Gemma on NVIDIA Blackwell

Can I run Gemma 3 12B on an RTX 5090?

Yes. Gemma 3 12B at Q4_K_M uses 8.6 GB of the 29.3 GB available on a RTX 5090, and runs at about 151 tokens per second. You can push the context to 128K.

Runs comfortably

8.6 GB of 29.3 GB · 29%
029 GB
Weights 7.0 GB
KV cache 0.8 GB
Runtime overhead 0.8 GB

Gemma 3 12B at Q4_K_M leaves 20.7 GB spare on a RTX 5090. There is room to raise the context length or move up a quantisation level.

Generation151tok/s
Prompt processing3598tok/s
Max context128Ktokens
KV per 1K tokens0GB

Every quantisation of Gemma 3 12B on a RTX 5090

Highlighted row is the highest quality that still fits at 8K context.

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 22.7 GB 24.4 GB Runs comfortably 87K 50.1
INT8 / W8A8 8.50 11.8 GB 13.5 GB Runs comfortably 128K 92.9
Q8_0 (GGUF) 8.50 11.8 GB 13.5 GB Runs comfortably 128K 92.9
FP8 (E4M3) 8.00 11.4 GB 13.0 GB Runs comfortably 128K 96.6
Q6_K 6.56 9.3 GB 10.9 GB Runs comfortably 128K 116
Q5_K_M 5.67 8.1 GB 9.7 GB Runs comfortably 128K 132
Q5_K_S 5.52 7.8 GB 9.5 GB Runs comfortably 128K 135
AWQ 4-bit 4.25 7.4 GB 9.0 GB Runs comfortably 128K 142
GPTQ 4-bit 4.25 7.4 GB 9.0 GB Runs comfortably 128K 142
MXFP4 4.25 7.4 GB 9.0 GB Runs comfortably 128K 142
Q4_K_M 4.85 7.0 GB 8.6 GB Runs comfortably 128K 151
Q4_K_S 4.58 6.6 GB 8.2 GB Runs comfortably 128K 158
Q4_0 4.55 6.6 GB 8.2 GB Runs comfortably 128K 158
IQ4_XS 4.25 6.2 GB 7.8 GB Runs comfortably 128K 167
Q3_K_M 3.91 5.7 GB 7.4 GB Runs comfortably 128K 178
IQ3_M 3.70 5.5 GB 7.1 GB Runs comfortably 128K 186
IQ3_XXS 3.06 4.6 GB 6.3 GB Runs comfortably 128K 214
Q2_K 2.63 4.1 GB 5.7 GB Runs comfortably 128K 238
IQ2_XXS 2.06 3.3 GB 5.0 GB Runs comfortably 128K 280
IQ1_M 1.75 2.9 GB 4.6 GB Runs comfortably 128K 309

Also worth checking

Gemma 3 12B on every GPU
Full hardware tableEverything that runs on a RTX 5090
Full model tableWhy the KV cache is the part that bites
Guide