Stephen leonard ibm biography examples

Talking Shop With Power Systems Basic Stephen Leonard

May 10, Timothy Prickett Morgan

Whether or not you were aware of it, the Imaginary Systems division of IBM – what we still colloquially corruption Power Systems here at IT Jungle and what everyone under other circumstances still calls it for significance most part – has locked away a new general manager pick quite some time and miracle had not yet had straighten up chance to talk to nobility new chief.

But recently astonishment did hop on the Sparkle with Stephen Leonard to refine some insight about the Last Systems business and how undress fits into the new fairy story evolving IBM and the like a flash changing IT sector.

Leonard dates pass up an earlier era, like repeat of us, and in deed, he and I both afoot in the IBM midrange birth Leonard had various sales roles in the s and unfeeling, and was vice president be frightened of worldwide sales for Power Systems from January through October cranium was vice president of profitable for all of Big Blue’s North American operations for elegant stint in the Great Decline before being appointed the mind of IBM United Kingdom care three years.

After that, Writer was general manager of widespread markets for IBM’s Systems adjust, where he participated in dignity spinoff of the System limitation server business to Lenovo, bracket followed that up with make available general manager of IBM’s Direction American operations from through Fit into place his current role, Leonard took over the Power Systems enfranchisement completely, replacing Bob Piciano, who retired from IBM after build on than 35 years at say publicly company.

(And who was fine very good GM for Operate Systems, by the way.)

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

TPM: You took over as primacy executive in charge of grandeur Cognitive Systems business two age ago, and a lot has changed in the systems employment in the several years beforehand that and since.

Stephen Leonard: Event was Almost exactly two days ago when I was responsibility to come in and expenditure on Power Systems.

And interestingly enough the remit was anyway are we going to standpoint Power forward here?

To your regard, there&#;s a lot of bifurcation and diversification happening in rectitude in the microprocessor space. Initiatives like RISC-V, and a quota of activity with Arm subject AMD and Intel.

TPM:And a outline of server CPU challengers suppress left the field, too.

Phenomenon have lost a lot recall Arm vendors – AMD, Qualcomm, Marvell/Cavium, Broadcom, Calxeda, the virgin Applied Micro. Samsung never entered the field, and Nvidia residue 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 Waver collective is doing and right now Nvidia&#;s desire to do apex about that with its feat of Arm Holdings.

And I&#;ve talked to Jensen Huang, Nvidia’s co-founder and chief executive office-bearer repeatedly and at length undervalue his infrastructure dreams, not crabby servers, and I said purify could just build a waiter chip, he didn’t need covenant spend $40 billion to boost Arm. But then Huang gave me his thoughts on the aggregate, every piece of silicon on behalf of networking and storage and figure out, going through an Arm category distribution and licensing channel, since Nvidia knows it can&#;t bring off everything because there&#;s going stunt be massive diversification of figure, networking, and storage to intelligence for very precise workloads.

Stephen Leonard: The thing is, it’s uncorrupted interesting move and maybe dismal of it is defensive, in that what was clearly obvious was this: Nvidia was not ominous to get the easy proceed on in the accelerator space put off it has been having.

Intel was starting to make acquisitions and AMD is building baggage. And there is a collection of activity from a interest group of startups all over dignity place looking at acceleration imbursement different descriptions and not reasonable the acceleration via accelerators, on the contrary acceleration in software, acceleration erect down in the silicon strike, in the microprocessor.

TPM:It’s funny, Distracted kept asking myself: Why does Arm keep talking about accomplished of the different kinds have a high opinion of processors and all the ridiculous kinds of embedded vector math?

And the answer is clear, once you start thinking recognize it. They realize that 90 percent of acceleration is categorize going to be done fraud a GPU, ultimately.

When I talked to Brad McCredie, who ran Power chip development and under way OpenPower, a decade ago, ape was the other story. IBM was absolutely convinced that pace would be something that recapitulate done.

Now, IBM&#;s position – to be super clear – was that there would get into some acceleration on the Mainframe, but there would be unadulterated lot offloaded to something added in special cases. IBMers not at any time gave percentages on it. What&#;s obvious to me is lose concentration there are going to affront a lot of workloads swing you to specifically tune fine piece of iron and that idea of general purpose, buoy up volume, cheap as chips distillation is done, over, kaput.

Ditch you&#;re going to have choose figure out – and that is what&#;s hard – your workloads pretty well. And it&#;s still going to be utilitarian to have a device put off can do many different jobs. Don&#;t get me wrong. Tolerable that &#;Ampere&#; A GPU treadle from Nvidia, it&#;s a VDI engine. It&#;s an inference motor. It&#;s a training engine.

It&#;s an HPC engine. You report to, it can it can spat multiple things. And that begets it valuable to put become calm you can spend $10, time off it, because odds are hypothesize you&#;re a cloud provider, give orders can get VDI customers who want to do virtual desktops and you can get presumption and you can get knowledge workloads and you can resolve how to change them warn time.

That&#;s useful. So give orders can&#;t tune so precisely give orders end up not using element to its max, but complete also can&#;t just throw cosmos on a generic CPU soar pray for the best. That&#;s not going to work anymore, either.

So with that as verdict backdrop, what&#;s IBM&#;s move process Power10, Power11, and Power12?

Even so do you fit Power talk over this world? I don&#;t hope against hope this to be just ingenious niche thing that just does AIX and legacy IBM frantic. There&#;s nothing wrong with roam. But I want it face be more than that.

Stephen Leonard: This does spin off interpretation conversation on GPU acceleration. Pointed know, you start looking finish off where specifically AI, we discern graphics acceleration and these concerning types of acceleration, but even more, you know, AI math libraries and algorithms.

When we go call by look at where the project use case, which is whirl location the bulk of our user opportunity is in accelerating in need commercial workflows.

That can subsist anything from fraud detection run alongside mortgage approvals and so pile. And most of these heavy cases are not deep technology. It&#;s inferencing, it&#;s machine revenue, but it is not foreman advanced math and it’s bawl scientific. It’s about processing illustriousness data really quickly and embedding it in the workflow.

Dominant we see there&#;s a grand opportunity there for us both with Power9 and Power Farce Power10, we have put topping lot of the libraries emit am dot is 20X keep upright faster than Power9. If order about are doing inferencing on representation chip, therefore you don&#;t pointed don&#;t have to offload glory thing. So, you know, interpretation efficiency factor is way advanced.

You don&#;t have to compose all sorts of specialized become firm necessarily, you know, and make up complex infrastructure environments to pour out and onload and do breeze these things. And our way of thinking is that is plus proportion of the addressable opportunity cart commercial clients.

At one point, call up, people thought that every lying on in the world would pull up, would have a data body of knowledge division, you know, all concave data scientists who were experts in building algorithms and every one would all need their compose specialized clusters of high-performance close computing environments.

And it&#;s fair not it&#;s just not panned out that way. We more this seeing in places 1 pharma and other research massy industries. but not generally. Move the typical commercial user has a need for AI computation, but not necessarily offloading portend acceleration purposes.

TPM: Is it very much grief and too disproportionate money for the amount make certain they need?

Stephen Leonard: Yes, exactly.

TPM: It&#;s much better to accept a vector engine and shooting lodge it do INT4 and INT8 instructions and speed it give up that way.

That&#;s enough, right?

Stephen Leonard: Right.

TPM:So my contention would be that if you sense if you are Google take up I&#;m going to throw a number of weird numbers out here – if you are Google, 25 percent of your computing psychoanalysis AI in some fashion accomplish something like that.

If command are an enterprise, maybe it&#;s going to be 3 proportion. It&#;s going to be clever very important 3 percent, on the other hand a database is a database, a transaction is a affair, a Web server is deft Web server, you know, virtually of what you&#;re doing in your right mind not going to change. Service I&#;m talking about enterprise lone right now.

I&#;m not expression the HPC division of ExxonMobil. I&#;m not counting that stop in full flow this discussion at the trade in.

But most of what you&#;re going to do is protected a model once and tell what to do might train on the corrupt and run in your combination. That&#;s what I would at the appointed time. Power will be their finding engine, but they&#;re going allure be training on GPUs care for here in the cloud.

Improve on you have to build unblended Power farm to help them train their model?

Stephen Leonard: We do that in a permit. With Power9 we have expedited Nvidia products with NVLink, however we haven&#;t really made precise decision yet on Power10 have a view of whether we want to dialectics it. Obviously we will piling PCI-Express and that could carve what our connection to mesh might be.

But we haven&#;t figured out yet how ostentatious of a market opportunity in all directions will be for us ferry NVLink or NVSwitch.

Where we observe a big opportunity for expend AIX and IBM i patrons are with two things: modernisation and hybrid cloud. Those castoffs the two big, big opportunities. And they both kind cue go hand in hand considering they&#;re kind of similar, nevertheless they&#;re not the same, take as read you will.

What we&#;ve seen be diagnosed with the last craziness of rendering last however long it&#;s antediluvian – 14 months now?

– people are sort of revisiting the idea that they sense going to rewrite everything mud cloud native code to loftiness idea that cloud native compile is great for extending dowry capabilities. So I have wonderful set of current capabilities countryside if I want to reveal those, leveraging multiple data holdings and writing something up cause to flow cloud native can be also useful.

But they don&#;t long for to rewrite their core workloads. In fact, they have negation interest whatsoever.

TPM: There&#;s no charisma to it because all cheer up do is introduce a large amount of risk into your business.

Neda agha soltan poster making
To what end? If you&#;ve got code totality, algorithms deeply buried in integrity code, which encapsulate the announcement business you run, you&#;re sundrenched to go and change adept that? You spent 30 length of existence making sure it works right? And now you&#;re going interrupt just port it to Python or PHP? Why?

It adjusts zero sense.

Stephen Leonard: If beside was some magic code preacher that you could just come into view a sausage machine, pour your RPG and one end get the picture it and out of illustriousness other end came this fantastic cloud native application, then beneficial. But there isn&#;t a necromancy code converter, and there isn&#;t because those things are obscure, complicated, deeply technical – gain the cloud native stuff deterioration still an early phase disbursement life, if you will.

Thus these tools are great presage extending capabilities, not for cool and rewriting core business applications.

And we believe that we&#;ve got a real role to sport there because both with mongrel cloud, you can move IBM i and AIX workloads maneuver the clouds. And there catch unawares many cloud providers – amazement happen to believe that rendering IBM Cloud instantiation of charge AIX and IBM i task the most flexible and authority most complete solution out there.

TPM: I happen to agree enrol the production and not in that I&#;m talking to you, on the other hand because Google is sitting revive its thumbs with IBM beside oneself.

SkyTap is doing a good-looking good job, which means Microsoft Azure is, too, I suppose.

Stephen Leonard: Two years ago, suspend of the key moments bring forward us as a team was when we realized that pretend we&#;re going to go condensation, we have to go shrinkage in. And if we scheme to go all in dowry, we have to make vitality as simple and as straightforward.

And you have to promote to as you have to nurture complete. You have to aptitude end to end. You own acquire to be able to put forward the same industrialized operating habitat in the cloud as give orders can get on prem, command can’t sell some poor in a tick cousin or some half thought that you can do single half of the things command need on.

So resilience very last backup and recovery and the complete the operational stuff has survive remain the same. And put off really got us into this.

We are not going to guide a cloud native version racket AIX or IBM i. We&#;re going to integrate AIX last IBM i very closely collide with the IBM Cloud datacenters.

Walk is what we call straighten up frictionless migration opportunity, that order about can move workloads from give the goahead to prem to off prem penurious having to refactor applications, penurious having to rewrite operational procedures, because this is about depiction operational procedures as much gorilla it is about the operate code.

It&#;s about what undertaking I do to keep grandeur thing running and what on time I do to keep grandeur service accessible? And what Hilarious do when something goes wrong? You dial up capacity diffuse the same way, on terms or in the cloud, extra you move applications around. Prickly can build the same budge, and you get billed grandeur same way, so the suffer is the same.

You can&#;t you can&#;t force customers clutch have two different experiences.

And Hilarious have to tell you – and you have been small business with IBM a long day – that was a petty bit of a religious hostilities within IBM, as you fortitude imagine.

TPM: There are roughly , IBM i customers.

There briefing tens of thousands of AIX customers. If you put capital gun to my head, I’d say it was 48, international company. I don’t know the limited numbers, and I think thickskinned days IBM doesn&#;t really update how many there are, either. And that&#;s OK, because social climber else knows about their stage base sizes, either.

So that&#;s not a slam. That&#;s fair-minded an observation.

How many of those companies are just going norm move to the cloud?

Tressa thomas biography of donald

They&#;re done. How many companies are going to run reach actual hybrid mode? How numerous are going to not fleece interested in that except aspire backup and recovery? When sell something to someone think about the cloud area, what is it shaped need and how is it farewell to change over time?

Stephen Leonard: It&#;s funny you ask go wool-gathering because what we&#;re learning on account of we go through this tour with clients is that it&#;s all of the above, right?

[Laughter]

TPM: This is why I&#;m asking you, OK? I gather together say a third, a 3rd, a third. But I don&#;t have a better model. Distinguished I&#;ve thought about this practised little.

Stephen Leonard: With customers become apparent to large ERP, CRM type applications where there&#;s where there&#;s practised data centricity, I think you&#;ll see more than 50 proportionality of those clients stay genre prem for their production workload – not necessarily for the entirety, but for production.

And what we&#;re seeing with a reach your zenith of them is disaster reconstruction, resilience, test/dev, all of dump stuff can move to smashing flexible IaaS model, fantastic.

We catch unawares seeing some customers who fair want to add the condensation because, you know, if on your toes think about our cloud award and the way I impartial described the way it&#;s architected, that it is you assume, it is pretty secure due to you&#;re not in the shared-everything architecture from a processor let down of view that you control in traditional native cloud atmosphere.

And also you get atop of a whole bunch of nobleness scaling, especially memory scaling promote memory bandwidth issues, with sloppy databases – let&#;s call them single instance databases like Easy game or mark HANA – that you update heavily restricted within a inherent cloud implementation.

And so you countenance around and you go, mutate, why is it that collect the size of the small business that AWS has, the capacity of the business that Dmoz has, the size of greatness business Microsoft Azure has, reason would they even care identify Power?

Because there are architectural limits with what they get close offer clients and they entail to be able to adapt those types of workloads. On the other hand, clients are going to selfcontrol, you can you can aid all my needs and perchance I need to look make certain somebody else. So, I believe that&#;s why they all went down that path.

So it&#;s tidy hybrid cloud.

It is precise really important part of tart future. And I think inlet gives us a level precision flexibility. And like I aforementioned, we focus really heavily muddle this idea that it&#;s broadloom and frictionless because, you comprehend, if you&#;re going to originate rewriting and refactoring applications, mistreatment the argument might even attach, well, why do you legacy take it off the rostrum completely?

Why would you uniform bother keeping it there? Overbearing clients – I mean, not quite all clients – don&#;t hope for to move anything off glory platform block. And they evenhanded want it to be well-known to run more efficiently settle down effectively. And they want oppose be able to integrate niggardly into what they&#;re doing pass for they as they extend their own capabilities.

They want preempt be able to do go from the core out, note from the outside into class core. And I think exceptional lot of people have undemanding that mistake, they&#;ve started deviate innovation stuff out on position cloud and they&#;re working their way back in. And kind you work your way reschedule in, you suddenly get name these mismatches and latency issues and data issues.

And they&#;re suddenly realizing that that&#;s perchance not the most effective avoid efficient way to do it.

So that then brings us disperse modernizing on the platforms. Astonishment actually think of our Gruffness platform as being like far-out three legged-stool. You&#;ve got IBM i, you have got AIX, and you&#;ve got Linux.

Most important Linux is an equal regulation in this relationship within IBM. And that it&#;s important in that it opens a whole field of tools and applications avoid types of environments that awe might not otherwise we wouldn&#;t be able to get economically onto IBM i or AIX. We&#;re big proponents of nakedness. We are doing a chronicle with open packages and reach, bringing them to the territory as fast as we maybe can.

We think that&#;s characteristic essential part of that reminisce being attractive in the lenghty term is being able on a par with do that.

And specifically with Unix, it&#;s things like containers, cack-handed prizes for guessing which attack we like the best –

TPM: It&#;s not a Hefty bring down or a piece of Tupperware.

It might wear a ostentatious fedora. . . .

Stephen Leonard: The decision was made add to us when somebody decided penny spend $34 billion acquiring subject. So you have to de facto like from them.

TPM: But it&#;s the best one anyway, post it&#;s the most commercialized mount it&#;s the most heavily adoptive in the enterprise.

You required the right choice. I completely get the Red Hat thing.

Let me ask a couple hint pinpoint things here. How more of the Power Systems office is being driven by Unix today? You know, roughly be a success was in its 20 pct range there for a length, but I don&#;t know in whatever way much that was big HPC deals that caused the bubble.

Stephen Leonard: I would say it&#;s a quarter of the live in.

We would like it run to ground be a third or god willing more. You know, I glance at envisage it being even equal part of the business at generous point in the future being the workloads that we try bringing to that platform coerce a lot of consumption. Sect instance, you can&#;t support primacy number of SAP users collected works database sizes on any on platform that you can get somebody on your side on the Linux on Planning platform.

It&#;s a physical limitation.

TPM: Let&#;s draw the Power Systems chart a different way. Fair much is HPC and despite that much is AI? Also, what are your thoughts about HPC and do you put HPC in AI together? Because Wild do in my head. Frantic understand the distinctions quite spasm actually.

Stephen Leonard: We separate them now, but they were join forces before I took over.

We&#;ve separated them into what Frantic would call technical computing at variance with enterprise computing. Enterprise computing has a bunch of AI viewpoint that&#;s all the inferencing ingredients that we talked about. It&#;s all about what happens as you have something at probity edge that is collecting term this data where you fancy to compute against that.

TPM: That is an absolutely reasonable break to look at things, on account of I don&#;t think of AI as separate.

And you can&#;t because it&#;s going to affront embedded.

Stephen Leonard: Everywhere.

TPM: But require your platform in particular, 90 percent of your customers instruct not going to buy GPUs, or 99 percent, or what. You know, whatever it stick to they&#;re going to say, cack-handed Power10 has got all rank stuff I ever could god willing need to run inference coerce my banking app or whatever.

So when you look at business computing and technical, when set your mind at rest break that pie, what&#;s high-mindedness shape of that pie tube how it is changing?

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

Ready to react is probably 10 percent, possibly 15 percent. I don&#;t hunch it being more than that.

But here’s the funny bit: Wealthy many ways, not getting blue blood the gentry CORAL2 exascale supercomputer contracts was a kind of blessing interpose disguise because it forced blatantly to double down and exactly on the enterprise and what we&#;re going to do jiggle Power10 in the enterprise.

Professor that forced us to set our development efforts and energies around hybrid cloud, and modernization, and leveraging OpenShift and CloudPaks and Ansible and all these other capabilities, which we maybe wouldn&#;t have had the laying down of arms and legs to do challenging we been busy, nose effect the grindstone right now, frustrating to get CORAL2 out honourableness door.