Intel's Larrabee: A killer blow for AMD


Recommended Posts

It's a silly sounding name, Larrabee. But it must fill AMD 's heart with terror. It's the codename, of course, for a whole family of new processors being cooked up by Intel . And it promises to add graphics insult to AMD's existing CPU injuries.

Frankly, things are bad enough for AMD already. Since launch last summer, the Core 2 processor has been pistol whipping AMD's Athlon CPUs into burger meat. Meanwhile, AMD's upcoming quad-core competitor, broadly known as Barcelona, looks like a pretty unambitious effort. It will certainly have to be some chip to take on Intel's upcoming 45nm die shrink of the Core 2 chip. Factor in recent reports of a launch delay for Barcelona and I'm beginning to get the fear about AMD's ability to compete.

Then there's the spectacular fashion in which the wheels have come off AMD's recently acquired ATI graphics subsidiary. ATI's all new flagship graphics DX10 board, the Radeon HD 2900 XT was very late, extremely underwhelming on arrival and possibly a bit broken. The midrange variants of the Radeon HD range don't look much healthier: they've been sent back to the fab for a respin. Not a good sign.

In that context, the emergence of the Larrabee project from Intel is just further proof of how far ahead of the game Intel appears to be at moment. For the uninitiated, Larrabee is an all new multi-core processor design that majors on floating point power.

The full feature set hasn't been revealed as yet, but an official Intel document turned up on a university website recently that reveals several fascinating new details.

Try these specs for size. Larrabee will be available in configurations ranging from 16 to 24 with clock speeds as high as 2GHz and raw performance in the 1TFlop range. The latter figure is approximately 40 times more than an existing Intel Core 2 Duo chip. Yup, you read it right. 40 times. And the first Larrabee chips are pencilled in for as soon as 2009.

Of course, floating point power is just one part of the overall PC processing equation - Intel will be retaining a conventional CPU roadmap for general purpose duties based on the existing Core 2 family.

But Larrabee will take Intel into brand new markets. Significantly, the document confirmed that a variant with full 3D video rendering capability is on the cards. As we reported earlier this week, the rumblings on the rumour mill suggest the chip could be a joint effort with Nvidia.

Either way, the most fascinating aspect of the Larrabee GPU is the expectation that it could be the first graphics processor to combine both traditional raster graphics with more advanced ray-tracing techniques.

Without getting bogged down in the details, suffice to understand that raster graphics are a bit of a kludge when it comes to simulating lighting. Ray-tracing is the real deal. Ask any 3D graphics professional what they think about ray tracing on GPUs and they'll tell it's a matter of when rather than if.

Of course, AMD and ATI will know perfectly well that ray tracing is the future. But what must be really worrying is that it presents Intel with the perfect inflection point to enter the graphics market. ATI and Nvidia have refined raster graphics to the point where other companies, including Intel, simply can't compete. But a new age of ray-traced graphics will level the playing field and might just hand Intel a chance for the total domination of the PC platform it so dearly desires. Jeremy Laird

Source

Larrabee on Wiki

post-1520-1182279078_thumb.jpg

edit-

so does this mean integrated graphics are going to own now?

Well, each new chip owns a previous generation chip :) So in a sence, they have owned all the time, but to answer your question, I think it means that Intel is very serious abou grabbing a GPU market share and with a chip this strong, I'm sure they will build a standalone video card :)

Ok, I really don't understand... Is it a GPU or a CPU?

The technology behind GPU and CPU has always been the same. In a way, a GPU has always been a CPU with graphics related instructions dedicated to graphics only whereas a CPU would do all the functions.

This is what makes it special:

Larrabee is different from its predecessors in that it uses a derivative of the x86 instruction set for its shader cores instead of a custom graphics-oriented instruction set, and is expected to be more flexible. In addition to traditional 3D graphics for games, Larrabee is also being designed for GPGPU or Stream Processing tasks; for example to perform ray tracing or physics processing, perhaps as a component of a supercomputer
I hereby dub thee...CORE 3!!!

Intel will probably name Wolfdale/Yorkfield/Penryn that ;)

Um, is this real? Lol :p

Yes it is :) Here is more info.

Its essentially mimics how current graphics processors work with a number of execution units executing code in SIMD fashion. So yes it can be considered a GPU oriented more towards general purpose computing.

glad i havent upgraded from s939 yet.

And AMD dying off would bring us back to the 80s and early 90s - most of you are too young to remember how hard Intel ****ed on us all back then.

No competition - no innovation - no choice - who loses? You do.

But equally.... Intel shouldn't be holding back in case the good products they come up with kill AMD. It's all on AMD's head to raise their game.

But equally.... Intel shouldn't be holding back in case the good products they come up with kill AMD. It's all on AMD's head to raise their game.

:yes:

Intel holding back would be the same as AMD going bust. AMD or someone else has to compete.

This topic is now closed to further replies.
  • Recently Browsing   0 members

    • No registered users viewing this page.
  • Posts

    • Micron reveals AI companies are spending billions to lock up its memory years in advance by Karthik Mudaliar The demand for more memory is far from over, and Micron is turning the AI-driven memory shortage into a much more predictable business. The company has revealed that it has signed 16 strategic supply agreements backed by roughly $22 billion in customer deposits and other financial commitments. The contracts cover DRAM and NAND deliveries over several years, with some running through 2030. With the AI boom, demand for high-bandwidth memory (HBM) has grown so quickly that large customers are now prepared to help finance future production in exchange for a guaranteed supply. According to Micron’s latest financial results, the company received commitments worth about $22 billion across its new agreements. Around $18 billion is expected to arrive as cash deposits, while the rest will come through other financial arrangements. Micron says the agreements could generate approximately $100 billion in future contracted obligations. They cover around 20% of its expected DRAM shipments and one-third of its NAND shipments during their respective terms. It should be noted that although AI infrastructure is the main force behind the current shortage, not all 16 agreements with Micron involve AI companies. Micron said the customers also include consumer electronics and automotive businesses, two sectors that increasingly compete with data centers for the same manufacturing capacity. HBM is consuming an increasing share of that supply. Unlike conventional desktop or server RAM, HBM stacks multiple memory dies vertically and places them close to an AI accelerator. This gives GPUs and other AI chips access to data at much higher speeds, but it also requires more complicated manufacturing and packaging. Micron says its 12-layer HBM4 memory is now shipping in high volume for a lead customer, with samples also supplied to other companies. The chipmaker has already generated more than $1 billion in HBM4 revenue and says the product is ramping twice as quickly as its earlier HBM3E generation. Samsung has similarly warned that the memory shortage could continue into 2027 and beyond. Consumer memory companies have also had to address sharp increases in DDR5 pricing, suggesting the effects are already reaching beyond the data center. For consumers, that could mean the AI memory crunch lasts longer than expected, even as manufacturers invest heavily in new production.
    • XnConvert 1.112 by Razvan Serea  XnConvert is a cross-platform batch image-converter and resizer with a powerful and ease of use experience. All common picture and graphics formats are supported (i.e. JPG, PNG, TIFF, GIF, Camera RAW, JPEG2000, WebP, OpenEXR) as well as supporting over 500 other image formats. Also available within the batch operations include rotating, adding of watermarks, adding of text along with many image-adjustment features such as brightness, shadows and more. Among the features included are: Batch adding of files and folders Support for drag and drop of files Batch rotating, cropping, resizing and more Adding of photo masks Preserving or removing image metadata in conversions Multipage image file support (i.e animated GIF, APNG, TIFF) Command line integration via NConvert Filters - such as 'Blur', 'Gaussian Blur', 'Emboss', "Sharpen' and much more Effects - such as 'Old camera' and much more Download: XnConvert 64-bit | Standalone | ~30.0 MB (Freeware) Download: XnConvert 32-bit | Standalone Links: XnConvert Website | Screenshot | Release Announcement Get alerted to all of our Software updates on Twitter at @NeowinSoftware
    • Microsoft updates Visual Studio Code with chat cost tracking and multi-agent chats by Paul Hill Microsoft has just launched Visual Studio Code 1.126, its latest weekly release. This time, the company has focused on letting you see the total cost of chat sessions to spot expensive conversations; enabling multiple chats per session that run side-by-side in one agent host Copilot session; and letting you browse new folders safely in restricted mode. We have now reached the stage where free AI in IDEs is coming to an end. To help you keep track of your costs, VS Code now lets you see the entire cost of a chat session, rather than just individual turns. This should give you more transparency about which sessions consume the most credits, so you can better manage your usage over time and spend less. For those of you using the Agents window, you know it is possible to run and manage multiple agent sessions at once. In this update, a Copilot session started from an agent host can hold several chats at once. Explaining how this feature works, Microsoft writes: Finally, from this update forward, Microsoft will remove the pop-up when opening an untrusted folder. When you open a new folder now, it will automatically open in Restricted Mode. You will see a banner that lets you manage the trust level of the folder. Microsoft has made this change so that it’s easier to start inspecting code without giving it trust right away. If you have VS Code, you can check for updates within the app now to get this new version. Otherwise, you can download it from the Visual Studio Code website.
    • Anthropic accuses Alibaba of using 25,000 fake accounts to copy Claude's capabilities by Karthik Mudaliar Anthropic has accused Alibaba of using nearly 25,000 fraudulent accounts to extract capabilities from Claude on a huge scale. According to a report from Reuters, Anthropic told US lawmakers that operators linked to Alibaba and the company’s Qwen AI team generated 28.8 million exchanges with Claude between April 22 and June 5, 2026. That is a lot of Claude conversations, but Anthropic says this was not ordinary chatbot use. The company believes the accounts were part of a coordinated effort to collect answers that could help train or improve rival AI systems. The alleged campaign reportedly focused on some of Claude’s most valuable skills, including software development, multi-step reasoning, and agentic tasks. In practical terms, that means getting an AI model to plan and complete work across several stages rather than simply answering a single question. This is called 'distillation,' where AI companies use outputs from a larger model to train a smaller and cheaper one. The smaller model learns to imitate useful parts of the more capable system without needing the same amount of computing power. The distillation process isn't automatically suspicious, but the problem comes when one company gathers another provider's outputs without permission and at an industrial scale. Also, this does not mean Alibaba obtained Claude’s source code, model weights, or original training data. Instead, Anthropic claims the accounts repeatedly asked Claude carefully designed questions and collected the answers. Those answers could then be used as training material for another model. Anthropic has made similar accusations against DeepSeek, Moonshot AI, and MiniMax earlier this year. As Neowin previously reported, Anthropic said those three companies collectively generated more than 16 million Claude exchanges through roughly 24,000 accounts. Anthropic says the new campaign produced almost twice as many exchanges in a matter of weeks. Anthropic reportedly told lawmakers that the campaign could help Chinese AI developers approach the capabilities of its Mythos Preview model. Mythos is focused on advanced cybersecurity work, including finding and exploiting complex software vulnerabilities. via Reuters | Photo via DepositPhotos.com
    • An Indian manufacturer that assembles roughly one-third of Apple's iPhones and supplies semiconductor components to Tesla confirmed Monday that attackers had stolen and publicly published a 630-gigabyte cache of confidential files — including engineering blueprints stamped "TRADE SECRET," a 52-page quality inspection document for iPhone circuit board components, and cryptographic certificates that security experts say could be weaponized in follow-on attacks. https://www.techtimes.com/articles/319019/20260624/apple-tesla-supplier-tata-electronics-confirms-630-gb-data-theft-iphone-specs-dark-web.htm
  • Recent Achievements

    • Rookie
      krychek57 went up a rank
      Rookie
    • Grand Master
      Jaybonaut went up a rank
      Grand Master
    • One Year In
      Philsl earned a badge
      One Year In
    • Dedicated
      Scoobystu earned a badge
      Dedicated
    • First Post
      Tom Schmidt earned a badge
      First Post
  • Popular Contributors

    1. 1
      +primortal
      441
    2. 2
      +Edouard
      176
    3. 3
      PsYcHoKiLLa
      133
    4. 4
      Michael Scrip
      79
    5. 5
      Xenon
      77
  • Tell a friend

    Love Neowin? Tell a friend!