Odd performance measures trend

Over the past year or so a lot of people have started moving towards some performance measurements that seem silly to me. Can somebody explain them to me or join me in my effort to educate people .

Performance per core:

Is it a valid measure since most things that max out 1 core tend to use many cores now and even more in the future. If I offer them a Bugatti Veyron or a Nissan GTR and all they care about is performance they are saying they would pick the GTR because it has a better 0-60 time per cylinder completely ignoring that the Bugatti is a much faster car but it has 2.5x as many cylinders so all common performance benchmarks for it divided by 2.5 make the Nissan better.

Performance relative to clock speed:

This would be valid if their processor can reach a frequency at which they perform evenly but almost always this is not the case. Take the same car argument above but now it's like these people are saying I'll take the GTR because its 0-60 time relative to horsepower is better(not valid until you consider that around 1000 horsepower than GTR can get to 60 in the same time). So, once again they choose a much slower car unless they plan to spend $80000 on mods.

What happened to performance per watt and performance per dollar? Is it going extinct?

The A8-3870K thread kicked this off when I noticed somebody use those 2 odd(to me) performance measures to call the APU amazing. If it's amazing then the i3 2120 with 80%-100% of the performance of the APU with just 2 cores is phenominal. I'm not knocking the APU though, as far as budget builds go it's amazing, I've considered selling my FX-6100 machine just to play around with one!

Maybe I've got it all wrong though, thoughts?

Is it a valid measure since most things that max out 1 core tend to use many cores now and even more in the future. If I offer them a Bugatti Veyron or a Nissan GTR and all they care about is performance they are saying they would pick the GTR because it has a better 0-60 time per cylinder completely ignoring that the Bugatti is a much faster car but it has 2.5x as many cylinders so all common performance benchmarks for it divided by 2.5 make the Nissan better.

Click to expand...

It's a very valid measurement. Cores aren't like cylinders - a program can't just be spread out along all the cores like horsepower in an engine. A program has to be specially designed to take advantage of multiple cores, and there are a lot of tasks that just don't fit that pattern. So it matters very much how fast each individual core is. And it's also possible that a program may be able to parallelize its workload to 2 or 4 cores, but not 16. It matters a lot in that case too.

This would be valid if their processor can reach a frequency at which they perform evenly but almost always this is not the case.

Click to expand...

Sure. But it does help immensely when you're trying to figure out whether to choose that 3.2GHz i5 or the 3.4GHz FX CPU. If they both have the same number of cores, for example, it's hard to make a judgement as to which is better. The i5 might perform better despite the lower clockspeed, so clockspeed becomes deceptive. It's also helpful for comparing results once you're overclocked. It doesn't matter if I can overclock a CPU to 6GHz if it's slower at that speed than something else at 4GHz.

What happened to performance per watt and performance per dollar? Is it going extinct?

Click to expand...

I still see these around, though "performance" is getting harder to measure in aggregate, largely for the reasons I mentioned above.

perhaps it's just that we newbs do not understand the details.
i see that there is a compitition going on for intel chips, I don't understand the details of how the cpu's with lower clocks are posting higher scores than the ones with higher clocks.
I race cars so i understand that this can and will happen.
truly undrstanding the details of how? I just don't know.
you seem to be of an older school of overclocking where you had to know all the little ins and outs of a cpu and a motherboard, I have just started this fun and as I see all around, we all only know to jack the multi with little understanding of even knowning what that is.
perhaps the manufacturers of the cpu's and motherboards have made it all too easy and now we can just be rocketman with out haveing to know, much less understanding what we have just tinkered with.

For johan.

To your first point, that is fair. I had considered it and felt that things using 1 or a less than optimal number of cores while also having a noticeable difference in time due to for example a 10% better per core performance were very niche at this point in time especially for power users(while the number of people using the argument to their advantage isn't niche at all and oddly tend to be power users). It's odd that I'd forgot that side of the argument though because I used to stand on the other side of the fence in the Pentium D days.

I was considering the second one from what somebody with a knowledge level above using IE with MSN as their home page would say. I had a customer choose a Q6600 over a Phenom X4 940 with no intention of overclocking because performance relative to clock speed was better. Performance relative to price seems more relevant to nearly all arguments of this type unless you don't have to spend money. Though as you said and has often been true it is hard to determine aggregate performance.

"It's also helpful for comparing results once you're overclocked. It doesn't matter if I can overclock a CPU to 6GHz if it's slower at that speed than something else at 4GHz" That's what I was getting at.

Anyway I kicked the day off in a bad mood and what you said was good, it gave me a perspective other than my own.

caddi daddi said:

perhaps the manufacturers of the cpu's and motherboards have made it all too easy and now we can just be rocketman with out haveing to know, much less understanding what we have just tinkered with.

Click to expand...

Quite the opposite of the automotive industry isn't it? My friends 383 rebuild was 100x easier to figure out than the insanity on my KA24DE.

The performance per core comes down to how efficient each core is with code, it's all in the chip design. AMD's current gen chips have been designed to give a higher overall performance when all cores are used, however they shared components in each module to save on chip size and power useage.

The downside to this is lower single threaded performance. Lower in fact than their previous generation chips.

When more cores are utilised their approach actually works quite well, you just need to understand that most software simply hasn't been written to make use of the additional cores, and likewise many types of software would show no benefit to actually do so. Likewise coding to use a higher numbe of threads is more difficult and drives the development time and cost of that software.

On a personal note I rather agree with AMD's way of doing things. Most of the tasks not overly suited to being highly threaded don't really need a massive amount of grunt. On the other hand, many of the things that are suited to being run in a ghighly threaded way would probably perform better if run off a video card instead..... Thats part of the reason I believe so strongly in AMD CPU/GPU mashup, particularly their successor to the upcoming Trinity APU's

I had a customer choose a Q6600 over a Phenom X4 940 with no intention of overclocking because performance relative to clock speed was better.

Click to expand...

That's not bad data, it's simply a bad decision.

johan851 said:

That's not bad data, it's simply a bad decision.

Click to expand...

Lol, not really. He took into account the data he had without taking into consideration the other platform advantages. I'd call that misinformed.

this is very different than my race cars all the way down to how voltage effects things, to cool an electric motor you up the voltage to reduce amp draw, to cool a cpu you reduce voltage and up the amp draw.

I raced the ka24de solid cam bucket engine for a while in a 91 240, I can go through one in my sleep, the chevy smallblock makes me scratch my head.

Yup, its fair to say computers have absolutely nothing in common with cars, only that its easier to explain computers to someone who knows nothing about technology in the terms of an engine..... Bizarre eh?