Techstrong Gang – December 4, 2024

Hey, we're at reinvent and as usual, AWS is making a lot of promises. What's real, what's not? You are watching Textron Gang.

Hey everyone, it's Alan Shimmel. It's uh, Wednesday and we are here having an, an amazing time at AWS reinvent, along with 60,000 other people thrown up and down the Las Vegas strip. You know what's interesting is I I the gambling tables are not crowded.

No, not at all. So I don't know if that means geeks don't gamble. I Don't know what the official numbers are, but there's definitely, in my opinion, less people here than there was last year.

Well, according to the keynote yesterday, they said 60,000 people. Yeah, I don't sense less. Seems like just as Much if not more.

But you know, I can actually get places and, and on time and I haven't been late for a thing yet. So that was not the case last year, that's for sure. Well, maybe you're just a little lighter in your steps this year.

You're a little, all that working out is painful little experience from last year Flying. It's true. I might actually know where I'm going.

There you go. Well, okay, we'll give you credit. Could be You don't have that Florida man disease going.

Yeah, that's probably it. Hell, anyway. Anyway, this is Mike Ard.

He's our chief content officer and he is, he is man on the spot at AWS If there's something going on, he's there. He's to my left, to my right Futurum VP analyst. He's, he's more cerebral about things.

They come to him. Mitchell Ashley. Hey Mitchell.

Hey Mike. How are you? I'm good.

Before we go any further, I gotta just say one thing. I heard a rumor, and it's only a rumor that the Czar of Silicon Valley actually got off his throne and drove down here. I, I saw the czar.

You did see him playing eyes on his Oh, so it's real. He got, he's got a sighting. He's only here for the day though.

Is the day out? I'll bet. Well, we'll talk about it off camera.

I mean, we, we've got, we didn't, you'll know her, right? We didn't prepare for a royal visit. No.

Well, we didn't tell him to be here, So I take it we're not paying for this. Uh, his, uh, I think, dang. I think he's on the fu bickle for that one.

Okay. Welcome. John s he's there all day.

All Maybe you'll take us to dinner. Okay. Moving on though.

We're happy John is here. Actually, we will try to hook up with him. So, sorry he couldn't make it to our, to John Swartz.

That's what I'm referring to. Sorry. He couldn't make it to our gang show today.

But it's been an interesting, uh, AWS reinvent yesterday was sort of the big opening salvo, uh, keynotes. And we, we had the AWS, um, CEO up here. And I will tell you, you know, this is the first time I've seen him live like that.

And it's what I expected. This is a guy who's been working the, the innards of AWS engineering. Um, I didn't think it was particularly dynamic, his presentation.

I thought when Andy Jasi came out, it, it definitely stepped up two steps in terms of dynamic dynamicism. Well, it being more dynamic. Well, Andy, Andy reserved the, the best stuff for himself though.

He was the one who was talking about the new models and all that. Well, Nova, but, and we're gonna get into Nova. My question, is Nova an AWS offering or is that an Amazon offering?

Is that something consumers download? Is it just for developers? But, we'll, we'll talk more about Nova in a moment, but there were, there were several big announcements.

One, one set certainly was around Nova. One set was around silicon, uh, you know, a new, a second gen and a third gen promise as often they do here at AWS reinvent. And then the third thing was Q developer, which sounds like where if Nova is aimed more at an individual user, Q is aimed at your developer, Tim.

Absolutely. Yeah. Um, and assorted other sundry items in three plus hours worth of keynotes.

There you go. Let's talk about that a second. I'll be honest, I have a front row seat here in our suite at the wind.

I live stream the keynotes on this big TV behind us. It's a great way to watch a keynote rather than sharing the room with 60,000 people. Um, or worse yet being shov off into the press room or the analyst room or the bathroom or wherever it is you wanna watch this.

Um, but still three hours of keynotes is a lot of keynotes. Keynote, did you sit through the whole three, Mike? Tell the truth.

I did. I did. I sat through the whole thing.

Uh, I might have took a bathroom break during the storage presentation because I knew there wasn't any net new there, so it was Okay. Well they brought affordable storage to the masses. Yes.

I again, A buck again, once again. Yes. I, I, I sat to that one.

Um, but let, let's, let's talk about the big stories, gentlemen. You mentioned Nova. So let's kick right off with that.

So Nova will be available through what they call the Amazon bedrock service, which is this, um, place where you can mix and match models as you see fit. There'll be ones from them, there's six in the initial launch. There'll be a bunch more coming behind that.

And they're talking about multimodal a lot and those types of capabilities. Um, they'll be available alongside philanthropic, which Amazon has put $4 billion into. And then they also will have Mistral and all these other folks there that you can mix and match these things.

Now I kind of feel like it's similar to a lot of the things that Amazon does where they offer everything, but they kind of, you know, push you towards their preferred thing. Um, and you know, I think we're seeing the same thing on the process across the board and data. It's the Amazon way And, and, and the whole nine yards that go with that.

So sometimes, you know, I wake up in the morning and Amazon's kind of a distributor of different technologies and then on the other hand there's like, here's our preferred solution that we built ourselves or that we tweaked and we think that this is the way you should go. I just dunno how many people are going the full bite on Amazon versus, um, you know, thinking through about how much they might get locked into that. I was talking to Nvidia about that very issue where they were like, you know, we love Amazon and God bless them, but you know, customers are keen and aware of the portability issues, especially post that VMware issue.

And their point of view on it was, you know, our stuff is designed to be more portable and if you can also split up the training and the inference engines a little bit more as you see fit. And you know, so they weren't like quat in their boots over the fact that Amazon had a bunch of models and was gonna build their own training and processors as alternatives to GPUs. But they were like, you know, it's co-opetition and it's finest.

There Was a lot of jumping in the pool where there are other people. There are and and muddying waters. Yeah.

I, I was almost waiting for, and this is included in your Amazon Prime membership. I'm serious. I'm wait.

Yeah, well, I mean we'll probably hear he's seriously like, here, here you go. Go ahead and use the, it is Included in your enterprise license agreement, right. And, you know, you can determine how many resources you wanna allocate in credits to various services, Image generation and the whole, you know, any format, any so that that's would all that why Yeah.

Put in prime, right? I mean Nova more than the other things we're gonna talk about. And I think that's why I felt that Andy Jassy was the one who did the Nova uh, rollout was aimed more at a consumer level, the same people who were paying 20 bucks a month for Jet GPT or Claude or, or any of those.

When you peel back the release though, it's part of that bedrock service. So they're aiming that hard at ML lops folks and build Those. Yeah.

But what are ML ops folks doing, creating images, creating videos for social media? Well They're, they're using the APIs for that stuff to create next generation application. So it's, you know, through Bedrock.

It's not just you as an end user, it's you. So I can't go on as an end user and just subscribe to Nova. Like they Have a something called Play Rock where you can go in as an end user and kind of play around with anything you want.

But it's not for building production applications for Where you're going is they want you to embed it within your applications. You know, it isn't to say, want you to use z figure out, well use these APIs, make this part whatever app to make AI empowered or AI empowered applications. It's part of Bedrock.

Let me just give a quick plug. I'm actually doing a center stage webinar 'cause someone else couldn't make it. Who with some A AWS partners on, on building and you harnessing AWS for AI empowered applications and I actually we're gonna Yes, December 11th and uh, we'll be discussing NOVA as part of that.

I guess I be Yeah. Did I tell you I was doing it because someone couldn't make it? I heard that.

Yeah. There's a no show. It's okay.

Oh, I, I just don't know if, you know, showing up with a bunch of models is gonna make you supplant open AI and even Microsoft those places. I mean, but let, let's, I think it's easier 'cause it's there, you know, right. Kind of thing.

It's built into Bedrock. I can just, you know, Bedrock's built into AWS not aws. This is right.

This is, this is their mo They, you know, I used to say, I used to say they give you 80% of the functionality for 20% of the price, but you know what? They've gotten better. Maybe you get 85% now.

But the bottom line is, is they give you, it's like using open source software. You could use the open source software. So what if you don't have a gui?

Right? And, but you could use that open source software and create your own interface. You can do things with, it may not be as easy to use, it may not be as powerful as the, you know, the open core features built into some open source.

But it's perfectly functional. And I think what AWS has here is perfectly functional, kind of their equivalent of the Google perpetual beta be dealing. I get them a product, I mean not the same thing, but they have their own models.

They'll continue to tweak it and improve it so that it remains, I dunno if the word's competitive but viable. Yeah. But There is this IRS race underway about, you know, how big can I build the next trillion parameter LLM.

And apparently that promises to be smarter and have better reasoning capabilities as it gets exposed to more data sets. Well size, If you haven't learned anything in life size matters. Yeah.

But there's a contradiction in that because you know, it really isn't the silver bullet, not the partway awesome. But there is it the silver bullet of the next greatest, biggest model or there was also the messaging around use whatever model you want and we have this new kind of multi interface that will give us your prompts and we will, I forgot to turn it used for it. I'm not even sure We might care how big the model is at the end of the day.

I kind of not impressed by this because, because they're mans mansplaining. They're, they're, they're saying small is, they're saying small is now beautiful and these domain small language models and they have new tools that make it easier to build a small one off the big ones so that, you know, I don't to do all that for you. Right.

And so I think, you know, as far as the average enterprise is concerned, I think they're gonna have a lot more of these smaller domain specific LLMs and what was used in the backend to make 'em, you know, is mildly interesting. Do you Really care? No, but let me tell you what is more than mildly is very interesting in creating these LLMs and training them.

What process are you using is very interesting. And it's very interesting to both your friends at Nvidia and our friends here at AWS and that was a big announcement here today, right? You know, at first, and I forgot the name of their chip.

Titanium, titanium, not Tanium, Tanium cranium boy, I forgot it too. Glad I let you go first. Yesterday's Intel girl from Intel, did anyone mention Intel?

Let's not go there. Um, but Tanium, so the, the initial release of Traum, you know, all the benchmarks for it was how it outperformed X 86 processors and God bless them. Wow.

They, they, they said it outperformed the existing GPUs that they have on the Amazon site today. Right? Yeah.

They weren't comparing it to weren't specific. Right. So they were, and No, but they, they also, the the metrics he gave were for X 86 is too, for the TRA one.

Yeah, the TRA one. Yeah. It's talking about traum one, they've now pivoted two and they announced TRA two, which is somewhat available right now And, And will be available 30 To 40% more efficient than GPUs.

That's the GPU killer. Right? Right.

The Nvidia play theoretically, you know, I'm not, I didn't test them, but that being said, clearly they pivoted from X 86 processors being the competition to GPUs being the competition. And because they wouldn't do that. And they, and they specifically mentioned two use cases, right?

One was in training and creating your models and one was in, not in the roof, what do they call it? Inferencing. Inferencing.

They, they think X 86 is in the rear view mirror because of graviton and the arm processors, one of the things They said Graviton was the other one they mentioned. Yeah. One of the things they said two days ago is that half of all their new workloads are running on Graviton.

And it's like yeah, that's a huge Amount and that is an 86 killer. Yeah. So it is what it is, but well it doesn't, it didn't do Intel any good.

That's for sure though one may say, Hey if, if Amazon designs a GRAVITON or a a Tanium, someone has to build those chips, you need foundries, let's have foundries built in the US that that could be a new intel or a part of Intel or it also depend on who might buy that from Intel and Well Yeah. And all those graviton trips are based on arm and arm principle was always that very same thing. It was like we design it but we Could build it anywhere.

And, and, but let's be fair, you know, there are some who say that if AWS ever came up with an original idea on these things, we'd kill over. Right. Google is doing that.

Microsoft is doing that. Oracle, everybody's looking to build their own silicon right now. Yeah.

I wouldn't, I don't know who's decided to do that first. And it might have been, you know, these things go back in time. Uh, you know, AWS was pointing out that the nitro thing that they built for virtualization is a chip that they built to run virtualization themselves specifically.

And that's how that whole thing functions and it's what they call their key differentiator. But um, so I think Amazon's been thinking about building custom silicon farm. I think they all dreamed if you're a cloud provider and you said one of my big expenses, well electricity, air conditioning real estate.

Well what about the equipment that CapEx, you know, chips? Yep. There you go.

That's where they come to it. So I think that, you know, they have some engineering chops on the silicon. Uh, I'm just not quite clear, you know, they're saying that they built theirs that so it scales better than a general purpose X 86 server would and they're probably right.

But I couldn't tell you for certain how much original engineering went into that versus, you know, did they tweak the design and the workflow and the data flow for a particular, well actually now that I think about it, two days ago they were spending about an hour and a half going through that various thing about how they optimize where the data runs, a processor and everything else. But you know, isn't everybody else doing something similar? Probably, I gotta move us along here.

Okay. But, um, graviton, traum one, they announced TRAUM two, which is on par with today's GPUs. Yeah.

G and then they, I guess I would call it a pre-announcement. Yeah. They probably have a different name for it earlier this year went too with three, which is really being pegged to meet or exceed the latest Nvidia, which I thought was interesting.

They started off with an Nvidia partnership background and then talked about how this chip's gonna be better than the latest than Nvidia chips. Yep. Put in that their infrastructure right there, Nvidia, They're stuck in between.

Right. So I've got this titanium two that lines up with somewhere between the existing g uranium traum somewhere between the existing GPUs and this Blackwell thing that Right. Video's coming out.

And then I'm pre announcing something for late 2025 that exceeds Blackwell theoretically. And it seems like they wanna leapfrog guys We're, we're in Vegas, they wanna cover the whole roulette board. There You go.

Play red and black and just hope it doesn't come out. Zero or double zero. Zero down zero.

The only thing that is kinda not accounted for here is this portability argument. Um, a lot of people in training models in the cloud, but the data that they need to run it against isn't in the cloud. It's in the data center that they run some Until they can get it into their S3 buckets, mware systems in the cloud, then guess what's coming with ware so Many, how many of those snowball appliances do I have to buy to get all my data from that on-premise environment?

And you're Like, go out once and then you're help. I Think we more than once Where AWS God damnit, um, let's move along. So Traum one, traum two is out.

Tra three's coming. Maybe your GPU of the future lives right here. We've got Nova Light, Nova Pro, it uses all of the usual suspects as a it Comes with some extra Smear.

Yeah, an extra smear whole lot with a scooped out Bali. But um, the other big announcement or area of announcements for me, 'cause there was some S3 stuff, but we'll cover that another time. Was the queue developer stuff.

Mitch, I know you were, I saw you sitting on the edge of your seat. Yeah. That kind of stuff after two and a half hours you Yeah.

Infrastructure. Infrastructure. Oh, development.

Yeah. Here we go. But you know, look, I felt like this was a Michael Corleone day.

Today's the take care. We took care of all family business. We're going after Microsoft, VMware and those damn mainframes.

It was them all along. Mm-Hmm. Well, you know, we're we're, we've gone past the chat bot, we've gone, we know we've gone past the Yeah, we can generate and help complete code and we're in this cycle, right?

And when are we, when are we gonna do something to really kind of add add direct value to the bottom line? And what's the biggest impediment? It's toil, it's the backlog, it's all the things that we don't have time to do that we get pulled off of business income generating revenue generating activity.

So their announcement, they first talked about, you know, being able to do code reviews and generate test cases and that's not new stuff. We've heard that from Atlassian, Broadcom, you know, so they're entering the phrase there. But where it really got interesting to me was, okay, let's talk about modernizing your applications.

Let's start with, we'll help you upgrade Java. Okay, that's mildly interesting. Let's talk about do net windows, there's your Microsoft kind of kind.

net, we'll put you onto, you know, Linux operating system and you know, a different language and something that, well, well licensing costs here, here's what it is. App modernization is AWS code for get rid of people or, or, or products we don't like and come onto our licensing fee infrastructure. So, so They did take a direct shot though at Microsoft and they said specifically it's you can you run t net apps on Linux so that you don't have to pay for windows?

Exactly. Yeah. And I don't, they weren't converting T net yet into something else that's, were, I'm sure that's not No Q developer takes t net and moves it into other languages.

And they took a direct shot at VMware licensing too. That was the other, they said get rid of these license notes. Yeah, I'd help you migrate off of VMware.

Which to me connect the dots is, okay, well if you're gonna move your mainframe where there other things that are running on VMware, um, then why not have your data in the cloud too? Isn't that where you're gonna want it if you're moving more of the apps to AWS? So that gets you into the AI ecosystem.

Yeah. Now you got your data, but, but here's what I occurred to me listening to the VMware specific piece was, I, I get it, people are upset with Broadcom licensing, but are you, if you are already going through this transformation app modernization, are you moving, are you going both feet in and, and moving to a Kubernetes container-based architecture and using maybe serverless and Kubernetes or, or are you just taking, you know, same meet the new boss same as the old boss? I still have a hypervisor.

Well, or, or the way I looked at it was I'm changing programming languages, I'm changing databases, I'm changing kind of the envi operating system that I run the app in. Then there's the, oh, I also, now I wanna take it a step further. Let's do microservices and serverless and service mesh and containerized.

So re-architect the application at the same time. So it's, I think it's an either, you can go either path. I don't know that they talked about re-architecting the app to microservices.

Did they, in the announcement? I don't, they didn't use the word microservices there. There's a semi-religious argument about it.

So on the one hand, first you gotta convert VMware VM into an open source vm and then you can encapsulate the open source vm and then that horizon it and that becomes a big ass micro and then that talks to a little micro services. And that's kind of how they're theoretically doing it. When I was scratching my head about it and I was like, well if I look at all this application modernization and I got an existing app and I'm gonna go to all that effort, wouldn't I just write a new app and call it, even start over?

So let, let me stuck there, been caught down that path multiple times in my career. That is the, uh, quickest career ending move you can do is sign up to rewrite the billing system. Yeah, I've seen that happen.

It's right outta Phoenix project. Oh it it is, it is the payroll system, you know, career, career ending move. And I think that's the big challenge.

It's why SW is, why there's so many mainframe applications still around is uh, they are extremely costly. They fail and they take a long time to fail. Yeah.

So I think it, if it really proves out that they can do this, get you off the mainframe, we have that ability with ai. There's a lot of things to go into. I think what's missing is okay, yeah, I can transform it for you, but that's just, that's just the code, that's just the structure of it.

How do you get it off of the mainframe and move it into a completely different environment? So we, we went from, from a binary process, we went from going to a, a cloud native from VMware to coming off mainframe. And, and I'm gonna tell you something, my mom always told me never discuss religion and politics and you know, I don't discuss either.

So, um, so flipping that off, he says it this morning, he says, We'll be back tomorrow folks. He has to sign that agreement every morning. But we do say, if you are gonna go through the trouble or coming off VMware, you might as well go whole hog and and and go microservices and and do that whole thing.

Now the mainframe though is a whole nother issue. There have been people claiming that they're gonna migrate you off the mainframe pretty much since the mainframe was developed. You go tell some CFO that you're gonna walk away from your $750 million investment in that piece of big iron to go move to a great CapEx thing at the cloud and wait till you don't, I already get enough cloud bills.

So I I think, you know, if you think this spells the demise of the mainframe, rumors of its demise is a grossly under overstated and the people that we're in the mainframe know how fraud it is to try to Yeah. This just replacing a database that you're using and moving to something else. No, I mean it's impressive.

They're gonna take your cobal and it was interesting, they said this, they said, we're not gonna take your COBOL app and put it in the cloud. We're gonna take your COBOL app app and give you documentation for it that allow you to make some decisions, which will Reduce the amount of time and half, which is to say that it already takes two years now it might only take one year. And so I was like kind of like, guys, it's nice but it's Two years is really four to five.

Yeah. Well, And and I don't think Be, yeah, I don't think I know anybody who, you know, maybe one or two people claim that they have successfully made a migration off the mainframe, but everybody else I know complains about, you know, the kids from the system integrator who moved in for a year and it cost them a fortune to have those kids just to document the code only to discover that they couldn't move it in the first place. Agreed.

Agreed. Next step. Yeah, I think the people that I wouldn't expect AWS to understand what it takes to modernize a mainframe application.

Not that they can't, not that they can't get that, but that's a whole, whole nother environment. Look, that is such a massive shift. It's not an easy check and, and and is and you're gonna do it just to do it because Penny, for penny dollar for dollar, it's still the most efficient, secure platform that a lot of organizations have.

For what? It's a workhorse. Yeah.

And if the, what is it truly do you always wanna replace scales, wanna upgrade it? Something Just the the fear factor that I hear people responding to is they're worried that they cannot find enough programmers. However, who is that in the age of ai?

We may not Worry about it. We don't have to. It'll write it for you and, and not to, and the mainframe runs things other than cobalt.

But write it in pl one. Yeah. Anyway, guys, I gotta pull the plug.

Oh, we, we've got a full day today of more live from reinvent some stuff from last reinvent and some great content coming at you on text on tv, Mike Mitchell. And I'll be back tomorrow with an AWS Wrapup and um, a reinvent wrapup. And we hope you're enjoying our coverage here out here in slightly cold, chilly Las Vegas, but not as cold as it was last year.

Until then, though, this is Alan Shimmel for Text and Gang. Have a great day everyone.