AMD's Turbo Core Technology
Thuban and Turbo Core
While AMD will not be releasing their actual 6 core desktop processors for a few weeks yet, they did want to release a few teasers about one of the bigger features they are including in this chip. AMD has had 6 core server chips for some time now, but because these parts are significantly larger in terms of die size than the previous “Shanghai/Deneb” based quad core 45 nm parts they were not exactly predisposed for the desktop due to clockspeed and TDP issues. Not only are these parts larger, but all those extra transistors running at multiple GHz speeds increased the TDP of the 6 core parts as compared to the smaller quad core Phenom IIs and Opterons. Certainly AMD could have released 6 core parts shortly after the server introduction, but these would not have had the impact many were hoping due to their relatively low clockspeeds which would cause the chips to stumble when running single threaded or lightly threaded loads.
Istanbul is the code name for the server version of the 6 core processor, and it was a slightly tweaked version of the previous 45 nm quad cores. Not a huge amount of work was done on the architecture, but rather some simple optimizations which should increase performance to a very small degree, and perhaps more importantly work on the power and thermals of the part. The desktop version is now codenamed Thuban, and it is again slightly changed from the previous Istanbul in a number of ways. While clock for clock performance is the same, AMD has added a major new feature that is not available in the Istanbul based chip.
Power and Heat
Before diving into the Turbo Core feature, we need to take a look at what AMD has done to improve the thermals and power draw of their upcoming Thuban part. Current Istanbul parts are rated by ACP rather than TDP, but we can somewhat extrapolate TDPs. The top speed processor is 2.8 GHz and it looks to have a TDP of around 125 watts. When we consider that a year and some months ago, the top Phenom II X4 940 was running at 3.0 GHz with a TDP of 125, we can see that AMD has been working to get power draw down.
GLOBALFOUNDRIES (GF) manufactures all of AMD’s processors right now, as it was previously the manufacturing arm of AMD. Now that it is spun off, we see that their overall process philosophy is basically unchanged when it comes to CPUs. AMD could not afford to do what Intel does with process tech, and so they had to take a “small steps” method to improving their lines. Instead of spending billions to create a new, advanced process at once (as Intel does) and leave it basically unchanged throughout its useful existence, AMD integrated new technologies into existing nodes. So when AMD would make a jump to a new process, it was not nearly as large in terms of thermals and transistor switching performance as compared to what Intel experienced. Those processes would however improve sometimes dramatically throughout its useful existence. GF is essentially keeping that method intact.
Thuban will stay in the 125 watt TDP range, even though it looks like we will see 3.0+ GHz speeds. GF apparently has included a new materials technology in the process line that it uses for Thuban. GF has been a leading manufacturer of SOI (Silicon on Insulator) in the industry, and they have relied upon that to minimize gate leakage. It seems as though GF has added another wrinkle to their 45 nm SOI process, and that is the further use of low-k dielectrics. This again adds to the efficacy of SOI by lowering leakage. It does not improve upon transistor switching performance though.
This is not a huge technological feat, as quite a few other processes around the world utilize low-k dielectrics. But this is another good example of how GF is able to inject new technologies into current process nodes to effectively keep up with Intel without spending the billions needed to develop a whole new process node from scratch. Also, GF does a lot of other work behind the scenes in terms of their process technology which again refines the node and allows for higher speeds and better thermals even on existing products. The improvements made by these process engineers have helped AMD to release their current quad cores in the revision C3 variant which now has a maximum clockspeed of 3.4 GHz all the while keeping in the 125 watt TDP envelope.
These improvements also allow the upcoming Thuban chips to run several of their cores at 3.6 GHz. Which leads us to our next topic.
Intel released a turbo mode with their i7 processors, which dynamically adjusted the clockspeeds of the individual cores, as well as totally powered down unused cores. This was a very complex, but effective feature which maximized processor performance across a variety of usage scenarios. AMD is releasing something similar, but far simpler, with their upcoming Thuban based parts.
The beauty of Intel’s technology was the granularity of control. If one core was being used, it would be clocked upwards quite a bit, while the other cores were shut down. If two cores were being used, those two would see a boost while the others were powered off. If all four cores were in use, but the top TDP was not being exceeded, then all four cores would receive a clockspeed boost.
Unfortunately for AMD, they did not have the time to integrate such a complex design into their chips, and instead have taken a rather simplistic approach to boosting clockspeed when all six cores are not in use. The processor is essentially divided into three parts. The uncore is clocked at its usual speed and does not change or power down. The six cores are essentially divided into two processor blocks of three a piece.
When the processor determines that three or less cores are being fully utilized, it will place the underutilized cores into a lower power, low clockspeed state. In this case the cores will be clocked at 800 MHz and will have their voltage decreased. The other three cores will be placed into a boost mode, with their clockspeeds increased as well as their voltages per core increased. AMD is designing these cores to be run between 400 MHz and 500 MHz faster. A theoretical six core clocked at 3.2 GHz would then see a boost of three cores up to 3.6 GHz.
By downclocking the other cores and cutting their voltage, AMD is able to keep the processor in essentially the same TDP envelope. Unfortunately, AMD does not have the ability to totally power down and shut off those cores, unlike Intel. The big key here to take away is that this process is invisible to the OS and applications that the CPU is running. It is a totally controlled in hardware, so at most a BIOS update is needed for motherboards. No special drivers or software (such as AMD’s OverDrive) are needed to cash in on this performance tweak.
This should be a performance boon for single threaded, or lightly threaded applications. When combined with the extra performance features that Thuban has as compared to the older Shanghai/Deneb based cores, we should see a nice overall boost for the Phenom II line.
AMD also will be releasing a new quad core based on the revised architecture, and it too will have the Turbo Core functionality. There is currently some conflicting information if this new quad core is in fact a new, standalone design or if it is simply a six core processor with two cores disabled. The information that I have is that it is in fact a new quad core design, and if that is the case then we may be seeing the entire Phenom II X4 (and smaller) lineup get a change. Even if this is a new native quad core design, some of the new X4’s will most likely utilize the 6 core chip that either did not bin well with all six cores, or has one or two of the cores totally defective.
While this is not exactly a whiz-bang feature which will attract consumers in droves, it is a nice update which will have a fair impact on overall performance and usability for a variety of users. The two extra cores that the X6 packs in will improve the performance of those “mega-taskers” while the speed boost in lightly threaded content will be an added plus for those who do not hammer on all six cores 24/7.
A full review featuring these new processor will be available in a few weeks time. Stay tuned!