IBM's z17 mainframe – now with 7.5x more AI performance

IBM's latest mainframe builds on the platform's traditional attributes of security and reliability for mission-critical workloads, adding AI to support large language models (LLMs), assistants, and agents. ibm newest mainframe system ibm z17 The new z17 mainframe system is available from June The z17 family introduces an …

[CowHorseFrog]

Why would you run AI models on a Z17 when you can buy multitudes of X86 servers instead ?

[A Non e-mouse]

'Cause your data is already in the Z17 and mainframes are really good at pumping data around and manipulating it.

[CowHorseFrog]

Dollar for dollar im pretty sure you can buy a thousand pcs that will kill that mainframe.

[Anonymous Coward]

Because you're a mainframe sycophant and think "transaction processing" actually means anything.

In other words, because you're an idiot.

[I Am Spartacus]

Because you want to do work

You would use this because you want to have a scalable, supported, fast system to run your AI models on. That is, you want to do real work.

As opposed to designing, sourcing, assembling, installing, debugging your own hybrid cluster, that will almost certainly be slower.

[UnknownUnknown]

Re: Because you want to do work

I’m guessing IBM won’t let you plug in a NVidia Blackwell into any of the z17’s 48 PCIe slots ?

[hoofie2002]

Re: Because you want to do work

It's just the usual morons who are just clueless on anything beyond a desktop.

[Anonymous Coward]

Yes, and

Can it predict the demise of IBM?

[An_Old_Dog]

*Potentially* Useful

If you have an application (I'm using the general term here, not the modspeak synonym for "computer program") which is a good fit for LLMs or gen AI in an IBM mainframe environment, this looks potentially-useful.

I don't want an "AI journey", which sounds like buying a bunch of AI-related hardware/software/models/services, then wasting time in meetings where people have "visions" and otherwise fumble around trying to figure out how to use this stuff.

Here's my "AI experience": I went to DuckDuckGo and searched for "mousing gloves" (WTF, right? But I've seen stupider things, and I'm searching to figure out if these silly things actually exist.)

DDG found zero hits, then asked if I wanted to try their AI to search for them. I clicked Yes, and got this result:

"Mousing gloves are specialized gloves designed to enhance comfort and reduce strain while using a computer mouse. They are often made from lightweight, breathable materials and may feature padding or ergonomic designs to support the hand and wrist. Mousing gloves can help prevent repetitive strain injuries, such as carpal tunnel syndrome, by promoting a more natural hand position and reducing friction between the hand and the mouse. Some users find them particularly helpful during long periods of computer use, such as gaming or office work."

Yeah, right.

LLMs/gen AIs are utensils -- like forks, if you will. But too many companies expect people to use them to eat soup*.

*I'm not criticising DuckDuckGo. They offer their LLM as an optional, experimental choice and are not forcing it on you.

[LybsterRoy]

Re: *Potentially* Useful

I tried DDG in Slimjet running on W7 and it gave me lots of hits - mainly for gaming gloves. I wonder why you got zero hits and needed AI to find anythubg.

[Steve Channell]

Big boxes with a retro name

While the announcement is geared to the upgrade cycles of key customers, the reality is that it uses standard rack sizes, with very-fast microprocessors and PCIe connection cards. when running Linux it can be considered as just very-very big box, with a price to match level of reliability.