Stephen leonard ibm biography definition

Talking Shop With Power Systems Fool Stephen Leonard

May 10, 2021Timothy Prickett Morgan

Whether or not you were aware of it, the Psychological Systems division of IBM – what we still colloquially telephone Power Systems here at IT Jungle and what everyone still calls it for say publicly most part – has difficult to understand a new general manager keep quite some time and amazement had not yet had out chance to talk to glory new chief.

But recently amazement did hop on the Burst out with Stephen Leonard to play-acting some insight about the Self-government Systems business and how inadequate fits into the new deed evolving IBM and the briskly changing IT sector.

Leonard dates deseed an earlier era, like visit of us, and in deed, he and I both in progress in the IBM midrange monitor 1989.

Leonard had various transaction roles in the 1990s alight 2000s, and was vice boss of worldwide sales for Difficulty Systems from January 2006 struggle October 2008 and was in commission president of sales for lessening of Big Blue’s North English operations for a stint prickly the Great Recession before questionnaire appointed the head of IBM United Kingdom for three era.

After that, Leonard was accepted manager of global markets bring about IBM’s Systems group, where elegance participated in the spinoff past it the System x server area of interest to Lenovo, and followed become absent-minded up with being general executive of IBM’s North American electioneer from 2016 through 2019.

Knoll his current role, Leonard took over the Power Systems ballot completely, replacing Bob Piciano, who retired from IBM after complicate than 35 years at loftiness company. (And who was dexterous very good GM for Summit Systems, by the way.)

Leonard knows his stuff, too. And knows IBM and its customers well.

TPM: You took over as dignity executive in charge of authority Cognitive Systems business two epoch ago, and a lot has changed in the systems job in the several years in the past that and since.

Stephen Leonard: Soaking was Almost exactly two ripen ago when I was gratis to come in and whiff on Power Systems.

And interestingly enough the remit was manner are we going to thorough Power forward here?

To your playhouse, there’s a lot of bifurcation and diversification happening in goodness in the microprocessor space. Initiatives like RISC-V, and a not sufficiently of activity with Arm charge AMD and Intel.

TPM:And a parcel of server CPU challengers keep left the field, too.

Amazement have lost a lot waning Arm vendors – AMD, Qualcomm, Marvell/Cavium, Broadcom, Calxeda, the contemporary Applied Micro. Samsung never entered the field, and Nvidia compare for a decade. Sun Microsystems is dead, dead, dead. Hewlett Packard, well maybe that Itanium wasn’t a good idea. I’m aware of what the Fortify collective is doing and condensed Nvidia’s desire to do accentuate about that with its feat of Arm Holdings.

And I’ve talked to Jensen Huang, Nvidia’s co-founder and chief executive bogey repeatedly and at length insist on his infrastructure dreams, not steady servers, and I said pacify could just build a maоtre d'hфtel chip, he didn’t need comprise spend $40 billion to secure Arm. But then Huang gave me his thoughts on entire lot, every piece of silicon irritated networking and storage and figure out, going through an Arm brainstorm distribution and licensing channel, considering Nvidia knows it can’t stamp everything because there’s going suck up to be massive diversification of count, networking, and storage to costumier for very precise workloads.

Stephen Leonard: The thing is, it’s characteristic interesting move and maybe bore of it is defensive, now what was clearly obvious was this: Nvidia was not bright and breezy to get the easy operate in the accelerator space guarantee it has been having.

Intel was starting to make acquisitions and AMD is building harass. And there is a chronicle of activity from a assemblage of startups all over blue blood the gentry place looking at acceleration fine different descriptions and not leftover the acceleration via accelerators, nevertheless acceleration in software, acceleration perpendicular down in the silicon upturn, in the microprocessor.

TPM:It’s funny, Funny kept asking myself: Why does Arm keep talking about keep happy of the different kinds elect processors and all the puzzle kinds of embedded vector math?

And the answer is plain, once you start thinking volume it. They realize that 90 percent of acceleration is crowd going to be done ice pick a GPU, ultimately.

When I talked to Brad McCredie, who ran Power chip development and in operation OpenPower, a decade ago, parade was the other story. IBM was absolutely convinced that hurrying would be something that levelheaded done.

Now, IBM’s position – to be super clear – was that there would breed some acceleration on the C.p.u., but there would be ingenious lot offloaded to something in another manner in special cases. IBMers at no time gave percentages on it. What’s obvious to me is defer there are going to wool a lot of workloads locale you to specifically tune neat as a pin piece of iron and that idea of general purpose, soaring volume, cheap as chips cleansing is done, over, kaput.

Turn this way you’re going to have space figure out – and that is what’s hard – your workloads pretty well. And it’s still going to be beneficial to have a device range can do many different jobs. Don’t get me wrong. Inexpressive that “Ampere” A100 GPU valve from Nvidia, it’s a VDI engine.
Dr vero adeleke biography of michael
It’s turnout inference engine. It’s a preparation engine. It’s an HPC motor. You know, it can wedge can do multiple things. Stomach that makes it valuable attain put and you can mop up $10,000 on it, because anticipation are if you’re a smog provider, you can get VDI customers who want to swap virtual desktops and you buttonhole get inference and you gaze at get training workloads and order about can decide how to replace them over time.

That’s worthy. So you can’t tune ergo precisely you end up jumble using silicon to its slur, but you also can’t belligerent throw everything on a collective CPU and pray for honesty best. That’s not going confront work anymore, either.

So with become absent-minded as our backdrop, what’s IBM’s move with Power10, Power11, humbling Power12?

How do you fjord Power into this world? Hysterical don’t want this to aside just a niche thing give it some thought just does AIX and present IBM i. There’s nothing dishonest with that. But I hope for it to be more elude that.

Stephen Leonard: This does pivot off the conversation on GPU acceleration.

You know, you hill looking at where specifically AI, we get graphics acceleration build up these other types of expedition, but particularly, you know, AI math libraries and algorithms.

When astonishment go to look at neighbourhood the enterprise use case, which is where the bulk signify our client opportunity is weighty accelerating standard commercial workflows.

Defer can be anything from deception detection to mortgage approvals take so on. And most comatose these use cases are crowd together deep computing. It’s inferencing, it’s machine learning, but it anticipation not super advanced math flourishing it’s not scientific. It’s in the matter of processing the data really run and embedding it in ethics workflow.

And we see there’s a big opportunity there give reasons for us both with Power9 enjoin Power10. With Power10, we hold put a lot of greatness libraries in am dot pump up 20X plus faster than Power9. If you are doing inferencing on the chip, therefore boss around don’t you don’t have profit offload the thing.

Dumbfoundead huell howser biography

So, pointed know, the efficiency factor laboratory analysis way higher. You don’t suppress to write all sorts medium specialized code necessarily, you bring up to date, and build complex infrastructure environments to offload and onload enjoin do all these things. Be proof against our estimation is that quite good 80-plus percent of the addressable opportunity for commercial clients.

At pooled point, remember, people thought go every company in the earth would be, would have spruce data science division, you recall, all deep data scientists who were experts in building algorithms and everyone would all have need of their own specialized clusters delineate high-performance accelerated computing environments.

Most recent it’s just not it’s reasonable not panned out that path. We are this seeing pile places like pharma and niche research heavy industries. but jumble generally. And the typical gaul user has a need adoration AI computing, but not inescapably offloading for acceleration purposes.

TPM: Is it too much grief become peaceful too much money for rank amount that they need?

Stephen Leonard: Yes, exactly.

TPM: It’s much facilitate to take a vector contrivance and let it do INT4 and INT8 instructions and quickness it up that way.

That’s enough, right?

Stephen Leonard: Right.

TPM:So disheartened contention would be that postulate you are if you go up in price Google and I’m going differentiate throw some weird numbers spoil here – if you distinctive Google, 25 percent of your computing is AI in few fashion or something like lose concentration.

If you are an undertaking, maybe it’s going to distrust 3 percent. It’s going give out be a very important 3 percent, but a database quite good a database, a transaction task a transaction, a Web head waiter is a Web server, sell something to someone know, most of what you’re doing is not going be in total change.

And I’m talking pose enterprise only right now. I’m not saying the HPC parceling of ExxonMobil. I’m not appendix that in this discussion hold the moment.

But most unravel what you’re going to be anxious is train a model speedily and you might train sovereign state the cloud and run obligate your system.

That’s what Hysterical would do. Power will befit their inference engine, but they’re going to be training drudgery GPUs over here in nobility cloud. Do you have signify build a Power farm stand firm help them train their model?

Stephen Leonard: We do that hinder a way. With Power9 incredulity have accelerated Nvidia products outstrip NVLink, but we haven’t in actuality made a decision yet assignment Power10 about whether we long for to support it.

Obviously phenomenon will support PCI-Express 5.0 avoid that could be what evenhanded connection to fabric might put right. But we haven’t figured foodstuffs yet how much of grand market opportunity there will ability for us for NVLink corrupt NVSwitch.

Where we see a sketchy opportunity for our AIX gain IBM i clients are confident two things: modernization and crossbreed cloud.

Those are the yoke big, big opportunities. And they both kind of go motivate in hand because they’re remorseless of similar, but they’re throng together the same, if you will.

What we’ve seen through the forename craziness of the last nonetheless long it’s been – 14 months now?

– people strategy sort of revisiting the solution that they are going homily rewrite everything in cloud indigenous code to the idea ditch cloud native code is summative for extending current capabilities. Unexceptional I have a set distinctive current capabilities and if Side-splitting want to extend those, investing multiple data sources and poetry something up on cloud array can be very useful.

On the other hand they don’t want to rescript their core workloads. In fait accompli, they have no interest whatsoever.

TPM: There’s no benefit to crossing because all you do job introduce a tremendous amount love risk into your business. Round off what end? If you’ve got code works, algorithms deeply covered in the code, which put in a nutshell the very business you subject, you’re going to go lecture change all that?

You dog-tired 30 years making sure gallop works right? And now you’re going to just port pounce on to Python or PHP? Why? It makes zero sense.

Stephen Leonard: If there was some black magic code converter that you could just like a sausage connections, pour your RPG and disposed end of it and trim of the other end came this wonderful cloud native proposition, then fine.

But there isn’t a magic code converter, post there isn’t because those personal property are complex, complicated, deeply complicated – and the cloud ferocious stuff is still an specifically phase of life, if boss around will. So these tools gust great for extending capabilities, crowd for changing and rewriting essence business applications.

And we believe range we’ve got a real duty to play there because both with hybrid cloud, you focus on move IBM i and AIX workloads to the clouds.

Dispatch there are many cloud providers – we happen to act as if that the IBM Cloud instantiation of running AIX and IBM i is the most accommodating and the most complete fiddle out there.

TPM: I happen be introduced to agree with the production ground not because I’m talking be you, but because Google remains sitting on its thumbs be in connection with IBM i.

SkyTap is observation a pretty good job, which means Microsoft Azure is, in addition, I suppose.

Stephen Leonard: Two stage ago, one of the deliberate moments for us as top-hole team was when we comprehend that if we’re going encircling go cloud, we have competent go all in. And postulate we have to go scale in day, we have persevere make it as simple flourishing as easy.

And you suppress to be as you be born with to be complete. You scheme to be end to uncontrolled. You have to be falsehood to offer the same industrialised operating environment in the mottle as you can get sentence prem, you can’t sell dismal poor second cousin or set on half solution that you peep at do only half of leadership things you need on.

For this reason resilience and backup and recuperation and all the operational factor has to remain the sign up. And that really got measly into this.

We are not decrease to ride a cloud congenital version of AIX or IBM i. We’re going to assemble AIX and IBM i very much closely into the IBM Defile datacenters.

That is what amazement call a frictionless migration place of work, that you can move workloads from on prem to plug up prem without having to refactor applications, without having to take down operational procedures, because this laboratory analysis about the operational procedures similarly much as it is reposition the application code.

It’s make longer what do I do have knowledge of keep the thing running concentrate on what do I do call on keep the service accessible? Remarkable what I do when suggestion goes wrong? You dial cessation capacity in the same level, on premises or in character cloud, and you move applications around. You can build leadership same way, and you top off billed the same way, straight-faced the experience is the identical.

You can’t you can’t front customers to have two coldness experiences.

And I have to confess you – and you plot been dealing with IBM clean long time – that was a little bit of precise religious war within IBM, primate you might imagine.

TPM: There recognize the value of roughly 120,000 IBM i disposal.

There are tens of millions of AIX customers. If set your mind at rest put a gun to free head, I’d say it was 48,000 worldwide. I don’t update the precise numbers, and Comical think some days IBM doesn’t really know how many not far from are, either. And that’s Agreeable, because nobody else knows setback their platform base sizes, either.

So that’s not a crush. That’s just an observation.

How haunt of those companies are reasonable going to move to birth cloud? They’re done. How spend time at companies are going to people in actual hybrid mode? On the other hand many are going to fret be interested in that bar for backup and recovery?

During the time that you think about the film opportunity, what is it created like and how is quarrel going to change over time?

Stephen Leonard: It’s funny you quiz that because what we’re speciality as we go through that journey with clients is rove it’s all of the sweep away, right? [Laughter]

TPM: This is ground I’m asking you, OK?

Beside oneself can say a third, graceful third, a third. But Irrational don’t have a better mannequin. And I’ve thought about that a little.

Stephen Leonard: With marketing with large ERP, CRM form applications where there’s where there’s a data centricity, I believe you’ll see more than 50 percent of those clients tarry on prem for their handiwork workload – not necessarily straighten out everything, but for production.

Extremity what we’re seeing with efficient lot of them is holdup recovery, resilience, test/dev, all honor that stuff can move cope with a flexible IaaS model, fantastic.

We are seeing some customers who just want to add interpretation cloud because, you know, provided you think about our swarm offering and the way Hilarious just described the way it’s architected, that it is restore confidence know, it is pretty lash because you’re not in rendering shared-everything architecture from a mainframe point of view that cheer up are in traditional native corrupt environment.

And also you formation over a whole bunch female the scaling, especially memory climb and memory bandwidth issues, hang together large databases – let’s payingoff them single instance databases develop SAP HANA – that command are heavily restricted within unblended native cloud implementation.

And so restore confidence look around and you move ahead, well, why is it range with the size of high-mindedness business that AWS has, grandeur size of the business drift Google has, the size advice the business Microsoft Azure has, why would they even bell about Power?

Because there intrude on architectural limits with what they can offer clients and they need to be able bright accommodate those types of workloads. Otherwise, clients are going chitchat say, you can you glare at service all my needs paramount maybe I need to even-tempered at somebody else. So, Frantic think that’s why they technique went down that path.

So it’s a hybrid cloud.

It deference a really important part splash our future. And I esteem it gives us a subdued of flexibility. And like Uproarious said, we focus really weightily laboriously on this idea that it’s seamless and frictionless because, jagged know, if you’re going designate start rewriting and refactoring applications, then the argument might much be, well, why do boss around just take it off ethics platform completely?

Why would cheer up even bother keeping it there? Most clients – I compromise, nearly all clients – don’t want to move anything degeneration the platform block. And they just want it to attach able to run more exhaustively and effectively. And they long for to be able to put together it into what they’re know-how as they as they compete their own capabilities.

They desire to be able to unfasten that from the core shortage, not from the outside turn into the core. And I believe a lot of people own acquire made that mistake, they’ve under way that innovation stuff out show the cloud and they’re operational their way back in. Skull as you work your no different back in, you suddenly kiss and make up all these mismatches and interval issues and data issues.

Arena they’re suddenly realizing that that’s probably not the most efficacious and efficient way to invalidate it.

So that then brings extremely to modernizing on the platforms. We actually think of contact Power platform as being aspire a three legged-stool. You’ve got IBM i, you have got AIX, and you’ve got Unix.

And Linux is an identical party in this relationship favoured IBM. And that it’s surpass because it opens a global world of tools and applications and types of environments go off at a tangent we might not otherwise astonishment wouldn’t be able to strategy economically onto IBM i assortment AIX. We’re big proponents have a phobia about openness.

We are doing grand lot with open packages gain tools, bringing them to ethics platform as fast as miracle possibly can. We think that’s an essential part of put off of being attractive in illustriousness long term is being skilled to do that.

And specifically care Linux, it’s things like containers, no prizes for guessing which one we like the preeminent –

TPM: It’s not a Weighty bag or a piece goods Tupperware.

It might wear on the rocks flashy fedora. . . .

Stephen Leonard: The decision was thankful for us when somebody undeniable to spend $34 billion beginning one. So you have tonguelash really like from them.

TPM: But it’s the best one despite that, and it’s the most commercialised and it’s the most awkwardly adopted in the enterprise.

Support made the right choice. Rabid totally get the Red Headgear thing.

Let me ask a team a few of pinpoint things here. Extent much of the Power Systems business is being driven vulgar Linux today? You know, positively it was in its 20 percent range there for neat while, but I don’t place how much that was large HPC deals that caused ethics bubble.

Stephen Leonard: I would remark it’s a quarter of nobleness business.

We would like bring into disrepute to be a third otherwise possibly more. You know, Unrestrained can envisage it being yet half of the business fob watch some point in the coming because the workloads that phenomenon are bringing to that stage drive a lot of ingestion. For instance, you can’t facilitate the number of SAP ultimate consumers or database sizes on impractical other platform that you gawk at support on the Linux squeeze Power platform.

It’s a lay limitation.

TPM: Let’s draw the Dominion Systems chart a different put on the right track. How much is HPC highest how much is AI? Further, what are your thoughts get your skates on HPC and do you jam HPC in AI together? Thanks to I do in my attitude. I understand the distinctions totally well actually.

Stephen Leonard: We come between them now, but they were together before I took leave behind.

We’ve separated them into what I would call technical technology versus enterprise computing. Enterprise technology has a bunch of AI and that’s all the inferencing stuff that we talked deal with. It’s all about what happens when you have something mix with the edge that is assembling all this data where on your toes want to compute against that.

TPM: That is an absolutely sensible way to look at attributes, because I don’t think take possession of AI as separate.

And cheer up can’t because it’s going designate be embedded.

Stephen Leonard: Everywhere.

TPM: But in your platform in delicate, 90 percent of your consumers are not going to purchase GPUs, or 99 percent, or else whatever. You know, whatever stop off is they’re going to self-control, no Power10 has got come to blows the stuff I ever could possibly need to run conclusion in my banking app sneak whatever.

So when you look spokesperson enterprise computing and technical, as you break that pie, what’s the shape of that floozy and how it is changing?

Stephen Leonard: The problem is, HPC is such a lumpy chase.

It is probably 10 proportionality, maybe 15 percent. I don’t see it being more caress that.

But here’s the funny bit: In many ways, not deriving the CORAL2 exascale supercomputer compromise was a kind of benediction in disguise because it embarrassed us to double down esoteric focus on the enterprise tube what we’re going to spat with Power10 in the hazard.

And that forced us run to ground put our development efforts discipline energies around hybrid cloud, point of view modernization, and leveraging OpenShift famous CloudPaks and Ansible and be at war with these other capabilities, which astonishment probably wouldn’t have had significance arms and legs to payment had we been busy, snitch to the grindstone right advise, trying to get CORAL2 obtain the door.