News Intel launches Gaudi 3 accelerator for AI: Slower than H100 but also cheaper

[Admin]

Intel formally introduces Gaudi 3 AI accelerators, claiming massive price and TCO advantages over Nvidia's H100.

Intel launches Gaudi 3 accelerator for AI: Slower than H100 but also cheaper : Read more

 

[JayNor]

Intel claims
"New Gaudi 3 AI accelerators offer up to 20 percent more throughput and 2x price/performance vs H100 for inference of LLaMa 2 70B"

 

[Findecanor]

Whoever owns the rights to Antoni Gaudí's legacy, I think they should sue Intel for the use of his name for something that is intended for stealing art.

 

[A Stoner]

It can have value even being slower if it scales well, and is energy efficient enough to allow more of them to work together to get the same or better overall results. I like nVidia more than Intel, but I like competition!

 

[JayNor]

hothardware says:
"According to Intel, Gaudi 3 delivers an average 50% improvement in inferencing performance with an approximate 40% improvement in power efficiency versus NVIDIA’s H100, but at a fraction of the cost."

 

[cyrusfox]

Not hard to beat Nvidia on cost, but in terms of software plug and play and performance + scalability, good luck.

Intel was forecasted to do $500 million only on Gaudi 3 in revenue for this year, paltry amount compared to AMD($4.5 billion), and miniscule compared to NVidia during this AI goldrush($100+ billion by my count). If Gaudi is good, it needs to show it on the market and not on a press deck. Go win some mind share and market share, especially outside of china, as China is largely blocked from many of the top AI solutions thanks to US government restrictions.

 

[Stomx]

Good memoty size but no FP64 , even no FP32 and 10x price than probably a bit slower Tenstorrent Wormhole GPU . This means unlikely it will find good applications in HPC.

 

[loverofGd]

I kind of seeing your info misinformation, because what I see from nextplatform differ from your claims.. In Intel's test, too, Intel Gaudi3 is MUCH faster than H100.. I doubt it is only for software..
Peak FP16/BF16 is much faster than H100 and even B100.. Nvidia's are 8 & 14, Gaudi3's is 14.68
In peak FP8 H100 is 16 B100 is 28 and Intel's Gaudi 3 is 14.68 again..
In peak FP32, H100 is 4 and Gaudi 3 is 2.39


These are the numbers that I read in the Next Platform, I added the link, please check it out from below..

https://www.nextplatform.com/wp-con...ntel-computex-gaudi-vs-nvidia-tnp-table-2.jpg