[11] Mike Clark, AMD Zen Chief Architect

Ian Interviews #11: Mike Clark, AMD Zen Chief Architect
We've heard from the Uncle of Zen, now it's time to hear from the father of Zen himself. Mike Clark is a Corporate Fellow at AMD, and has had his hands in almost every CPU from AMD since K5. He finished university in 1993 to work at AMD, and has been there ever since. Time to ask him about the Zen story, CPU design, and the future of AMD.
Written version:
www.anandtech.com/show/17031/...
00:00 Intro
00:40 Q1: Work History
05:22 Q2: When did Zen really start?
08:13 Q3: Ability to Pivot
09:17 Q4: Zen Funding during AMD Struggles
11:16 Q5: 8am Meetings with Keller
14:40 Q6: Father of Zen?
16:27 Q7: The Naming of Zen and Ryzen
19:06 Q8: About Skybridge
20:59 Q9: Sign-off and Low Hanging Fruit
24:46 Q10: More Money, More Roadmaps?
26:06 Q11: Zen 5, or Zen 8?
27:00 Q12: Zen A0 Silicon Getting Chilly
29:17 Q13: How Do You Find That Out?
31:08 Q14: Core Decode Width Limits in x86
32:52 Q15: Predicting workloads 4+ years out
34:28 Q16: What's the right amount of cache?
36:55 Q17: Virtual L3 considerations?
38:08 Q18: Stacked L3 Cache
39:44 Q19: Machine Learning in EDA Tools
40:38 Q20: How big can a CCX be?
43:27 Q21: Will Xilinx acquisition change Ryzen?
44:17 Q22: Limits of modern IPC on x86
48:16 Q23: Scalable Cores vs Hybrid Designs
49:54 Q24: What should users look forward to?
Mike Tuuk: / michael-tuuk-9846419
Tim Wilkens: / tim-wilkens-91169a
Jay Fleischman: / jay-fleischman-8421bb6
Leslie Barnes: / leslie-barnes-a2557b3
Will Walker: / will-walker-2322022
Welcome to the TechTechPotato (c) Dr. Ian Cutress
Ramblings about things related to Technology from an analyst
Patreon helps support this channel: / techtechpotato
Join the TechTechPotato Discord! All we ask is you sign up for Patreon for as little as $1.50 a month and your account will be linked - it only has to happen once, and then you can cancel if you wish. Come chat and hang out!
Support the channel by hitting that subscribe button, or using our Amazon affiliate link. Please note that TTP may receive a commission if you purchase anything through these links
AMD Ryzen 9 5950X (16C/32T): geni.us/5950X
AMD Ryzen 9 5900X (12C/24T): geni.us/5900X
AMD Ryzen 7 5800X (8C/16T): geni.us/5800X
AMD Ryzen 5 5600X (6C/12T): geni.us/5600X
AMD Ryzen 9 3950X (16C/32T): geni.us/Bench2020-3950x
AMD Ryzen 7 3700X (8C/16T): geni.us/Bench2020-3700x
AMD Ryzen 5 3600 (6C/12T): geni.us/d08ThJ4
Follow Ian on Twitter at / iancutress
Follow TechTechPotato on Twitter at / techtechpotato
#techtechpotato #amd #zen

[@TechTechPotato]

Written version: www.anandtech.com/show/17031/anandtech-interviews-mike-clark-amds-chief-architect-of-zen?

[@Zeee530]

any chance we could get this in audio?

[@GregDeocampoogle]

Zen is an incredible accomplishment and he seems like such a good guy.

[@muziqaz]

I love his answer about the what future holds for AMD, how he answered it, you can really feel the man is super excited about "Zen 5". When you ask other industry "insiders" about what to expect from the future product, you always get this generic answer. Not this time. Love these non marketing answers and interviews

[@kelownatechkid]

These are absolute GOLD. Thank you to Dr Cutress and Mr Clark!

[@RobBCactive]

What a master class in discretion! It'd be good to have him explain past stuff he can talk about for the historical record. Perhaps the news focus had that effect, talking more about the early Zen foundations or in general may have been more enlightening.

[@lightmanek]

Had all the processors Mike worked on: K5, K6, K6-2, K6-III, K7, and so on ... Great job as I had fun playing and benchmarking each of them ... BTW K6-III was amazing for games, kind of like the upcoming V-Cache Ryzens will be.

[@LawrenceTimme]

I like that this guy pushes the roadmap to take risks as that is exactly what is needed to drive AMD forward at a great rate. A lot of people are way too risk adverse and would end up making refreshes for 6 years straight like bintel.

[@misterpinkandyellow74]

Risks can backfire though, like the FX chips.

[@LawrenceTimme]

@@misterpinkandyellow74 better to take the risk and fail than be half arsed and fail.

[@50shadesofbeige88]

Fantastic interview, very insightful.

[@MissMan666]

Great interview and great questions asked Ian, much appreciated.

[@Nuk1945]

Great engineer! Humble but the real hero.

[@jonnyj.]

Man, this extremely cool as always ian. Your interviews with lead architects are absolutely amazing, and hands down the best thats available on youtube (and the internet, really). I wonder if you'll get a GPU architect to interview sometime soon. Ive always had a facination with GPU design, sometimes purely because of the size of things. Also, I've wondered if there are any applications which utilize 100% of the transistors on something like a 3090 that consumers can get (besides furmark). Most games certainly dont as far as I know. Power constraints for such a big die are also interesting, as is manufacturing. Compared to CPU architecture which I have at least a high level understanding of, graphics rendering always seemed like magic to me :D

[@joleif4970]

Very interesting interview! Thanks for this series!

[@miyagiryota9238]

One of your best interviews, thanks Ian!

[@happydawg2663]

Very nice interview, enjoying every minute

[@utubekullanicisi]

Didn't even realize this interview was 52 mins. All the way from beginning to end an extremely interesting conversation!

[@magnusgranlund3138]

This is the kind of episods i'm looking for! One in a hundred. Fantastic historic document.

[@BlahBleeBlahBlah]

Oooooh, looking forward to this!

[@MarcoComercibjt2]

This summer i studied stack based architectures. Most of them were old designs. But i went to a recent paper of an AMD fellow engineer, regarding stack cache on x86 and ARM designs. A cache with nearly register latency that adds to the L1 (and save L1 space by saving all stack data only in that cache and not polluting the L1). The article was very interesting and i wonder if a future AMD product will integrate such cache.

[@larryfromaustin9117]

Great interview; thanks Ian. Great job Mike.

[@call_me_stan5887]

Great stuff! Also I can't help but notice, that you are VERY good at presenting your interviewees with questions that are constructed in such a way, that by being denied, confirm something else, which happens to be the real querry. Very, very clever :) I can only imagine what happens off the record (probably even better, when the interview happens in person), all the stuff you learn that cannot be shared yet. Well done! Thank you!

[@TechTechPotato]

hehe

[@lordy1952]

Confirmed. Will be waiting for Zen 5

[@eyewhipz]

Thank you, Ian!

[@JakeDownsWuzHere]

you're a great interviewer! love this content

[@RonanAmicel]

Love it. Thanks Mike and Ian!

[@bhuvaneshchoudhary9093]

This guy was as important as Jim Keller in AMD success but a little less known

[@svennikolajsen6624]

Thanks for this interview

[@generalgrevous1984]

The man who brought Intel to its knees. Most impressive.

[@PeterPan-ev7dr]

After he got inspired and "debugged" by Jim Keller ..

[@shmookins]

Or the one that woke up the blue giant. :p

[@louisfriend9323]

When the blue giant was sleeping and free from the fab which was holding them back, thanks to Lisa.

[@herrxerex8484]

This is awesome content 👏

[@j340_official]

Great interview! Very downto earth and takes pleasure in his craft! Good luck to AMD! Zen 4 vs Raptor Lake! Can’t wait!

[@hardwareguru5877]

Cool guy and great interview. Thx.

[@Cypeq]

CPUs are so mind boggling complex i'm surprised lead architect for such a project seems like your regular dude.

[@RobBCactive]

Mike's smart enough to hide his irregularities 😉😉

[@velo1337]

great interview.

[@rektide]

"But that's engineering, right? I mean, it's tough. We know that engineers are good at smelling out bullshit, so you have to be very careful that you don’t give them completely impossible goals - they'll see that it's impossible, and they won't set themselves set up to fail. You can’t have easy goals either, so you have to find that nice balance of not impossible goals but really hard goals, and then tell them if we don't get there, it's still going to be alright as well. But we have to set these aggressive goals, and if get the engineers on-board, you'd be amazed at how hard they work to get it done." One of the greatest general descriptions of human & corporate motivation that I've ever heard.

[@Rocman76]

great interview, love the uarch stuff

[@rektide]

Wonderful interview. I guess the one topic I'm still wishing had some discussion around it, the fabric. I want a version of this talk for HyperTransport and Infinity Fabric & whatever else.

[@soufsoufs]

I've just finished reading the interview at anandtech, what a bad luck!

[@almostmatt1tas]

Thanks Ian!

[@shadow7037932]

Great interview! Ian, would it be possible for you to coordinate a round table discussion between Mike, Jim, Suzanne to discuss where they see the CPU tech will head in the next 20-25 years? Especially curious to know what they think of things like FPGA integration, ASIC integration to CPUs (for ML/AI), ARM vs x86 vs RISC, new lithography tech, etc.

[@TechTechPotato]

That's unlikely. People only come out for interviews when they've got stuff to talk about. In Jim's case it was Tenstorrent, with Mike it was 5 years of zen etc etc

[@shadow7037932]

@@TechTechPotato Gotcha. Makes sense. Thanks for the reply!

[@vincentwong3156]

Hi Ian, thanks for the awesome interview. Could you possibly analyze what is the IP that Xilinx has that could be used in amd soc in the future according to Mike?

[@blackknight50277621]

This has been my top tech channel alongside MLID for weeks now

[@t0scanelli]

You're asking questions I didn't even know I was asking. Great interview 👍🏿

[@Veptis]

Would have loved to hear about executive change (like CEO) when you work on such an important and big project

[@mesostinky72]

Wow. Great interview. I haven’t finished listening to all of it yet but I see in the time codes there’s nothing about Xbox and PS5 I would’ve liked to of heard all about what he had to say about AMD’s interactions with those companies and how he felt about the way they utilized AMD’s products.

[@RobBCactive]

Asking a current AMD CPU designer that, simply won't be enlightening, he's not wanting involvement in inter-corporate PR.

[@paulinuss]

Aaaaa. Instalike after seeing, who do you got out of amd basement :D

[@Speak_Out_and_Remove_All_Doubt]

Man what I would give to know what he knows!!

[@apefu]

Woho! K5 baaaby! My fist AMD.

[@StubbornProgrammer]

PotatoCam gonna potato - unfortunate, but on brand at least. Fortunately for us, your interview skills are FPS independent.

[@trjozsef]

Pro tip that I wish I had done before: have a phone on a tripod record you when doing online interviews.

[@kelownatechkid]

And record your own screen(s) - you can always use that as a last resort for any of the streams!

[@TechTechPotato]

Screen was recorded. That was the straw that broke the camel's back causing all the issues.

[@Speak_Out_and_Remove_All_Doubt]

I know you can read too much into things sometimes but is he more excited for Zen 5 because it's a bigger job than what Zen 4 is going to be???

[@spacechannelfiver]

Probably because Zen4 is taped out and he’s working on Zen5 or Zen6

[@plkh9602]

@@spacechannelfiver 4 > +8% 5 > ~+17%

[@AnotherLotte]

Oh yeah, I've always wanted to ask what ever happened to Hierofalcon, from the 2014-2015 era.

[@privatelanstream3555]

13:53 I do remember an AMD talk from 2012 about 3D stacking. The conclusion back then was that it was impossible, due to heat constraints. I think it had to be someone else put in charge than AMD engineers to make this happen, soemone like TSMC engineers. To me, I think I would set my engineers on an impossible goal, or hire a better ones. Also I live in a mindset of thinking, there are no impossible task, as an equivalent there are only tasks that require time or money.

[@Nuk1945]

I was hoping to see some Zen 8 diagrams on that whiteboard.

[@goodiezgrigis]

I think that room was scrubbed by crime scene cleanup crew prior to turning the camera on. 🤫

[@ttb1513]

@@goodiezgrigisYeah, I was zooming in and analyzing for any faint, faint dry erase marker remnants. It was a little confusing, having to separate the overlays of the zen7 and zen8 diagrams. 😂

[@LawrenceTimme]

This guy seems normal and down to earth.

[@swdev245]

Great interview. The decisions that lead to Bulldozer and his view on this would have been interesting. But maybe that's a touchy subject ;)

[@RobBCactive]

I doubt it, he doesn't seem to have his ego invested unrealistically in decisions. The dual integer shared fp idea looked reasonable given what most programs actually do, until you see the weight attached to fp in review benchmarks. Then one wonders how it stuck, especially as Athlon had had strong fp performance.

[@RyTrapp0]

The way I understand it, part of the basis of the architecture of Bulldozer was that AMD wanted to share processing load between the CPU and the GPU(this being a few years after the purchase of ATI of course). Why have the GPU sitting their basically unused for most of its life, only being spooled up under heavy loads like games? Leverage its advantages to share the load with the CPU in any task it could help in. This is where the "APU" term came from - it wasn't supposed to be a CPU with integrated graphics, it was supposed to be an Accelerated(GPU=accelerator) CPU. As far as I'm aware, the big failure was that no one else in the industry adopted this [rather forward thinking] concept, software programming was never adapted to work with hardware like this, so AMD ended up left with an inherently compromised CPU that was misunderstood. BTW, the media just will not let "APU" go when AMD stopped using it officially(aside from the occasional slip up in a slide or interview LOL; that said, I haven't paid attention lately, I wouldn't blame them if they've given up trying) with Ryzen, obviously because at this point it really IS 'just' a CPU with an iGPU, and not a GPU accelerated CPU. Then the media talks about how "AMD's marketing just can't make up their minds about what they're calling these things. Those dummies shouldn't have tried to give an iGPU a marketing name in the first place!", and it's like, "...well, it's actually... ...*sigh* nevermind..." Honestly, it's a bit unfair how harshly some forward looking advanced products can be punished by the world. If Bulldozer and the shared load computing was too ahead of its time, so be it; but does AMD have to be remembered for "being incompetent" in this era, when they could be remembered for trying to move the industry forward, stepping out on that limb, and just failing to succeed? I mean, being sued because customers assumed they were buying AMD's version of an "Intel-style" x86 CPU architecture is rather asinine; why should anyone take such risks when this is what failure looks like? But I'll digress... ArsTechnica has a pretty good article on the design philosophy and beliefs about the future of the industry that went into Bulldozer's design... arstechnica.com/gadgets/2011/10/can-amd-survive-bulldozers-disappointing-debut/ I wouldn't be surprised if Bulldozer was a touchy subject for Mike - just not for the reasons why most would assume(it would bother me if something I poured myself into for years was completely misrepresented historically to the point of being a joke).

[@PretentiousStuff]

yes

[@alb9229]

Me : What's your min spec Ian ? Ian : How about Zen chief architect ... One word , FOKIN NOICE MATE !!!!!

[@807800]

Did you edit out the second last segment?

[@TechTechPotato]

It was rearranged a bit in thr written version for flow

[@807800]

@@TechTechPotato My mistke then.

[@trjozsef]

6:25 He can't say names because of headhunters.

[@TechTechPotato]

Yeah that's usually the case I know

[@JonMasters]

Did you upgrade your camera?

[@TechTechPotato]

True potato quality now

K5 burnt me so hard i never looked back.

[@LawrenceTimme]

Just imagine leaving AMD thinking it was going to fail and looking at them now. 😬😬

[@mix3k818]

Could've sworn the title said "Intel Zen" for a while...

[@TechTechPotato]

'Ian Interviews' does have a lot of letters in common with Intel. Especially with no serifs on the I that looks like an l. Not my fault, unfortunately.

[@trowachess]

I'm here for that Beautiful Potato

[@Phenylalanin1979]

Great interview Ian. Only grip I have is why the hell does no one at AMD have a headset with a basic mic?

[@RenRenification]

Pretty sure this guy is Wesley Crusher and Jim Keller is his warp bubble thought alien.

[@Pulver]

How it's possible that this kind of interviews getting downvotes?

[@goodiezgrigis]

Intel engineers...

[@0deepak]

WHERE CAT PIC?

[@jimmahT]

Who thinks Ian was just hiding a bad hair day? :P

[@philipp594]

So sad that you did not give us the chance to provide you some questions we care about. I wouuld have loved to know why zen3 is having so many io issues (USB-Dropouts, PCIe 4.0 issues and unstable infinity fabrics / whea 19s) while using the zen 2 io die. Nice interview tho, love to see content like. Keep it going.

[@TechTechPotato]

He deals with the core, not so much the SoC part of it. Different disciplines.

[@philipp594]

@@TechTechPotato He must know someone how feels responsible. communication from amd about these issues has been meeh. AMD at least has to stay winner of the hearts when alder lake out-tdps them.

[@DripDripDrip69]

WHEA 18 for me, only trigger when running heavy instructions(AVX AES SHA) have to set +5 curve optimizer to core 10 for it to go away.

[@TechTechPotato]

@philipp594 I highly doubt he's going to phone someone, call them down, get them to answer live in an interview

[@philipp594]

@@DripDripDrip69 Whea 18 is normal instability. Mostly caused by too low or too high soc voltage.

[@darrylr]

Nice. But you missed the "When is Zen 3 Threadripper Pro shipping" :-)

[@damagepy]

Don't chip designers have a CPU simulator where they simulate everything to the millionth of second how data and electricity travels within the chips, factoring in the lenghts+turns (for conductivity etc), bitflips, down to the fraction of the frequency (and maybe even simulating heat) and see how it performs, maybe eventhrowing in a full motherboard simulation as well... and then(if simulated results kinda match the real performance) letting an AI with machine learning experiment and find out the fastest (even if a core takes up a full chip, with cache everywhere along the way) or most efficient per surface or the all-in-all best layouts? I'm thinking about it for years, it's definetely what I would do so they surely have something similar...

[@AmrikSadhra]

Generally speaking, performance models do not simulate at the physical layer. EDA packages allow optimisation of power and timing corners of the design. The key is to have a model that's flexible enough to easily modify (to deliver results ahead of RTL), but correlated enough to accurately represent performance of hardware. Modellers parameterise the models, and can 'sweep' a variety of configurations for a number of benchmarks. Often, we know where the performance gains/bottlenecks are but chip design trades off power, performance and area.

[@cedricvillani8502]

Ahhh too the truth of the matter at 8:20 lol what Mike really meant was, “uhh well, without going into too much technical mumbo-jumbo we don’t actually manufacture our own chips, speaking of chips, I smoked a pretty fat Doobie right before this rainbow potato head interview. MMM Listening to you Mr. Potato head makes me really want some Fish-n-chips now!!, they only feed me sushi and bat’s…. This isn’t live right?”

[@jjoonathan7178]

y no sram dimms?

[@bitzelijoschaevci3444]

Too expensive

[@gerryjamesedwards1227]

Cache is king, or so I hear.

[@virtualpilgrim8645]

So, if they know where they're going why don't they just build it now?

[@TechTechPotato]

Building stuff takes time. It's one thing to have an idea and a layout, another to actually implement it in silicon and get it to a working state.

[@hammerheadcorvette4]

"The father of Zen"

[@orlovskyconsultinggbr2849]

AMD should have gold on the latest cpu which would allow better heat dissipation.

[@roncable6792]

intel ceo: now bring this guy to intel , bribe him or poison him. i want him gone. evil laugh.........

[@cedricvillani8502]

Oh poor Mike, doesn’t even realize he’s being Punk’d by a rainbow eating potato!

[@GregFerro]

This guy is smart , but doesn’t know what a microphone is. Quite difficult to listen to someone with such poor audio quality

[@MonSteh]

I'm listening from a phone speaker... No issue. The content is what's important imo (also CC works just fine).

[@call_me_stan5887]

Eh, you find a way if you really want to listen to what the guy is saying ;) Not every laptop or phone has a top of the line mic super array.

[@christheswiss390]

This video is only mildly informative if your reflect on what is actually being said, but it's clear Mike Clark is a very dedicated and loyal employee as well as a bright and "take charge" and "rally the troops" kind of systems architect. AMD is very fortunate to have him as a key contributor on board! What I'm VERY befuddled about is how a multi-billion dollar entrprise such as AMD both let's completely untrained engineers and apparently untalented speakers talk to the public as well as how such large enterprises seemingly can't afford to give their best and brightest employees even the most basic media or public speaking training. Why do such large firms not invest in their best people to allow them to shine more in public? Mike Clark's consistent rambling, repetitiveness, his language patterns more resembling a high-school student (using "like" and "you know" as filler words seemingly constantly) than an employee in an important leadership position, unclear messaging, unstructured answers, poor story telling capabilities and the typical engineer's desire to keep talking, rather than making a point and then letting the interviewer take it a step further, imho makes for a slightly cringeworthy interview with quite a lack of actual substance. With just three days of communications training, Mike's interview would be WORLDS better and much more informative to the viewers. Perhaps something he should include in his coming review session with the powers that be at AMD... However, your videos are still very informative - thank you!

[@shauna996]

Lisa Su uses a lot of “you know” filler also. Might be something I can look for in great thinkers.

[@christheswiss390]

@@shauna996 Like, yeah, you know? Like, I'm sure "great thinking" is the driver for underdeveloped speech patterns. It just CAN'T be lack of training. No way. Never. 😂

[@Speak_Out_and_Remove_All_Doubt]

There are downsides to not having a clear public roadmap with up-to-date timetables and realistic product expectations. Sure I know a company doesn't like to set a date in case it slips and it makes them look incompetent or give performance expectations away too early because it tells the competition where they will be in X months but on the flip side of this, I am likely going to buy an Alder Lake system (depending on Intel's pricing) and I only buy a system once every 5 years or so. If AMD had a very clear roadmap of what they had coming up, when it was going to be released and what kind of performance I could expect then they might be able to get me to not buy Alder Lake and wait for one of their products because I know Zen 3D will be out on 15th Jan and it will be 18% faster in single thread and 14% faster multi threaded workloads in general, then a 6nm Zen 3 refresh with reduce pricing in June and then Zen 4 if coming Oct/Nov and it will be around ~30% faster than Zen 3 and ~12% faster than Zen 3D. But as it is Zen 4 might not even make it into 2022 at all or I might buy Alder Lake in Nov and then AMD release Zen 4 in March and I'm really annoyed with myself and AMD didn't get my money. I'm sure AMD won't lose any sleep over me buying AL but I can't be the only one that thinks like this???

[@Steve-Richter]

Is intel run by Indian engineers, whereas AMD has Americans designing their chips?

[@TechTechPotato]

Both have both. Modern chips are worldwide operations.

[@BrownHotPot0096]

your point being?

[@Steve-Richter]

@@BrownHotPot0096 just asking. Do Indian engineers hire other Indians, with the end product becoming overly complicated? And engineers of other backgrounds prefer to work elsewhere?

[@CMSonYT]

@@Steve-Richter I'm sure that's not a thing. Anyway Intel might count as an American company, but the reality is their teams are international. Chances are same for AMD.

[@saiyadulahmad2012]

@Steve Richter what has an overly complicated product got to do with Indian engineers specifically? Aside from Alder Lake, I didn't see Intel launch a complicated product for a long, long time. They always try to play safe when there's no competition.