Skip to content

Can A40 48G run Hy-Memory-30B-A3B?

Comfortable: GGUF Q8_0 needs 33.8 GB at 4K, batch 1; headroom 14.2 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 Q3_K_M15.5ComfortableVerified file
GGUF Q4_K_M19.5ComfortableVerified file
GGUF Q6_K26.2ComfortableVerified file
GGUF Q8_033.8ComfortableVerified file
GGUF Q8_0-expsIQ2_XXS9.9ComfortableVerified file
GGUF Q8_0-expsIQ1_M8.8ComfortableVerified file

Context capacity · GGUF Q8_0

ContextTotal GBVerdict
4K33.8Comfortable
16K35.1Comfortable
32K36.7Comfortable
128K46.6Tight

Generation speed

GGUF Q8_0: 230 tok/s · Bandwidth ceiling (not measured)

Other model / GPU pairs