by Anna Filatova , Ilya Gavrichenkov
09/24/2009 | 08:50 AM
When we were getting ready for the IDF Fall 2009, we anticipated to hear some exciting details about the upcoming 32 nm processors based on Nehalem microarchitecture. Yes, I am talking about the CPUs known as Westmere.
According to Intel’s “Tick-Tock” concept, that they have been actively promoting for the past few years, Westmere should become Nehalem’s die-shrink manufactured with new 32 nm process. Intel confirmed that they have no problems with 32 nm process, it has been certified and the first semiconductor wafers are already coming off the production lines. Within the fourth quarter they are planning to launch mass production. First mass production solutions using 32 nm cores will start selling in the very beginning of 2010. And this time Intel decided to focus primarily on desktop and mobile mainstream processors codenamed Clarkdale and Arrendale.
Our today’s short report is going to dwell on Clarkdale. As for Arrandale, it will offer very similar functionality but adopted for the mobile segment.
Note that the introduction of 32 nm technology into the desktop computer segment will start with general-purpose systems rather than high-end, where first 32 nm six-core Gulftown processors will only appear 6 months later. The first design of the 32 nm Westmere semiconductor die will contain only two cores, although they will support Hyper-Threading technology, which means that the operating systems will see Clarkdale and Arrandale as processors with four computational cores. This is what the processor die of Clarkdale/Arrandale solutions will look like:
As you can see, there is not that much cache here compared with the rest of the CPU. Don’t be surprised, first-generation Westmere processors with two computational cores will only have 4 MB of L3 cache onboard. However, the new manufacturing process will allow raising their clock frequencies: the top Clarkdale CPU will work at 3.46 GHz. Hyper-Threading technology together with high clock frequencies will allow Clarkdale processors to perform very successfully in the same field as quad-core Yorkfiled and even Lynnfield CPUs. However, they won’t completely oust Lynnfield CPUs from the scene that is why those of you who have already bought Core i7-800 and Core i5-700 processors won’t regret it.
In other words, 32 nm Gulftown and Clarkdale will successfully coexist with their 45 nm predecessors for quite some time after their official launch.
And it will partially happen this way because Clarkdale (and Arrandale) is not just a dual-core CPU. It is the first and the only processor with integrated graphics.
However, the graphics core will still not be integrated directly into the actual CPU die. Two dies – CPU and the graphics accelerator – will simply be put together onto a single processor circuit board.
Note that only processor die will be manufactured with 32 nm technology. The graphics core will still be made with 45 nm process. But you shouldn’t think that the CPU and the graphics accelerator combined on a single PCB will exist independently from one another. Intel engineers are still going to implement some kind of communication between them. The most interesting part of this interaction is the Turbo Boost graphics-processor technology called Graphics Turbo that will accelerate the graphics core when the CPU is not loaded with work too much. Keeping in mind that this dual-die implementation of GPU and CPU in Westmere processors will use the same cooling solution for both, this whole solution looks very sweet. However at this point they are planning to implement graphics Turbo Boost only in mobile Arrendale processors.
As soon as the graphics core will migrate into the same processor packaging as the actual CPU die, the entire system architecture will change dramatically. The launch of the new LGA1156 platform has already transformed the system core logic into a single-chip solution and the entire system now consists of two major microchips instead of traditional three: CPU and South Bridge. The introduction of embedded graphics part won’t change this structure. Clarkdale processors will be LGA1156 compatible and systems based on them again won’t need a North Bridge.
They will only add a KVM-Switch into the existing South Bridge and the pure video signal will be transferred from the CPU to the South Bridge via special Intel Flexible Display Interface bus. Other than that, nothing will change.
As you all know, the current generation mainboard with Intel’s integrated graphics core is G45. The new 5-series chipset for the upcoming Clarkdale processors will have another big chip onboard. But the overall mainboard PCB design is lower-dimension. It offers more freedom in the X and Y dimensions as well as in Z dimension without compromising the performance.
Today Lynnfield processors used P55 chipset. With the introduction of Clarkdale there will be also H- and Q-chipset series. H-series will be targeted for home users and Q-series - for business users, that is why the Q-series chipsets will also support vPro technology. Theoretically, Clarkdale processors should be able to work in existing Intel P55 based mainboards, but in this case you will need an add-on graphics card as well.
Just like in Lynnfield processors, the new CPUs will have an integrated memory controller and a PCIE controller therefore there will be no need for the North Bridge. By the way, specifically due to the PCIE controller Clarkdale based systems will also be able to work with discrete graphics accelerators. Actually, there is real practical value in using external add-on graphics cards, because the graphics core in Clarkdale processors is a direct successor to Intel GMA. In other words, it comes from the GPUs that used to be integrated into chipsets.
However, the new GPU will undergo some serious improvements. First, Intel Clear Video engine will be able to hardware decode two high definition video streams simultaneously. This feature may be used for Blu-ray playback with enabled picture-in-picture mode and minimal CPU utilization.
Also, Intel has completely redesigned the user interface for their graphics and media control panel. Now it looks overall better and more feature rich.
As for the 3D part, the number of shader processors will increase to 12 (Intel G45 only has 10 of them) and their operational frequencies will be between 733-900 MHz (Intel G45 supports 800 MHz frequency). The amount of video memory that the GPU will be able to take away from the system RAM will also increase to 1.7 GB. Overall, Intel promises that Clarkdale graphics will work about 1.4-1.5 times faster than the integrated graphics core in G45 based systems.
Coming back to the processor core, we have to point out that it will have only one significant microarchitectural modification compared with 45 nm Nehalem processors. Namely, Intel is going to introduce support for the new AESNI instructions set in their new Westmere CPUs. AESNI is a set of six utilitarian instructions pursuing only one goal: accelerating the AES encryption algorithm.
Intel claims that Windows 7 operating system already supports new instructions and right after Clarkdale launch, this OS should start using this new set for all of its encryption algorithms. Moreover, they will also find their way into all sorts of software applications, but, of course, only after it has been recompiled for the new instructions. For example, Intel demonstrated an upcoming version of WinZip archiving tool. During encrypted data compression on CPUs with AESNI support, it may speed up by up to 1.5 times.
So, this is what the base features of the new CPUs codenamed Clarkdale will look like:
At this point Intel doesn’t disclose any specific frequencies, but we can tell you that the frequencies of the new Clarkdale processors will lie in 2.8-3.46 GHz range. Moreover, they will be offered within the new Core i5 series as well as under Core i3 and Pentium brands. At this point we expect to see the following model lineup:
Click to enlarge
And in conclusion let us say a few words about the performance. Intel didn’t let use run any benchmarks on a Clarkdale based system ourselves. So, at this point we can only share those few numbers that were reported by the manufacturer:
According to SPEC CPU2006 test, Clarkdale processor working at 3.33 GHz will perform close to Intel Core 2 Quad Q9400. As we can see, even though Clarkdale is a dual-core CPU, it performs as fast as the previous generation quad-core working at 2.66 GHz. If we compare its performance against that of a 3.0 GHz dual-core processor, it will be almost 1.5 times faster.
As you know, most of SPEC tests have a memory component in them. Clarkdale delivers much higher memory bandwidth and provides a lot more headroom to work with compared with LGA775 platforms.
Therefore in SiSoft Sandra test we also see a massive memory bandwidth advance. So, if, for instance, you work with a lot of HD content and applications like Cyberlink PowerDirector, you can really see improvement in speed.
PCMark Vantage offers us a more optimistic prognosis for Clarkdale performance. Here we can actually compare the new CPU with the top quad-core Yorkfields. Clarkdale owes some of its success to the fact that PCMark employs operating system algorithms, and Windows 7 supports AESNI instructions. So, we will only see this advantage under Windows 7 OS. Under Windows Vista the result will most likely be lower.
And here are the numbers that will allow us to draw some conclusions about the graphic core performance (the tests were run in systems with integrated graphics). 3DMark Vantage clearly shows that Clarkdale GPU is really 1.5 times faster than the Intel G45 integrated graphics core. As for the processor performance, the dual-core Clarkdale again runs just as fast as quad-core Yorkfield.