llamaperf
Submit

DeepSeek V4 Flash 284B (13B active)

on NVIDIA A100 40GB · llama.cpp

NVIDIA hardware
Sep 7, 2026
Share
Throughput
16.1 t/s gen
Quant
Q8_K_XL (GGUF)
VRAM reported
40 GB

Summary

User reports all experts on CPU with only 15.8 GB of 40 GB VRAM used.
View source ↗