AMD’s Ryzen Looks Promising
Double the cores, double the fun, but no six-pack
I’VE BEEN HEARING about AMD’s Zen, now Ryzen, for what feels like forever. The Zen architecture was first mentioned in 2012, and ever since, the repeated refrain of AMD CPU apologists has been “just wait for Zen.” That wait is finally over, and Ryzen is now available. But as we go to print, I still don’t know exactly what AMD’s Ryzen lineup will be in terms of core counts, clock speeds, and pricing. It’s quite the contrast compared to AMD’s open discussion of the Ryzen architectural details, not to mention Intel’s routinely leaked product roadmaps.
Here’s what I do know. AMD is planning on a full release of Ryzen processors, all multiplier unlocked. Ryzen uses a quadcore module as its building block, with SMT providing eight threads, similar to Intel’s Hyper-Threading, and 8MB L3 cache. AMD claims a 40 percent improvement in IPC compared to its existing Excavator architecture, with clock speeds of at least 3.7GHz on the eight-core part. Architectural changes include a reworked, more efficient pipeline. Each core is a six-wide superscalar design (four-wide fetch), compared to the previous four-wide design, the branch prediction is more accurate, and cache latencies and bandwidth are both improved. The result is that Ryzen should be much more competitive with Intel’s CPU offerings.
Interestingly, AMD apparently won’t have a six-core model; instead, it will have an eight-core chip without SMT as its mid-tier offering. That means that at each product segment, AMD will deliver double the cores/ threads as Intel. Intel will almost certainly maintain its lead in per-core performance, but if 4C/8T Ryzen SR3 goes up against Core i3’s 2C/4T design, or 8C/8T Ryzen SR5 versus Core i5’s 4C/4T, even with slightly lower clock speeds, Ryzen should come out ahead in many benchmarks. We’ll have the full skinny next month.