Underdog-Saluki-27B

Qwen3.8 family · 27.0B params

1recorded run

22.6best tok/s

1verified

Evidence: 1 sourced run · 1 independent source · first recorded Oct 8, 2026. What these labels mean.

What the Underdog-Saluki-27B numbers mean

The fastest verified result on our board is 22.6 output tok/s on an DGX Spark at IQ2-mix with a 262144-token context. That is about 44.2 ms per generated token — 500 tokens takes roughly 22.1 seconds.

How much VRAM does Underdog-Saluki-27B need?

Underdog-Saluki-27B has 27.0B 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)	18.3 GB	24 GB
Q8_0 (8-bit)	33.8 GB	48 GB
FP16 (16-bit)	62.9 GB	80 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)

DGX Spark22.6

Models with a similar footprint

Model	Params	Q4_K_M weights
gemma-4-26B-A4B-it	27B	15.2 GB
gemma-4-26B-A4B-it-qat	27B	15.2 GB
gemma-4-26B-A4B-it-ultra-uncensored-heretic	27B	15.2 GB
Gemma4-26B-A4B-QAT-Uncensored-HauhauCS-Balanced-MTP	27B	15.2 GB
Ornstein3.6-27B-MTP-NSC-ACE-SABER	27B	15.2 GB
Qwen3.8-27B-MTP	27B	15.2 GB

Ranked by distance from Underdog-Saluki-27B's parameter count. Full benchmark board.

All benchmarks Methodology

GPU	Model	Quant	Backend	Ctx	tok/s	prompt tok/s	VRAM GB	Source
DGX Spark · 128 GB	Underdog-Saluki-27B	IQ2-mix	llama.cpp b11459 (f498f864f)	262144	22.6 ✓	—	—	src