Jim Keller's return to AMD suggests the company is at least considering ditching Bulldozer in favour of a revised K8 design - much as Intel did with Netburst.
AMD has rehired the man largely responsible for the design of the Athlon 64 and Opteron 64 processors - the first chips to feature a native x86-64 architecture - suggesting that the company is rethinking the whole Bulldozer approach to chip design.
AMD has successfully poached Jim Keller from Apple, where he was a director of the product architecture group specialising in the company's mobile chips following apple's acquisition of fabless low-power chip specialist P.A. Semi. Prior to that, however, Keller was an AMD man, having been part of the design team responsible for the highly successful Athlon 64 and Opteron 64 laptop, desktop and server chips.
Since Keller's departure, to Broadcom, then SiByte, and on to P.A. Semi before being acqui-hired by Apple, AMD appears to have lost its way. Even the company's most hardened fans admit that Bulldozer has been a bit of a damp squib - somewhat analogous to the missteps made by rival Intel in the design of the Pentium 4, which allowed AMD to gain significant ground thanks to its far more efficient designs - and while its Accelerated Processing Unit (APU) products are generating plenty of buzz at the lower end it is blocked out of the mobile market by the likes of Intel and ARM's multitudinous licensees.
With Keller back, it's possible that AMD's fortunes could turn around. A co-author of the HyperTransport specification and the x86-64 instruction set architecture, there's no denying Keller's experience or ability - and while he may have been working on mobile system-on-chip (SoC) designs for the last few years, he's unlikely to have lost his touch for efficient high-performance designs.
'
Jim is one of the most widely respected and sought-after innovators in the industry and a very strong addition to our engineering team," claimed AMD's Mark Papermaster in a statement on Keller's re-hiring. '[i]He has contributed to processing innovations that have delivered tremendous compute advances for millions of people all over the world, and we expect that his innovative spirit, low-power design expertise, creativity and drive for success will help us shape our future and fuel our growth.'
Keller's official role at the company is that of corporate vice president and chief architect of the microprocessor arm, reporting directly to Papermaster. Whether Keller's return suggests that, like Intel and Netburst before it, AMD is thinking about ditching Bulldozer and returning to an enhanced version of the K8 design remains to be seen.
12 Comments
Discuss in the forums ReplyThis pretty much. But will they ditch bulldozer? Nope, not if they follow their own roadmap. They've got a few more iterations of it. And truth be told, it isn't nearly as bad as other say it is. Outside of the relatively bad misprediction penalty and the abnormal voltage leakage spike (perhaps the Gate first issues) it's not a bad chip. Just a sub-par one in comparison to intel's offerings.
That said, AMD's APU's are still very good.
Nothing of the sort. It suggests they want the man who built Apple it's SoCs to build their own new SoCs.
*I only run a small business in order to make money. The above comment assumes amd have the same goal.
Again, i hope you are right, though.
the gpu will become the FPU in the future and the modular design is the idea behind bulldozer not the high clocks.
also the flex plex fpu is a nice idea they will probably make it wider like 512bit and split-able to 64 or 128 bit.
(maybe they break the module apart so you could use all the fpu for one core or share it between all of them. i havn't had computer architectures yet so i don't know if that is possible).
bulldozer is also a higher clocked design. maybe they hoped that it would be possible whit the new processes.
but it isn't completely like netburst .
in the netburst architecture the alus where double pumped so if the P4 ran at 3 ghz the alu ran at 6 ghz. that's why it used a lot of power.
It does mean, however, that two years down the line, or more, AMD might finally get their arse into gear and start producing chips worth half a damn.