Lynnfield Architecture
Today Intel is officially releasing the Lynnfield processor that many of us have been waiting for since the release of Nehalem last year. This time around we find better frequency scaling, lower prices and performance that will simply blow you away for the money and power it will cost you to run it. Intel’s dominance simply can’t be debated anymore.
IntroductionIntel’s latest (and possibly greatest) desktop processor launch is upon us today and with it finally come the answer to questions we, and we assume you as well, have had for months if not years. The processor codenamed Lynnfield has been on our radar since at least the middle of 2008 and has been slowly gaining steam as details of the CPU were revealed. Of course a large majority of the technology in the Lynnfield processor already exists in the current Nehalem-based Intel Core i7 processors; but Lynnfield has a few new tricks up its sleeve to make today’s release even more exciting for the average enthusiast PC builder, not the least of which is a lower price point.
Today I am going to walk you through the technical changes found in the Lynnfield cores, the various processor options available, pricing, platform considerations and course evaluate a metric ton of benchmarks and power consumption numbers to help you make the most informed buying decision possible.
Lynnfield Architecture – Nehalem plus or minus…
As noted above, the Lynnfield processors from Intel being released today are largely based around the Nehalem architecture that launched in November of 2008. My initial review of those processors includes just everything you would want to know about the design of the processors and the technical changes from the previous Core 2 architecture of CPUs. I will be briefly be going over some of those unique features here below but if you really want to take the deep dive into the world of the Nehalem architecture, you should really check out my previous review.
Die shot of a Nehalem processor
The Nehalem processor was the first monolithic consumer quad-core processor from Intel and the first to introduce several key new features that you might already be familiar with.
Key changes in the Nehalem architecture:
- Integrated memory controller – While AMD has been on the boat with integrated memory controllers on their Athlon line of processors (and now Phenom) for years, this was Intel’s first CPU on the desktop to move the primary memory functions from the north bridge to the processor itself. The advantage of this is drastically reduced memory latency with the drawback of a more complex CPU socket and slower memory progression times as making new chipsets was always easier than building new CPUs. The Nehalem architecture was the first to offer a triple-channel memory controller (compared to the dual-channel on AMD’s CPUs) and with it was able to push huge of amounts of raw memory bandwidth through the system.
- No more front-side bus – Again, first done by AMD’s Athlon 64 line of CPUs, Nehalem was the first Intel desktop CPU to remove the words “front-side bus” from the system. Instead of having that aged, slow data and performance bottleneck, Intel developed a point-to-point interface known as QPI (Quick Path Interconnect) that is used for communications between different processors on a multi-CPU system as well as between the Nehalem CPU and motherboard chipsets.
- Shared L3 cache – By introducing Intel’s first desktop shared L3 cache Nehalem was able to produce MUCH faster core-to-core communications on the same CPU. Previous dual-die quad-core processors from Intel used the front-side bus for communication, a process that was slow and prone to complication.
- Re-introduction of HyperThreading – The ability for one processing core to work on more than one thread of code at the same time is called SMT (simultaneous multi-threading), but Intel’s brand for their implementation was known as HyperThreading back in the day of the Pentium 4. The technology went away when we had the Core-based architectures but found its way back with the Nehalem processors effectively allowing a quad-core CPU to run eight threads all at once.
- TurboMode – This was Intel’s answers to the debate of processor choice: do you get the quad-core CPU with the lower frequency for improved multi-threaded performance at price X or do you get the higher clocked dual-core CPU for faster single-threaded performance at the same price X. Turbo Mode effectively allowed the Nehalem architecture to automatically overclock itself depending on the work load and thermal envelope available to it. If you are running just a single heavy thread then the Nehalem architecture would be able to overclock the frequency by 2 steps or so (266 MHz on average) as opposed to when the CPU was chugging on 3 or all 4 cores.
All of these features came at a cost though – the lowest priced Nehalem CPU was the Core i7-920 that launched at right around $300. Motherboards based on the X58 chipset (the only Nehalem-compatible chipset) were often priced over $250 and you needed to purchase three DIMMs to take full advantage of the triple-channel memory controller as opposed to the two needed for Core 2 or AMD-based systems.
The new Lynnfield core is an answer to that problem and comes in aimed squarely at the mainstream consumer and enthusiast.
The new Lynnfield core is an answer to that problem and comes in aimed squarely at the mainstream consumer and enthusiast.
The new Lynnfield CPU die is very similar to Nehalem
You can see in the die image above that Lynnfield and Nehalem are VERY similar – and in fact they are basically the exact same architecture with a few changes, removals and additions made to Lynnfield in order to differentiate the product lines. Lynnfield is still built on the Intel 45nm process that all the current Nehalem parts have been built on. Unfortunately, Lynnfield CPUs will be available under both the Core i5 and Core i7 brands, but we’ll get more into that later in the review.
The die size and transistor counts have changed with Lynnfield as well; all three desktop processors have a die size of 296 mm^2 and consist of 774M transistors on the Intel 45nm technology. What is interesting here is that Lynnfield is both larger and more complex than Nehalem that is 263 mm^2 and 731M transistors. Even though Lynnfield removes a memory channel from the die, the addition of PCI Express on the chip makes up the difference and then some. And even though Lynnfield’s die size is larger than Nehalem, the actual CPUs are smaller since the packaging no longer needs pin outs for the third memory channel or a high-bandwidth QPI connection.
Here’s what has changed:
The die size and transistor counts have changed with Lynnfield as well; all three desktop processors have a die size of 296 mm^2 and consist of 774M transistors on the Intel 45nm technology. What is interesting here is that Lynnfield is both larger and more complex than Nehalem that is 263 mm^2 and 731M transistors. Even though Lynnfield removes a memory channel from the die, the addition of PCI Express on the chip makes up the difference and then some. And even though Lynnfield’s die size is larger than Nehalem, the actual CPUs are smaller since the packaging no longer needs pin outs for the third memory channel or a high-bandwidth QPI connection.
Here’s what has changed:
- Dual-channel memory controller – Lynnfield does indeed still have an integrated DDR3 memory controller but here Intel has removed one of the channels from Nehalem to make the new core dual-channel. While this is theoretically a performance drawback, in real-world testing (as you will soon see) the third channel made very little performance difference in most consumer applications. By removing one channel Intel was able to make the CPU package smaller (less pins required), lower the cost of ownership (only two DIMMs needed now) while keeping performance very high.
- No QPI interface – While Nehalem used the QPI interface to communication with the north bridge of the motherboard, Lynnfield uses a connection called DMI. DMI is much slower than QPI (2 GB/s versus 26 GB/s or so) but in nearly all cases that bandwidth will be more than adequate for the amount of data moving between the processor and chipset even in a worst case scenario. Again, with this change, Intel was able to lower the pin count for Lynnfield in relation to Nehalem.
- Integrated PCI Express 2.0 – A first for a consumer processor, Lynnfield actually takes 16 lanes of PCIe 2.0 and moves them onto the die of the processor itself. That means that the graphics cards (or any other PCIe devices) now communicate directly with the processor itself rather than through a north bridge or chipset controller. The advantage here is for a lower cost chipset though I am betting that performance advantages from this are going to be minimal to non-existent. This is basically another move towards a more highly integrated platform on the Intel CPU.
- Split HyperThreading integration – HyperThreading is still around on the Lynnfield processors, but it will only be enabled on SOME of the product offerings. While the die size and transistor counts will be exactly the same, Intel will basically just flip off the HyperThreading capability in order to create a market differentiation and pricing segmentation. More details on what CPUs have what on the next page.
- Larger Turbo Mode differentiations – While Lynnfield was being prepared Intel’s engineers found out how to get a little more frequency out of the Nehalem architecture and you’ll notice that when you see the larger Turbo Mode improvements on Lynnfield. Essentially, the cores have been fine tuned in a way that will allow them to auto-overclock higher than even the most expensive Core i7 processors available today.