IBM z Processors Climb to 5.5 GHz
More GHz, please.
The core of IBM's venerable mainframe architecture, the z196 chip, is currently shipping as a 5.2 GHz quad-core processor. The next product generation will apparently climb to 5.5 GHz, according to a report published earlier this month by the Wall Street Journal.
There was no clock speed information on the next Power chip, currently called Power7+. Power7 runs at up to 4.14 GHz today and IBM says that the next generation will be 10 to 20 percent faster and it is more than likely that a slight upward adjustment of the clock speed will arrive as well.
IBM's zEnterprise servers have a starting configuration price point of about $75,000 for the z114 (up to 14 processors at 3.8 GHz) and scale into the hundreds of thousands of dollars for z196 systems with 24 processors/96 cores. A single 5.2 GHz quad-core z196 processor, called central processing complex (CPC) is rated at a power consumption of about 300 watts.
In comparison, a 4.14 GHz Power7 chip is rated at 190 watts thermal design power.

It's like in cars or pretty much any other machine... the higher in performance you get the more unproportional the required power goes.
Like making a car go from 0 to 40... any small engine can easily do that. Make it go from 160 to 200mph takes some major hardware upgrade :-)
Bravo, IBM.
...
Contribute to global warming much? I though performance per watt was supposed to go up not down
But we all wish it could run Crysis! =D
It's like in cars or pretty much any other machine... the higher in performance you get the more unproportional the required power goes.
Like making a car go from 0 to 40... any small engine can easily do that. Make it go from 160 to 200mph takes some major hardware upgrade :-)
That is only for a few specialized tasks. For most computing you need a CPU. Otherwise why would anyone give a rip about Intel and AMD CPU's? We'd just get an Atom and only worry about which nVidia or ATI GPU we have.
Depends entirely on the task. Also, Kepler doesn't have great CUDA performance.
Even if that were the case, the companies that still use these do it because porting the programs that run on these machines and transitioning the infrastructure would be really expensive.
In the meantime please feel free to hum a little ...
Bravo, IBM.
You could throw anything at those cards, but much of it would run slower than it does on your $1000 commodity PC. GPU acceleration is great for certain specific tasks, but at GP-CPU type tasks they would be VERY inefficient. Big Iron is still a player for a lot of reasons, code base is one but not the only one... the architectures of those platforms make them good at their market focus areas. Sure you can cluster smaller servers for things like big DB apps... but by the time you get done putting a comparable cluster together you're may find yourself in a similar price point. $75k isn't a lot of money for an enterprise platform (though I'd be REALLY surprised if you could actually get a Z system for the 'base price' and be able to do much with it).
The purpose of a Mainframe is to be extremely reliable while performing high IO transaction processing tasks. They are not designed or meant to run simulations.
Mainframes and Supercomputers are two different things