Rumor around the 'Net is that AMD's Bulldozer follow-up, codenamed Vishera, could drop in early Q4 this year. The new AMD 8350, as its known, will supposedly be a four-module/eight-core design clocked at 4GHz (4.2GHz Turbo Mode) with a 125W TDP. Vishera could potentially deliver what last year's Bulldozer didn't; AMD's latest CPU was widely panned for failing to meet most of its performance, clockspeed, and power consumption targets.
Given that Vishera is based on the already-launched Trinity core, we can make a few predictions concerning the chip's performance -- and whether it can measure up.
AMD's Trinity incorporated two technologies that substantially reduced power consumption -- hard flip-flops and a resonant clock mesh. Vishera will assuredly do the same, and this, combined with further improvements to GlobalFoundries 32nm process, makes a 4GHz target frequency at a 125W TDP sound feasible. The low turbo mode, if accurate, implies that the power consumption curve is still quite steep.
4GHz would give AMD a bit of a PR opportunity (Intel's fastest official CPU topped out at 3.8GHz in the P4 era) and would finally distance Vishera from Thuban. One of the criticisms of Bulldozer was that at 3.6-3.9GHz, it often struggled to keep pace with (or fell behind) AMD's K10-based Thuban at 3.3GHz. The low-level improvements baked into AMD's second-gen Bulldozer, combined with an 11% base frequency boost, would finally distinguish the two products -- at least most of the time.
Roughly six weeks back, Anandtech's Johan De Gelas posted an insightful deep dive
into Bulldozer's branch prediction penalties, cache miss rates, and cache latencies. His work strongly indicates that slow caches aren't the only thing plaguing the first-gen Piledrive's design, and cast light on where AMD might try to improve future products. Vishera is based on Trinity, but AMD didn't disclose much in the way of front-end improvements when it described the chip prior to the mobile launch
earlier this year.
If Vishera is Trinity + L3 cache, our expectations for its performance are similarly modest. If AMD went back to take advatage of higher TDP levels and optimized certain parts of the front end, it could deliver a CPU better able to challenge Intel's Sandy Bridge / Ivy Bridge desktops. Given the way the desktop market continues to shrink, we suspect AMD will likely take the former route and save significant revisions for Streamroller. Steamroller is the third-gen APU currently targeted for sometime in 2013. To date, AMD has shared very little about this part, including its release date or which foundry is manufacturing it. Similarly, there's no word on whether or not the company will make an eight-core third-generation part.