back to article IBM's z12 mainframe engine makes each clock count

When you charge hundreds of thousands of dollars per core for an engine that is designed to run full-out all day doing online transaction processing and all night running single-threaded big batch jobs, you have no choice but to believe in higher clock speeds and doing anything and everything to boost that single-thread …

COMMENTS

This topic is closed for new posts.

Difficult

"...The top-end EC12 has z12 engines that are 25 per cent more powerful, at around 1,600 MIPS..."

With all those IBM engineers struggling to build a new high performing Mainframe cpu, it is a heavy burden to try to keep up with the old back compatibility from 1960 or so. That is maybe the reason the new Z12 Mainframe cpu gives only 1,600MIPS? And the previous IBM Mainframe z196 cpu from 2011, was dubbed the fastest cpu in the world, by IBM. Thus, the Z12 which is faster. should cement the IBM claim "worlds fastest cpu" even more. Probably, IBM will soon claim the z12 is the fastest CPU in the world.

http://www.engadget.com/2010/09/06/ibm-claims-worlds-fastest-processor-with-5-2ghz-z196/

An old Intel Nehalem-EX cpu gives 400 Mainframe MIPS - under software emulation using "TurboHercules". Software emulation is 5-10x slower than running native code. If Nehalem-EX could run IBM Mainframe software natively, it would give 2000-4000MIPS. This is faster than the "worlds fastest cpu" Z12. Here are sources:

http://en.wikipedia.org/wiki/Hercules_emulator#Performance

Thus, if you had 20 of the old Nehalem-EX, they would give 40,000-80,000 MIPS, matching the biggest z12 Mainframe. For a fraction of the price. Of course, the IBM Mainframe have better uptime and better I/O - no doubt. But cpu performance lags behind, even though they are "the worlds fastest cpu".

1
7
Anonymous Coward

Re: Difficult

So with a Hercules Emulator, a measly 200 socket Nehalem PC and a little help from Kebabbert, you have almost got yourself a mainframe.

IBM must be very afraid.

3
1
Anonymous Coward

Re: Difficult

Your last sentence said it all. Mainframe is not built for raw CPU. It is an I/O monster for bank transactions and the like. IBM makes supercomputers if you want to crack the human genome. Buses are slow and Ferraris have horrible seating capacity. Pick the tool for the job.

4
0
Pint

Re: Difficult

Exactly!

I've worked on midrange boxes for a while and they are infinitely superior in terms of I/O and uptime than any x86 based machine. They are just workhorses. Couple with this the software architecture of the applications that run on these beasts and you have a platform designed from the ground up to perform data processing - which is what they're meant for. Our machine can support thousands of users, run ERP, crunch data like no tomorrow, serve web pages and applications, work as a MTA, do file and print services, etc. etc. all at once. Yes, you may use a 5250 terminal to administer and get down and dirty with it (there is a GUI but I'm a die hard!) but it works. All the time. No skipping a beat, no coughing and spluttering. Just working.

You can just can't compare a spanner and a screwdriver - they're two completely different tools!

And no, you probably couldn't play Crysis on it!

1
0
Anonymous Coward

Re: Difficult

"I've worked on midrange boxes for a while and they are infinitely superior in terms of I/O and uptime than any x86 based machine."

Agree, all of the x86 vendors and Intel go on about their amazing CPU (which is actually not that amazing), but the problem is that very few applications are CPU dependent. Most needs tons of cache and I/O (DB, application servers, etc). That is where mainframe and mid-range, Power Systems, are golden. They are also just tanks.

0
0
Facepalm

Re: Difficult

So with a stack of 9V batteries and a few belt grinder motors and a little help from Adam Savage, you have almost got yourself a hybrid car. Toyota must be very afraid.

Anyone see the problem here?

0
0
Anonymous Coward

Re: Difficult

#So with a stack of 9V batteries and a few belt grinder motors and a little help from Adam Savage, you have almost got yourself a hybrid car. Toyota must be very afraid.

Anyone see the problem here?#

Yes. Faulty sarcasm detector.

0
0
Thumb Up

Re: Difficult

Oh. Good then. Seen too many PFKs making comments like the original and being serious!

1
0
Anonymous Coward

Mainframe doesn't cost "hundreds of thousands per core"

List on the entry z114 is about $100,000 for the hardware. When you add all of the software, then mainframe might be hundreds of thousands per core... but x86 is in that range too after the OS, VMware, Oracle DB, management utlities, replication, etc licensing. A z strictly with RHEL IFLs is really inexpensive. Do the research.

0
0
Anonymous Coward

Combined heat and computing

Frankly, I don't understand the mainframe racket. It makes very little sense for someone with a general computing background. And I get the feeling IBM keeps it that way deliberately. By not giving numbers, by giving confusing numbers (those MIPS figures seem absurdly low, but maybe everyone else is giving absurdly high figures instead, who knows), by being deliberately obtuse, by not allowing anyone to compare chips, or even by fudging the materials they do give out. Like those pretty pictures. A 96kB cache just about the same size as a 1MB cache? Or even the year numbering in that other graph. Reminds me a teensy bit of the predicted itanic sales graph, though that was different, like a few shades more epically spectacular in its wishful thinking. Anyway. I get the feeling that since IBM's mainframes are so obviously something to be sold to large enterprise managers, who as a rule have no clue whatsoever on technical things, any technical things, take your pick it doesn't matter they won't have a clue, that IBM can go full-out in car salesman mode (not 'used', no, this stuff is new, after all) and prattle happily about impressive-sounding numbers regardless of their meaning and their marks will happily sign up for another fleecing. And what a fleecing it is, they've certainly cornered their market.

I'll give them that not merely delivering the latest on time, but ahead of schedule, is that much more impressive. Of course, they've been doing this for a while and they know exactly what their customers want, so hey, why not give it to them. Then crank the price a bit. Sure, why not.

This isn't to dis their technical prowess. They've successfully prevented me from getting a handle on that, so I can't comment. I can comment on what they do show, and, well, there's dogs and ponies in it.

0
5
Silver badge
Paris Hilton

What are you getting at here?

> those MIPS figures seem absurdly low

Could be because they are S/360 instructions or some other old CISC instruction set

> A 96kB cache just about the same size as a 1MB cache?

Could be because the transistor count per cell is way higher for L1 than L2

> the year numbering in that other graph ... wishful thinking

It's a picture of the past, not the future

> IBM can go full-out in car salesman mode

IBM sells well sure. If the buyer is ready to pay the price because he cannot or won't deal with commodity hardware, what's the problem? Faster machine keeps your existing software infrastructure running and that's worth more than the pricey upgrade in most cases. If you want to have x86 you know where to find it, but chances are you don't want it in the first place.

As someone once said,

"Commodity hardware is like violence, or XML - if it doesn't solve your problems, you are not using enough of it"

3
0
Anonymous Coward

What I'm getting at here?

IBM is being deliberately obtuse and opaque. Their numbers don't make sense to me. This annoys me.

I have no use for your "could be"s. You don't need to repeat to me what I just observed. I'm not interested in x86. I'm not interested in your platitudes. I want to know what this is about. And it's not telling me in a way that makes sense to me. I can't relate it to what I already know without going "that's not right". This again annoys me.

0
0
FAIL

Re: What I'm getting at here?

I'm afraid the fact that you don't understand it fails to prove that it doesn't make sense. But that doesn't seem to stop you from at least implying that anyone who buys a mainframe is ignorant. Suggest you do some research beyond your x86 world until you do understand: then you'll have something meaningful to say.

In other words: there are lots of us who *do* understand it, and *can* prove the value. The information is out there. And no, IBM isn't "deliberately obtuse and opaque": you're just not well-read enough to understand. Sorry.

0
0
Anonymous Coward

"Commodity hardware is like violence, or XML - if it doesn't solve your problems, you are not using enough of it"

For performance, yes... for reliability or manageability, not really.

0
0
Anonymous Coward

Re: What I'm getting at here?

"I want to know what this is about."

Short version: http://searchdatacenter.techtarget.com/feature/Traditional-mainframe-systems-vs-x86-software-mainframe

Long version: http://www.redbooks.ibm.com/Redbooks.nsf/RedpieceAbstracts/sg248050.html?Open

0
0
Silver badge
Happy

I have to wonder

Could you fry an egg on those chips? I like mine with crispy bacon (and "sunny side up", as they say in the US).

Oh bloody hell, I am hungry now.

1
0
This topic is closed for new posts.

Forums