Google beats chip and server makers to the future
Who needs oil and gas when you have text ads?
Analysis Google's orgiastic, eccentric acquisition of start-up PeakStream must scare the major players in the server processor and hardware universe. An ad broker has eaten a potentially super-valuable, industry-wide asset with no greater ambition than self-gratification in mind. As a result, high-end server applications could hobble along for years to come.
The bleak scenario presented above goes too far, if you're wedded to near-term thinking. PeakStream, a two year old start-up, made software tools that helped ease the flow of single-threaded software across fancy, multi-core parts such as today's x86 chips from Intel and AMD and tomorrow's GPGPUs (general purpose GPUs) from Nvidia and AMD/ATI.
Advocates of the technology portrayed PeakStream and still independent rival RapidMind as godsends that would allow developers to continue coding in single-threaded models instead of enduring the arduous task of crafting parallelized code. Meanwhile, critics complain that the technology exists as little more than a short-term fix for a much larger problem and relegate such tools to the "niche" high performance computing market.
Over the past two months, PeakStream held talks with a number of vendors to discuss a possible acquisition, according to our sources. We don't know all the exact companies PeakStream talked to, but it's safe to assume the list included folks such as Intel, AMD, Nvidia, Sun Microsystems, IBM, HP and Microsoft. The chip and software crowds seemed the most obvious buyers for PeakStream's technology since they could use the tools to help out the widest possible audience.
Instead, the PeakStream engineers who made the transition to Google will now help Google.
Our most recent chats with relevant parties confirm that Google has no intentions of selling the PeakStream tools to chip, server or software makers. Of course, you knew that already, since Google is primarily in the ad and search business.
In addition, we're hearing that Google has little more than passing interest in crafting code to run on GPGPUs - the most immediate promise of an independent PeakStream's technology. Rather, Google ate up the PeakStream talent to develop better multi-threaded code capable of traversing x86 chips.
We've managed to track down a number of former PeakStream employees now at Google. One of the new employees worked on a just-in-time (JIT) compiler for multi-core x86 chips while at PeakSteam. Another employee, an ex-Transmeta fella, also worked on compilers for x86 systems, while another worked on a random number generator algorithm for GPUs and CPUs with a special emphasis on high performance computing software.
Google's acquisition of such talent provides an impressive example of how smart the company seems to be and how it will continue to outflank the likes of Yahoo! and Microsoft. Seriously, can you imagine Yahoo! or Microsoft having the foresight to acquire a company like this in order to improve their search and file systems? It would never happen.
Well beyond the search set, Google now shakes up the hardware industry as a whole on an unprecedented scale.
For example, many of you have likely heard that Google consumes the most processors and hard disk drives after the top server vendors. Google uses the components to craft its own servers rather than buying gear on the street like other red-blooded service providers. In addition, it has crafted so-called White Trash Data Centers of its own where shipping containers are filled with servers and transported around the world, while the likes of Sun Microsystems and Rackable try to turn this model into an actual business. Lastly, we understand that Google plans to make its own switches and other networking gear.
And now Google buys pricey software tools that seem like they might be handy. Were Google so flush with cash five years ago VMware might have turned into a helpful Goobuntu partitioning add-on rather than an industry standard piece of software driving data center spending. In another year, Google may well purchase Sun just to add some Java talent or gobble Red Hat because it needs more Linux support staff.
It's easy to make too much of the PeakStream buy, trust us, but the reality is that we're left with just one real player in the single-threaded-to-multi-threaded code market - RapidMind.
Ideally, server customers would enjoy the presence of at least two, competing start-ups. PeakStream and RapidMind could have pushed each other to develop better products and to nurture a very immature market. The end result may have been a plethora of HPC applications tuned for GPGPUs and multi-core chips. We'd find more oil and gas; we'd create safer cars and we'd fatten the pockets of Merrill Lynch's traders.
Instead, server customers need to pray now that RapidMind can do most of this work on its own and avoid being acquired by another greedy vendor. It would sure help too if an engineer somewhere could create a magical machine capable of teaching thousands of developers parallel programming methods overnight.
If you think we've gone too far, just think back to this article when you're paying $10 per gallon for gas in 2009 because Exxon has failed to find any new reserves. You can have a pained giggle too over that stack of worthless GPGPUs in the corner of your data center.
Enjoying a quick search on Google may help ease the situation, but then again it may not. ®