community pool / model answer

DeepSeek-V4-Flash-0731-GGUF

chat · unsloth/DeepSeek-V4-Flash-0731-GGUF

33.9 tok/s · model median · 7 runs in the model record.

Reference rig	Quantization	Median result	Basis	TTFT (ms)	Peak VRAM	Max context	Engines
RTX 3090 24GB ×4	2-bit	41 tok/s	reported	18,774	90.3 GB	131,072	llama.cpp
RTX 3090 24GB ×7	4-bit	42.4 tok/s	reported	916	146.5 GB	262,144	llama.cpp
RTX 5090 32GB	1-bit	17.8 tok/s	reported	250	- GB	2,048	llama.cpp
RTX 5090 32GB	8-bit	10.4 tok/s	reported	271	- GB	2,048	llama.cpp

Context tested

The table above is the tested context for each exact rig and quantization cell. Empty fields remain empty rather than being filled with an estimate.