The dependence on RAM has grown enormously due to the demand for language models and artificial intelligence, which has led to availability issues and rising prices in the market.
However, Google has presented a possible solution to this crisis: TurboQuant, an innovative compression technology that reduces the RAM memory required for AIs to operate effectively by up to 80%, without compromising speed or accuracy.
TurboQuant, from Google, reduces RAM consumption by 80% for AI models
TurboQuant consists of an algorithm that modifies the way data is stored and processed, using a two-step technique. First, it simplifies the activation vectors, reducing the complexity of the instructions that an AI needs.
The second step involves the elimination of errors through the QJL technique, which allows each vector address to require significantly less memory space. This results in drastically lower RAM consumption, opening new possibilities for the development of more powerful AIs.
The announcement of TurboQuant has already impacted the market, evidenced by declines in the stocks of companies like Micron and Western Digital, which recorded drops of 4% and 6%, respectively. Although these are not alarming crashes, they indicate that investors fear a possible slowdown in RAM demand from large tech companies.
However, caution is needed. According to the Jevons Paradox, when a technology is more efficient and consumes fewer resources, its use may increase. If TurboQuant meets expectations, companies could leverage this efficiency to develop more advanced AIs that, unexpectedly, could end up requiring even more RAM.
Thus, although TurboQuant could provide temporary relief for users looking to increase their RAM, uncertainty about market evolution and component availability persists.