RTX 2080 Ti 22GB mod joins Ampere unlock: 1.85x tensor throughput, open-sourced
bdsqlsz · x · 2026-10-25
KBlueleaf confirmed the NVIDIA tensor-compute unlock now works on an 11GB→22GB modified RTX 2080 Ti, extending the earlier RTX 3090 result, with the KohakUnlock repo open-sourced (Turing source, ROM analysis, Ampere repro notes).
- Unlocked 2080 Ti 22GB: mmapeak FP16→FP32 accumulation jumps from 59.9 to 111.0 TFLOPS (1.85x); MAMF median up 53%
- RTX 3090 at 360W power wall: 75T → 135T measured in PyTorch
- Unlocked 2080 Ti at 250W nearly matches a locked 350W 3090 in mmapeak/mamf-finder
- Repo intentionally ships no firmware blobs, ROM dumps or flashing utility; research-only, fail-closed design
Related event: KohakUnlock Unlocks Hidden Compute Power on RTX GPUs(3 posts)→
More from Infra
- One Blackwell GPU trade-off: 60,000 TPS shared vs 1,000 TPS for a single user — FinanceYF5 · 2026-10-25
- 125B Qwen3.8-Flash-Next hits ~20 tok/s on 16GB VRAM via direct NVMe SSD offloading — MaxDev0 · 2026-10-25
- DeepSeek may unlock consumer GPU matmul performance with memory-bandwidth-first architecture — teortaxesTex · 2026-10-25
- Brex data: agent compute rentals surge, Baseten tops fall 2026 fastest-growing vendors — FinanceYF5 · 2026-10-25
- SpaceX monetizes ~$50B per GW of AI compute with 1-year payback vs CoreWeave's $20B — elonmusk · 2026-10-25
- DeepSeek insider: infra is the big brother and can veto research decisions — teortaxesTex · 2026-10-25