customer service FAQ

The Core i7 (formerly known as Nehalem) is currently the highest performing desktop CPU on the market. Instead of using a shared host bus to connect with the rest of the system and also the memory, the Core i7 uses separate interfaces for the memory and the rest of the system. The new system interface uses the Quick Path Interconnect or QPI. Much more important is that the memory subsystem is now directly connected to the processor through a highly integrated structure called the “Uncore”, featuring three separate memory controllers as well as a large 8 MB L3 cache.

Because the bottleneck of the old system interface bus is no longer in the way, the Core i7 can freely access memory at much higher bandwidth than possible ever before. This does increase the load on the memory subsystem, simply because more data are exchanged between the CPU and the memory in the same time when compared to the Core2 or any other system architecture. At the same time, the better data transfer is also the reason for the superior performance of the Core i7 – as long as high quality DDR3 memory is used that can withstand the onslaught of transfer requests.

The Core i7 features three memory controllers that are acting largely independent of each other. Each memory controller should interface with at least one DDR3 DIMM, that is, any given system needs at least three modules to utilize the capabilities and full benefits of the architecture. Intel-brand boards only feature four memory expansion slots of which three can be used for triple channel configurations. Other, third-party boards can feature 6 memory slots, all of which can be populated in multiples of three to get the full bandwidth.

As in all systems, more memory is always better but there are limitations on the level of the operating system, in that 32-bit versions of Windows or Linux can only address up to 4 GB, meaning that for triple channel operation the system is stuck with a maximum of 3 GB. 64-GB Operating Systems love memory and we recommend 6 GB as the minimum.

The Core i7 memory system is running the memory controller at the frequency of the “Uncore, which has to be at least twice the frequency of the memory bus itself. Compared to a memory controller running at 400 MHz – which was the fastest official speed on the previous Intel architectures, the memory controller running at up to 1600 MHz is up to 4 times faster than before and it does not have to share the bus with other devices. Overall throughput always depends on the applications running in a multitasking environment and cannot be compared from one scenario to the other but, in two simple words, the answer is "a lot!".

The memory supply voltage also powers the receivers and senders on the memory controller with a voltage plane known as VDDQ. Because the Corei7 is manufactured on a 45 nm node, the logic is sensitive to overvoltage and Intel defines the upper safe limit for the Core i7 as 1.65V.

Core i7 can be overclocked on air to roughly 4.1 GHz, anything much higher than that requires extreme refrigeration using non-conventional methods such as liquid nitrogen. Currently, the world record using liquid Helium is at 5.5 GHz.

Using standard memory benchmarks like SiSoft Sandra you can measure the overall bandwidth achieved by your memory. Compared to the Core2 architecture, the Core i7 delivers more than 3 x the memory bandwidth to the CPU. For example, at 1066 MHz, the memory bandwidth in triple channel configuration is about 20 GB/s and at 1333 MHz, it is almost 25 GB/s.