Skip to content

Can H100 80G run Ripple-65B-Preview?

Comfortable: GGUF Q6_K needs 60.9 GB at 4K, batch 1; headroom 19.1 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 IQ1_M20.0ComfortableVerified file
GGUF IQ2_XXS21.6ComfortableVerified file
GGUF IQ2_XS23.8ComfortableVerified file
GGUF IQ2_S27.2ComfortableVerified file
GGUF IQ3_XXS29.9ComfortableVerified file
GGUF IQ3_S34.5ComfortableVerified file
GGUF IQ4_XS41.1ComfortableVerified file
GGUF Q4_K_M41.2ComfortableVerified file
GGUF Q5_K_M50.8ComfortableVerified file
GGUF Q6_K60.9ComfortableVerified file
GGUF Q8_072.6TightVerified file

Context capacity · GGUF Q6_K

ContextTotal GBVerdict
4K60.9Comfortable
16K61.8Comfortable
32K62.9Comfortable
128K69.5Comfortable

Generation speed

GGUF Q6_K: 37 tok/s · Bandwidth ceiling (not measured)

Other model / GPU pairs