Hi, this is Wayne again with a topic “AMD has Come a Long Way”.
Of course, we were going to have some questions about your time at AMD. Uh Hector asks uh, you returned during a challenging time, which you’ve alluded to just now um. What was the morale like? What were was it was it fight? Was it flight uh resignation, ooh apathy yeah, so when I joined there, so so so so I worked at AMD. I I liked the like, when I was in AMD in 98 99, something like that um I, like the culture of it’s a teamwork. Culture, um Apple was much more of a hard Edge. Do the best Excellence. You know reward the top kind of culture um. So I was I was intrigued. I knew I thought they were going bankrupt when I joined um. It was closer than I thought they had fired a third of the people. I think we ultimately laid off over half the company um, but the people that were there liked each other um.
You know the expression, the rats Le the sinking ship there weren’t many rats at AMD when I got there, they had all left and the people that were there were often techically, very good and they were good to work with so, and I thought it would be Fun to figure out how to turn the company around yeah. It didn’t work. You know like I just get another job. It’S like like. I wasn’t really worried about job so so I went in there sort of knowing you know. What’S going on, there were some things that, to be honest, were worse than I thought yeah. Nobody talks about the rat that um voluntarily swims back to the sinking ship. So I put you’re trying to say that. Well, I call SP is Spade. Well, I was friends with Mark paper master and I said I wanted to lead the CPU team and they wanted me to be the architect, and I thought I could ever pull this off.
If I’m the architect – and somebody else manages this to people because building a great product takes a combination of architecture, teamwork, organization, all kinds of stuff and uh, I hadn’t managed like 500 people before, but I read a book about it. It’S not that hard to be hon. I read 10 books about it. I hired a consultant who was excellent about this kind of stuff.
Okay, so it’s a little bit hard. Then yeah 10 books took a couple of weeks, but uh it was really fun and then well. Uh Rory Reed was CEO at the time and I told him I had to cancel all the projects and start over he’s like yeah. Nobody cares, you know if we have like a 50 % of the competition or 53 and uh. You know go ahead and do it. So I had a lot of freedom and then a lot of people had a lot of good ideas, and it wasn’t just me like so we kind of unlocked it.
You know and I’d say a lot of people didn’t believe in the project. Some of them didn’t believe it right till it was finished, really cuz. They were so used to AMD, losing yeah well hold on a second. How does that happen? Okay, this is something that maybe is does not. Engineers are very determined people.
They can work on something they know is never going to work but they’re having fun doing their part and they just Soldier on, like it’s, it’s okay but hold on a second. And maybe this isn’t your problem because you know you can tell just from the blunt honesty that you you haven’t spent a day in marketing in your life um, but oh I’m great at marketing. I could I can sell you your own shirt. That’S a book about it: it’s not that hard, okay, but help me out here, because this is something that blows my mind is a product will arrive and US monkeys who basically are just like. I don’t know: ooh ooh, ooh ooh Run game.
You know measure frames per second are sitting here, going, hey guys, you got the pricing way wrong, you’re at you’re at you’re, at 80 % of the competition and your price, 20 % higher, and you get people who presumably talk to people who worked on the bloody Thing and they’re like oh really, and you just kind of go: where does this disconnect come from? How could you possibly be help me with this? How could you possibly be an architect or or someone working on Zen? You got your nose right up against this thing and you go. I don’t know, maybe this thing I have no idea. How does it happen? How can we tell and they can’t I’m sorry, what question did you asking? Okay, the question you were saying some of the people working on Zen up until the very end you go yeah, they didn’t, they didn’t believe in it how’s that even possible because they’re seeing they well. They they’ worked on the previous product.
That wasn’t any good, and you know they just assumed this wouldn’t be, I don’t know, but they got. I would tell people what we’re doing and they would look at me like Jim. We could never do that, we’re not that good or so I don’t know so. Here’S a funny thing so a friend of mine told me this years ago, so every company will tell you.
We only hire the best we’re the smartest people in the world and what he said is at 100 people. You can have a like excellent group and at a th people you can be above average and at 10,000 all companies are average, like it’s true just by statistics. Now, there’s a question about whether you lead from the top or the middle, like there’s a bunch of management theories about this and then there’s there’s a lot of problems with how you do things and then there’s this uh risk reward.
If you have an okay design and you want to make it 10 % better, it might be really hard. But if you do a new design G 30 % better, you can do it, but the risk of that’s way higher and so people make bad risk reward. Tradeoffs, like the existential risk of being 50 %, the performance, your competition was 100 %, and yet they were doing lowrisk 5 % moves right.
So we said: hey, let’s build Zen to be just as fast as I think it’s we started out. I said we’ll beat Haswell, which was the processor competition at the time and which, by the way, was shooting behind the duck a little bit. Cuz we assumed Intel would keep moving but as that was my next question, was how much how much did in handy, save AMD by just stagnating, like that, I mean 50 %. Okay, that’s a lot! Now, a good design, hey and – and somebody said uh. He always told me I was I I did not bet on luck enough cuz. You know I I do what I can and I assume everybody else is doing what they can, but I I didn’t see that coming so I think that was pretty handy for them.
Yeah a little bit so tell me right now, but it was good design and it was clean. So one of the things people don’t realize is when we built in you know we had a pretty clean architecture and redid the Catal and methodologies, the flows and stuff, and then you know Zen: 1, 2, 3. 4. They were able to make pretty good progress.
Then what happens is at some point: progress starts to slow down. Cuz you’re really need to do a big either a from scratch or a big rewrite, and then you know to get on the next curve and that’s that’s that’s one of those complicated things now tell me this um right now. Realistically, the 800 lb gorilla in the space is is not Intel. It’S it’s Nvidia um or unless you disagree, in which case I would love to I’d, love to hear about it.
Okay, good, so we’re on the same page and it’s still founder. L and Jensen is a really smart guy. So it’s not only a are you shooting behind the duck? Are you shooting ahead of the duck where what what are you guys targeting cuz you’re trying to disrupt Nvidia? Essentially, if I I’m not mistaken, no, I don’t care about Nvidia. Okay, then tell me tell me what duck you’re aiming at there’s so many it’s a huge AI is a huge Market. Nvidia builds very expensive, very high performance, very high power products that people like right with very high growth margins, turns out. There’S a big market for smaller AI engines, open source software licensable IP chips they can buy and put in their own products.
Like that’s, not a hundred billion Doll Market. I don’t need that I’d die and go to heaven at 500 million in Revenue. So so I’m building products for other people. Now some of our products, I think, are really effective and you know we’ll see, but it’s going to take a while to you know to do that and I have literally more business right now that I can deal with and you know we’re working on. You know delivering hardware and software, and you know we’ll see what happens now, um and and then the other piece is, I think the AI revolutions just started from a computer architecture, point of view and also there’s going to be an interesting revolution in how we build General purpose Computing, so one thing I want to do this is J personally is I want a really good, Ai and CPU design that I can then iterate on as software and models and a whole bunch of things change, and so we designed with conscious intent, like Our AI engine is clean and simple right. Our software stack, you can go, read it read it yourself, right, it’s pretty straightforward, we’re getting really good performance on it and we have a whole bunch of stuff coming in the next six months.
That raises the bar on it, but if we had to say hey, there’s this new model go rewrite the software, I don’t have 2,000 people 20 years of technical debt of software. I have a 100 great people. Writing software and the software Stacks clean and same with the CPU RS 5 CPU. Is it’s going to be super fast, but it’s a brand new design with brand new architecture and it’s really clean and if we want to you know radically change it.
I could do it, I’m not stuck because you know somebody’s CPU, that’s been iterated on for 10 years, where three4 of the code was written by people who don’t even work there anymore are like, like we have we own our own stuff, which is pretty fun so I’Ll give you a funny example. So everybody told me you know there was a big debate about autonomous driving, should it be driven by a c program that makes the decisions or an AI model, and the Assumption was the AI model. Is this murky? You know thing that inputs go in and outputs come out and you don’t know what the AI code’s doing and they had it exactly wrong. To C program was 5 billion lines of code written by 100 people over five years.
They had no idea how that cogram worked, but the weird part is: it doesn’t, have a proper loss function, whereas the AI model, you trained it with a known data set, and when you train it, you know exactly what its error properties are so which one’s better. The AI model you built yesterday from scratch, with a known data set with a known error, function where the C program written by a whole bunch of people over time that nobody knows how it works. I don’t know so. One of the things I want to do is you know, build the next generation of computing in a world.
That’S changing fast! So I’m not worried about the 800 lb gorillas because they don’t they don’t move as fast. Speaking of that, um yeah sorry go for it. It’S a fun thing: uh, do you see other competitors in the risk five space as almost like teammates helping legitimize risk 5 or you see yeah yeah, oh yeah. Definitely yeah so like like R.
Just got Ras just got funding the it’s company’s run by a friend of mine. They got some really good designers. I wish them the best. Civ is a good company.
Christ was one of the original, pretty good Berkeley guys, as you pointed out, that helped build the RIS live architecture. Uh Andy is a really great company. They’Re CEOs he’s a character. I really like of a lot they’re driving they’re driving real stuff.
I know the vantana guys yeah, like the CPU Market’s huge yeah right and, to be honest, you know having five so x86 was the original open source architecture right. They licensed it like six, seven companies and the reason it built the8. 6800 652 um was because those were single Source, proprietary, architectures and x86. The original 886 was open.
Now now it’s it’s become proprietary with two. You know duopoly, you know controlling it, but it was the open architecture. That’S why it won. It didn’t win because it was better. All those CPUs were were crap um and small and arbitrary .