Intel unveils Gaudi 3 AI accelerator, says it beats Nvidia's H100
Intel made a splash at the kickoff of its Vision 2024 conference by introducing the Gaudi 3 AI processing chip. This new chip is tailor-made to turbocharge AI tasks on a large scale within businesses, boasting superior speed compared to Nvidia's H100 GPU, the current gold standard in the industry.
Built upon the same architecture and core principles as its
predecessor, Gaudi 3 distinguishes itself by leveraging TSMC's cutting-edge 5nm
manufacturing process, a significant leap from the 7nm process utilized in
Gaudi 2. This enhancement translates to heightened efficiency and performance.
During its presentation, Intel presented benchmark
comparisons pitting Gaudi 3 against Nvidia's leading H100 GPU, showcasing
performance disparities across various large language models such as LLAMA2-7B,
LLAMA2-13B, and GPT-3 175B. Impressively, Gaudi 3 outpaced its competitor by up
to 1.7 times in speed.
Furthermore, Intel highlighted Gaudi 3's exceptional power
efficiency, asserting that it surpasses Nvidia's H100 by up to 2.3 times in
this regard.
Equipped with up to 128GB of HBM2e memory, Gaudi 3 boasts an
impressive bandwidth of 3.7 TB/s and operates within a power envelope of 900W.
During its presentation, Intel presented benchmark
comparisons pitting Gaudi 3 against Nvidia's leading H100 GPU, showcasing
performance disparities across various large language models such as LLAMA2-7B,
LLAMA2-13B, and GPT-3 175B. Impressively, Gaudi 3 outpaced its competitor by up
to 1.7 times in speed.
Furthermore, Intel highlighted Gaudi 3's exceptional power
efficiency, asserting that it surpasses Nvidia's H100 by up to 2.3 times in
this regard.
Comments