This is something of a surprising announcement and even one that may bother Intel a bit. AMD has already shipped the first silicon (presumably engineering samples) for its latest Trinity processor design to board-maker ASUS. In addition CEO Rory Read announced to AMD’s investors last week that Trinity has achieved a doubling of performance per watt and a raw 25% performance increase over previous generations. If accurate - and trust us companies mislead investors one heck of a lot less than their customers - Trinity will be a fairly game-changing experience, especially in the crucial mobile sector. The improvements in the also-shipping Brazos 2.0 are more evolutionary, with a slightly faster ‘Bobcat’ CPU, sporting ‘Turbo Core’ (dynamic clockspeed dependent on the number of active cores) as well as USB 3.
The two new models of processor being shipped are both ‘Accelerated Processing Units’ (APUs), a combination of CPU and GPU that’s proven extremely popular in laptops and is gaining some traction in desktops. The most groundbreaking APU of the two is the ‘Trinity’, a successor to Llano, and will be available in A4, A6 (dual core) as well as quad core A8 and A10 models. Meanwhile the creatively named Brazos 2.0 is the successor to... well the first Brazos, a popular notebook/netbook part that competes fiercely with Intel’s Atom. Trinity and Brazos are currently geared solely towards further improving AMD’s fortunes in APUs - the one area where it is at least on fairly equal footing with Intel. Those looking for a higher powered enthusiast CPU after the Bulldozer debacle will have to wait a while longer for the improvements to filter through.
We say ‘debacle’ as the mix of PR circus, outright lies and overall lukewarm reception associated with Bulldozer’s release compares pretty unfavourably to Charlie Sheen’s coca-leaf derivative fuelled “I’m winning... honestly!!” campaign. Bulldozer did finally give AMD an all new Architecture (its first in thirteen years) but it was generally no faster than its Phenom II predecessor, bringing AMD mostly embarrassment on release. Thus Bulldozer’s only benefit seems to have been in giving AMD a much newer (and therefore improvable) design to work from. It’s with the Trinity APU that we’re seeing the first fruits of such a campaign to improve Bulldozer and by its own account it seems AMD has been pretty successful.
The CPU side of Trinity represents a crucial first outing for thus updated ‘Piledriver’ core and it’s considered important to see these kinds of improvements here both for Trinity itself and, as mentioned, other designs that will use the Piledriver core. The design sees several tweaks to Bulldozer’s design by AMD itself, accounting for some of the 25% performance increase over Llano (which used a modified Phenom II CPU) but is more interestingly the first implementation of a rather nifty new third party technology called ‘Resonant Clock Mesh’, licensed from Cyclos Semiconductor.
Clock Meshes: Totally Resonant with Today’s Crowd
Okay; a tl;dr version for our chronic ADD readers (Atomic cares). Resonant Clock Meshes hypothetically can improve performance, while substantially decreasing power usage. A more thorough explanation for those of us without access to extensive pharmacopeias follows...
This is all about getting a CPU to run on schedule, which turns out to be a surprisingly complex achievement. The traditional method is called a ‘Clock Tree’ and involves simply feeding how fast a CPU should run, in one end of the chip and hoping it reaches the other end correctly. This assumption requires that designers incorporate a delay into the CPU cycle, allowing slower regions to catch up - the delay is known as ‘Clock Skew’ (an option more extreme overclockers will be familiar with). As speeds increase over ~2GHz this has become more and more of a drag on performance; Cyclos’ estimate being that, in addition to consuming large amounts of electricity, Clock Trees and the resultant Clock Skew reduce performance by about 10%.