Apple has revealed that it does not use NVIDIA’s AI accelerator GPUs for its own AI model. A document released by the manufacturer itself reveals that they rely on Google’s TPUs (Tensor Processing Unit) to train their data.
Apple Foundation Models (AFMs) use TPUv4 and TPUv5 chips, both those seen in servers and those housed in devices – with a design that works both online and offline, as was demonstrated during WWDC 2024 in June.
The AFM for servers is the manufacturer’s largest LLM and remains online at all times. The documents reveal that it is trained with 8,192 TPUv4 chips – separated into eight divisions of 1,024 each, all connected through a data center network.
According to Apple, its AFMs gather training data for AI from the Applebot web crawler as well as several “high-quality” licensed datasets. They also claim to use carefully chosen code, mathematics, and public datasets.
See more:
Artificial intelligence on Apple devices
On Arm devices, Google clusters with TPUv5 are used by Apple – being trained by 2,048 chips in just one division.
Obviously, the model will not reach the scale of learning and knowledge seen in servers, but the manufacturer points out that knowledge distillation techniques have been optimized for performance in smaller models – ensuring its efficiency.
Benchmark tests have been performed, showing that in some aspects the choice is right – as its AFMs (both servers and devices) are among the best in following instructions, using tools, writing and more. See the graph below, related to the writing techniques of the Apple model:
Apple, Google e NVIDIA
Considering that several companies like Dell, OpenAI and even millionaires like Elon Musk have chosen NVIDIA GPUs for artificial intelligence, it is curious that Apple prefers to use Google TPUs to continue progressing with its LLM.
It’s worth noting that AMD is also preparing new Instinct accelerators like the MI350 and MI400, just as Intel has been teasing its Gaudi 3 – offering more market options, perhaps even more economical, than the leader NVIDIA.
Another detail worth mentioning is that Microsoft itself has also avoided acquiring their GPUs, with several reports that they are producing their own chips to deal with artificial intelligence and accelerators in the near future.
Source: Tom’s Hardware
Join the Adrenaline Offers Group
Check out the best deals on hardware, components and other electronics that we found online. Video cards, motherboards, RAM and everything you need to build your PC. By joining our group, you receive daily promotions and have early access to discount coupons.
Join the group and enjoy the promotions
Source: https://www.adrenaline.com.br/apple/apple-apela-para-google-para-nao-usar-gpus-nvidia-ias/
