Gemma on NVIDIA Ada

Can I run Gemma 3 27B on an RTX 4070?

Not at Q4_K_M — it needs 17.5 GB against 10.5 GB available. You would need 2 of these cards.

Won't fit

17.5 GB of 10.5 GB · 167%
018 GB
Weights 15.6 GB
KV cache 1.1 GB
Runtime overhead 0.8 GB
Over the limit 7.0 GB

Short by 7.0 GB. You can run it with 34 of 62 layers on the RTX 4070 and the rest in system RAM, at roughly 4.49 tok/s — usable for batch work, painful for chat. A smaller quantisation or a shorter context is usually the better trade.

Generation4.49tok/s
Prompt processing445tok/s
Max context0tokens
KV per 1K tokens0GB

Every quantisation of Gemma 3 27B on a RTX 4070

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

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 51.0 GB 53.0 GB Won't fit — 0.86
INT8 / W8A8 8.50 26.8 GB 28.7 GB Won't fit — 1.90
Q8_0 (GGUF) 8.50 26.8 GB 28.7 GB Won't fit — 1.90
FP8 (E4M3) 8.00 25.5 GB 27.4 GB Won't fit — 2.03
Q6_K 6.56 20.9 GB 22.9 GB Won't fit — 2.72
Q5_K_M 5.67 18.1 GB 20.0 GB Won't fit — 3.43
Q5_K_S 5.52 17.6 GB 19.5 GB Won't fit — 3.61
Q4_K_M 4.85 15.6 GB 17.5 GB Won't fit — 4.49
AWQ 4-bit 4.25 15.5 GB 17.4 GB Won't fit — 4.52
GPTQ 4-bit 4.25 15.5 GB 17.4 GB Won't fit — 4.52
MXFP4 4.25 15.5 GB 17.4 GB Won't fit — 4.52
Q4_K_S 4.58 14.8 GB 16.7 GB Won't fit — 4.86
Q4_0 4.55 14.7 GB 16.6 GB Won't fit — 5.03
IQ4_XS 4.25 13.8 GB 15.7 GB Won't fit — 5.67
Q3_K_M 3.91 12.7 GB 14.7 GB Won't fit — 6.71
IQ3_M 3.70 12.1 GB 14.0 GB Won't fit — 7.54
IQ3_XXS 3.06 10.2 GB 12.1 GB Won't fit — 13.1
Q2_K 2.63 8.9 GB 10.8 GB Won't fit 4K 23.4
IQ2_XXS 2.06 7.1 GB 9.1 GB Runs comfortably 24K 37.8
IQ1_M 1.75 6.2 GB 8.1 GB Runs comfortably 35K 42.9

Also worth checking

Gemma 3 27B on every GPU
Full hardware tableEverything that runs on a RTX 4070
Full model tableWhy the KV cache is the part that bites
Guide