# Zagita tool doubles RTX 30/40 GPU compute, driving used prices up

Digest AI · Hardware & Compute · published 2026-10-24T22:16:55Z

Canonical: https://digestai.news/story/zagita-tool-doubles-rtx-30-40-gpu-compute-driving-used-prices-up

## Summary

A new open-source Linux tool called Zagita has unlocked the full BF16/FP16 compute rates on NVIDIA's RTX 30 and RTX 40 series GPUs. By using a kernel module that applies specific speed selectors before the driver binds, the tool doubles the instruction issuing speed for FP/BF16 FMA operations on GA102, GA103, GA104, and GA106 chips without requiring vBIOS flashing or hardware mods. Users report up to an 80% boost in compute capabilities, with the RTX 3090 Ti seeing BF16 performance jump from 87.5 TFLOPs to 157 TFLOPs.

In practical AI LLM workloads, the unlock delivers a 33-36% faster prefill speed and reduces time-to-first-token latency by 25%, though decode performance remains unchanged. One user reportedly achieved 381 tokens per second on a Qwen3.8-27B model using an RTX 3090, a significant increase from the baseline 82 TPS. The tool currently supports only Linux, with Windows support not yet available.

This performance uplift has triggered a sharp rise in the used market. The RTX 3090 24 GB, previously around $1,000, has climbed past $2,000, while the RTX 3090 Ti approaches $3,000. RTX 3080 models are now selling for $900-$1,100 on eBay. While the tool makes older cards viable for AI inference and fine-tuning, it is making them less accessible for gamers due to the price surge driven by AI demand.

## Key points

- Zagita tool doubles BF16/FP16 compute on RTX 30/40 GPUs via Linux kernel module.
- RTX 3090 Ti BF16 performance increases from 87.5 TFLOPs to 157 TFLOPs.
- Used RTX 3090 prices have surged past $2,000 due to increased AI demand.

## Why it matters

This tool significantly extends the lifespan of older NVIDIA GPUs for AI tasks, making high-performance inference more accessible. However, the resulting demand is distorting the used GPU market, driving prices up for both AI enthusiasts and gamers.

## Sources

1. [CMP170HX inspiration unlocks 30x/40x also, up to 2X Prefill speedup --> NVIDIA RTX 30 & RTX 40 GPUs Get Impressively Fast In AI LLMs As Compute Doubles With New Unlock Tool But Prices Surge Too](https://wccftech.com/nvidia-rtx-30-40-gpus-get-faster-in-ai-llms-with-new-unlock-tool-but-prices-surge-too) (wccftech.com, 2026-10-24)

Part of the developing story: [Reviving Older GPUs Through Software Optimizations](https://digestai.news/thread/cachyos-kernel-module-may-unlock-rtx-3090-showing-15-performance-boost) (2 stories)

## Cite

Digest AI, "Zagita tool doubles RTX 30/40 GPU compute, driving used prices up", 24 October 2026, https://digestai.news/story/zagita-tool-doubles-rtx-30-40-gpu-compute-driving-used-prices-up

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/zagita-tool-doubles-rtx-30-40-gpu-compute-driving-used-prices-up.json