The substantial, if not indefinite, delay of Trinity desktop parts, as well as server-centric focus of the remaining Piledriver chips that are expected – like 16-core 'Abu Dhabi', raises some questions.
It's good 3 months now since the notebook flavour of AMD 'Trinity' Fusion APUs arrived, including the high end K10 series. Their GPU performance is pretty darn well good for the die and process used, however the CPU core performance was kind of just OK, still somewhat lower than the Sandy / Ivy Bridge parts, including those in the same price band. Why? Well, it seems the "Piledriver" core design, that Trinity pioneered in an actual product, isn't that much of an advantage over the faltering predecessor, Bulldozer, in per-core performance.
For the mobile market, that can be still forgiven if there's power efficiency to throw in, which Trinity does seem to have – the laptops do provide decent 6 – 10 hour battery life. If AMD was able to tweak in the process and the design further, and enable a 17 – 20 W 'Ultrathin' Trinity with all 4 CPU cores i.e. two dual-core blocks plus full GPU, integrated, it'd be a very decent competitor to Intel's Ultrabook designs.
However, the desktop PC market has more of performance focus, especially in the CPU department – since discrete GPUs are more common on desktops, anyway. Now, there's no 'Vishera', the Piledriver-based 4-core block (8-core in AMD speak or 4-core multithreaded in Microsoft Task Manager) replacement for the infamous 'Zambezi' Bulldozer desktop out yet, and the desktop Trinity flavours, some of which our guys have in samples, are also postponed without clear, if any, release date. Only the "Abu Dhabi' dual die lower clock 16-core total server chip should be out next month. Why?
Well, one reason could be that, maybe, there is not enough improvement in per core performance from "Bulldozer" to "Piledriver" to justify the new product release, as it wouldn't improve the competitive position against Intel in the desktop segment at all. Whether it is 'instruction per clock' IPC or pure clock frequency, both have to contribute to the end performance to justify the product.
The other reason, linked to the first one, can be – likely – that AMD now has to place its hopes to the upcoming 'Steamroller' core early next year, coupled with the 28 nm process. I guess we can accept that and give the AMD CPU team more time to do that migration, as long as it is real soon now. Else, another excuse this time next year, to maybe wait for 'Excavator' to fix things, would be really unacceptable…
Leave a comment