Benchmarking OSCAR2 KV Cache in llama.cpp
giveen · reddit · 2026-07-19
OSCAR2 KV Cache Benchmark in llama.cpp
The author tested the oscar2 KV cache solution in llama.cpp on an RTX 5090, comparing it against f16/f16 and q20/q20. For Qwen3.6-27B, the generation speed of oscar2/oscar2 is approximately 42 t/s, which is about 35% faster than q20 while closely matching the performance level of f16.
They also tested performance across different context lengths: from 4K to 1M, the generation speed of oscar2 remains mostly flat without obvious degradation. In contrast, f16 quickly hits OOM at longer contexts, whereas oscar2 continues to run. The overall conclusion is that this cache solution perfectly balances speed and VRAM usage in long-context scenarios.
More from Infra
- Nearly half of Claude revenue now flows through cloud marketplace commitments, up from 11% — BenBajarin · 2026-10-26
- Anthropic ARR goes vertical after price cuts: tokens up 4x, $100B+ ARR back in play — GavinSBaker · 2026-10-26
- Alex Banks argues power caps GPU supply, real risk is demand cooling first — thealexbanks · 2026-10-26
- Analyst: Ant has the most fungible AI workload of any model provider today — BenBajarin · 2026-10-26
- Anthropic S-1: training infra edge cut per-task cost 40% and doubled revenue growth — BenBajarin · 2026-10-26
- Rewriting scaling law study: optimal synthetic data share rises from 10% to 32% with compute — heghbalz · 2026-10-26