IBM Reveals World's Fastest Microprocessor: z196

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.

Camikazi

Distinguished
Jul 20, 2008
1,405
2
19,315
[citation][nom]mikem_90[/nom]Big Blue Boasts big business brain for biotech? But biome brooding becons big-time bits too![/citation]
So how much spit is on your monitor after saying that?
 
G

Guest

Guest
Actually I work on Mainframes daily. They are excellent for very high I/O bandwidth that does NOT require any use of the main engine (engine is IBM's term for CPU). All the I/O work is handeled by a separate I/O subsystem so the actual Mainframe CPU only does real work. The CPU design is such that it is made from the ground up to run multiple concurrent workloads, hence the large caches. The Hypervisor is z/VM and it can host z/OS, z/VSE, other mainframe operating systems and LINUX!!

Many businesses I've seen has used the Mainframe to server consolidate about 40 cpu cores to one mainframe cpu. This saves a bundle in licencing and software costs.

Did I mention it runs Linux like a champ!
 

jecastej

Distinguished
Apr 6, 2006
365
0
18,780
Yummi!!! is goog to know this technology is possible, running and evolving if even on a few companies. Some day some of this may filtrate to our humble desktops. And again is not about the raw power. It is about what is possible only with this technology. From this point of view there is nothing more cost effective.
 

togenshi

Distinguished
Sep 18, 2008
47
0
18,530
[citation][nom]iam2thecrowe[/nom]so...all those specs are nice, but can someone give me a comparison?? faster than an intel 8 core??? 2 intel 8 cores?[/citation]

IBM hardware would destroy the Intel at in-order processing. Pretty much every bank in the world is run on IBM servers. Nothing can beat the reliability of IBM computers. Nor their support for that matter.

But you also pay in the ass for such luxuries.

If it were up to me, I would make sure that no business operates without a IBM server handling its specialised tasks.
 

vexun11

Distinguished
Dec 17, 2009
719
0
18,990
Codename Manufacturing Process Cores Transistor Count Die Size
AMD Phenom II X6 1090T Thuban 45nm 6 904M 346mm2
AMD Phenom II X4 965 Deneb 45nm 4 758M 258mm2
Intel Core i7 980X Gulftown 32nm 6 1.17B 240mm2
Intel Core i7 975 Bloomfield 45nm 4 731M 263mm2
Intel Core i7 870 Lynnfield 45nm 4 774M 296mm2
Intel Core i5 670 Clarkdale 32nm 2 384M 81mm2
AMD Phenom II X4 965 Deneb 45nm 4 758M 258mm2
 

g00ey

Distinguished
Aug 15, 2009
470
0
18,790
Intel's Xeon Nehalem-EX octocore has 2.3 billion transistors and Tukwila quadcore has 2 billion transistors, both put on 596 mm2 dies.

IBM's CPU may be very fast but I have a feeling that a well equipped server with two Nehalem-EX would do pretty much the same job if not even better than an IBM Mainframe at the fraction of the cost. There are many flavors of Linux, and then you have Free/Net/OpenBSD and Solaris. From what I know Solaris/OSOL has very good support for virtualization on different levels such as Zones and VirtualBOX. I believe Zones are what tantamounts to the virtualization capabilities of the z/OS.

But that's my two dimes....
 

besterino

Distinguished
Mar 19, 2010
16
0
18,510
Funny that certain companies spent so much for IBM hardware considering that - according to so many posters here - a couple of AMD / Intel chips would do the job so much better. Boy did these companies not do their homework right... and lucky IBM that apparently so many tech guys out there are either plain stupid or love IBM a great deal for spending (or having a company spend) so much money on stuff they do not need.

Irony off.
 

g00ey

Distinguished
Aug 15, 2009
470
0
18,790
But releasing a CPU claiming it to be the fastest in the world without releasing any kind of benchmarks that can be compared to other CPUs is naught but pure hypocrisy!
 

vexun11

Distinguished
Dec 17, 2009
719
0
18,990
[citation][nom]besterino[/nom]Funny that certain companies spent so much for IBM hardware considering that - according to so many posters here - a couple of AMD / Intel chips would do the job so much better. Boy did these companies not do their homework right... and lucky IBM that apparently so many tech guys out there are either plain stupid or love IBM a great deal for spending (or having a company spend) so much money on stuff they do not need.Irony off.[/citation]

Then why would it say that it's the worlds fastest if there are server chips with more transistors?
 

pcfxer

Distinguished
Dec 12, 2007
37
0
18,530
[citation][nom]phatboe[/nom]What exactly are these Z-Series mainframes used for and what OS is it running?[/citation]

IBM zOS, IBM RHEL (Red hat linux) and IBM's Virtual OS Host thing.

You can run java applications, COBOL (legacy mainframe stuff), databases, websites, EVERYTHING.

These mainframes are a thing that puny-brained desktop users can never fathom. One can have redundant processors set aside on each processor board or the processors can be assigned to specific tasks, like differing types of I/O channels and such.
 

blackened144

Distinguished
Aug 17, 2006
1,051
0
19,280
[citation][nom]iam2thecrowe[/nom]so...all those specs are nice, but can someone give me a comparison?? faster than an intel 8 core??? 2 intel 8 cores?[/citation]
I would be interested in seeing how it compares to the clusters we are running here at work. In 2 racks we run 160 servers - Quad Xeons with 12gb mem each, thats 640cores and 1.92TB of mem compared to the possible 96cores and 3TB RAIM from IBM.
 

spitcher115

Distinguished
Aug 10, 2009
3
0
18,510
[citation][nom]g00ey[/nom]Intel's Xeon Nehalem-EX octocore has 2.3 billion transistors and Tukwila quadcore has 2 billion transistors, both put on 596 mm2 dies.IBM's CPU may be very fast but I have a feeling that a well equipped server with two Nehalem-EX would do pretty much the same job if not even better than an IBM Mainframe at the fraction of the cost. There are many flavors of Linux, and then you have Free/Net/OpenBSD and Solaris. From what I know Solaris/OSOL has very good support for virtualization on different levels such as Zones and VirtualBOX. I believe Zones are what tantamounts to the virtualization capabilities of the z/OS.But that's my two dimes....[/citation]

I guess that's why all the major banks, insurance companies, telco's, etc... use IBM mainframes to do their daily computing for them and not server array's?

I'm a z/OS systems progammer for a telco here in Canada and we are currently running a z900 mainframe and we will be upgrading to a z196 in 2012. The shear bandwidth and I/O that a mainframe is capable of across multiple independant Logical Partitions (LPAR's) can be astounding. We run our telco on approx. 700 MIPS and the top of the line z196 can put out in the neighbourhood of 48000 MIPS (I am probably off a little here but IBM just gave us a presentation on Wednesday past and I know it blew my mind).

Do you know how much raised floor space/cooling/electrical is required in a datacenter for servers to replace one mainframe and DASD? With the growing cost of real estate this is a real concern for larger businesses. Not to mention the cost of downtime that can be associated with server farms unless redundant arrays and SAN's are used (we have greater than 99.99975% uptime on our z900, the industry standard is referred to as maintaining 5 9's). If you believe that 1 or 2 or 10 Intel/Unix/Linux bases servers can replace one of these then you need to do some more research and get the facts. Pundits have been reporting on the death of mainframes since the early 1990's and yet here we are...

 

g00ey

Distinguished
Aug 15, 2009
470
0
18,790
[citation][nom]spitcher115[/nom]... We run our telco on approx. 700 MIPS and the top of the line z196 can put out in the neighbourhood of 48000 MIPS (I am probably off a little here but IBM just gave us a presentation on Wednesday past and I know it blew my mind). ... If you believe that 1 or 2 or 10 Intel/Unix/Linux bases servers can replace one of these then you need to do some more research and get the facts. [/citation]
I'm sure that the IBM mainframe does a good job but compared to what? Does the performance justify the price or am I better off buying a much cheaper server from a competitor and run an emulated mainframe if I need to?

MIPS is a terrible way of measuring performance which is rarely used nowadays but if we take your claim of 48 000 MIPS I'm not impressed and here is why: A single-core Pentium 4 of a fairly new generation (supports x86_64) operates at about 10 000 MIPS. After looking into different benchmarks here at Tom's I estimate that a modern Core 2 CPU does twice that per core i.e. ~20 000 MIPS. This means that a modern average consumer-grade Quad core CPU from Intel does ~80 000 MIPS altogether (4x 20 000) which is almost twice as much as the z196.

I know that the IBM mainframes historically have been known for good I/O performance but hardware changes and IBM today may even have been overtaken by its competitors. It is also a fact that most modern Linux, *BSD and Solaris based servers have very good uptimes nowadays.

The purchase decisions of those companies using IBM mainframes may be based on lobbyism or what is known as "consumer-loyalty". As a buyer I want the facts and there are plenty of good benchmarking tools specifically designed for measuring server related performance traits such as IOPS and transactions etc. Just look at spec.org, use their benchmarks and lay out the facts.

I have a hard time finding true facts that talks in favor of IBM mainframes so I would say to anyone representing IBM; "Please enlighten me instead of b*llsh*tting around about how fast the z196 CPU is!".
 

pozaks

Distinguished
May 12, 2010
89
0
18,630
So with all this dismissal of running a VMWare cluster being in any way comparable, you IBM guys know that you can do live logical and storage migrations and alter the CPU/RAM of a live partition now, right? Some of your assumptions seem to be implying that we're still on esx3.5. Just how much benefit are you getting over a few bladecenters on netapps w/ redundant FCOE, which would still be an order of magnitude cheaper and roughly the same physical footprint that your z-beasts?
 
Status
Not open for further replies.

TRENDING THREADS