// Home / News

IBM touts "world's fastest processor" in $1m system

By Stewart Mitchell

Posted on 2 Sep 2010 at 11:40

IBM will start selling what it claims to be the world's fastest computer chip on 10 September.

The z196 processor is a four-core chip that contains 1.4 billion transistors on a 512mm-square surface and runs at 5.2GHz, which, according to IBM, will be needed to meet the demand from increased web-based banking and retail technology.

The mainframe processor also packs in 24MB of Level 3 Cache memory, making use of IBM's embedded DRAM (eDRAM) technology. The company says eDRAM allows it to place dense DRAM caches, or components, on the same chips as high-speed microprocessors to improve performance.

The chip is designed to sit in IBM's latest mainframe monster, zEnterprise System, which is in itself an impressive piece of kit.

The core server in the zEnterprise System - the zEnterprise 196 - contains 96 of the new microprocessors, and is capable of executing more than 50 billion instructions per second.

To put that in perspective, IBM said the zEnterprise 196 runs 17,000 times more instructions than the Model 91, the high-end of IBM's popular System/360 family, could execute in 1970.

More importantly, the company said the microprocessor technology included new software to optimise performance of data-heavy workloads.

IBM said the result was 60% improvement in performance in intensive database tasks and a 66% boost for Java workloads.

Anyone wanting to get their hands on this technology will need deep pockets, however, with the zEnterprise System expected to start at $1 million when it goes on sale next week.

Subscribe to PC Pro magazine. We'll give you 3 issues for £1 plus a free gift - click here

From around the web

User comments

While a 17,000 times increase sounds impressive, over 40 years it isn't actually that great. Taking the common misinterpretation of Moore's law, that compute power doubles every 18 months, as a guideline, we'd expect an increase of about 134,000,000 (2^27) times in that period. Obviously this is a horribly skewed back of a beermat calculation, but it illustrates the point.

By flyingbadger on 2 Sep 2010

Really?

Are you sure it's "512mm x 512mm surface" and not 512 mm2? It's difficult to imagine the cpu more than 20 inches wide.

Other than that, it would be interesting to see how it compares against much cheaper intel's X7560 :-)

By Lomskij on 2 Sep 2010

Obviously they didn't use the new chip to do the calculation

Assuming these chips can do one instruction per cycle (and if they are super-scalar that the minimum you'd expect) the 5.2GHz x 4 cores = 20 Bips per chip x96 chips = 2 trillion ips. If they are talking about 96 cores per server (24 processors) then its 500 Billion ips.

By milliganp on 2 Sep 2010

"Up to 96 cores segmented to up to 80 processor units, up to 14 system assist cores and 2 spares" - So it's actually closer to 400 billion, not 500 :-)

Anyway, I believe these cpu's are water cooled, no? Somehow I have a sneaking suspicion that similar cluster of xeons or magny cours would run linux at the same speed but several times cheaper.

By Lomskij on 2 Sep 2010

@ Lomskij

You're quite right - that would be quite a manhole cover.

By SMitchell on 2 Sep 2010

Leave a comment

You need to Login or Register to comment.