Hey Tom's editorial staff,
I have a somewhat ambitious idea for an article I would like you to consider.
It is my belief that many would appreciate an apples to apples performance comparison of as many CPU architectures as possible.
What would differentiate this article from your current reviews and charts is in how it would be performed.
Instead of comparing by price point or market segment, I envision this article matching as closely as possible all variables except the CPU architecture.
To that end I would propose that all CPUs are tested with only one core enabled, at an identical clock rate with all turbo modes disabled and with RAM Data Rate and Timings matched (when it is not possible to directly match RAM types).
Additionally, CPUs that support SMT could be tested again with SMT enabled under a threaded workload.
This would allow you to accurately show any improvements in implementations throughout different generations.
Excluding oddball platforms (Atom, Brazos), this testing could be done with a minimum of motherboard swaps and reinstalls.
A P45 or x48 based LGA775 platform would take you from a P4 through the 45nm Core 2 arch, an AM2+ platform could take you from an Athlon x64 to Phenom II and an additional LGA1366, LGA1156 and LGA1155 platform would round it off.
At the end of such an article you would be able to definitively quantify relative IPC performance between dissimilar architectures and different generations of similar architectures.
Additionally, it would be possible to create, and maintain upon the release of new architectures, a new chart showing the Performance Per Clock of each arch for simple reference.
I believe an article such as this would be extremely enlightning and quite true to Toms roots.
Thanks for reading and considering this proposal.
-outlw
I have a somewhat ambitious idea for an article I would like you to consider.
It is my belief that many would appreciate an apples to apples performance comparison of as many CPU architectures as possible.
What would differentiate this article from your current reviews and charts is in how it would be performed.
Instead of comparing by price point or market segment, I envision this article matching as closely as possible all variables except the CPU architecture.
To that end I would propose that all CPUs are tested with only one core enabled, at an identical clock rate with all turbo modes disabled and with RAM Data Rate and Timings matched (when it is not possible to directly match RAM types).
Additionally, CPUs that support SMT could be tested again with SMT enabled under a threaded workload.
This would allow you to accurately show any improvements in implementations throughout different generations.
Excluding oddball platforms (Atom, Brazos), this testing could be done with a minimum of motherboard swaps and reinstalls.
A P45 or x48 based LGA775 platform would take you from a P4 through the 45nm Core 2 arch, an AM2+ platform could take you from an Athlon x64 to Phenom II and an additional LGA1366, LGA1156 and LGA1155 platform would round it off.
At the end of such an article you would be able to definitively quantify relative IPC performance between dissimilar architectures and different generations of similar architectures.
Additionally, it would be possible to create, and maintain upon the release of new architectures, a new chart showing the Performance Per Clock of each arch for simple reference.
I believe an article such as this would be extremely enlightning and quite true to Toms roots.
Thanks for reading and considering this proposal.
-outlw