What type(s) of performance improvements to expect between architectures using QPI versus a FSB

computer-architecturecpufsbperformance

While comparing Intel's two new i7 processors, I noticed that the only difference between them apart from the L3 cache size and CPU speed is that one has QuickPath Interconnect, and that seems to account for most of the massive price difference between the two ($600 vs $1000). I'm not interested in shopping advice, but I do wonder what the advantages of QPI are.

I've read the Wikipedia article and I understand that QPI essentially replaces the Front-Side Bus. The article says that the first QPI implementation achieved double the throughput of the contemporary FSB, but it's not clear that this increase in throughput results in an increase in performance in real-world situations.

I understand that a faster FSB enables faster processors and memory, so theoretically replacing the FSB with something faster (QPI) would also enable faster processors and memory. However, I haven't seen anything indicating there are significantly faster components that are only usable with QPI. In other words, a fast FSB seems to be "good enough" for all intents and purposes.

Is a FSB still considered a bottleneck anymore by anyone who isn't trying to sell me something? If yes, when does QPI offer a performance improvement over a FSB? QPI is apparently not used for memory access like a FSB is; what effect does that have?

Edit: I'm looking for an answer that can say "When gaming and communication with the video card is the bottleneck, you'll see X sort of improvement. When doing CPU-intensive operations like compiling, you'll probably see no effect unless it's also memory-intensive" or something like that. Theoretical maximums and performance under contrived situations isn't useful, nor is the exact percentage performance gain given specific hardware and a specific task like the first loading screen in Crysis :P. A happy medium, please! You can support generalizations with specific test results, though, of course.

Mega bonus points if you can also address the same issues for HyperTransport.

Best Answer

The QuickPath Architecture increases the bandwidth between physical processors, memory and I/O in (at least) three ways:

  • The memory controllers are moved from the chipset into the CPU, shortening the path from each processor to RAM, and preventing data from going a back and forth through a central, shared chipset to get at memory.
  • There are connections/paths added to the processors to pass data directly between them, again eliminating data trips though a central chipset to share data between CPUs.
  • There are now multiple paths to the chipset, which is what attaches you to other I/O.

These are connections are called QuickPath Interconnects (QPI), and are considerably shorter, wider and faster than traditional FSB connections.

QPA's mesh-like (point-to-point) architecture will also allow great future expandability without the bottleneck of overloading a single FSB.

"When will I notice the improvements?"

You should notice an increase in performance in anything that passes data between CPUs, to and from system Memory, and to and from system I/O. YMMV. :)

See here for best info out there about it.

Related Question