KitGuru KitGuru.net – Tech News | Hardware News | Hardware Reviews | IOS | Mobile | Gaming | Graphics Cards
Advanced Micro Devices on Wednesday confirmed that it is developing an all-new high-performance x86 micro-architecture code-named “Zen”. Rory Read, chief executive officer of Advanced Micro Devices, admitted that the current-generation “Bulldozer” micro-architectures could not improve AMD’s positions on the market of central processing units, so, at present AMD pins hopes on the next-gen x86 micro-architecture. He also said that the company will remain on the market of x86 server chips.
“Everyone knows that Bulldozer was not the game changing part when it was introduced three years ago,” said Rory Read at Deutsche Bank 2014 Technology Conference. “We have to live with that for four years. But [for] Zen, K12, we went out and got Jim Keller, we went out and got Raja Koduri from Apple, Mark Papermaster, Lisa Su. We have built and are building now next generation graphics and compute technology that customers are very interested in.”
Nothing particular is known about the “Zen” and the “K12” micro-architectures except the fact that both are developed by teams headed by Jim Keller, who led the development of AMD’s ultra-successful Athlon 64 and Opteron (K8) processors in the early 2000s. Both architectures are developed with high-performance applications and servers in mind.
“There are very few people who know how to create server chips,” said Mr. Read. “Jim Keller has a lot of experience in that space. You get Zen and K12. […] ARM architectural leadership, x86 architecture big leap forward…”
The chief executive officer of AMD stated that before starting to develop the next-generation high-performance architectures, the company talked to its customers to find out whether they actually wanted new server chips from AMD. Apparently, all replied positively.
“We have had very good conversations with the OEMs in the server space and with the customers,” said Mr. Read. “We tested it because if we did not see that acceptance we would have ended the [server] business a year ago. But we saw that opportunity and we saw the opportunity to rebuild that and to do it in a more consistent way and I think the next generation technology will position us. Now we have to execute and deliver.”
The only thing currently known about AMD’s Zen is that it will drop clustered multi-thread (CMT) design in favour of more traditional simultaneous multi-threaded (SMT) design. This may result in decrease of the amount of cores inside AMD’s processors, but will increase their efficiency. Both Zen and K12 will likely be compatible on the platform level and both will use similar memory controllers and other common blocks.
Two the key reasons of relatively low performance of AMD’s current-generation microprocessors and accelerated processing units are believed to be their CMT design (which involves so-called “modules”) as well as relatively low performance in single-threaded operations. Each CMT module inside AMD APUs/CPUs is identical to a dual-core processor in its integer power, and to a single-core processor with SMT capability in its floating-point power; besides, each module shares certain resources between two “cores”. As a result, in many cases AMD’s eight-core microprocessors cannot outperform Intel’s quad-core chips that have the same number of floating point units (FPUs) which are more efficient compared to AMD’s FPUs.
As reported, both K12 and Zen are due in 2016.
Discuss on our Facebook page, HERE.
KitGuru Says: The word “Zen” can be approximately translated as “absorption” or “meditative state”, according to Wikipedia. Not a bad word to call a high-performance x86 micro-architecture… What is important, of course, is how well will it perform. Unfortunately, this will be evident only two years down the road.
How much better AMD would have been with a 32nm Thuban in the beginning and latter with better revisions of that chip?
The webcast can be found here: http://www.media-server.com/m/p/kebrnbpc as soon as it’s posted.
Also, AMD needs to tread carefully, what I’m reading there is “We’re betting the farm on server chips because we can’t compete on the desktop”, and considering AMD’s fame for not delivering the performance it promises and once they fail, skewing their marketing towards the value segment, it’s just a “seeing is believing” game.
For a company who shed it’s entire manufacturing division just to focus on design, it has done very poor design.
actually Rory said a whole lot more than you mentioned in your article… http://seekingalpha.com/article/2482085-advanced-micro-devices-amd-ceo-rory-read-presents-at-deutsche-bank-technology-conference-transcript
I think you may be projecting a bit too far on that.
In this context and with this information release the focus is on server chips because of the known demand and industry competitive designs. On the Consumer PC market they were not directly competitive and had to be the value option, on the server side they were not the underdog to Intel they were on the consumer side. This is an instance of leading with the best foot, they know, as they stated themselves, that bulldozer was a flop, and they want to reinvent their lineup coming from the server side.
I get the feeling you will see the new Zen cores in servers then make a quick leap to APUs and the eventual merging of FX into the APU socket with Zen based higher core count no GPU CPUs with the FX monicker, and the APUs continuing with better CPU architecture to go along with the currently leading IGP solution.
On bulldozer specifically, I believe it may have been on emulators.com, but there was a run down of why bulldozer didn’t deliver like was promised, and it wasn’t the idea or the foundation it was the implementation and components and design of the internals that let it down. Had they gone full bore into bulldozer and gone with the best available options rather than the most affordable they would have had the performance to directly compete with Intel, but they apparently wanted the price edge enough to cut corners that bit them on performance.
Hi Hal,
Yeah I posted later that Rory said a lot more than what was covered in KG’s article. The whole presentation is a bit rehearsed for the investor audience. Lots of reassuring and “assertiveness” going on… a bit “we know where we’re going and we’ll surprise the world” kinda stuff.
– Zen/K12: AMD has always done this… start on the servers and then move the architecture to the desktop (as opposed to Intel’s desktop first, then server). If they fail on servers, they won’t have a desktop solution.
– Carrizo (the new APUs): which I actually noticed has some details published on wccftech, it seems that AMD is just pushing its current line-up and not reinventing anything. Hexacore on 20nm sounds good and might bring the power efficiency AMD needs, but I don’t see it as a contender… if the info on WCCFTECH is to be trusted, it’s got 6 cores, 512KB per core like Athlon II, @4.4GHz (A10-8890K) and some architectural changes that are a throwback to prior architectures.
The comment about the value-for-money segment being a “dollars and cents” business is quite true, but then again, why did he mention it if that is all AMD has (and will have) for a rather long time? I’m sure AMD is living on borrowed time thanks to the console market, in 3-4 years that won’t help them with Intel’s dominance in performance-per-watt.
I also found it quite strange that there was almost no information about ARM-based CPUs.
thoughts?
I actually think a lot of what they already have published is going to be modified or expanded soon. I think we are still being told a lot from the wind down of previous gen rather than the truth of what next gen is going to bring.
I mean they brought back Jim Keller and the amount of time he’s put into this so far does not jive with the schedules released as of yet. I think they may have a big announcement possibly by the end of the year, more likely middle of next year on a big change. Possibly including the ARM info. More thoughts on a lot of that can also be found in the posts on emulators.com, but Intel and AMD both kind of sidestepped an avenue for real robust enterprise engineering and I think they may be getting back around to it.
As for the APUs I think until 2016 we won’t see any real news, rather we will see the lame duck winding down of the old design. There are improvements to be made, and there will still be competition, but I think AMDs big push, and big attempt to retake a competitive edge across the board will be coming with the design of Zen/K12, and their assimilation of ARM into server applications. Whether its robust low level virtualization, remaking their bulldozer based module paradigm into a more conventional hyperthreaded core setup, or if it will be merging x86 and ARM with current APU/GPU tech for a CPU/ARM/GPU HSA setup with a wider spectrum of reliability, efficiency, flexibility, and performance. I don’t know. But, I really hope to see them push something new with some real advanced thinking.
The moves they made over the last 5 or 6 years made a great foundation for HSA and a robust multi-faceted approach to enterprise computing. And I really want to see something new and interesting. Heck imagine something like ARM cores integrated into a normal x86 architecture with GPU interconnect for amorphous workload and on the fly multithreading where at one second you could have generic single thread CPU function at 5 watts, then huge dozens of threads cycling up and down as the need for parallel workloads changes, shunting work from ARM to x86 to GPU as complexity dictates, reclocking and retasking on the fly. Massive performance improvements with the capacity for huge power savings.
If you’ve not seen the emulators.com site his blog postings are very informative and reference as far back as P3/P4, and goes very in depth into cpu architecture and instruction sets. And a lot of what he wanted to see happen ten years ago is now starting to come to pass as the lines between enterprise computing and consumer hardware are blurring.
So five years with the 990FX Chipset and Steamroller is forgotten forever, because they know CMT or so-called “modules” were not doing what they anticipated. Plus I remember then Windows 7 didn’t have the so-called Module support when Bulldozer first hit the market. When Microsoft got around to supporting their so-called Modules it still didn’t matter as the results were the same.
Nevertheless still looking forward to their Zen K12 Architecture in 2015 I mean 2016.
PhoneyVirus
https://twitter.com/PhoneyVirus
https://phoneyvirus.wordpress.com/
probably enourmously, an 8 core 32nm Thuban with some refinements would have probably destroyed Bulldozer/Piledriver, at least in Single Threaded apps.
Thanks for posting. On page 2…a lot of information here.
I’ve seen finegrained preemption on the carizzo apu… if that is true then hsa is complete an theoretically instructions could trigger context switches. If this comes with carizzo or not we’ll see but it would be a huge step forward. Imagine executing all simd & media encoding etc. instructions on the gpu automatically…
Nope. We already saw “K10” on 32nm, that was Llano, and even though it had a 6~7% IPC increase on average over Thuban, it’s poor clock range offered no advantage over a moderately-OC’d 45nm K10 chip. People forget that “K10” i.e. 10h were nothing more than a refined K8 design, itself having it’s origins way back in ’99. They’d already hit a wall in terms of what they could extract from the design.
But still a possibility, the original Pentium M design came back from the P6 design that dates from 1995, and even the latest Core i7 has it roots based on the Pentium M design which was improved later as a Core Duo, then Core 2 Duo and so on.