With the release of the new core i5 and i7 processors from Intel, this week a predictable mess of confusion has emerged on the internet as to the place and use and applicability of the various processors to an individual PC configuration. This is perhaps not surprising when most PC buyers will only ever see a model number and a clock speed and will not understand some of the mechanical differences between the various parts.
If you want the bottom line, skip to the bottom. The rest of this gets reasonably geeky but attempts to explain why the differences in the processors and platforms matter.
The "Dashboard" Reference
I put together a quick comparison table to help reference back to when we start talking about the various lines of processors. Credit goes to Intel for providing useful data for each of their processor lines.
The "Old" Core i7 and i7 Extreme
The first parts to market were the "old" core i7 which utilize a socket with lots and lots of pins (1366 to be exact). These pins are necessary because intel created an architecture which relies on focusing the processor itself on doing computational tasks and asks other parts in the system to take on things like reading and writing to memory, talking to graphics cards, dealing with disk accesses, etc.
Here, we can see that the processor itself (the CPU) talks to only one thing on a regular basis: the memory. The i7 processors which are already on the market (the "9 series") provide the capability to use 3 separate channels of DDR3 memory, each of which can be "talked to" at up to 8.5 Gb/second. Practically any other component of operating a computer, whether graphics, video, et al, is handed off to another part within the platform.
The platform itself is comprised of two separate components, the first is the Input/Output hub which provides connectivity to the controller as well as the high speed interface to be able to work with PCI express graphics cards. You will note in the table before, I just wrote X58. The x58 chipset is able to support up to "36 lanes" of PCI Express 2.0 accessibility. A "full speed" PCI Express slot is 16 lanes at maximum capability. The reality is that almost nothing fully utilizes the amount of information which can pass over 16 lanes of PCI Express connectivity.
More commonly these days, a motherboard may choose to provide only one of those slots at 16 lanes and may choose to provide one or more slots at 8 lanes. In this way, a motherboard can offer up to 4 slots at 8 lanes for graphics cards, or 2 of 16 for graphics cards and another of 4 for a super-high-speed network interface card, etc. There is a lot of flexibility available in the architecture and it is up to the manufacturer of the motherboard to choose just how much capability and what kind to offer for video chipsets.
This flexibility comes at a cost. The processor has to talk to a communications hub who then has to work with a video card to get work done. Each step in the process has a miniscule amount of time that a signal takes to go back and forth, an extra step that in processor time can become more costly depending on how fast your computer is trying to do something and how often information is having to make that trip between the processor and the graphics card.
The second component that we see on the motherboard to comprise the platform outside of the processor is the I/O controller hub. This is the chip that will do things like working with disks, provide software RAID, provide slots for sound controllers, USB, on-board LAN support, and talking to the BIOS. When you use a second chip for the motherboard, this is another component that costs money to produce and integrate. Its another step between information going back and forth to peripherals (like your hard drive) to the processor and back.
As an added challenge, the interface between the ICH and the IOH is at only 2GB/s versus the CPU/IOH connection at 25.6 GB/s. When you consider that each of 6 SATA ports is capable of talking to devices at 3Gb/s and that the network port also has a certain amount of traffic available to it, with just a couple hard drives and some added network traffic, the theoretic limit of the connection to the ICH becomes a practical limiting factor on moving data to and from the processor efficiently.
The processing platform itself is not without its assets. Although the turbo boost version which is integrated into the earlier i7 9-series products is older and therefore unable to assist with older single-threaded applications, the 9-series does offer hyper-threading which allows the processor to switch back and forth between more individual processing tasks than it actually has hardware to handle. This speeds up a system because no single processing task is going to run all of the time. When a processing task is waiting for data to go somewhere or come back, for instance, hyper threading allows another task to use the physical processing resources during that time.
The "New" Core i7 and i5
The biggest difference with the new core i7 is that the processor now directly takes over the work that the Input/Output Hub used to do for us in the X58 platform but because of the limited space physically available on a processor die, we give up a good bit of flexibility to get the performance boost that having these functions "closer" to the logic core brings.
The processor is able to directly talk to the processor and the memory in the system, not only not handing this function off to another component, but the processor is actually set up to be able to talk to the memory faster than the 9-series per-channel (remember that there are less channels of overall memory, however). On-die memory access can talk to each channel at 10.6 GB/s for an overall throughput to memory of 21.4 GB/s. This helps make up for the lack of a third channel on the 9-series processors which allows the "old" processors a total throughput of 25.5 GB/s.
When we look at the PCI express graphics capabilities, we run into a significant limitation of this new platform. The processor can only work with a maximum of 2 video card slots AND a maximum of 16 lanes of PCI express communication. This results in really only two choices for the motherboard. EITHER a single video card with the full 16 lanes of traffic OR two video cards with 8 lanes of traffic each. The implication here is that triple SLI which is possible in 9-series platforms with the intel X58 chipset is not possible with the new core i7 processors.
Essentially the rest of the chipset platform is identical to the X58 ICH capabilities except that more channels of lower speed PCI express communication are available (8 channels in the P55 vs 6 channels in the X58 ICH) and more USB ports are available (14 in the P55 vs 12 in the X58 ICH). For practical purposes, however, these additional capabilities are limited in impact as the connection between the processor and the P55 chipset is limited to 2GB/s, the same limitation of the X58's ICH and well below the practical communications throughput of an active network connection and a couple active hard drives.
One interesting thing that you will note in the chart above is that the new i7 and i5 processors are actually better at working with older single-threaded applications. Much of what is out on the market right now is built to work with only a single physical processor. As a result, there are many programs which will see a performance boost by the processor "bursting" to a faster cycling speed to do more work in a second that the 9 series does not perform as efficiently.
Summary
- "Old" Core i7
- Has more connections to talk back and forth with memory and graphics cards
- Huge degree of flexibility
- Can handle more computing tasks simultaneously
- Better overclocking potential.
- Turbo Boost for old programs not as efficient
- More expensive, uses more components to provide the overall system
- "New" Core i7
- On-Die Memory and graphics assets allows for faster connections back and forth to most common system resources
- Improved turbo boost to speed up old programs
- Supports Hyper Threading to handle more computing tasks simultaneously
- Comparatively inexpensive
- Limited to two connections to memory and total of 16 lanes to graphics cards
- Chipset connection no faster than "old" IOH/ICH connection.
- "New" Core i5
- On-Die Memory and graphics assets allows for faster connections back and forth to most common system resources
- Comparatively extremely inexpensive
- Lack of hyper threading limits simultaneous tasks and makes a small reduction in the amount of improvement that turbo boost can make with old applications
- Limited to two connections to memory and total of 16 lanes to graphics cards
- Chipset connection no faster than "old" IOH/ICH connection.
Additional resources to read up on the subject:
Toms Hardware - Intel Core i5 and Core i7: Intel's Mainstream Magnum Opus
http://www.tomshardware.com/reviews/intel-core-i5,2410.html
Video @ http://www.youtube.com/watch?v=qY_4thU6-9k
The official Intel Processor Selection Page
http://www.intel.com/consumer/learn/processors/index.htm?iid=learn_all+processors
The official Intel X58 Express Chipset Page
http://www.intel.com/Products/Desktop/Chipsets/X58/X58-overview.htm