NVIDIA · 24 GB GDDR6X · 936 GB/s bandwidth ·
Google · 4.3B · model ctx up to 128K
Yes — the RTX 3090 24GB (24 GB GDDR6X) can run gemma-3-4b-it. At Q4 with the default 8K context it needs ≈ 4.15 GB VRAM (weights 2.37 GB + KV cache 0.78 GB + 1 GB runtime overhead), which fits within 24 GB. Decode at Q4/8K: ≈ 152.7 tok/s (estimated, batch 1). FP16 (≈ 10.38 GB) also fits, at ≈ 42.0 tok/s (estimated). Estimates come from memory-bandwidth math; rows tagged measured override estimates. Relative ranking is more reliable than absolute tok/s. Methodology.
Fit & speed by quant and context
Quant
Context
VRAM needed
Fits?
tok/s (decode)
Q4
4K
3.76 GB
Yes
152.7est.
Q4
8K default
4.15 GB
Yes
152.7est.
Q4
32K
6.49 GB
Yes
152.7est.
FP16
4K
9.99 GB
Yes
42.0est.
FP16
8K default
10.38 GB
Yes
42.0est.
FP16
32K
12.72 GB
Yes
42.0est.
Fits = weights + KV(ctx) + 1 GB overhead ≤ 95% of VRAM. Measured
anchors are context-agnostic; the fit verdict is recomputed per context. A missing tok/s
means the model is far beyond this card (offload-only territory).