How fast does Gemma 4 26B-A4B run on the NVIDIA DGX Spark (GB10)?

Gemma 4 26B-A4B runs at about 69.9 tokens/s on the NVIDIA DGX Spark (GB10) (measured, Q4_K_M, CUDA, llama-bench). Based on 1 measured run(s), each linked to its public source.

Does Gemma 4 26B-A4B Q4_K_M fit in the NVIDIA DGX Spark (GB10)'s memory?

Estimate: the Q4_K_M file is 16.8 GB and needs about 18.2 GB with an ~8K context; the NVIDIA DGX Spark (GB10) has about 96.0 GB usable, so it fits.

Gemma 4 26B-A4B › NVIDIA DGX Spark (GB10)

Gemma 4 26B-A4B on the NVIDIA DGX Spark (GB10)

Gemma 4 26B-A4B runs at about 69.9 tokens/s on the NVIDIA DGX Spark (GB10) (measured, Q4_K_M, CUDA, llama-bench).

Best measured: 69.9 tokens/s generation and 2,657 tokens/s prompt processing (Q4_K_M, CUDA, FA on).

For comparison, our size-scaling estimate for this combination is 57.3 tok/s.

Every run

Quant	Backend	Generate tok/s	Prompt tok/s	Test	Build / by / notes	Source
Q4_K_M	CUDA FA on	69.9	2,657	tg128	5d3a4a7 Mean of 5 runs, batch 2048.	shamily/gemma4-llama-dgx-spark (GitHub)

tg128 = generating 128 tokens; pp512 = reading a 512-token prompt (llama-bench). LocalScore averages several prompt and output lengths, so its numbers run lower than tg128. "decode, llama-server" = single-stream generation speed measured against a running llama-server (no speculative decoding).

Check prices

Amazon Newegg

As an Amazon Associate I earn from qualifying purchases.

No card yet?

Try it on a rented GPU (RunPod)

Affiliate link: if you sign up, RigBench may earn a referral credit. It never affects rankings or numbers.

All results for the NVIDIA DGX Spark (GB10) · Gemma 4 26B-A4B on other hardware

Embed this result

HTML

<a href="https://rigbench.pages.dev/model/gemma-4-26b-a4b-it/on/nvidia-dgx-spark-gb10/"><img src="https://rigbench.pages.dev/badge/gemma-4-26b-a4b-it--nvidia-dgx-spark-gb10.svg" alt="Gemma 4 26B-A4B · NVIDIA DGX Spark (GB10): 69.9 tok/s (Q4_K_M, measured)"></a>

Markdown (GitHub README, forums)

[![Gemma 4 26B-A4B · NVIDIA DGX Spark (GB10): 69.9 tok/s (Q4_K_M, measured)](https://rigbench.pages.dev/badge/gemma-4-26b-a4b-it--nvidia-dgx-spark-gb10.svg)](https://rigbench.pages.dev/model/gemma-4-26b-a4b-it/on/nvidia-dgx-spark-gb10/)