Jim Keller Leaves AMDby Ian Cutress on September 18, 2015 4:05 PM EST
- Posted in
- Jim Keller
AMD’s history has been well documented, especially given several reorganizations in the early part of this decade along with changes in senior staff and how both its market share in CPU and GPU markets is progressing. Today we have learned that one of those senior staff, the head of the CPU group Jim Keller, is to leave AMD effective September 18th (today). Readers may remember that Jim Keller was a recent re-hire in 2012, tasked with leading AMD's CPU group and helping the company develop new core processor architectures in order to bring AMD's architecture in line the competition.
Jim Keller has worked at AMD before, most notably developing the K7 and K8 processors that formed the basis of much of AMD’s success at the turn of the century. This includes assisting in the generation of the x86-64 instruction set that would form the basis of many of the x86 based computers people used today. At other points in time Jim has also spent several years each at Apple helping design their A4 and A5 SoCs as well as at DEC on Alpha processors, giving him a wide degree of experience in CPU development that AMD has been tapping during his latest tenure there.
As a re-hire at the top of the CPU chain, Keller's latest project at AMD was to develop the next generation of high performance processors for AMD and to build a team around the concept of PC performance. This was announced as a rapid departure from the module design of Bulldozer-based cores sharing parts of a processor and towards a new base architecture called Zen. Other projects in the pipeline at AMD CPU group include ARM-based AMD processors (K12), an ARM counterpart of sorts for Zen that is set to launch later on.
As for the big question, the state of Zen, along with confirming that Keller is leaving the company today, AMD is also officially reiterating that their roadmaps are still on course, with Zen set to come to market in the latter half of 2016 and a first full preiod of revenue to be reported in 2017. Given the long (4+ year) design cycles for a modern high-performance CPU, at this point in time all of the "heavy lifting" on Zen development should be done. With only a year or so to go before launch, the rest of Keller's team at AMD will be focusing on fixing bugs and bringing products to manufacturing.
As a result while the loss of Keller is certainly a significant one for AMD, Keller's architecture work on Zen should already be complete, which is likely why we are seeing him leave at this time. And as a quick aside to give you an idea of CPU development timelines, by comparison, Jim's work on K8 was done over 3 years before K8 shipped in 2003. Consequently the biggest loss for AMD here shouldn't be Zen-related, but rather that they won't have Keller's talents to call upon for further refinements of Zen or for a post-Zen architecture.
Meanwhile leadership of the CPU architecture team in Keller's absence will be turned over to CTO Mark Papermaster, who will be leading the group as they wrap up work on Zen. AMD is calling Mark the "acting leader" of the group, so this is likely an interim posting while AMD looks to find or promote someone to lead the CPU architecture group on a permanent basis. Otherwise as we're approaching the end of the fiscal quarter, AMD is in their quiet period, so AMD is limited in what they can say at this time. I suspect we'll hear a bit more on the plan for the final year of Zen development in the company's Q3 earnings release, which will be on October 14th.
Finally, it will be interesting to see if and when Keller will pop up next in the industry. Given his history of switching jobs to work on new CPU projects and his high level of skill which has allowed him to so freely move between companies, we may yet see Keller show up on another CPU project in the future. On the other hand after having worked for AMD twice and Apple, Keller has certainly earned an early retirement. In the meantime with the launch of Zen closing in for AMD, all eyes will be on just what Keller and his team have put together for AMD's next generation CPU.
Top image (from left): Mark Papermaster (CTO), Dr. Lisa Su (CEO), Simon Segars (CEO of ARM), Jim Keller
Post Your CommentPlease log in or sign up to comment.
View All Comments
Sttm - Friday, September 18, 2015 - linkGood luck to him.
Hopefully this doesn't mean Zen was a failure. Really want to replace my 2600k with one next year, as I have grown rather dissatisfied with Intel's mobile prioritized offerings and disgusted by their racist and sexist hiring bonuses.
Gondalf - Friday, September 18, 2015 - linkhe! i don't think Zen will be able to overclock like your 2600K, even with around the same IPC, the old Sandy likely will remains faster. These finer nodes are not great in high clock speeds.
Sttm - Friday, September 18, 2015 - linkIf they can't match my overlocked 2011 i7 in 2016 with a brand new architecture, they should just fold their cpu division.
flyingpants1 - Saturday, September 19, 2015 - linkThat is either a very stupid joke, or just very stupid.
TheJian - Wednesday, September 23, 2015 - linkNeither. Very smart. What is the point in designing cpus that LOSE money hand over fist? If you can't top Intel when they are wasting 1/2 (more actually) of their die on GPU, and YOU are stripping out yours, moving to an new process, moving to FINFET etc, you have FAILED.
It's not mhz here, he's talking the PERF (no matter the clocks) of a 2016 chip vs. 2011 overclocked. You should be able to beat that given that you (if chosen wisely) should be spending the EXACT size (roughly...LOL) of Intels CPU+GPU on ALL CPU. They will fail if they don't make a die at least 75% of Intel's die size, and they will lead for a few years if they match Intel's top i7 die size (cpu+gpu sides). Intel's best hope at that point is to strap two together with stripped gpus and hope 8 cores can do something to stem the bloodshed until they can make massive cores again and even this would take a while to tapeout etc. We know 8 doesn't work vs. 4 already (see AMD vs. Intel now...LOL), so going MASSIVE for AMD on 4 cores will be tough to beat for a while as we also know Intel stopped making a ton of designs at once when AMD quit. You can speed up engineering a bit by throwing more money at it, but not 4-5 YEARS.
I sincerely hope AMD doesn't try to match Intel by simply hacking off gpu and merely matching the size of Intel's CPU ONLY side of their current tech. That would be pointless and give ZERO pricing power, to which intel's response will simply be PRICE CUTS until they put out a better design a few years later to get price power back in their house (basically not too different from last time, just AMD was constrained also allowing Intel to not price cut massively). This is why I really hope they looked at intel's WHOLE die and said, "we'll do that size but PURE CPU" and charge $50 over Intel's cpu at every value or more on high end stuff (make hay while sun shines!). If you're WINNING you can do that and make money again. If you have a BETTER product you can price higher than Intel, if you match them, they just price you down until you can't sell yours.
One more point about money here; if they can't BEAT Intel, they should SELL the company now before it's worthless, and hopefully to someone with billions behind it that can actually put out better tech. IE, R&D dropping for 4yrs not good, losing people (30% of your engineers, now Keller, first Meyer and company etc) will kill you over time that are badly needed to innovate. I can't believe they didn't fight to keep this guy for a decade+. Fire all of marketing, and keep people like this, get Meyer back etc. It seems to me, AMD keeps cutting the wrong people. Meyer was right back in ~2011, when he said you need a KING in product line before going to crap like APU, etc. Comic they fired him, then now go back to his strategy that has ruled for Intel (cpu), Nvidia (gpu), Qcom (modem) etc. You can't branch out from a position of weakness and bad balance sheets. AMD incorrectly chased things that were NOT their core products, like consoles, apu, etc, instead of PURE CPU, GPU and DRIVERS! Jack of all trades in semi's sucks, when KINGS of all trades blow you away. Master something first, or die.
AMD has lost 6B in 12 yrs. You should get the point. If they had listened to Dirk, we probably would have had ZEN in 2011 and likely wouldn't have left the cpu competition. Instead they went custom chip crap, consoles (totally robbed from ZEN like cpu for 3yrs, drivers too, took 8 months to put out a driver since Dec 2014) etc, and forced out the guy who said they needed a ZEN...LOL. That sir, is STUPID.
Unlike before when AMD had a winner for 3yrs, this time they literally could manufacture all they need (capped at ~20% last time) for the most part and make a HUGE dent in Intel while they're distracted by ARM, modems, gpus (in cpu but still R&D going there too) etc. Back then (2000), you could still maybe get fired for going AMD in enterprise, but not so today since most people now know they are the same as far as compatible products goes. So if you put out a BETTER Intel, you can sell the crap out of it today, manufacture the crap out of it to meet those needs, and finally make a few billion to provide for future R&D and engineers or at worst massively pay down debt so you don't have to make 200mil/yr just to break even.
I really doubt Intel has a HUGE 4 core die waiting in the wings, as they've needed to cut everywhere to fund the 4B+ losses per year in mobile while trying to keep up with samsung, tsmc etc. We are now seeing how that has hurt them (IE, everyone basically meeting at 14 or at worse 10nm). AMD could actually be on the same process (essentially) as Intel while hitting with a new, hopefully HUGE (IPC and size wise), cpu core tech that might take Intel 3-5yrs to compete with for real. The shoe really could be on the other foot for a while. IE Intel having 8 core that only wins a small scant few things where more cores matters, while AMD rocks in anything 4 core or less across the board (like Intel now). Intel probably has enough production to cherry pick 5ghz models if AMD only goes 1.25x Intels' cpu side or less and aims at Intel price or lower (stupid). But if they went 1.5-2x Intel cpu side, they should have pricing power for at least a few years and possible cause Intel to have financial problems.
Can you imagine if Intel had price pressure to sell loser products in cpu (their cash cow funding their 4B loss in mobile yearly, along with fab funding too) for 3yrs at this stage in the ARM game/fab game as arm keeps moving to a 500w PC like box with NV discrete gpus (or AMD at some point too for gpu - I mean a full PC box w/85w ARM SOCS or discrete for high end stuff)? They were able to afford it when it was just AMD and they couldn't produce enough to steal any real Intel market share to hurt their profits, but that would NOT be the case today. GF and TSMC could be used to produce ZEN, and maybe samsung (since GF/Samsung share all processes now) if samsung saw a money maker in it. Samsung will need to replace Apple's loss (A10) with something right? ZEN or GPU's is about all you can do to take up all of apple's soc loss.
transphasic - Thursday, September 24, 2015 - linkWell said on all counts. The truth is, AMD is in serious trouble, and most everyone knows it.
They are bailing water out of a sinking ship financially, and unless they merge with Samsung ASAP, then one has to wonder just to much time they have left to exist with a very large infusion of capital (preferably in the Billions of dollars).
They have made costly mistakes at the Executive Level in the decision-making processes, and have thought too short-term in their R&D planning to make any headway against Intel, if even that was their original goal anyways.
Lucky for them that they have the Console Market with Sony PS and Xbox to help them for the time being financially, or else they would have been DOA for a long time now.
AMD's GPU nightmares that stem from bad Management decisions and gross incompetence haven't helped them either. Anyone who bought their very poorly constructed 7970m GPUs knows full well of what I am talking about here. Very bad P.R and it's why a great many went back over to Nvidia.
AMD had better hope that ZEN is the quick Fountain of Youth for them, because without a merger with Samsung, AMD is toast.
MrSpadge - Friday, September 18, 2015 - linkIt doesn't have to overclock the same as his Sandy. Better IPC, more cores and good clock speed can easily enough. Otherwise we'd be all be running 2 GHz Pentium 4 overclocked to 4 GHz and complain that a recent Broadwell i7 doesn't reach higher clocks.
fluxtatic - Saturday, September 19, 2015 - linkI don't even need it to beat SB. I'd be good with it beating Phenom II. I picked up a FX-6300 a while back. Since I was dropping it in as a replacement for the PII X3-720 I had, I thought the time was right to do a little head-to-head using wPrime. Core-for-core, the 720 beat the 6300. Not by huge amount, but I had the 720 mildly overclocked, at 3.5GHz, while I left the 6300, which for me will turbo up to 3.8GHz, at stock.
I'm not going to let myself get too excited this time. When Bulldozer was about to drop, I had grand plans of finally building the ridiculously over-the-top, water-cooled, custom-case beast I've had in my head for years, in celebration of AMD getting back in the game. Then BD released...
I won't buy Intel for a few different philosophical differences I have with them, but AMD sure as hell makes it hard to be happy backing the underdog.
Oxford Guy - Saturday, September 19, 2015 - linkThe 8 core chips, fully enabled and at a reasonable clockspeed are a good budget option for some workloads even today. At one point you could get an 8 phase motherboard and an 8320E for $133.75, tax included.
Samus - Friday, September 18, 2015 - linkThis guy seems to have a history of methodically building a CPU development system inside a company, and once the teams are built and the groundwork for his first-gen architecture is put into motion, he moves to another company to do the same thing. He built AMD's comeback before, maybe he'll do it again. He built Apple's ARM CPU division and left, and it's been strong since. So with the right management and engineering talent, they could continue doing what he (Jim) puts in place.
Like the article said, hopefully they don't fall apart again without him, because the K8 was literally milked dry with no real architectural improvement since the initial Athlon 64 (they coasted on the same core CPU design for nearly 8 years!)