Phenom II X6: A Family Of Two
AMD is launching two Phenom II X6 processors: the 1090T and 1055T.
Both models are architecturally identical. That is to say they’re monolithic hexa-core designs with 128KB L1 cache per core (64KB instruction and 64KB data), 512KB of L2 cache per core, and a shared 6MB L3 cache.
This is, of course, a different approach than Intel’s. With its six-core Gulftown design, Intel increased the size of its shared L3 in order to maintain what its architects considered an optimal ratio between cores and cache. Because AMD isn’t also shifting to a smaller manufacturing process, it really can’t follow suit and keep the Phenom II economically viable at the same time. But as we’ll see in the benchmarks, simply increasing core count is enough to give the X6 a significant boost in threaded apps.
As I mentioned in my Phenom II X6 preview, AMD is sticking with its 45 nm process here. As a result, die size increases from 258 square millimeters to 346. And while AMD is hesitant to divulge official transistor counts, it admits that its Thuban design is closely related to existing hexa-core Istanbul-based Opteron CPUs. From that, we can safely assume the count lands around 904 million, up from approximately 758 million. That’s not as large an increase as Intel’s six-core design because Gulftown’s L3 is transistor-heavy. So, you can argue that AMD is getting the best return in performance for the added complexity. But while Intel was able to fit its 6-core/12MB L3 design into a smaller piece of silicon than the quad-core model that came before, AMD’s 346 mm² die represents a fairly large increase.
Though AMD won’t comment directly, we know through a handful of motherboard vendors that there will be Thuban-based quad-core CPUs following in the wake of the 1090T and 1055T. They’ll likely be armed with Turbo CORE as well, and apparently it’ll be possible for certain lucky buyers to unlock the two disabled cores. I don’t get excited about core unlocking for two reasons, though. First, it’s chance-based—you can’t tell if the processor you buy had cores turned off due to defect or simply to meet demand. Second, AMD’s price structure is so tightly packed that it makes the most sense to buy the CPU you want/need right off the bat, rather than roll the (wait for it)…dice.
|Header Cell - Column 0||Clock Frequency||Price||HT Speed||Lith. Node||Total L2 Cache||Shared L3 Cache||TDP||Voltage||Max. Temp.|
|Phenom II X6 1090T||3.2 GHz (3.6 GHz with Turbo)||$295||4 GT/s||45 nm SOI||3MB||6MB||125W||1.125-1.4V||62 C|
|Phenom II X6 1055T||2.8 GHz (3.3 GHz with Turbo)||$199||4 GT/s||45 nm SOI||3MB||6MB||125W||1.125-1.4V||62 C|
|Phenom II X4 965 Black Edition||3.4 GHz||$185||4 GT/s||45 nm SOI||2MB||6MB||125W (w/ rev. C3)||.825-1.4V||62 C|
New Tech, New Names
In a world where every piece of technology-turned-marketing needs to be rolled into the name somehow, AMD follows in Intel’s footsteps by tacking another letter onto its product naming scheme.
Intel has the Core i7-980X, which, dissected, means absolutely nothing informative.
AMD now has its Phenom II X6 1090T and 1055T. At least the II is indicative of a second-generation Phenom family. The X6, of course, tells us that these processors are hexa-core—that’s fairly easy for most folks to figure out. The 1090 and 1055 are arbitrary performance indicators, and the T tells you these CPUs are Turbo CORE-enabled.
The 1090T runs at 3.2 GHz by default (down from the Phenom II X4 965’s 3.4 GHz). When Turbo CORE is active, it can hit speeds of up to 3.6 GHz, though as our experimentation on the next page shows, this isn’t something seen on a regular basis. It sports a 2 GHz (4 GT/s) HyperTransport interconnect and is rated for a 125W TDP.
The 1055T operates at 2.8 GHz by default and runs at up to 3.2 GHz in Turbo CORE mode. It also boasts a 125W TDP, along with the same HyperTransport connection able to move up to 16 GB/s bidirectionally.
Both CPUs sport the same DDR2/DDR3-capable memory controller rated for DDR2-1066 and DDR3-1333 (officially). Performance-oriented memory modules listed in AMD's online database can be pushed at up to DDR3-1600, though, and simultaneously force lower latencies and overclocked northbridge freq./voltage settings.