Skip to content

Can DGX Spark 128G run Gemma 4 31B IT?

Comfortable: GGUF Q8_0 needs 35.7 GB at 4K, batch 1; headroom 92.3 GB.

Text input, FP16 KV cache, one card. Unified-memory capacity uses the GPU guide’s usable limit; negative headroom may require host offload. Speed source settings may differ from this 4K estimate.

Measured runs

No measured run for this pair. Sizing is an estimate, not a runtime test.

Quant memory at 4K

QuantTotal GBVerdictWeight sizing
GGUF Q4_K_M22.1ComfortableVerified file
GGUF Q4_020.1ComfortableVerified file
GGUF Q8_035.7ComfortableVerified file
GGUF Q3_K_M18.3ComfortableVerified file
GGUF Q5_K_M25.3ComfortableVerified file

Context capacity · GGUF Q8_0

ContextTotal GBVerdict
4K35.7Comfortable
16K36.8Comfortable
32K38.1Comfortable
128K46.4Comfortable

Generation speed

GGUF Q8_0: 7–9 tok/s · Derived range

Other model / GPU pairs