The First IBM Mainframe For AI Arrives

An anonymous reader quotes a report from ZDNet, written by Steven Vaughan-Nichols: Mainframes and AI? Isn't that something like a Model-T Ford with a Tesla motor? Actually, no. Mainframes are as relevant in 2022 as they were in the 1960s. IBM's new IBM z16, with its integrated on-chip Telum AI accelerator, is ready to analyze real-time transactions, at scale. This makes it perfect for mainframe mission-critical workloads such as healthcare and financial transactions. This 21st century Big Iron AI accelerator is built onto its core Telum processor. With this new dual-processor 5.2 GHz chip and its 16 cores, it can perform 300 billion deep-learning inferences per day with one-millisecond latency. Can you say fast? IBM can.

Anthony Saporito, a senior technical staff member for IBM Z hardware development, said "One of the Telum design's key innovations is we built an AI accelerator right onto the silicon of the chip and we directly connected all of the cores and built an ecosystem up the stack. Through the hardware design, firmware, the operating systems, and the software, deep learning is built into all of the transactions." According to Patrick Moorhead, Moor Insights & Strategy's chief analyst, "The AI accelerator is a game-changer. The z16 with z/OS has a 20x response time with 19x higher throughput when inferencing compared to a comparable x86 cloud server with 60ms average network latency."

The new model z16 also includes a so-called quantum-safe system to protect organizations from near-future threats that might crack today's encrypted files. This is done with the z16's support of the Crypto Express8S adapter. Built around a CCA cryptographic coprocessor and a PKCS #11 cryptographic coprocessor, it enables users to develop quantum-safe cryptography. It also works with classical cryptography. If you want your data and transactions to be safe both today and tomorrow, this deserves your attention.

Stealth Dupe!

[Kunedog]

Or is it the first dupe . . . with AI!!!

Re:

[DrMrLordX]

Someone needs to turn a golden wrench so BeauHD can access all the RAM IBM installed in his mainframe.

Re:

[AleRunner]

Hey, it's been more than 24 hours since this story was originally posted. Can't expect the "editors" here to remember things that far back.

Maybe they need to buy an IBM mainframe with AI to detect duplicates? To be fair, this is a report of a ZDNet story gushing about the AI features of the new IBM mainframes whilst the original article was just an announcement of the z16. By Slashdot's standards, where the same story on the same site can get duplicated that's not a dupe at all. In fact, given that I just read the headline of the previous story, I can even say I learned from this that "AI" (should that be deep learning - this still isn't gene

This dupe could have been avoided

[ItsJustAPseudonym]

This dupe could have been avoided, if Slashdot were using the new IBM z16 mainframe [slashdot.org].

Re:

[raftpeople]

Watson was marketed as a solution before it actually was a solution (for healthcare), which caused it to be not very good (that's the polite phrase). Typical large corporation, without any real understanding of what is really required to succeed, just start spitting out some marketing slogans and hope someone make a purchase.

Re:

[iggymanz]

Gotta love "quantum safe encryption" when its unknown what crypto might fall to some future quantum computer... not that there exists a QC that could even factor a 15 digit number yet if ever

Re:

[zeeky boogy doog]

The largest composite number factored by a quantum computer, that we know about, is over 1 trillion. And that was 3 years ago.

https://www.newscientist.com/a... [newscientist.com]

Re:

[iggymanz]

That number had 13 digits. Read what I posted again.

Re:

[DrMrLordX]

What about dupes? Why didn't you post this on the original story? Or maybe you're just blind, or what to help BeauHD look less incompetent.

Re:

[DrMrLordX]

Want, not what. Bleh. Hey at least I fix my mistakes.

Re:

[cstacy]

I thought Watson could infer all that and bring WorldPeace at the same time...

I'd be afraid to trust that -- Watson might actually put world peace in jeopardy!

Oboldie: The Forbin Project

Re:

[aitikin]

Watson might actually put world peace in jeopardy!

Not sure if pun intended [slashdot.org] or not, but appreciated nonetheless.

So advanced!

[localroger]

It can make an incorrect inference a billion times a day.

Re:

[account_deleted]

Comment removed based on user account deletion

Re:

[DrMrLordX]

Please don't make serious replies to dupes. Please.

Re:

[account_deleted]

Comment removed based on user account deletion

Re:

[DrMrLordX]

Yes.

Obligatory

[ZiggyZiggyZig]

Why didn't they call it HAL-9000? And is it going to be deployed in the ISS?

So this is not a mainframe...

[tofus]

...but a BRAIN-frame?

Reads like a press release

[omnichad]

So ZDNet appears to just be publishing press releases unedited these days. Tell me this wasn't 100% written by IBM's marketing department.

Just *love* all the "mainframes, so 20th century"

[whitroth]

20 years ago, some crazy at IBM, using IBM's VM (which they've had since the seventies) maxed out a mainframe at 48,000 instances of Linux, all running at the same time. The same mainframe ran just fine with 32,000 instances.

Now, how many servers do you need to run 32k vm's of Linux, without being bogged down? Pulling a number out of my hat, a thousand? At, oh, speaking as someone who configured and ordered servers for 8 or 9 years (until 2019), we're talking well over $100k. Oh, wow, with all that, a mainf

Re:

[zeeky boogy doog]

I think that would be really dependent on what an "instance" is.

I could probably throw a solid 1 or 2 thousand "instances" of an absolute bare-minimum Linux VM onto one of our 128-core 512G-ram systems that occupies all of 2U, but I won't kid myself into thinking they'd be doing anything useful with 384M of memory and an average of 1/8 of a core.