Hardware Corner Mar 2026: RTX 3060 12GB Qwen3 8B Q4_K decode @ 16k ctx
42.0 tok/s (decode, per-request, batch 1, KV depth 16k)
Fields
id
benchmarks/hc26-rtx3060-12gb-qwen3-8b-q4k-tg-16kMetric
decode_tok_s
Value
42
Unit
tok/s (decode, per-request, batch 1, KV depth 16k)
Model
Qwen3 8B Q4_K
Methodology
THIRD-PARTY LAB MEASURED (Hardware Corner per-GPU LLM benchmark page, Updated Benchmarks: March 2026). llama.cpp, single GPU, weights fully in VRAM. Context length is the KV depth the table column names. Token Generation column = batch-1 per-request decode tok/s, NOT aggregate across concurrent requests. Prompt Processing column = prefill tok/s at that context depth. Engine build hash, CUDA version and exact prompt/output token counts beyond the context-depth bucket are not published. Where the page header quotes a single '14B @ 4k' figure that disagrees with the per-model table, the per-cell table value is the one recorded.
Measured by
third_party
Reproducible
yes
Thermal state
unknown
notes
Single RTX 3060 12GB GDDR6, 360 GB/s as stated on the page. Ampere, PCIe gen4. format_id is the nearest standard GGUF; the page labels the quant Q4_K. Earlier records for this card cover a different model/context pair and are not restated here.
Notes
Single RTX 3060 12GB GDDR6, 360 GB/s as stated on the page. Ampere, PCIe gen4. format_id is the nearest standard GGUF; the page labels the quant Q4_K. Earlier records for this card cover a different model/context pair and are not restated here.