heads, 8 KV heads, head dim 256; sliding window 1024
KV cache per token (FP16)384.0 KiB (sliding layers capped at the window)
Max context131,072 tokens
Licensegemma (gated: accept terms on Hugging Face)
Released on HFMar 1, 2025
Downloads (30d) / likes405,670 / 918

Sources: Hugging Face model API · config.json (unsloth/gemma-3-12b-it-GGUF) · GGUF files · GPU specs from vendor pages (list). Data fetched Nov 17, 2026. Spot an error? Report it.