Michael Dolan Dot Com

Monday, January 14th, 2008

Toshiba Spurs Engine: Cell processors in laptops and TVs … could it be an x86 and Power/Cell race in the datacenter?

There’s always a flood of news coming out of CES, but one that I find interesting that also impacts the server and technology market is a bit of news from Toshiba. As this video shows, Toshiba has embedded a Cell BE processor (yes, the same ones in PS3s) into a laptop and a television. The reason is that the Cell processor is far more capable of handling video rendering and stream computing than a typical general processor such as you find from Intel or AMD.

Now you may watch the video and then wonder ‘what the heck does this imply for the server market?’ and you’d be right to ask. Interestingly, Scott Handy from IBM recently did an interview with The Register and gave everyone the answer. The processor marketplace for desktops, laptops, mobile devices and servers requires huge volumes to provide economic returns to manufacturers. There is a huge capital investment that goes into every generation of processor not to mention all the IP that must be created to advance from one generation to the next. Similar to software sold on CD, producing the first chip may cost $1B and the second chip $500M, and it only gets cheaper as your volumes increase. In order to maintain a processor business profitably, chip makers live off volumes (hence the huge battles between Intel and AMD). Now the problem facing Intel for Itanium and Sun for SPARC is that these server technologies have no consumer application that will drive additional volumes. In terms of volume, the server market is a spec in chip consumption.

Yet the processor market is a rat race – vendors need enormous chip volumes to provide funding for future chip designs and whoever has the most funding, tends to produce the winning chips (see Intel for years despite competition from AMD). In fact one might point to Itanium as Intel’s decline – a chip that never saw the volumes needed to make it a competitive chip. Then as Intel was divesting profits from its x86 chips into Itanium chips, it suddenly found itself behind AMD. For a good grin, take a look at IDC’s forecasts for Itanium and the ‘revisions’ to those estimates. Heck, probably the only person to have worked around Itanium and advanced their career was Ashlee Vance.

Now look at Sun and SPARC and you see what used to be a high volume SPARC or UltraSPARC platform with many 1-4 way SPARC servers going into every datacenter giving Sun ‘good enough’ volumes to invest in future generations of processors. Then comes along Intel and Linux and Windows wiping out low end SPARC volume servers and cornering SPARC processors into a lower volume mid-high end server market. Without the funds for future generations of chips, Sun has to resort to stopgap measures craftily using old SPARC core designs in a Niagara multicore processor configuration. Regardless, that still doesn’t create high volume – the non-x86 server market is actually relatively low volume.

When you look at Power processors in everything from Mars rovers to cars, servers, printers and today it’s used in all 3 gaming systems (Xbox 360, PS3, and Wii). The gaming market alone consumes tens of millions of processors (far more than the server market). The Cell BE processor in the PS3 and IBM or Mercury Cell blades actually have 1 power core (PE) and 8 synergistic processing units (SPEs). The PE acts as a ‘router’ for instructions and data sent to the SPEs. So an application developer write much of the code similar to a power architecture when targeting a Cell based system. (Now I’m oversimplifying processor architectures a ton here, I’ll admit – but the point is not to debate the differences between PowerPC/POWER and Cell PE/SPE units… look at the big picture).

Now, going back to servers you can understand that winning in servers requires a high volume chip that funds future chip development and generates economic profits for the manufacturer. Without funds to invest in future chips, you lose the race. Once you view it all in this light, it becomes easier to see why the Power, ARM and Intel/AMD x86/x64 architectures are far ahead of other RISC or other architectures – they have volume. POWER6 was no small ‘speed bump’ – POWER6 is a fierce chip with raw compute power and system throughput that leapfrogged its competitors. It took economic profits reinvested into chip design over years to create it. Cell is no different. And with Cell and Power having volumes in non-server markets including now potentially Toshiba laptops and TVs, I’ll leave you to place your bets on which architectures will be around in the server market for years to come.

Oh, and I should also note that Linux is the only OS that runs on both the Cell BE and POWER processors ;-) Guess which OS is likely to be around in your next generation server, television, printer and … well if only we could ‘fix’ the desktop/laptop market reliance on a different OS…