HTs take on the Future of CPU rating

[YossI]

You know the subject so I'll make it short:
What's your thoughts of future CPU rating ? its obvious that the public won't go read benchmarks before they buy a CPUs ? and MHz alone should stop being considered as an absolute rating of a CPU.

And its clear why: the average person would pick up a Pentium 4 2.4GHz over an Alpha EV7 1.3GHz even if they would have cost the same? obviously such a situation wouldn't have been encountered in real-life ? but it's clear we need some sort of CPU rating?


How can you rate a CPU? A suit of benchmarks would be system dependent? and benchmarks like SPEC do not reflect performance on average?

Frankly ? I don?t know what should be done ? but that?s what HT is for, isn't it?

 

[AndyHui]

Your comparison between a Pentium 4 and an EV7 is not valid, as they are used for completely different things.

Would you accept a comparison between an Athlon and an Apple G4?

 

[YossI]

True ? I used the EV7 in-order to emphasize the problem and to avoid making compression between competing products in-order to keep the debate at the technical level and unbiased.

 

[Sohcan]

Why reinvent the wheel? The entire microprocessor industry already relies on SPEC to judge CPU performance.



<< and benchmarks like SPEC do not reflect performance on average? >>

Perhaps not for PC desktop performance, but that's not the role the suite is supposed to occupy. SPEC is comprised of real programs, whose creators were gracious enough to donate the source code to SPEC (in the case of programs that weren't already open sourced). The programs include (among other things) compression, C compiling, FPGA and VLSI place and routing, Perl interpreting, OO database system, differential equation solver, MESA 3D graphics libraries, computation fluid dynamics, and numerous scientific simulations (high-energy physics particle accelerator, meteorological, seismic wave propogation, crash simulation). It is absolutely invaluable for judging enterprise, workstation, and server computing performance, as well as an important research tool for computer architecture and compiler development. Lastly, and most importantly, it is the most useful benchmark suite designed to test single processor performance that is cross-platform compatible and open-sourced.

As far as PC desktop performance is concerned, I really don't see any impetus for AMD and Intel (and to a lesser extend Via and Motorola) to come together and agree on a standard benchmark suite (and one that has to be updated every few years). I'm sure Intel's marketing department is quite secure with the P4, as is AMD with the XP performance rating.

 

[Shalmanese]

What I would like is an adjustable benchmarking system from which you could assign weights for different apps that you use, either manually or automatically by it sitting in your taskbar and logging info. That way, you could determine which CPU/RAM/VideoCard combo would be best for you.

 

[rimshaker]

I personally think that pure raw speed will become less and less important in the future. The true MHz rating will be just part of the standard spec sheet. I see more emphasis being put on work done per clock cycle, and efficiency of overall microprocessor design. I mean cmon, let's get real even today... Aside from hardcore enthusiasts, JoeSixPack out there won't need some 2.4GHz cpu over, say, a 1 GHz cpu will he? This rat race with pure MHz speed leads to a dead end IMO.