Intel Talks Falcon Shores Flub, Merges Habana Gaudi Roadmap

Falcon Shores
(Image credit: Intel)

Intel originally planned for its Falcon Shores chips to have both GPU and CPU cores under the hood, creating the company's first 'XPU' for high performance computing. However, its surprise announcement a few months ago that it would pivot to a GPU-only design and delay the chips to 2025 left industry observers shocked — this leaves Intel unable to compete with AMD's Instinct MI300 and Nvidia's Grace Hopper processors that both feature a blended CPU+GPU design.

Today Intel provided some of its somewhat questionable rationale behind the decision to pare back its plans for Falcon Shores to make it the GPU-only successor to the Xeon Max GPU series. Intel also sketched out some of the early details of its new GPU-only Falcon Shores design, which we'll cover further below.

Intel also issued a new HPC and AI roadmap that doesn't show a successor to the Gaudi3 processor — instead, Gaudi and GPUs merge with the Falcon Shores GPU as it takes on the mantle of Intel's premiere HPC and AI chip. Intel told us that it 'plans to integrate the Habana and AXG product [GPU] roadmaps,' but the details of the integration are slight.

The Gaudi compute architecture varies significantly from a standard GPU, so it doesn't appear that its compute architecture could be entirely integrated into the GPUs. As such, Intel could be incorporating smaller parts of the Gaudi design, like its networking interface or other IP blocks, into its GPUs. We're told that Jeff McVeigh, the VP and GM of Intel's Accelerated Computing Group, will provide further details today. As a reminder, Intel paid $2 billion for Habana Labs and killed off products from its $350 million Nervana acquisition to focus on the Gaudi chips. 

Falcon Shores

(Image credit: Intel)
Paul Alcorn
Editor-in-Chief

Paul Alcorn is the Editor-in-Chief for Tom's Hardware US. He also writes news and reviews on CPUs, storage, and enterprise hardware.

  • JayNor
    "However, the CXL interface provides only 64 GB/s of throughput between elements"

    based on what limitation in CXL?

    Are you saying, based on a 16 lane PCIE5 bus?
    Reply
  • JayNor
    The IPUs are taking over a lot of traditionally CPU owned territory. Perhaps Intel is getting schooled by google on the GPU only design of the initial Falcon Shores.

    No mention yet of optical, which was hinted by R. Koduri.
    Reply
  • I think it is up to 64 GB/s in each direction over 16 lanes.

    32GT/s x16 for CXL 2.0

    64GT/s x16 for CXL 3.0
    Reply
  • PlaneInTheSky
    lol @ the AI product line

    "We're going to sell lots of AI chips! ...right John?"

    "I think it was a fad"

    "a fad?! what about all those people using Bing with chatGPT?"

    "No one is using Bing sir"
    Reply
  • JayNor
    Diamond Rapids is expected in 2025 timeframe, and with PCIE6/CXL 3.0 according to some leaks. It would make sense to delay the GPU a year to introduce the new CXL and PCIE6, if that's what they're doing.
    Reply