Nemotron-Cascade-2-30B-A3B

Nemotron family · 32.0B params

1recorded run

95.4best tok/s

1verified

Evidence: 1 sourced run · 1 independent source · first recorded May 5, 2026. What these labels mean.

What the Nemotron-Cascade-2-30B-A3B numbers mean

The fastest verified result on our board is 95.4 output tok/s on an Radeon AI Pro R9700 x3 at Q8_0 with a 812-token context. That is about 10.5 ms per generated token — 500 tokens takes roughly 5.2 seconds.

How much VRAM does Nemotron-Cascade-2-30B-A3B need?

Nemotron-Cascade-2-30B-A3B has 32.0B parameters. These estimates use the same formula as the VRAM calculator: weights, plus about 15% for a moderate-context KV cache and runtime overhead, plus a small engine floor.

Quantisation	Estimated VRAM	Smallest card that fits
Q4_K_M (4-bit)	21.5 GB	32 GB
Q8_0 (8-bit)	39.9 GB	48 GB
FP16 (16-bit)	74.4 GB	96 GB

Long contexts and large batches raise the real number — read the smallest card as "possible with a reduced context". Measured throughput is on the benchmark board.

Fastest GPUs for this model (best tok/s)

Radeon AI Pro R9700 x395.4

Models with a similar footprint

Model	Params	Q4_K_M weights
NVIDIA-Nemotron-3-Nano-30B-A3B	32B	18.0 GB
NVIDIA-Nemotron-3.5-Lightning-30B-A3B	32B	18.0 GB
Olmo-3-32B-Think	32B	18.0 GB
gemma-4-31B-it	33B	18.6 GB
GLM-Z1-32B-0414	33B	18.6 GB
Laguna-XS-2.1	33B	18.6 GB

Ranked by distance from Nemotron-Cascade-2-30B-A3B's parameter count. Full benchmark board.

All benchmarks Methodology

GPU	Model	Quant	Backend	Ctx	tok/s	prompt tok/s	VRAM GB	Source
Radeon AI Pro R9700 x3 · 32 GB	Nemotron-Cascade-2-30B-A3B	Q8_0	llama.cpp b9025-1-ga267d945e	812	95.4 ✓	—	—	src