Unheralded Larrabee plots AMD and Nvidia takedown

Intel by any other name


A shadowy organization called Larrabee Development Group has set a most ambitious goal: unseating AMD/ATI and Nvidia as the largest producers of high-end graphics chips. And it might just succeed.

It may seem unfathomable for an unknown such as Larrabee to knock off two of the most powerful processor companies. That is until you realize that Larrabee is little more than a weak disguise for Intel. The chip giant has, in fact, ramped up its graphics efforts in recent weeks to create a product code-named Larrabee theoretically capable of besting AMD and Nvidia's best kit.

Intel over the past couple of months has peppered job boards with ads for the Larrabee Development Group - a company "working on the next generation of leading- edge graphics technologies. Towards this end, LDG is seeking hardware graphics engineers and/or architects (for lead position) in all aspects of graphics development. This includes component architecture, micro-architecture and performance simulation development."

(The chip giant didn't try terribly hard to disguise its ties to Larrabee as you can see from this job ad.)

Recently, Intel has stepped up the Larrabee effort even more with our sources saying the project has shifted from bet hedging to a whole-hearted push. Insiders have also confirmed that Intel has amassed "a crack team" of graphics chip experts behind Larrabee.These engineers may very well hold Intel's future and the future of PC and server chips in their brains.

GPGPU Inside

So-called GPGPUs or general purpose graphics chips have become all the rage in the processor industry. Such chips take advantage of the tremendous performance gains made by the likes of ATI and Nvidia by tweaking GPUs to handle non-graphics software loads. So far, the technical computing crowd with math-heavy software have shown the most interest in the technology. ATI and Nvidia have obliged by rolling out what amount to test boards that can help foster the software work.

Eventually, however, GPGPUs could stretch to handle a broader set of software applicable to consumers and business customers.

AMD's purchase of ATI, and Nvidia's apparent reluctance to be acquired have left Intel as an odd man out in the GPGPU race.

Without question, you can see AMD and Nvidia hoping to raise the prominence of GPUs in the broader processor landscape. This pursuit could take the form of GPGPUs alone or it could evolve into a push that places the most attention on the GPU and then presents standard x86 chips as secondary engines. Such a future would not be in keeping with Intel's historical interests given the company's focus on complementary rather than standalone graphics.

Larrabee has the potential to bring Intel in step with rivals by giving it either a full-fledged GPGPU board or simply a graphics-rich add-on to its current CPU lines. Intel appears to be leaning toward the latter option if CEO Paul Otellini's comments at the recent Intel Developer Forum are any indication. When asked if Intel planned to mimic AMD by adding more graphics oomph to its CPUs, Otellini responded with a curt, "The short answer is ‘yes.’ That’s the only answer I’m giving. Next question."

Coding me softly

On one level, there's a real sense of urgency behind the Larrabee project.

GPGPUs garner plenty of attention today and will earn even more press by early 2008 when AMD and Nvidia rollout products that support double precision (64-bit) floating point operations. The move to double precision should give GPGPUs an even greater performance boost on high performance computing workloads.

However, the dash of realism that seems to be missing from all the GPGPU hype is the software problem at hand. A vast army of coders will need to embrace the GPGPU model and re-craft software to suit the processors. Such efforts always take longer than expected.

So while ATI and Nvidia may have hoped to relegate the CPU to a secondary status in the near-term, such a shift could take years and years.

Exactly how Intel is reading this software transition remains to be seen.

The company has been punished over the past three years by falling behind on 64-bit extensions, dual-core chips and a growing emphasis on low-power processors. And yet it seems to have realized these mistakes having now taken the lead on both four-core product and marketing. In short order, the Larrabee Development Group should be revealed as a proper body within Intel - at which time we'll all gain a better sense of the chip maker's intentions. ®


Other stories you might like

  • You need to RTFM, but feel free to use your brain too
    But I was only following the procedures!

    Who, Me? Monday is here, and with it a warning that steadfast determination to ignore instructions might not be such a silly thing after all. Welcome to Who, Me?

    Today's story comes from a reader Regomized as "Sam" and takes us back to his first proper IT job following his departure from the education system.

    Sam found himself on the mainframe operations team for a telecommunications company. The work was, initially, pretty manual stuff. The telco wasn't silly, and had its new recruits start by performing offline duties, such as gathering tapes and job tickets for batch runs, handling payslips, "basically anything involving a bit of leg work," he told us.

    Continue reading
  • Tropical island paradise ponders tax-free 'Digital Nomad Visa'
    Live and work in Bali, pay tax at home

    The government of Indonesia has once again raised the idea of creating a "digital nomad visa" that would allow foreign workers to live and work in the tropical paradise of Bali, tax free, for five years.

    The idea was raised before the COVID-19 pandemic, but understandably shelved as borders closed and the prospect of any digital nomads showing up dropped to zero.

    But in recent interviews Sandiaga Uno, Indonesia's minister for Tourism and the Creative Economy, said the visa was back on the drawing board.

    Continue reading
  • Small in Japan: Hitachi creates its own (modest) cloud
    VMware-powered sovereign cloud not going to challenge hyperscalers, but probably won't be the last such venture

    Hitachi has taken a modest step towards becoming a public cloud provider, with the launch of a VMware-powered cloud in Japan that The Register understands may not be its only such venture.

    The Japanese giant has styled the service a "sovereign cloud" – a term that VMware introduced to distinguish some of its 4,000-plus partners that operate small clouds and can attest to their operations being subject to privacy laws and governance structures within the nation in which they operate.

    Public cloud heavyweights AWS, Azure, Google, Oracle, IBM, and Alibaba also offer VMware-powered clouds, at hyperscale. But some organizations worry that their US or Chinese roots make them vulnerable to laws that might allow Washington or Beijing to exercise extraterritorial oversight.

    Continue reading
  • Beijing probes security at academic journal database
    It's easy to see why – the question is, why now?

    China's internet regulator has launched an investigation into the security regime protecting academic journal database China National Knowledge Infrastructure (CNKI), citing national security concerns.

    In its announcement of the investigation, the China Cyberspace Administration (CAC) said:

    Continue reading
  • Cerebras sets record for 'largest AI model' on a single chip
    Plus: Yandex releases 100-billion-parameter language model for free, and more

    In brief US hardware startup Cerebras claims to have trained the largest AI model on a single device powered by the world's largest Wafer Scale Engine 2 chip the size of a plate.

    "Using the Cerebras Software Platform (CSoft), our customers can easily train state-of-the-art GPT language models (such as GPT-3 and GPT-J) with up to 20 billion parameters on a single CS-2 system," the company claimed this week. "Running on a single CS-2, these models take minutes to set up and users can quickly move between models with just a few keystrokes."

    The CS-2 packs a whopping 850,000 cores, and has 40GB of on-chip memory capable of reaching 20 PB/sec memory bandwidth. The specs on other types of AI accelerators and GPUs pale in comparison, meaning machine learning engineers have to train huge AI models with billions of parameters across more servers.

    Continue reading
  • Zendesk sold to private investors two weeks after saying it would stay public
    Private offer 34 percent above share price is just the thing to change minds

    Customer service as-a-service vendor Zendesk has announced it will allow itself to be acquired for $10.2 billion by a group of investors led by private equity firm Hellman & Friedman, investment company Permira, and a wholly-owned subsidiary of the Abu Dhabi Investment Authority.

    The decision is a little odd, in light of the company's recent strategic review, announced on June, which saw the board unanimously conclude "that continuing to execute on the Company's strategic plan as an independent, public company is in the best interest of the Company and its stockholders at this time."

    That process saw Zendesk chat to 16 potential strategic partners and ten financial sponsors, including a group of investors who had previously expressed conditional interest in acquiring the company. Zendesk even extended its discussions with some parties but eventually walked away after "no actionable proposals were submitted, with the final bidders citing adverse market conditions and financing difficulties at the end of the process."

    Continue reading
  • Singapore promises 'brutal and unrelentingly hard' action on dodgy crypto players
    But welcomes fast cross-border payments in central bank digital currencies

    In the same week that it welcomed the launch of a local center of excellence focused on crypto-inspired central bank digital currencies, Singapore's Monetary Authority (MAS) has warned crypto cowboys they face a rough ride in the island nation.

    The center of excellence (COE) was established by the Mojaloop Foundation – an open source effort to create payment platforms to make digital financial services accessible to those without access to banks. The COE aims to "accelerate financial inclusion in emerging markets" through hackathons, workshops and pilot projects while examining expanded CBDCs payment capabilities."

    Singapore's sovereign wealth fund has invested in Mojaloop, and MAS chief fintech officer Sopnendu Mohanty serves as a board advisor and the authority provides representatives to the Foundation's working group, alongside folks from the Bill & Melinda Gates Foundation, Google, and more.

    Continue reading
  • Software-defined silicon is coming for telecom kit later this year
    Startup EdgeQ believes pay-for-what-you-use model will make 5G transition more cost-effective

    Interview While the IT industry waits to see if and when Intel will introduce software-defined silicon in Xeon CPUs, one startup us is moving ahead with plans to bring a pay-for-what-you-use pricing model to the telecom market with its "base station-on-a-chip" later this year.

    Silicon Valley-based EdgeQ, which is led by Qualcomm and Intel executives, announced last week that it has begun sampling an EdgeQ-based 5G small cell and OpenRAN PCIe accelerator card for base stations with telecom operators and equipment makers.

    Things are apparently moving smoothly enough for the startup that Adil Kidwai, EdgeQ's head of product management, told The Register that its RISC-V-based chip will appear in mass-manufactured products like small cells and base station accelerator cards "by the end of this year."  

    Continue reading

Biting the hand that feeds IT © 1998–2022