Startup trains AI models with gaming GPU setup under $100K

GPU
(Image credit: Shutterstock)

The heart of a good AI system is a well-trained model. This requires a great deal of resources and power which can end up costly for investors and everyone else involved. One startup is shaking things up a bit by training its model using more affordable hardware. Fastino is reportedly using less than $100,000 (USD) worth of gaming graphics cards to handle its training needs. This stands out against more commercial operations like Elon Musk's giant xAI, which has seriously intense power requirements.

TechCrunch first reported on Fastino and its use of low-end gaming chips.

According to Fastino, this was accomplished by developing architecture for their AI model that is "task-specific" and designed to function on a smaller scale. This change has made for a faster model that has a noticeable degree of accuracy. The best part, however, is the cost effective benefits of training their system on cheap gaming hardware.

Fastino may have less than $100K invested in GPUs but it has already started raking in money from investors that see the potential. Most recently, Fastino acquired $17.5 million from Khosla Ventures following $7 million raised late last year by both M12, Microsoft's venture capital fund, as well as Insight Partners.

Fastino isn't the only company to innovate on the AI hardware front with relatively affordable development options. Tiny Corp released what it calls TinyBox last year with six repurposed AMD Radeon 7900 XTX graphics cards specifically for AI processing.

So far, Fastino has yet to share any performance data, so we can't confirm exactly how well their new architecture stacks up against other AI systems. The team has released statements suggesting noteworthy performance, touting its ability to provide a complex answer within just milliseconds. Without more information, it's hard to tell how well it will perform against other big names in the AI field.

Follow Tom's Hardware on Google News to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button.

Ash Hill
Contributing Writer

Ash Hill is a contributing writer for Tom's Hardware with a wealth of experience in the hobby electronics, 3D printing and PCs. She manages the Pi projects of the month and much of our daily Raspberry Pi reporting while also finding the best coupons and deals on all tech.

  • Notton
    While I don't care if they snap up all the manufactured e-waste that are the 8GB cards, it'll send the wrong message to Nvidia and AMD that those are popular cards.
    And if they snap up all the 16GB~24GB cards, well then it's war.
    Reply