Craziness: In general I agree with your comments (and bumped you), but I must point out the QPI and IMC do have a real impact (if small) on single socket systems.
The IMC will reduce memory access latency showing an improvement nearly across the board. The problem is, Intel already got the memory access latency down so low with the X38/X48 that there isn't that much room for improvement.
The QPI separates the CPU frequency from the rest of the system. It also provides a full duplex path (as opposed to the half duplex FSB) to the rest of the system. Further, memory accesses will no longer be traversing the same path as the rest of the system data. For single socket (monolithic cpu) systems, this boils down to a relatively small improvement due to (again) latency. If Intel connects two quad cores on one package using QPI, it will be just as useful as it is in multi socket systems (in multithreaded apps of course).
So, when you bundle a few small improvements from IPC, QPI, HT, and other core optimizations, you get a somewhat meaningful clock for clock improvement. Keep in mind, AMD's Athlon64 chips were only about 25% faster clock for clock than the AthlonXPs. The improvements everyone is looking for will be more apparent when the clock frequency scales and Intel puts two of these in one package. Of course, software developers have to fully embrace multithreading to benefit from more cores, but that is where the industry is going, so they'll either do it or their products will stagnate as CPU frequencies don't seem to be going up much anymore. On that note, if the test sample is running 2.93GHz, where do you suppose these chips will scale to?
The IMC will reduce memory access latency showing an improvement nearly across the board. The problem is, Intel already got the memory access latency down so low with the X38/X48 that there isn't that much room for improvement.
The QPI separates the CPU frequency from the rest of the system. It also provides a full duplex path (as opposed to the half duplex FSB) to the rest of the system. Further, memory accesses will no longer be traversing the same path as the rest of the system data. For single socket (monolithic cpu) systems, this boils down to a relatively small improvement due to (again) latency. If Intel connects two quad cores on one package using QPI, it will be just as useful as it is in multi socket systems (in multithreaded apps of course).
So, when you bundle a few small improvements from IPC, QPI, HT, and other core optimizations, you get a somewhat meaningful clock for clock improvement. Keep in mind, AMD's Athlon64 chips were only about 25% faster clock for clock than the AthlonXPs. The improvements everyone is looking for will be more apparent when the clock frequency scales and Intel puts two of these in one package. Of course, software developers have to fully embrace multithreading to benefit from more cores, but that is where the industry is going, so they'll either do it or their products will stagnate as CPU frequencies don't seem to be going up much anymore. On that note, if the test sample is running 2.93GHz, where do you suppose these chips will scale to?
Facebook
Twitter
Instagram