The Man Behind AMD's Zen Microarchitecture: Jim Keller | Hacker News Reader

The Man Behind AMD's Zen Microarchitecture: Jim Keller

en.wikipedia.org

99 pointsgeezerjay42 comments

Zen is the work of a huge team of talented engineers. To single one out as "the man behind Zen" seems very wrong. I don't know what Jim Keller's contribution to Zen was (and without a blog or autobiography or similar from someone well placed inside the team, then neither do most commentators), but if he did work on the Zen architecture, it's hard to believe that he would have accomplished much without the help of a good team. Keller is the main AMD engineer singled out for praise on The Internet, while the hard work (and given that Zen is such a success, it's surely the result of a mountain of hard work) of everyone else is mostly ignored.

No doubt, but on the other hand, look at his resume. Every project he has led has been a huge success, and when he left AMD for a time, they floundered. Lots of people work hard and contribute, but the value of a top leader is undeniable.

In our 5000 years of recorded History, the Leader gets the glory and/or the sword. Now a days, they do get Golden parachutes but attributing success to Leaders (and in many cases rightfully so) has not changed.

And Apple was more than Jobs and Woz, but they get all the credit.

From what I have read, Woz is in a league of his own. If anything he doesn't get the credit he deserves.

It's true. https://www.theymadethat.com/things/iwe/apple-macintosh-128k

It's just so hard to track any history related to people who are not super famous.

Reading Woz's book, it seemed like he knew what every gate on every chip on the first Apple computer... if not gate, at least every chip.

The strange thing about Woz is that I don't know what he worked on at Apple after the Apple II

https://www.theymadethat.com/people/31em5b/steve-wozniak

The main contribution of a technical leader is calling bullshit on things that are bullshit. Left to their own devices, large teams of people will produce a lot of things, but someone qualified needs to decide which of those things actually need to be produced, and how to make them work when you put them together.

So while you are right, and he is not the only person "behind Zen", it would also be wise to not underestimate how crucial his contribution was. Dude very obviously knows his shit.

Arguably his A64/x86-64 work has been more impactful (so far). Imagine a young Jim Keller from Penn State University, makes you wonder how many SV firms today would toss out that resume in favor of a less... generic... institution. Even worse, he works for DEC, which is a boring big company. This guy is like vomit in the mouth of startup culture.

That is because the average startup culture thinks everything other than web/mobile is not interesting. All the system architecture guides I see these days consist of load balancers, webservers and databases as if that is all that matters.

Startup culture you describe is more entrepreneurship than high-tech. Business idea and execution makes the difference, not several years of technological edge over competition.

Biotech, medicine, electronics, wireless technology, signal processing, chemistry and material science startups work in different environment. Get good education (PhD is requirement in many areas), work in a firm or research group until you are at top of the game and have learned all practical skills, then start a business.

You can see it in the Rust community. So much of the focus has been on http servers instead of codecs, operating systems, daemons, file systems, etc.

To be fair, in a sense it's less surprising when the most notable project in Rust, from relatively early on, was a web browser.

That seems pretty uncharitable, there's more than a few OS projects happening in the Rust space and I know a bunch of people spend time poking at low-level stuff(myself included).

The above poster didn't say that there weren't any OSs or low-level stuff (anyone who frequents HN would know that) they said that majority focus is elsewhere.

To be fair, the DEC Alpha team was highly respected in the CPU architecture industry.

It's a shame how going through the hassle and politics of getting PhD (from a reputable school) is now almost a requirement for the cutting edge.

Getting a PhD is honestly such a bad deal financially and socially. I'm sure the prospect of sinking ~5yrs into some laser focused sub-field for garbage pay scares away a lot of potential--not including academia's useless bureaucracy cough GRE.

Depends on the field though. Although a good school they will pay for your Phd tuition and give you a small stipend(~1500K), which is not great. But there are ways for making up for the lost pay; I do know of a few Data Science Managers who earlier dropped out of their Phd and earned good sums. Some began consulting. Also the big tech firms are recruiting studying Phds to do research for them.

Losing out on an early five years of investing is huge. A ~$30k/yr stipend is not going to be enough to do anything with.

> But there are ways for making up for the lost pay; I do know of a few Data Science Managers who earlier dropped out of their Phd and earned good sums.

Isn't this supporting my claim that a PhD is a bad deal, if people decide they are better off completing the program?

I've also heard the story so many times. Once people have a good idea they realize that completing a PhD is just getting in the way. Instead of researching and studying for ~5yrs they leave and start a company/building a product.

> It's a shame how going through the hassle and politics of getting PhD (from a reputable school) is now almost a requirement for the cutting edge.

Nowadays, a PhD is literally dedicated to carry out original research, which normally entails a curricular part that serves to help a candidate get up to date on the state of the art. Conducting original research is a natural requirement to be on the cutting edge of any scientific or technical field.

> a PhD is literally dedicated to carry out original research

*for ~$30k/yr for five years

> Conducting original research is a natural requirement to be on the cutting edge of any scientific or technical field.

I'm not saying that a PhD won't prepare you. Just that it's inefficient in terms of stress/time/money.

> *for ~$30k/yr for five years

Perhaps that's a US thing. In most european countries (if not all), not only are the costs significantly lower but each state sets up research programs where PhD candidates are actually paid to pursue their studies. Those PhD positions are limited and in some places the pay tends to be somewhat low, but still PhD candidates do get paid for their work.

There are also programs where private companies can participate in PhD programs as part of their R&D investment, and enable their employees to pursuit PhDs while working for them.

I believe the OP is saying that you get paid ~30k/yr for five years which is pretty on the nose with that being the case in the US/Can and european countries as well.

At least for STEM PhDs

I've submitted this thread after stumbling on this HN thread on High End CPUs

https://news.ycombinator.com/item?id=14986105

Among the comments, a user pointed out Jim Keller's contributions to the CPU industry. Trully fascinating read.

Too bad he went to work at Tesla. It's nice to see that people are interested in the self driving car thing but, as in the case of Chris Latner, isn't it more fun / more leverage to work on the stuff that enables all this?

There's a lot of challenging & low level work to be done in the stuff Tesla is working on, and a lot of it will essentially be key in reinventing the auto industry. So I don't think it's necessarily fair to dismiss what he might be doing now as beneath what he's done in the past.

He basically building off his work at AMD, there was an announcement about a month ago that Telsa will use Zen chips for their self driving cars, surprise surprise. Who would have thought the guy that was lead the effort to design Zen has opted to use those chips in his new project at a another company.

Even if he hadn't helped in their design, it's a great chip for the massively parallel computing needed for smart cars.

The fact that (vs Intel) using it knocks $500+ off the price of the car (that's a couple percent of the model 3 selling price) and uses less power under typical loads makes it an even better proposition.

Better not drive fast on the same roads over and over, the car might crash!

Do you have source on Tesla using AMD chips?

Dr. Lisa Zu, CEO of AMD, is the real "man" behind the Zen microarchitecture. Not only is she an engineer who was deeply involved with the product, she lead the company through a tremendous turn-around that made Zen possible and put into place all the platform and support pieces the resulting Ryzen/Epyc products would need.

I find it interesting that when Apple brought out the iPad neither Keller nor Ive got credit, Steve Jobs did. I don't even know the project leads or primary engineers behind Windows 95, Bill Gates got most of the credit. Couldn't name for you the other geniuses behind SpaceX or Tesla, I just hear about Elon Musk. Does the dark lord get credit for Oracle's awful products? No, Larry Ellison is the name we hear. Good or bad, the CEO usually gets the credit... except in this case.

I wonder if it has something to do with Dr. Zu being a woman.

That said, obviously Jim Keller is a skilled chip designer. It would be nice to see more of the supporting cast and geniuses behind innovations get some spotlight instead of all credit flowing to the top (unless maybe it's a woman).

There were articles a while back mentioning AMD's success with Jim Keller in the past and hopes that his return would bring that kind of success again: https://www.computing.co.uk/ctg/news/2344223/amd-to-design-n...

I think assuming that the reason he's getting credit is specifically to slight a woman being CEO is baseless.

I completely agree, she deserves more credit.

BTW, I have always seen her name written as Lisa Su. Why do you write Zu?

K7, K8 (x86_64), A4, Zen... Only that.

>Only

Those are impressive feats, K7 and K8 made AMD truly relevant in the early-mid 2000's. Zen seems to be emerging as a strong contender to the current and potentially next generation of Intel CPUs with a lot more left to optimise and refine in the architecture.

I think he was being facetious. That is a remarkable list of accomplishments.

I said that because in the other thread K7 wasn't mentioned. K7 being the glory days of AMD at the forefront of the market (instead of being the slightly faster and cheaper intel compatible).

Basically Keller was involved, if not more, pushing the market enveloppe 4 times in 20 years.

I want this man to work with Chuck Moore.

You forgot Pwrficient, and he worked on some of the DEC Alpha processors as well.

Afaik Michael Clark was the chief architect for Zen, Keller was above him though.

Yup. I think what Keller's main talents are in putting together and leading a great team.

... and is this a guess? Have you met him, worked with him?

Reply on news.ycombinator.com