11 Comments

Summary:

Tilera, one of many companies trying to build specialty chips or systems for cloud and web-scale computing, received a strategic investment today from Broadcom. But even as the investment validates Tilera, does the cloud need its own specialty chips and gear?

Intel's single chip cloud computer

Tilera, a startup building chips that contain anywhere from 16 to 100 cores, said today it’s raised $25 million in a third round of funding from investors including Broadcom. Chips made by Tilera, which we named as one of five multicore statups to watch two years ago, are aimed at boosting performance and energy efficiency for networking and cloud computing, which is likely why Broadcom invested. But as Tilera spends more time emphasizing the cloud and big players like Intel do the same, we have to ask: Do cloud computing and web-scale computing need their own chips?

Broadcom likely wants an edge should Tilera’s multiple RISC-based (rather than Intel’s x86) processors set fire to the cloud computing world as equipment companies attempt to develop power-efficient chips that can be adapted to specific workloads. For Broadcom, an investment in Tilera is a direct challenge to Intel’s dominance in the data center computing space, as well as a bet on faster networking chips.

Tilera has advantages in cloud computing because its chip architecture allows for a lot of lower-power processors to talk to one another using an interconnect technology that doesn’t cause bottlenecks. In plain English, Tilera has figured out a way to get a lot of cores to talk without having to pause to listen to one another, which slows things down as you add more cores. A Tilera executive told me last year that if just 10 percent of cloud computing or web-scale customers took a chance on the startup’s architecture, it could succeed.

But while Tilera, which started developing its chips in 2004, may have the lead when it comes to building massively multicore chips with a mesh-interconnect, Intel smells an opportunity as well and as such is building out what it calls a “single-chip cloud computer” with 48 cores for the cloud computing market. There are also systems vendors trying to solve similar problems for those needing energy-efficient web-scale computing, such as SeaMicro and Smooth-Stone.

A key problem in all of these endeavors is figuring out how to get the multiple chips or cores to function together in such a way that performance scales linearly with the addition of each new core rather than tapering off as the communications between the cores or chips becomes overloaded. Intel and Tilera are hoping to do this on the chip itself, while systems vendors are trying to do it with a better box.

Related GigaOM Pro Content (sub.req’d):

We Can Call It A Cloud, But It’s Still Hardware

  1. I think the interesting question here isn’t really about the specific details of their technology… it is about the dynamics of the marketplace. Transmeta and others have attempted the “we have a better architecture for the future that is drive by ” and they have all failed. Is that because Intel is just better? Because they have so much money and marketplace power that they can prevent anyone else from getting traction? Because no one trend is so overriding that it completely changes the processor marketplace? I’m not sure – but we haven’t seen anyone ride one of those trends to success…

    Share
  2. Intel has made some attempts to gain inroads in the line-speed network processing market with Nehalem-based platforms, but ultimately their general-purpose processors just don’t pack the throughput punch necessary for this sort of application. Tilera’s main threats are Oracle (nee Sun)’s CoolThreads processors and Cavium’s Octeon (at the low end), along with a few other specialized processors. IBM’s recently-disclosed PowerPC-based network processor and Intel’s “single-chip cloud” also play in this space; if either of those makes it to market, it could put a significant dent in Tilera’s ambitions.

    Tilera does seem to have ambitions in the broader network application market, but these ambitions are probably doomed from the get-go. The advantage of a commodity architecture is the easy spring-board for development. Few cloud service providers will stop to rearchitect their application for maximum throughput after it hits big; rather, they’ll look for more effective ways to scale their existing architecture. If Tilera really does want to play here, it should offer a low-cost developer kit to enable startups to target applications to their processor. Otherwise, they’re depending on businesses to make the leap to depending on their platform without having had the opportunity to discover the advantages in a low-risk environment.

    Share
    1. Your last point about attracting startups to architect their apps to Tilera before they get big is a good one. Thanks!

      Share
  3. Lina Inverse Monday, March 8, 2010

    Intel’s “single-chip cloud computer” is more of an experiment that’s using the term cloud for its current hype. The chip has no provision for cache coherency and they’re only fabricating about 100, which will find homes in universities and research labs where we will see if this approach is useful.

    Share
    1. The lack of cache coherency isn’t necessarily a deal-killer in this application, provided that there are still high-speed message-passing interconnects on the board. Tilera also encourages the use of their built-in message-passing fabric for application.

      Share
  4. If anything, the cloud needs cheaper and (even) less power hungry chips. And not necessarily more cores. The Cell processor is out there for a while now and with few takers (in the cloud space).

    The big server farmers (Google, facebook, twitter, ..) happy with their versions of Map-Reduce, probably prioritize power bill reductions. I don’t mean they wouldn’t want cheaper hardware, but then it really needs to be cheaper, while still being “commodity” for the software they need to run.

    On the consumer devices side, at the risk of being proven utterly un-imaginative, the question that begs to be asked is – why would anyone need 48 cores? That brings us back to the classic problem of parallel computing – Is it (still) worth it?

    Share
    1. I’ve looked at clouds from both sides now,
      From up and down, and still somehow
      It’s cloud illusions i recall.
      I really don’t know clouds at all.
      LOL sorry Cheese I couldn’t resist

      Share
  5. Tilera has figured out a way to get a lot of cores to talk without having to pause to listen to one another, LOL women perfected that concept a long time ago :P

    Share
  6. [...] IBM made to push it into the HPC market as well as into the data center, but it turns out the market didn’t want the Cell [...]

    Share
  7. [...] IBM made to push it into the HPC market as well as into the data center, but it turns out the market didn’t want the Cell [...]

    Share

Comments have been disabled for this post