You’ve got to hand to AMD; for as long as we can remember Team Red has chosen Intel’s Developer Forum to share information about its latest products with the press.
This year however, AMD may have stolen Intel’s thunder altogether.
We say this because in San Francisco this week AMD shared information about its long-awaited processor refresh, Zen. Sadly the wait for Zen has been extended until 2017.
Until now, the most information we’ve managed to get from AMD about the chip was a demo of sorts during E3 in which the chip was shown to be running DOOM at 1080p and 60 frames per second.
Strictly speaking that isn’t a true display of what the CPU could be capable of but what AMD shared this week is.
Power without the demand
There are a number of improvements in the Zen architecture, most notable of which is the improved power efficiency of the chip. AMD is switching to a high-density optimised 14nm FinFET process, not unlike Intel’s Skylake CPUs.
The benefit of this is that AMD would be able to:
a) squeeze more performance out of the CPU at the same power draw as Excavator or;
b) get the same amount of performance we saw in previous chips but at a lower power.
This power efficiency was the key driving factor behind Zen’s development and AMD constantly hammered this point into its engineers according to a report by Anandtech. The result is that AMD claims Zen will exhibit a 40% total efficiency gain over the last generation.
More efficient cores
While on the topic of efficiency it would be amiss not to mention simultaneous multi-threading (SMT). We know that Intel has been using this tech for some time and now AMD joins the party.
Essentially SMT supports two threads per core meaning we could see octa-core CPUs with 16 threads when Zen releases. The benefit of SMT for the end-consumer is that the core instructions per clock are improved. Which tasks will benefit from this newly supported feature however, remains to be seen.
More cache, more performance
Perhaps the biggest change in Zen’s architecture is its cache hierarchy. The L1 data cache is double the size it was in Bulldozer and it uses asymmetric load/store implementation to access work flows that are used more often to improve overall efficiency.
As to the L2 and L3 cache, AMD has placed lower level caches closer to the core with a view to improving single thread performance. The downside that usually accompanies this is that cores will have to check back with each other to insure that the data they are using is still relevant.
This adds a process to the mix which could create a potential bottle neck but AMD has said that the cache bandwidth to each core is five times that of previous designs. Let’s hope that’s enough to stymie any bottlenecks.
Let the competition, begin
The pay-off with all of this is that AMD might have a chip on its hands that may just shake Intel, a bit. Granted, it will be hard for AMD to challenge Intel for the space at the top of the market that it has been enjoying for a number years but it could mean seeing more AMD processors along side Intel’s in PC stores.
What’s important about Zen though is that AMD is showing real improvements. At the event in SF, Team Red ran a Blender rendering demo using Zen which out performed an octa-core, 16 thread Broadwell-E CPU.
With the release of Zen now slated for next year AMD has said that it will reveal more information about the CPU as the release date draws nearer. This should include concrete performance numbers as well as the most important spec in an AMD chip, thermal design power.
It’s also worth noting that Intel is still sitting on its next CPU family, Kaby Lake. The follow-up to Skylake has also been delayed to 2017.
Whether you side with Intel or AMD though, 2017 is shaping up to be a great year for PC builders.[Source – Anandtech]