1/100th of the cost: CPU startup Tachyum claims that one of its processing units can rival dozens of Nvidia H200 GPUs — with a 99% saving that could turn the AI market on its head if true

Key Takeaways:

– Tachyum will start mass production of its Prodigy Universal Processor this year, which combines CPU, GPU, and TPU functionalities.
– The 192-core 5nm processor delivers significantly higher performance compared to existing processors for cloud workloads, GPUs for high-performance computing (HPC), and GPUs for AI applications.
– Tachyum’s goal is to transform hyperscale data centers into universal computing centers capable of handling AI workloads efficiently.
– The processor can dynamically switch between computational domains and eliminate the need for expensive dedicated hardware for AI workloads.
– Tachyum claims that one of its processors can match the AI training performance of a sophisticated array of 52 Nvidia H200 GPUs.
– Tachyum plans to release Prodigy 2, a 3nm processor using PCIe 6.0 and CXL, along with high-bandwidth memory (HBM) 3 RAM in 2026.
– Tachyum has received a major purchase order for building a large-scale system and has released white papers for developers to understand the processor’s capabilities.

TechRadar:

Tachyum has confirmed it will begin mass production of its Prodigy Universal Processor later this year – the product which combines the functionality of a CPU, GPU, and TPU in a single unit, and threatens to turn the AI market on its head.

The 192-core 5nm processor delivers 4.5 times the performance of the best processors for cloud workloads, can be up to three times better than GPUs for high-performance computing (HPC), and can be six times more effective than GPUs for AI applications. 

Source link

AI Eclipse TLDR:

Tachyum has announced that it will start mass production of its Prodigy Universal Processor later this year. The processor combines the functionality of a CPU, GPU, and TPU into a single unit and aims to disrupt the AI market. The 192-core 5nm processor is said to deliver 4.5 times the performance of the best processors for cloud workloads, up to three times better than GPUs for high-performance computing (HPC), and six times more effective than GPUs for AI applications. Tachyum first teased its universal processing component in 2022 and demonstrated its ability to emulate x86 applications in December 2023. The company claims that just one of its processors can match the AI training performance of 52 Nvidia H200 GPUs, at a fraction of the cost. Tachyum also plans to release Prodigy 2, a 3nm processor, in 2026. The company has received a major purchase order to build a large-scale system and has released white papers to provide developers with more information about the processor.