Core i5 661, the new and the old

Now before we start throwing names and numbers, there are couple of things that should be crystal clear to anyone who wants to understand this review content. It would seem that both Intel and AMD (GPU makers not excluded) have a mission in disorienting and confusing its buyers and generally population that like to be "in touch" with current IT technology with their naming schemes. Depending your source of information you can read a couple of different names, all referring to one product, or one two, or product line or... you get the picture. Here's a simple guideline to the meaning of some names in text that will follow:

• Nehalem - name for microprocessor architecture, 45 nm process node
• Bloomfield- microprocessor core codename, based on Nehalem architecture
• Lynnfield - microprocessor core codename, based on Nehalem architecture
• Core i7 900 Series - brand/model/SKU name for processors based on Bloomfieldcore and Nehalem architecture
• Core i7 800 Series - brand/model/SKU name for processors based on Lynnfieldcore and Nehalem architecture
• Core i5 700 Series - brand/model/SKU name for processors based on Lynnfieldcore and Nehalem architecture

• Westmere - name for microprocessor architecture, 32 nm process node
• Clarkdale - microprocessor core codename, based on Westmere architecture
• Gulftown - microprocessor core codename, based on Westmere architecture
• Core i7 980X (not yet released) - brand/model/SKU name for processors based on Gulftown core and Westmere architecture
• Core i5 600 Series - brand/model/SKU name for processors based on Clarkdale core and Westmere architecture
• Core i3 500 Series - brand/model/SKU name for processors based on Clarkdale core and Westmere architecture

In addition to the naming schemes above you should also be familiar with two desktop platform names Intel will be using for next couple of years. Both platforms are named after the pin count of its corresponding processor. The current LGA 1366 platform consists of Core i7 900 Series processors paired with an X58 chipset based motherboard, while the cheaper and less feature packed LGA 1156 platform consists of Core i7 800 and below series processor paired with P55/H5x chipsets. That makes our Core i5 661 test sample LGA 1156 based, and to use all of its features you'd need a H55/H57 chipset based motherboard but it will work just fine on most current P55 based motherboards (check your CPU support list), maybe a BIOS update is required.

If you are to distinguish one Core model from another, the list above is a must-know, with that obstacle out of the way we can finally talk about Westmere, Clarkdale and Core i5 661, which are more or less the same thing but with different meanings. I consider myself leaning on the optimist side, so we'll start of with good points in Core i5 661, and that would be the Clarkdale CPU core.



Clarkdale brings a few technological advances that will affect the market a great deal in the near future. The core is built using Intel's brand new 32 nm process node, second generation high-k metal gate and for the first time Intel is using immersion lithography during die production. For Intel this translates to better yields, smaller core dies and low production costs, and hopefully this means cheaper prices for the average user as well as better performance, power consumption and overclocking results. Intel will be using this new 32 nm process node until 2011 when their 22 nm process node is scheduled to launch, so there is a lot of time to fine tune production, up the frequencies and lower the power consumption.

Clarkdale's CPU core consists of two physical logic cores in one die with both of them supporting Intel's Hyper Threading technology and that makes it "effectively" a Quad Core processor. It reports four cores to Windows and applications, but in reality each core processes up to two threads each, utilizing all of the core's potential and performance. Improvements are noticeable across the board with Hyper Threading, but when dealing with heavily multithreaded and demanding applications four physical cores are always a better choice than two additional software cores. The other part of Clarkdale is the IGP or integrated graphic processor that holds both memory and PCI-E controllers, more on that part later.



Each of the two CPU cores in Clarkdale has 256 KB L2 cache memory and a total of 4 MB L3 shared cache memory. That's half of what Lynnfield has, and a third of what is available on Bloomfield processors. Clarkdale is also missing some key features that made Nehalem architecture so famous: integrated memory controller and PCI-E bus controller. Although they are physically close, and on the same package, performance wise it's almost the same as if they were implemented on the motherboard. The small 32 nm node production and absence of IMC and PCI-E bus controllers make the CPU part quite compact. With a die size of just 81 mm², it is over a third smaller than the Lynnfield core. Clarkdale's 45 nm IGP part adds another 114 mm² of die size, adding up to a total of 195 mm².

As for the architecture of the CPU cores, with some tweaks here and there to accomplish the 32 nm transition, the CPU cores are still based on the same core design used in Intel's first Nehalem processors, apart from smaller cache and the memory controller being dumped off core again. What's new is seven added instructions called AES-NI, all of them used in encryption/decryption tasks. For most users this is of minor importance, but large companies that encrypt every byte written to hard drive will be happy with the new features.



Turbo Boost still plays a big role in Clarkdale processors, overclocking it up to 3.60 GHz and that's a quite a boost for single threaded applications. With medium load on cores, Core i5 661 will work at 3.46 GHz, scaling down to 3.33 GHz once the 87 W TDP is reached. Now, I did play around with the Turbo Boost feature a little, and it started out as a simple experiment but it quickly became obvious that Core i5 661 will almost never work on its stock 3.33 GHz frequency, no matter how much load you put on cores. No complaints from me there.

Like in Nehalem based processors, Turbo Boost in Core i5 661 Westmere based models works on the same principle. Base clock BCLK always stays at 133 MHz which is multiplied by x9 to get ~1.2 GHz for the idle frequency, by x25 to get stock 3.33 GHz and Turbo Boost then kicks in with multipliers of x26 and x27 to get 3.46 GHz and the maximum of 3.60 GHz. Keep in mind that different models will work with different multipliers to get different values. On paper it looks great, but in reality most of the applications and games today know how to use at least two cores, so most of the time you'll be getting 3.46 GHz from Turbo Boost. Still not bad, but it would seem Turbo Boost makes a whole lot of sense with Quad Core processors. You have horsepower (physical cores) when it comes to heavy multithreading and speed (Turbo Boost-ed one/two or three active cores) when it comes to some light threaded applications.