AMD’s new CPU architecture, codenamed “Zen”, has been in development for couple of years and should be arriving in Q4 2016. Reports suggest the development started in 2012 lead by legendary CPU architect Jim Keller who built AMD’s K8 CPU architecture featuring first 64bit desktop CPU called “Athlon 64” which dominated Intel’s offerings at its time. Keller returned to AMD to help to develop a new architecture after AMD was struggling to compete with their disappointing Bulldozer architecture against Intel latest offerings. But what does Zen really means for AMD and its future?
Zen, AMD’s attempt to get back into high-end CPU market
Zen is AMD’s new CPU architecture developed completely from scratch, it is built on new 14nm FinFET process and will feature up to 8 cores. Zen is also integrating previously external motherboard components and will utilize new AM4 socket which is unified for both future CPUs and APUs, which integrates previously external motherboard components thanks to Zen’s SoC-like nature, and support DDR4 memory. AMD also moved from Clustered Multi-threading model to traditional Simultaneous Multi-threading model. Zen will bring completely new low latency cache subsystem, which was one of the main issues with Bulldozer architecture, to help increase the performance. It will also support all latest motherboard features such as M.2 drives, USB 3.1, NVMe as well as new security features. All this combined should bring over 40% Instruction Per Clock increase over current Carrizo architecture and big efficiency improvements thanks to 14nm FinFET process.
Recently a Corporate Vice President of worldwide marketing at AMD John Taylor told ITWire some interesting new info about Zen. Taylor says that Zen will compete with Intel’s latest offerings on performance, power and specifications and not just price. Taylor is also confident that with Zen AMD will change any negative perceptions about AMD customers may have. As an example of this he pointed at recent support for giants such as HP, Dell and Lenovo and their high-end laptop offerings.
This confidence is nice to see on AMD’s side considering AMD hasn’t updated their High-end CPU offerings in long time. With Zen AMD could finally offer some proper competition to Intel’s high/top-end offerings such as 6700k or upcoming 6950X and replacement for their older Opteron server CPUs where AMD reported several Zen design wins already in their Q4 2015 earnings call. Not only that but AMD will also introduce Zen based APUs in 2017 along with their new Polaris GPU architecture which might be able to replace low-end GPU segment where GPUs like R9 360 and GTX 950 are present. Rumors also suggest that Zen APUs will power future next generation consoles.
How much does Zen need to catch up?
Over 40% IPC improvement is nice but is it going to be enough? To put this into perspective we’ve prepared single-core Cinebench R15 benchmark comparison where we used Athlon X4 845 which features Excavator architecture which is AMD’s best IPC showcase against i7 6700k featuring Skylake architecture which is Intel’s best IPC showcase. Both CPUs were clocked the same at 3.8GHz. One might argue that Cinebench is not the best benchmark for comparing CPUs due to it being a synthetic benchmark and doesn’t represent real world performance, which is true, however Cinebench is one of the best representations of how much of a difference there really is when it comes to IPC.
Because we clocked the CPUs the same we can see the actual IPC difference between those 2 CPUs. The IPC difference is about 42%. If AMD delivers at least 40% improvement over Excavator like they claimed in their numerous presentations we can conclude that Zen actually has a chance of being good competitive CPU against Intel’s latest offerings.
Sadly apart from this we don’t know almost anything else about Zen architecture from official sources but even with the limited information not to mention this is just ONE benchmark and more so it’s a synthetic benchmark so you should take it with a grain of salt because in some situations Zen might gain more performance than in the others but I still consider Cinebench as good comparison between architectures because it scales well with clockspeeds, core counts and architecture improvements.