RTX 3090 Ti Qwen3-8B Q4 token-gen @4K (Hardware Corner)
124.0 tok/s
Fields
id
benchmarks/comm-rtx-3090-ti-qwen3-8b-q4-tg-4kMetric
decode_tok_s
Value
124
Unit
tok/s
Model
Qwen3-8B
Methodology
THIRD-PARTY MEASURED (Hardware Corner llama.cpp llama-bench table). Q4_K-class GGUF, CUDA, batch-1 decode @4K context. Sibling row at 16K ctx measures 126.94 (hwcorner-rtx-3090-ti-qwen3-8b-q4kxl-decode-16k); ~+9% over the 3090 on the same test.
Measured by
third_party
Reproducible
yes
notes
24GB GDDR6X, 1,008 GB/s. Used street ~$900-1,000 Oct-2026 - same money as a 3090 for ~9% more TG.
Notes
24GB GDDR6X, 1,008 GB/s. Used street ~$900-1,000 Oct-2026 - same money as a 3090 for ~9% more TG.