Fired-up Pat Gelsinger shoots from the lip at Qualcomm and Nvidia
Intel boss reveals first 18A silicon will run next week as he undercuts H100s on price and dismisses X Elite metrics
Computex Intel CEO Pat Gelsinger has used his keynote address at the Computex conference in Taiwan to fire back at competitors Qualcomm and Nvidia, and reveal a product he thinks will make his assertions more than words.
"Unlike what [Nvidia CEO] Jensen [Huang] might have you believe, Moore's law is alive and well," Gelsinger declared before announcing the Xeon 6 range of datacenter CPUs. We've already covered that silicon in depth. The TL;DR is that Intel's manycore Xeons are here, starting with a 144-core model, along with a strategy shift to use only efficient cores, and a 288-core unit is on the way.
The CEO also took a swing at Nvidia, revealing the price Intel will charge system builders for baseboards equipped with eight of its Gaudi accelerators: $65,000 for the Gaudi 2 and $125,000 for Gaudi 3. Gelsinger bragged those prices beat the price of comparable packs based on Nvidia's accelerators by two thirds and a third respectively and mentioned the H100 model – very much last year's model for Nvidia.
Intel's pitch for your AI workloads, however, doesn't take Nvidia head-on. Gelsinger suggested Intel processors and accelerators can usefully be applied to retrieval-augmented generation – an emerging AI workload that gives generative AI models access to on-prem data.
Gelsinger noted that 80 percent of enterprise data remains on-prem where it is "unmonetized." The CEO wants to change that, suggesting Gaudi bundles plus servers packing Xeon 6 processors can do the job.
That pitch came with another to use Xeon 6 to consolidate server fleets – a concept illustrated by an on-stage racking and stacking session in which an Intel employee slotted six servers into a rack and claimed they possessed computing power to match a full rack of machines packing Xeon 2 CPUs.
Gelsinger urged enterprises to do likewise, claiming power savings alone will make the move worthwhile.
- Intel details how Lunar Lake PC chips deliver 120 TOPS
- Intel challenges AMD's Epycs with a 144 e-core Xeon
- Intel reportedly seeking $11B to finance facility in Ireland
- The chip that changed my world – and yours
The Intel CEO's attack on Qualcomm came in the form of a smackdown of the performance claims for its recently-launched X Elite SoCs.
Qualcomm CEO Cristiano Amon yesterday insisted that his silicon represents the most powerful PC engine ever.
Gelsinger disagreed, claiming the Lunar Lake SoCs Intel launched today trounce Qualcomm's fare on all metrics – and declared rivals' claims that the x86 architecture can't match others in terms of power consumption are "a myth."
Lunar Lake already proves that, Gelsinger insisted, before teasing the late 2024 debut of Arrow Lake processors that bring the Copilot+ PC's 40-TOPS-or-more spec to the desktop.
He also told the Computex crowd that next week will see the first Panther Lake silicon powered on – an important moment, as those products are the first to be built on Intel's most advanced 18A process.
"It represents the culmination of all we have done since I came back to Intel," he declared, before promising "Panther Lake will be coming to life in a powerful way in 2025." Intel officials later told The Reg Panther Lake's TOPS count will "dwarf" that of Lunar Lake.
And if Panther Lake isn't a hit, but Qualcomm's Arm CPUs are?
"I want to be the foundry," Gelsinger said, referring to his plans that mean Intel is now a chipmaker-for-hire and can therefore win even if it loses. ®