CPU Benchmarking: Debunking the AMD vs Intel Comparison

[Saint]

If you go to AMD website, it always uses Intel's CPU to benckmark its own processors, try to impress you that AMD is as good as/ or even better than Intel.

However, intel's benchmark is on its own processors, never use AMD to benchmark.

I think intel is more gentleman!

 

[Adam]

Well, as far as my computer architecture textbook is concerned, the best results for benchmarking boils down to:

Ss/P = (Is/P)*(Cs/I)*(Ss/C)

Where:
I = Instruction
Is = Instructions
C = Clock Cycle
Cs = Clock Cycles
Ss = Seconds
P = Programme

From Chapter 2 of Computer Organisation & Design: The Hardware/Software Interface, by David Patterson and John Hennessy.

There are two important things to remember when comparing Intel and AMD. AMD has more gates, and more pathways dedicated to certain specific operations. This means it can get more done each CPU cycle. The way they're doing it is to work out the statistics on which types of binary operations are done most, and try to stick dedicated pathways into the chip for those operations, meaning they get shoved through very fast. Intel, on the other hand, has a history of trying to be more generic, having fewer pathways which can combine into complex pathways, but this takes more CPU cycles. A nice example of the difference is my little AMD K6 200mHz; when it came out, with 8.5 million gates, the equivalent Pentium had only 4 million gates. However, that all said, both Intel and AMD (and everyone else) have been incorporating RISC (the dedicated pathways I was talking about) into their chip design in recent years anyway. They're all doing it. AMD has more experience in that particular method for PCs, but I personally think they are both rather similar now in that regard. It doesn't take much to work out how to do it, and I'm sure the brainiacs at Intel know how to do it very well. After all, they invented microprocessors.

The other really important thing to keep in mind is that the entire 80x86 computer archiecture is old and crap. We're stuck with it because it is all over the world, but there are newer and better architectures around. I quite like the MIPS architecture, but that too is getting a little old. It's still way better than 80x86, but I'm sure something newer and better will be out there soon.

 

[russ_watters]

Saint, test comparisons are only valid if people use the same tests. If Intel has a test that a lot of people use, it makes perfect sense for AMD to use that test on their own chips.

 

[Adam]

Unfortunately Intel will use a test which performs best for Intel chips. AMD will do the same. The best test is to get a third, independent party to run common applications, such as MS Office and Quake.

 

[BoulderHead]

The way I consider AMD's handling of this matter is simply that they lost the war with Intel over brute clock speed of CPU's and needed to find another drum to pound. If Intel disappeared from off the face of the Earth this afternoon would AMD them speak in this manner;

This processor is equivalent to an Intel 4GHz, that is, if Intel were around to make such a device...

 

[Adam]

That is simply not the case. AMD was pursuing a RISC-like philosophy of design 6 or 7 years ago; Intel only started following suit very recently. There competition hasn't been about brute clock speed most of the time because AMD was simply using a faster technology, which is why, for example, an AMD 200 would be faster than a Pentium 200. Now they are both usign the same ideas though. And given Intel's massive research and development capability, it would not surprise me if they now shoot ahead of AMD in terms of improving what they already have. However, I do think that to keep up with Intel, AMD will be pressing hard to develop new technological advances.

 

[MacTech]

Originally posted by Adam
That is simply not the case. AMD was pursuing a RISC-like philosophy of design 6 or 7 years ago; Intel only started following suit very recently. There competition hasn't been about brute clock speed most of the time because AMD was simply using a faster technology, which is why, for example, an AMD 200 would be faster than a Pentium 200. Now they are both usign the same ideas though. And given Intel's massive research and development capability, it would not surprise me if they now shoot ahead of AMD in terms of improving what they already have. However, I do think that to keep up with Intel, AMD will be pressing hard to develop new technological advances.

and that is why you should just get a mac.

 

[russ_watters]

and that is why you should just get a mac.

Heh, not bloody likely. [/british accent]

 

[BoulderHead]

Originally posted by Adam
That is simply not the case. AMD was pursuing a RISC-like philosophy of design 6 or 7 years ago; Intel only started following suit very recently. There competition hasn't been about brute clock speed most of the time because AMD was simply using a faster technology, which is why, for example, an AMD 200 would be faster than a Pentium 200. Now they are both usign the same ideas though. And given Intel's massive research and development capability, it would not surprise me if they now shoot ahead of AMD in terms of improving what they already have. However, I do think that to keep up with Intel, AMD will be pressing hard to develop new technological advances.

I’m not sure I get your drift. Did not computing speed historically center around clock speed rather than the amount of work performed? Did not AMD break this tradition?

The next opportunity for AMD came with addressing their weakness, their FPU. AMD was in a lose-lose situation, they weren’t able to produce enough processors to compete with Intel solely on clock speed, and at the clock speeds they were currently at, they weren’t able to produce high enough performing parts to make a significant difference in the market share.
Taken from;
http://216.239.53.100/search?q=cach...tle+history+amd+intel&hl=en&start=20&ie=UTF-8[/URL]

Why the change? I submit it was triggered by the inability of AMD to match clock speeds with Intel. Read this;

The XP 2400+ is also the first AMD CPU to break the 2GHz barrier - internally the CPU runs at 2GHz even though it is rated at 2400 or 2.4Ghz. This apparent rating discrepancy is because Athlons run more efficiently that Pentium IVs at any given clock speed. AMD decided for marketing reasons to rate their CPUs in terms of Pentium IV equivalence rather than their true speed which would have made them appear slower.
Taken from;
http://www.compulink.co.uk/~davedorn/reviews/hard/pcsystems/athlonxp2400.htm


So there it seems to be; They couldn’t match clock speeds but because the processors could perform quite well none the less AMD employees a new marketing strategy. It is a confusing strategy that seems to leave them in a position of needing Intel in order to have a frame of reference;


The Athlon XP processor, unlike all previous AMD processors, are numbered strangely. It's unofficially called the "Intel equivalence numbering". For example an Athlon XP 1800+ processor is really a 1.53 GHz (the 1800+ is meant to signify what Intel Pentium processor you would need to compare it with). An Athlon XP 1600+ is a 1.4 GHz processor. This numbering system still has the potential to mislead, so be careful. An 1.8 GHz Athlon processor is obviously a lot faster than an Athlon XP 1800+.
Taken from;
http://216.239.37.100/search?q=cach...+amd+equivalence+speed&hl=en&start=2&ie=UTF-8[/URL]

Notice the ‘unofficial’ thrown in there. I think officially they tie it back to one of their own earlier processors, but come on, we know what they ‘really’ mean, haha

Clock speed is the bane of AMD´s life, especially when it comes to marketing processors to retail customers.

It has Athlons performing better than higher-clock rated P4s, so what is it to do? It introduces part numbers to imply equivalence with Intel PCs. So the Athlon 2100 desktop CPU, launched today, clocks in at 1733MHz and not the 2.1GHz which such a number would indicate if it were an Intel part. And it has something called the True Performance Initiative, to promote a more meaningful measurement of - er - performance - which retailers can market, and which customers feel comfortable with.
Taken from;
http://www.theregister.co.uk/content/3/24415.html