Hardware1 hOctober 29, 2026, 10:59

Modder unlocks Tensor Core limits on older Nvidia RTX 20, 30 and 40 GPUs

A modder published gpu-unlock on GitHub, removing Nvidia's half-speed cap on Tensor Core FP16, BF16 and TF32 math with FP32 accumulation in GeForce RTX 20, 30 and 40 Series cards. AI performance gains reach 65–100%, but the mod is stable only on RTX 30 and does not affect gaming.

Lowyat.NET

Open source

Modder unlocks Tensor Core limits on older Nvidia RTX 20, 30 and 40 GPUs

RTX 4090: 165 → 329 TFLOPS BF16, a 100% gain
RTX 3090: 75 → 125 TFLOPS BF16, a 65% gain
Stable only on RTX 30; RTX 20 and 40 remain experimental
No changes to CUDA cores, memory bandwidth or gaming performance

#Nvidia#GeForce#RTX

Open-source gpu-unlock lifts Tensor Core rate cap on RTX 30-series GPUs

Hardware

Modders port ringfenced DLSS features to older GPUs

Security

Nvidia ships security update for older GeForce GTX 700, 900 and 10 series GPUs

Hardware

Modder unlocks Tensor Core limits on older Nvidia RTX 20, 30 and 40 GPUs

Open-source gpu-unlock lifts Tensor Core rate cap on RTX 30-series GPUs

Modders port ringfenced DLSS features to older GPUs

Nvidia ships security update for older GeForce GTX 700, 900 and 10 series GPUs

Modders nearly double AI throughput on RTX 20/30/40 via firmware unlock