IBM Power 6 - A Winning Or Losing Strategy?
On the way to a meeting this moring, I was reading the FT. I always read all the IT and Biotech/Pharmaceuticals stories, so I was interested to see a story talking about IBM’s new Power 6 chip - due to be unveiled today. I hadn’t heard much about about IBM’s Power roadmap, so the story was a surprise. Contrary to what others in the chip industry have been doing of late, IBM has not only embraced multi-core in Power 6; they’ve also wound up the clock-speed too. Up to 6GHz in the lab.
The story in the FT left me with more questions than answers (it could be a fantastic chip, or it might not be - I couldn’t tell from the article). But there’s no doubt about it: IBM now has a rather different strategy to its competitors, such as Sun, in continuing to invest in increasing clockspeeds. As I read the FT story, this quote from a recent post on Sun CTO, Greg Papadopoulos’s blog came to mind:
Now here we are 20 years later. We have squeezed all of the blood from that stone. We’re done. Actually, we over-did it. Continuing to throw transistors at making single processors run faster is a bad idea. It’s kinda like building bigger and bigger SUVs in order to solve our transportation problems. As I said, a bad idea.
The point is - Sun sees an advantage in winding back the clock speed on its current chips (UltraSPARC T1 - previously known as Niagara) and future (Rock) multi-core chips. IBM sees an advantage in winding it forward. Time will tell who has the winning strategy (and the better technology)…
Asam Bashir wrote:
I’d like to see some estimates for power consumption when they are actually released. Think IBM counting on a niche market where power and it’s cost is not an issue. The G5 was a fantastic chip, but limited in application because of these considerations and it was because of this Apple switched.
You can make an analogy between the internal combustion engine and CPU’s. In the past it was always about adding more cylinders and valves to get more power out of an engine, now it’s very much different and modern engines use much more sophisticated technology like variable turbo chargers. Also in the same way horse power is meaningless without factoring the weight, Ghz is so old news compared to watts/cycle.
Posted 09 Feb 2006 at 3:03 pm ¶
simon wrote:
Oh yes - this is clearly a server chip - nothing to do with laptops. Nothing there that would have been of interest for Apple.
Competition for this chip, when it’s released in 2007, and rolled into product lines through 2010 will be offerings from Sun, and possibly (I guess) AMD. It may be that IBM has done something great with this chip. Or it may be that they’re being left behind as computing moves away being based around the microprocessor, and towards “systems-on-a-chip” i.e. complete multi-processor servers built on single pieces of silicon.
Posted 09 Feb 2006 at 5:09 pm ¶
Asam Bashir wrote:
Nothing for Apple? You forget Apple has been managing to sell a lot of its Xserve units which currently use the PPC G5 - This should be the last product that is supposed to be transitioned to Intel but that’s not certain anyway. Mac OS X will be built universal for the next few years so could easy use either x86-64 or a possible G6 (based on Power 6)
http://apple.com/xserve
Posted 09 Feb 2006 at 10:45 pm ¶
simon wrote:
Apple isn’t a serious player in the data center. I remember when Apple first launched Xserve years ago. Even Steve Jobs was humble. I don’t remember his exact words, but they were something like, “I know we don’t know much about the needs of the data center” And he was right. Although, they do look nice and shiny, stacked up in network cabinets.
However, the Xserve sure was a much needed offering in companies with Mac-based desktops. Prior to that, people had iMacs sitting on shelving in machine rooms, being things like mail servers.
Apple only has entry-level class servers. Not many companies are going to base their IT strategy around a server hardware/OS platform that: a) doesn’t get beyond “entry level”; b) isn’t a first-tier development platform for *any* important server-side piece of software; and c) doesn’t have amazing support options.
The point is: for entry-level, Apple is far more like to succeed at the if they use so-called “industry standard” architectures. I wouldn’t be at all surprised if Apple’s Intel-based Xserve hardware products didn’t turn out to be certified to run both MacOS X *and* Windows…
Posted 09 Feb 2006 at 11:13 pm ¶
Asam Bashir wrote:
The Virginia tech cluster, number 3 in the world for while, wasn’t a bad start, lots of other Mac OS X based super clusters since then. They’ve also been making a lot of progess in the life sciences with the bioinformatics cluster package they put together, http://www.apple.com/xserve/workgroupcluster/bioinformatics.html, also even Sanger centre for example uses Xserve storage systems. Take a look at the Xgrid technology sometime, http://www.apple.com/server/macosx/features/xgrid.html - drag and drop for massively parallel computation.
Posted 10 Feb 2006 at 5:12 pm ¶
Asam Bashir wrote:
Oh, and don’t forget the dedicated quicktime servers for broadcasting and streaming….
Posted 10 Feb 2006 at 5:12 pm ¶
simon wrote:
I don’t think Apple is anywhere in the compute farm business - has Pixar even bought an Xserve render farm yet? Don’t IBM, HP and the commodity Intel suppliers own this space?
You raise an interesting point about IT in the Life Sciences. I’ll do a separate blog entry on that.
Posted 10 Feb 2006 at 8:02 pm ¶