Like Intel, AMD obviously realizes that there are cases when all the processing cores inside their latest multi-core CPUs arenít being used. Even today, many software apps are only single or dual-threaded, leaving two core idling in a quad-core CPU, and four cores idling in the case of a six-core Thuban CPU.
To address this issue, AMD offered application profiles in their latest builds of Overdrive that would automatically OC the active processing cores when certain programs were run.
This software-based solution required AMD to constantly update their profile database in Overdrive to properly work, and of course it requires the end user to actually run AMDís Overdrive software at all times.
In other words, itís an inelegant solution to the problem. A hardware-based solution that requires no effort from the end user, yet still delivers comparable results is more desirable. In addition, Turbo Core can help make up for some of the clock speed deficit that Thuban suffers from in comparison to todayís Deneb-based Phenom II X4 processors.
With Turbo Core, Thuban is capable of watching the utilization of the CPUís six cores. If three or more cores arenít currently being used by the app, the CPU can reduce the power state on those cores, and Turbo up the three cores that are being used. The unused cores can be idled down to as low as 800MHz, while the active cores can be OCíed up to 500MHz on certain CPUs.
Notice that we said ďactive cores can be OCíed up to 500MHzĒ, and ďunused cores can be idled down to as low as 800MHzĒ. Itís important to note that these arenít 100% guaranteed conditions. To keep the Thuban CPU within its 125W TDP, the CPU canít always OC the active cores the full 500MHz. In some cases the active cores may just be clocked up 100MHz, or say 300MHz. We also noticed that the inactive cores donít always go down to the full 800MHz clock speed. Sometimes they may be running at 1000MHz. This is a little different than Intelís Turbo feature, which definitely does bump up the clocks the full 100% if certain conditions are met.
Also unlike Intelís Turbo Mode, AMD Turbo Core kicks in only when three or more cores are being used. This means in cases where four cores are being taxed, AMD Turbo Core doesnít do anything. It also doesnít completely shut down the inactive cores like Intel Turbo Mode does. As a result, it isnít as power efficient.
In other words, itís definitely still much less advanced than what Intelís offering with their latest CPUs, but itís definitely an improvement over relying on AMD Overdrive software. Itís an incremental step that will hopefully be improved when AMD launches their next-generation Bulldozer CPUs next year.
All AMD CPUs with Turbo Core technology will be denoted with a ďTĒ after the model number. So today weíre getting the 1090T and the 1055T CPUs, and in the future weíll supposedly see 1075T and 1045T processors if you believe the rumors.
With that discussion out of the way, letís go over the rest of the Thuban CPUís specs:
|AMDís Phenom II CPU Lineup|
|Phenom II X6 1090T BE||Phenom II X6 1055T ||Phenom II X4 965 BE||Phenom II X4 955 BE||Phenom II X4 945||Phenom II X4 925|
|# of Cores||6||6||4||4||4||4|
|Max Turbo Speed||3.6GHz||3.3GHz||N/A||N/A||N/A||N/A|
|L2 Cache Size||2MB||2MB||2MB||2MB||2MB||2MB|
|L3 Cache Size||6MB||6MB||6MB||6MB||6MB||6MB|
|NOTE: Excludes Phenom II EE CPUs|
If you recall our review of Intelís Core i7-980X six-core CPU, youíll remember that it ships with a larger 12MB L3 cache. Intelís engineers increased the size of the L3 to maintain what they believe is the optimal ratio between L3 cache and cores which is 2MB of L3 cache per core (although keep in mind that the L3 cache is shared amongst all six cores, so one core can use as much cache as it needs, including the full 12MB L3 for single-threaded apps). Intel was able to increase the size of the CPUís L3 cache without upping the chipís die size thanks to its smaller 32-nm manufacturing process. Thanks to 32-nm, Intel could add cache without making the chip bigger, and thus it isnít more expensive for Intel to produce (assuming of course Intel is getting equal yields which may or may not be the case).
With more L3 cache on board, the Core i7-980X was faster than the Core i7-975 even in cases where only four cores were being used. This is because the extra cache gave the CPU a performance boost in apps that werenít threaded for six cores.
Since Thuban is built at 45-nm, AMD doesnít have this luxury. The CPUís die is already 88 square millimeters larger than Deneb. This comes from the two extra cores. If they wouldíve bumped up the CPUís L3 cache, the chipís die size would be even larger and they wouldnít be able to price it aggressively and still make money. As you can see in the chart above, it therefore ships with the same L2 and L3 cache configuration as quad-core Deneb Phenom II CPUs.
While AMD wonít confirm it, the rumor mill says that AMD will eventually roll out quad-core Thuban CPUs later this year. These parts will merely have two of their processing cores disabled. Presumably the first CPU will be the Phenom II X4 960T. Rumors have also suggested weíll see a 940T part at some time as well.