Spring Hill NNP-I inference parts to live on, Spring Crest NNP-T is toast
Intel has axed Nervana’s in-development NNP-T AI training chip, code-named Spring Crest, as it goes full-steam ahead with Habana’s technology.
The decision is, no doubt, a blow for the folks at Nervana. The once-promising deep-learning startup was swallowed by Intel for $350m in 2016. Led by its cofounder Naveen Rao, who is currently corporate veep of the Artificial Intelligence Products Group at Intel, the unit has faced numerous setbacks in getting its dream hardware – the NNP-T, and the inference-focused NNP-I aka Spring Hill – out into the field.
Now, after repeated delays and name changes to Nervana’s products, Intel has given up on Spring Crest altogether, and killed it off before it could ship. The processor was demonstrated as recently as early December in Vancouver, Canada, at the NeurIPS machine-learning conference.
Spring Hill, meanwhile, isn’t completely dead, yet. Customers who ordered the neural-network-math accelerator should still get their hands on the silicon, we’re told. It reminds us of the fate of the Xeon Phi family.
Crucially, the move comes after Chipzilla snapped up Habana Labs, an AI hardware startup, last year for $2bn. Habana is known for its training and inference chips: Gaudi and Goya, respectively. Unlike Nervana, though, the Israeli startup has been successful in getting its components to market. In short, Habana won, and Nervana lost.
Here’s Intel’s statement to The Register in full on the matter, which suggests its customers preferred Habana to Nervana:
After acquiring Habana Labs in December and with input from our customers, we are making strategic updates to the data center AI acceleration roadmap. We will leverage our combined AI talent and technology to build leadership AI products.
We will bolster the current and next generation of Habana Goya and Gaudi with Intel’s AI hardware and software innovations. The Habana product line offers the strong, strategic advantage of a unified, highly-programmable architecture for both inference and training. By moving to a single hardware architecture and software stack for data center AI acceleration, our engineering teams can join forces and focus on delivering more innovation, faster to our customers. As part of this update we plan to deliver on current customer commitments for the Intel NNP-I inference accelerator (code-named “Spring Hill”) and cease development of the Intel NNP-T (code-named “Spring Crest”).
This roadmap decision aligns to Intel’s AI Strategy and our commitment to deliver heterogenous AI solutions that fit our customers’ evolving power and performance needs – from the intelligent edge to the data center.
Thus, it appears the training accelerator Spring Crest has been canned in favor of the more beefy Gaudi. The latter chip has 32GB of memory built in, a memory bandwidth of 1TB per second, and guzzles up to 200W of power.
Nervana promised Spring Crest would pack 60MB of on-die memory plus 32GB of HBM2-2000 RAM stacked on top, and a core frequency of 1.1GHz, all on a 680mm2 silicon die. The chip was expected to feature 27 billion 16nm transistors capable of hitting 119 trillion operations per second (TOPS), consuming up to 250W.
Nervana staff have been quietly leaving the team, too, with resumes winging their way to other semiconductor makers, we’re told, a sign that the plug has been, or is about to be, pulled on the project. You can read more about the history Nervana and its chip architecture, from 2016, at our sister site, The Next Platform.
Earlier today, analyst Karl Freund lent more weight to the theory that Nervana’s processors just weren’t good enough, clearing the way for Habana: “Apparently, Intel received feedback from its engineers and from large customers that the second Nervana designs, code-named Spring Hill and Spring Crest, just didn’t pass muster for these high-performance workloads.
“I suspect these customers pointed to Habana as the preferred platform that can compete with Nvidia.” ®