Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 



Forgot your password?
typodupeerror
×
Intel AI Hardware

Intel Won't Bring Its Falcon Shores AI Chip To Market (techcrunch.com) 20

During the company's fourth-quarter earnings call Thursday, Intel co-CEO Michelle Johnston Holthaus announced that Intel has decided to cancel its Falcon Shores AI chip. Instead, it'll opt to use it as an internal test chip while shifting focus to Jaguar Shores for AI data center solutions. TechCrunch reports: "AI data center ... is an attractive market for us," Holthaus said during the call. "[B]ut I am not happy with where we are today. We're not yet participating in the cloud-based AI data center market in a meaningful way ... One of the immediate actions I have taken is to simplify our roadmap and concentrate our resources." The focus instead will be on Jaguar Shores, which Holthaus called Intel's opportunity to "develop a system-level solution at rack scale ... to address the AI data center more broadly."

Holthaus tempered expectations for Falcon Shores last month, when she implied that it was an "iterative" step over the company's previous dedicated AI data center chip, Gaudi 3. "One of the things that we've learned from Gaudi is, it's not enough to just deliver the silicon," Holthaus said during Thursday's earnings call. "Falcon Shores will help us in that process of working on the system, networking, memory -- all those component[s]. But what customers really want is that full-scale rack solution, and so we're able to get to that with Jaguar Shores."

"As I think about our AI opportunity, my focus is on the problems our customers are trying to solve, most notably the need to lower the cost and increase the efficiency of compute," Holthaus said. "As such, a one-size-fits-all approach will not work, and I can see clear opportunities to leverage our core assets in new ways to drive the most compelling total cost of ownership across the continuum."

Intel Won't Bring Its Falcon Shores AI Chip To Market

Comments Filter:
  • The problem is... (Score:4, Informative)

    by Rei ( 128717 ) on Friday January 31, 2025 @05:59AM (#65132111) Homepage

    ... if you're not within spitting distance of NVidia on performance, people aren't going to use your chips even if you could give them away for free. GPUs are only a portion of server costs. Server costs are only a portion of datacentre costs. In addition to (minor) opex, you also have lots of power consumption. So any decrease in performance competitiveness really starts to stack up quickly against the selection of your product.

    It's not worth releasing something that's not market-viable.

    • If you could save some 10s of millions on hardware with the only cost a few percentage increase in training time I can gaurantee they'd be used. The cost vs efficiency/speed equation operates in most areas of life.

      If intel could bring a chip to market that is maybe only 90% the speed of Nvidea GPUs yet costs 50% of the price then they'd be on to a winner.

    • by necro81 ( 917438 )

      if you're not within spitting distance of NVidia on performance, people aren't going to use your chips even if you could give them away for free

      There is an argument to be made for availability, though. It's hard to get the latest and greatest from NVidia - demand far outstrips manufacturing capacity. This leads to higher prices, sure, but much like the RPi shortage during the pandemic, it leads to situations where some customers can't get anything close to the quantities they need. So, some RPi custome

    • Also, if you don't make your shit work out of the box with common frameworks, and make it really obnoxious for people to get started using your tech, then no one will. Hey AMD, looking at you...

  • ... of the iAPX 432, or the i960, or XScale, or Itanium, or Xeon Phi, or Arc GPUs, or ...

  • I'm curious where Intel thinks the reasonably mature software support and developer familiarity is going to come from if they just decide to release nothing.

    If their own press release says that they don't have the 'full-scale rack solution' that customers want I'm not going to doubt them on that point; just wonder why they aren't still offering it in smaller units: previous marketing material talked up 'compute density in x86 socket', which suggests a 500w or less part; and even if it's more than that th
    • Look at the reception the "AI PC" has received on slashdot. Maybe not the best barometer, but there is serious doubt about the utility of spending even a small amount of extra money on a PC to get "some" ability for it to do AI calculations, when it makes so much more sense to do it server-side in almost all cases.

      Particularly since the leading AI application currently is conversational AI, like ChatGPT. PC's are so, so far away from being able to host ChatGPT. If that threshold were cleared...

      • I wouldn't expect to see this thing as part of a typical client system; it's too big and too niche for that(also, Intel has a different IP block they use for the 'AI PC' purposes); but what puzzles me is why they are taking the "no full-scale rack solution - no release" position.

        By way of example; Nvidia wants you to go all in with 'DGX SuperPOD', a rack or racks, full of SXM GPUs, nvswitch, and Nvidia NICs; but if you aren't feeling that spendy they'll still sell you downclocked, lower TDP, versions in
  • by MTEK ( 2826397 ) on Friday January 31, 2025 @07:58AM (#65132225)

    Good luck, Intel.

  • by Junta ( 36770 ) on Friday January 31, 2025 @09:06AM (#65132341)

    Intel has taught their customers to be very dubious of any novel Intel product. Particularly in this space they keep launching products that demand the user do things entirely differently then abandon the product and move on to another different thing and ask their customers to start over. There's the corpses of Intel initiatives floating around the HPC space where no one wants to touch them since they are all dead platforms, despite institutions paying Intel a tidy sum to support the adventure.

    Also, a recurring obsession rears its head again:

    "develop a system-level solution at rack scale ... to address the AI data center more broadly."

    But what customers really want is that full-scale rack solution,

    Intel sees that the big boys make "rack scale" solutions but keep failing to understand that doesn't mean they want some wildly exotic approach to do so. Even something as tame as Infiniband is something they tend to avoid. Meanwhile Intel keeps coming back with their interpretation, which over time including announced and unannounced efforts have fixated on things like having a whole rack manifest as one OS with a single kernel instance (which is completely uninteresting to everybody) or building a natively PCIe SAN (when software approaches are more trusted and cheaper) or being able to have a sea of CPUs, memory, and storage with the ability to reassemble freely, which might be a little nice but not *that* interesting and certainly not at the huge price points they imagine it to be worth. Meanwhile they lose out to much more mundane competitive efforts that generally only scale to like maybe 16 GPUs in any "weird way" (SXM) and glue everything together with generally just really good ethernet using IP techniques to create compelling fabric topologies.

    They might have got fixated on nVidia's talk of rack scale NVLink, but there's not been huge signs in the market that is going to be big, while it's very easily proven that you need to at *least* get the stuff the customers actually care about mastered first before you go off on weird visions of what you think "rack scale" means.

  • Reading between the lines of buzzword bullshit bingo, Intel's current AI processors are unsalable because they are still on their old process and they don't have power efficiency.

    If 18A doesn't pan out, Intel-as-fab is over.

  • Who uses Intel's Compute Library for Deep Neural Networks (CLDNN) or OpenVino which will take advantage of this new architecture?

    This Google Trends Chart is just sad: https://trends.google.com/tren... [google.com]

    Even once Jaguar Shores is available, I cannot image anyone lining up to buy this. So what why is Intel making this investment? Is Intel doing this just to be able to say "me too" in the current AI stock frenzy?

Some people have a great ambition: to build something that will last, at least until they've finished building it.

Working...