>Umm, my math tells me its more like 29%...
You have a calculator that adds the numbers differently than AMD's calculator? I just took the number on the bottom line of AMDs table of test numbers. Do all of these numbers come out different on your calculator?
> When the hell didn't 22% not mean ALOT of difference in the PC business??
> People have spent 100's of $$$ on <5% differences.
True, people spend a lot. It means a lot in marketing, where the object is to get as high a price for your product as possible. It still is hard to notice the difference. I think people would be less inclined to spend that level of money if they could get an accurate impression of what the improvement is.
I remember when I switched from EDO simms to SDRAM. I could not see any difference. But it did show up on benchmarks. It was an expenditure that gained me nothing.
<<The model number just looks big because the Athlon Tbird's performance goes up slower as its multiplier increases.>>
>>
>I'm not sure I follow your logic here. Are you saying that Palomino and Thoroughbred utilize their MHz better?
Right. If we used a 486DX4 as a base, it would have to have a much faster clock to reach a given performance than an Athlon XP. If we pushed the multplier to the point an increase barely made a difference and used that as the base difference, the clock numbers could be pushed still further up. If we put that 486 clock as a model number on an XP, the numbers could be tremendously impressive. The difference between the Tbird 1400 and XP 2200+ in the performance test would still be 22% regardless
<<Another interesting thing is that a 2G P4 with DDR underperforms an XP 1600+. Using that P4 as a base, a 1600+ should be labeled 2000+. The 2000+ should be labeled as 2400+. Tell your friends that although your processor is labeled 2000+ it should be labeled 2400+.>>
>The 1600+ may beat the P4-2GHz in some benchmarks, ...
Correct. It also beats the P4-2GHz on the comprehensive performance number that AMD gets, and which they have on page 11.
> ... but it doesn't mean it would beat a 2GHz Thunderbird in those same benchmarks.
True. But if you adhere to the claim that XPs are rated by P4 performance the 1600+ will beat the 2GHz P4, and the XP should be labeled 2000+.
> It doesn't matter what the P4 performs, the PR rating isn't related to Intel performance. Otherwise AMD wouldn't pit their own 2200+ model head-to-head against the 2.53GHz chip from Intel. They'd rather pit their 1.8GHz/2533+ model if the PR was based on P4 performance.
True. But the table on page 11 gives the performance AMD measures for the processors I mentioned. From that table you can tell what a P4 can do and what an XP can do, regardless of what label is used.
If you look through the table, you can easily see the problem in relating XP and P4 performance. The numbers are different for P4s with DDR and RDRAM. Which should be used? Intel comes out with quad DR processors and still labels them P4s. Do you change the model numbers on AMD processors?
If I haven't gotten too tiresome already, let me point out what is wrong with a performance rating; why AMDs rating is going nowhere; and why it wouldn't matter if it did. Performance rating depends on what tests you do. Put more tests which emphasize one particular aspect and the numbers shift. What should the right proportion be? Why would Intel agree to a given standard if they could have a different one that made their product get bigger numbers? If a standard is agreed to, it will be Intel's standard, and AMD will do the agreeing.
Suppose AMD and Intel agree on a standard. The best one could do would only be statisically representative. Why would some one whose focus is on games want business apps included in the performance tests? Whatever games are chosen will be outdated in a few years anyway. Why would a business want 3D games included? The business apps get new versions regularly, so the test will be outdated regularly. More to the point, why would the Internet surfer care about either? Neither sort of test gives him much idea of what to expect. Few people use their computer in compute-bound situations. The Internet is responsible for the explosion in computer sales over the last several years. It is why grandma is in the computer store.
If a standard is ever agreed upon, then companies will design their product to get the best score on the test. The consumer will know exactly what the score is when they buy the computer. But will he know how his computer will actually perform? People who need high performance will have to look to other tests to find out. People who don't need high performance will just be subject to a different form of hype.
