This is my piece of investigative journalism for the month.
After following a link to Anands Dec 21/2000 article on the KT133 vs the KT133A platforms and reading it, I was struck by the huge difference in the gaming benchmarks in that article (1.0Ghz P3 vs Athlon) - as opposed to P4 1.7 review ... .
? ? ? fps ?Q3 Arena/Unreal T./Serious Sam/MBTR
A 760 2100 ... 143/102/79/69 ... 46% faster than 815
P3 815 ... 101/74/51/46
The P4 article showed a 46% advantage for the Athlon DDR in gaming vs the P3 which I thought was way off base considering how close the 2 platforms have been in the past. I?ve also seen several reviews where the 760 DDR was only showing a few % faster than a KT133A so the 46% number didn?t seem to make sense.
I don't know who at Anantech ran the gaming benchmarks in the P4-1.7 article, but after reviewing several other benchmark tests my analysis suggests something was really messed up for the P3. I mean the P3 benchmarks aren?t even remotely close to where they should be. I think I know what happened but bear with me for a moment while I first present the evidence. Consider the KT133 vs the KT133A test ...
... ... ... fps ? (Q3 Arena/MDK2/Unreal T./Expendable) ?
A 760 2100 ... 155/158/100/111 ... ... 3.8% faster than KT133A (5.8% over 815)
A KT133A ... 149/156/98/103 ... ... 1.9% faster than 815
P3 815 ... 151/150/98/98 ... ... 4.0% faster than KT133 (average of 4 benches)
A KT133 ... 140/146/93/98
The P3 beats the KT133, this result is expected and is what most benchmarks reviews at various hardware sites around the web showed. The KT133A picks up about 6% over the KT133 and passes the 815 by a small margin 1.9%. The DDR 760 picks up about another 4% to give it a 5.8% lead over the 815. A lead - but not a very big one.
If one goes to the P4-1.7 review, all of a sudden the 1.0Ghz Athlon DDR system has gone from a 5.8% advantage to 46%!. A gain of 40%! 2 of the 4 games are different in this review but the numbers in the 2 games that are repeated (Q3 and Unreal T.) for the P3 are completely off base compared to the previous review. Even if you use the same 4 games in the P4-1.7 review see Duron 900 review ... the P3 still beats the KT133 - so changing the game mix to that of the P4-1.7 doesn?t change things much.
... ... ... fps ?Q3 Arena/Unreal T./Serious Sam/MBTR
P3 815 ... 133/97/74/60 ... 0.4% faster than KT133
A KT133 ... 130/96/73/62
ANALYSIS:
The 760 DDR is only showing about a 10% advantage over the KT133 and a P3 815 is faster than the KT133. Logic dictates the P3 should be less than 10% slower than the 760 DDR. Also, one can see the Athlon on a KT133A beats a P3 by a couple of percentage points and the 760 DDR is only a few % faster than the KT133A. A 46% advantage for the 760 DDR over the 815 is not even in the ball park. Here are 2 more examples of benchmarks where even a 20% clock rate advantage for the Athlon DDR produces nowhere near a 46% advantage.
Aces? An Athlon 1200 DDR beat a P3 1.0 by 29% (800x600 on 4 games.)
Sharky ... Athlon 1200 DDR beat a P3 1.0 by 22.9% (2 games ?Quake 3/MDK2)
Well- what could have happened to the P3 in the review in question. Actually 46% is an interesting number. From Duron review we can estimate that the Athlon DDR should have about a 10% advantage in the same 4 games as the P4 review. Consider the following hypothetical gaming benchmark where an Athlon DDR has a 10% fps advantage over a P3 ? ?
A 1.0 760 DDR ? 100 ? ~10%
P3-1.0 (815) ? 91
Now the P3 has a 7.5 (x 133 = 1000mhz) multiplier and if you ran it on a 100mhz bus you would get 750mhz ?a 25% slower clocked processor (with a 25% slower fsb too). So lets reduce the P3?s score by 25% ? and ? look at the result ?
A 1.0 760 DDR ? 100 ? 46.5% ..advantage
P3-750 (815) ? 68.25
46%!!! ?..almost exactly the 46% advantage the Athlon DDR showed in the P4-1.7 review. I hardly think the fact that these numbers are so close is a statistical fluke and I can only conclude that whoever ran the P3 gaming benches ran them on a 750mhz P3 and not a 1.0Ghz P3 as stated in the Anandtech article. This is probably a case of sheer carelessness rather than some deliberate attempt to sabotage the P3, but it appears that someone needs to take a little more care in running benchmarks at Anandtech.
Finally, while it could be argued that gaming benchmarks don?t scale linearly with clock so my example isn?t relevant - I counter that a gaming benchmark won?t scale linearly if you change the MHZ and leave the fsb the same. When one changes the fsb & mhz together - a gaming benchmark will scale linearly with clock, or very close to it, which is what my example is doing. It?s reducing the clock rate and fsb by 25% ? (1000-750 and 133-100)
I think Anand needs to rerun those P3 benchmarks.
Do you agree?. Disagree ??..comments.
After following a link to Anands Dec 21/2000 article on the KT133 vs the KT133A platforms and reading it, I was struck by the huge difference in the gaming benchmarks in that article (1.0Ghz P3 vs Athlon) - as opposed to P4 1.7 review ... .
? ? ? fps ?Q3 Arena/Unreal T./Serious Sam/MBTR
A 760 2100 ... 143/102/79/69 ... 46% faster than 815
P3 815 ... 101/74/51/46
The P4 article showed a 46% advantage for the Athlon DDR in gaming vs the P3 which I thought was way off base considering how close the 2 platforms have been in the past. I?ve also seen several reviews where the 760 DDR was only showing a few % faster than a KT133A so the 46% number didn?t seem to make sense.
I don't know who at Anantech ran the gaming benchmarks in the P4-1.7 article, but after reviewing several other benchmark tests my analysis suggests something was really messed up for the P3. I mean the P3 benchmarks aren?t even remotely close to where they should be. I think I know what happened but bear with me for a moment while I first present the evidence. Consider the KT133 vs the KT133A test ...
... ... ... fps ? (Q3 Arena/MDK2/Unreal T./Expendable) ?
A 760 2100 ... 155/158/100/111 ... ... 3.8% faster than KT133A (5.8% over 815)
A KT133A ... 149/156/98/103 ... ... 1.9% faster than 815
P3 815 ... 151/150/98/98 ... ... 4.0% faster than KT133 (average of 4 benches)
A KT133 ... 140/146/93/98
The P3 beats the KT133, this result is expected and is what most benchmarks reviews at various hardware sites around the web showed. The KT133A picks up about 6% over the KT133 and passes the 815 by a small margin 1.9%. The DDR 760 picks up about another 4% to give it a 5.8% lead over the 815. A lead - but not a very big one.
If one goes to the P4-1.7 review, all of a sudden the 1.0Ghz Athlon DDR system has gone from a 5.8% advantage to 46%!. A gain of 40%! 2 of the 4 games are different in this review but the numbers in the 2 games that are repeated (Q3 and Unreal T.) for the P3 are completely off base compared to the previous review. Even if you use the same 4 games in the P4-1.7 review see Duron 900 review ... the P3 still beats the KT133 - so changing the game mix to that of the P4-1.7 doesn?t change things much.
... ... ... fps ?Q3 Arena/Unreal T./Serious Sam/MBTR
P3 815 ... 133/97/74/60 ... 0.4% faster than KT133
A KT133 ... 130/96/73/62
ANALYSIS:
The 760 DDR is only showing about a 10% advantage over the KT133 and a P3 815 is faster than the KT133. Logic dictates the P3 should be less than 10% slower than the 760 DDR. Also, one can see the Athlon on a KT133A beats a P3 by a couple of percentage points and the 760 DDR is only a few % faster than the KT133A. A 46% advantage for the 760 DDR over the 815 is not even in the ball park. Here are 2 more examples of benchmarks where even a 20% clock rate advantage for the Athlon DDR produces nowhere near a 46% advantage.
Aces? An Athlon 1200 DDR beat a P3 1.0 by 29% (800x600 on 4 games.)
Sharky ... Athlon 1200 DDR beat a P3 1.0 by 22.9% (2 games ?Quake 3/MDK2)
Well- what could have happened to the P3 in the review in question. Actually 46% is an interesting number. From Duron review we can estimate that the Athlon DDR should have about a 10% advantage in the same 4 games as the P4 review. Consider the following hypothetical gaming benchmark where an Athlon DDR has a 10% fps advantage over a P3 ? ?
A 1.0 760 DDR ? 100 ? ~10%
P3-1.0 (815) ? 91
Now the P3 has a 7.5 (x 133 = 1000mhz) multiplier and if you ran it on a 100mhz bus you would get 750mhz ?a 25% slower clocked processor (with a 25% slower fsb too). So lets reduce the P3?s score by 25% ? and ? look at the result ?
A 1.0 760 DDR ? 100 ? 46.5% ..advantage
P3-750 (815) ? 68.25
46%!!! ?..almost exactly the 46% advantage the Athlon DDR showed in the P4-1.7 review. I hardly think the fact that these numbers are so close is a statistical fluke and I can only conclude that whoever ran the P3 gaming benches ran them on a 750mhz P3 and not a 1.0Ghz P3 as stated in the Anandtech article. This is probably a case of sheer carelessness rather than some deliberate attempt to sabotage the P3, but it appears that someone needs to take a little more care in running benchmarks at Anandtech.
Finally, while it could be argued that gaming benchmarks don?t scale linearly with clock so my example isn?t relevant - I counter that a gaming benchmark won?t scale linearly if you change the MHZ and leave the fsb the same. When one changes the fsb & mhz together - a gaming benchmark will scale linearly with clock, or very close to it, which is what my example is doing. It?s reducing the clock rate and fsb by 25% ? (1000-750 and 133-100)
I think Anand needs to rerun those P3 benchmarks.
Do you agree?. Disagree ??..comments.
Facebook
Twitter