3 - Architecture

The Nehalem processor series is the first multi-core design from Intel to have its cores joined in a single die rather than to join multiple cores together, potentially speeding up operations with multiple program threads by cutting back on the time spent moving data between cores.

See, Intel's current quads are actually a pair of completely separate dual-core CPU dies glued together in a single processor package. From an engineering perspective, not very sexy. Compared to AMDs Phenom, not very sexy either as it already has all four cores efficiently arranged on a single die, but it does get the job done.

This is the reason that inter-core communications on the old Core 2 processor is a little intricate. Between two cores on the same die, bandwidth is massive and latency low. But if cores on separate dies need a chat, the communication path is routed off the CPU through the north bridge chip on the motherboard. Not ideal, so the new approach is definitely more elegant and efficient.

The new Intel Core i7 (Bloomfield) processors debut with the following features :

• Four processing cores
• Support for SMT (simultaneous multi-threading), allowing up to 8 threads to be processed simultaneously
• 32 KB instruction + 32 KB data L1 cache per core
• 256 KB L2 cache per core
• Large 8 MB L3 cache shared by all 4 cores
• An integrated memory controller (IMC) supporting three channels of DDR3 memory
  - Memory clock speeds of up to 1333 MHz
  - Memory bandwidth of up to 32 GB/s
  - Up to 6 memory sockets
• The new Intel QuickPath Interconnect (QPI) replaces the front side bus (FSB)
• Addition of 7 new SSE4 instructions
• Monolithic processor design (all four cores on a single die)
• Fabricated using Intel's 45 nm high-k process technology

What you need to remember is that the Core i7 architecture is Nehalem, and it will allow for a lot more products to be released in the next year or two. We will see higher clock frequencies, yet also 2, 6 and likely 8 core versions. This is just the beginning.

Architectural changes. The 731 million transistors encounting Core i7 processors are built up-on a 45nm fabrication processes, but itself that is nothing new as the lastest Core 2 Quad processors are on the same fabrication. Intel is already working on a move to 32-nm by the way. All Nehalem chips will pack 64KB of L1 cache and 256KB of L2 cache. Those are supplemented by a large pool of shared L3 cache, which in the case of the quad-core model is 8MB.

A good number of things have changed in it's architecture. The biggest being Quick Path, the move forwards an integrated triple channel DDR3 memory controller allowing to get the product heaps of memory bandwidth and the return of hyper threading (Pentium 4 era technology) , which this time really makes an astonishing difference. Well, that and a new processor socket, socket LGA1366 which is quite a bit bigger than LGA775.

Nehalem brings in some new improvements which signal important changes in the Intel approach to CPU and system scaling and DRAM memory management. Let's walk through the big three.