Powering the IT stack of every technology leader is a semiconductor. Those ingenious little bits of silicon continue to evolve; tiny, astonishing engines that fundamentally underpin how businesses can digitalise.
At the Intel Innovation event this week that evolution was on firm display.
“Moore’s Law” is not dead, insisted Intel CEO Pat Gelsinger as he took to the stage, citing advances in transistors — “we aspire from today about [sic] a hundred billion transistors on a single package; by the end of the decade a trillion transistors in a single package” — as well as in power delivery and lithography to support that view.
Yet Intel increasingly recognises that software is king.
“As we like to say at Intel now ‘software-defined; silicon-enhanced’: software comes first” said Gelsinger, adding (in one of his presentation’s more egregiously odd moments) that “my granddaughter, when she plugs in a USB stick into the computer, she says ‘thank you Papa’. You may like me but I care more about what she has to say.”
Intel’s press releases come in dense bunches that are read, regurgitated, and sometimes unceremoniously picked apart by a tightly knit coterie of semiconductor analysts, enthusiasts, and reporters. A lot of the news can get lost in the wash, so we thought we’d try and strip what we saw at the Intel Innovation day back to the basics.
Here are The Stack’s key takeaways from the Intel Innovation day, for the time-poor.
1: New GPUs for data centers
Intel announced its GPU Flex Series for cloud and data center workloads in August. Servers and other products from OEMs like Dell Technologies, HPE, H3C, Inspur, Lenovo and Supermicro are all coming soon.
In August Intel said these would be optimised for “media delivery and Android cloud gaming workloads.”
(Game streaming and other media streaming to devices works hardware really hard. Hardware that gets too hot, slow or expensive to run suits nobody. Chips optimised for performance and to avoid this are a win.)
The GPU Flex Series will now run popular industry AI and deep learning frameworks, including OpenVINO, TensorFlow and PyTorch, Intel said. The Intel data centre GPUs can be tuned via its oneAPI. (That’s a way to ensure development across different hardware architectures like CPUs, GPUs, FPGAs and AI accelerators.)
2: Cheap GPUs for gamers
Intel revealed the price of its new Intel Arc A770 GPU for gamers/desktop.
These will start shipping, from $329, on October 12. That’s considerably cheaper than rivals.
Intel touted “compelling content creation and 1440p gaming performance.”
Market rumour continues to circulate that Intel wants to get out of the consumer GPU market however amid a focus on the enterprise, but the long-awaited launch news and pricing was well-received this week.
3: Love for the “Universal Chiplet Interconnect Express”
It may sound like a ride at a steampunk fairground, but Universal Chiplet Interconnect Express (UCIe) is symbolic of a real evolution in the semiconductor world. It launched in March 2022 — with AMD, Arm, Google, Intel, Meta, Microsoft, Qualcomm, Samsung, and TSMC all pledging dedication to the new open industry standard.
UCIe launched amid a major shift towards building System-on-Chip hardware that comprises modular sets of interlocking “chiplets”. (AMD’s Zen 3 processors are one example of the approach: They take 7nm eight-core chiplets for the CPU/GPU components from TSMC and add I/O chiplet built on nodes from GlobalFoundries.)
At the Intel Innovator day CEO Pat Gelsinger enthusiastically talked up that approach.
“You can imagine something like this: ‘Hey I’m getting a few of my chiplets from Intel, but you know I have a few of these pieces that I’ve also developed on TSMC, and you know maybe my power delivery components are coming from TI and I have some of my i/o chips that I’m doing on Global Foundries. Intel has the most advanced packaging technology to assemble those together for me and be my supply chain manager’. So, you know, we’re going to bring all of those pieces together, enabling what we’d say is this next generation of the chiplet ecosystem.”
4: Intel Developer Cloud
As part of that wooing of developers, Gelsinger said that the “Intel Developer Cloud” would be expanded to give devs and partners early access to Intel technologies “from a few months up to a full year ahead of product availability” in a new highly limited beta project that may yet expand further: “During beta, selected customers and developers can try out and test many of Intel’s latest platforms in the coming weeks” he said, pointing to 4th Gen Intel Xeon processors (Sapphire Rapids), Habana Gaudi 2 deep learning accelerators, and Data Center GPU Flex Series.
“In the next decade, we will see the continued digitization of everything” concluded Gelsinger: “Five foundational technology superpowers — compute, connectivity, infrastructure, AI and sensing — will profoundly shape how we experience the world. Developers, both software and hardware focused, will build this future. They are the true magicians that advance what’s possible. Fostering this open ecosystem is at the center of our transformation and the developer community is essential to our success.”