This is an old archive page of HWZ prior to revamp. Please visit HWZ for the latest reviews and news.

 

 
» Articles
Athlon XP vs. Pentium 4 : The Ultimate 32-bit CPU Showdown
By CPU-zilla
Category : CPU
Published by Jimmy Tang on Sunday, 14th September, 2003


The Fastest Processor Is...

Now that we've seen the performance of each and every single processor running in all of the 19 benchmarks thrown at them, it's time to sum all of them up and give you the final rating. In order to give you a fair and tangible look at the performance of each of the processors tested, we aggregated results from the following benchmarks :-

  • BAPCo SYSmark 2002
  • SPECCPU 2000 v1.2
  • SPECviewperf 7.1
  • Lightwave 3D 7.5
  • WSTREAM
  • Cachemem (L2 cache results)
  • XMpeg 4.5
  • Futuremark 3DMark 2001SE Pro
  • Futuremark 3DMark 03 Pro
  • Unreal Tournament 2003
  • Jedi Knight II
  • Quake 3 Arena

    The reason why the above benchmarks were chosen was simple. These are well accepted benchmarks and they represent each facet of the processor's performance. In addition, the methodology used in these benchmarks were acceptable and we think they cover most of what users use today. Thus, the results from each of these benchmarks will play an equal role in calculating the final score.

    We summed up selected results from these benchmarks and they were then normalized to the slowest processor in this showdown, which happens to be Intel's Pentium 4 2.4C (2.4GHz is perceived by most to be slower than a 2500+).

    The results are then tabulated in the chart below.


    Although the Athlon XP had its fair share of ups and downs in each of the 12 benchmarks chosen above, it still did not managed to beat any of the new 800MHz FSB Pentium 4 processors. We were quite disappointed with this outcome as we really hoped to see the newer 400MHz FSB Athlon XPs take their place in either the third or fourth rankings.

    There's really not much we can say now but we have to really give the performance crown to Intel. In fact, it was quite a shame that the Pentium 4 2.4C made it past the Athlon XP 3200+.

    Looking at the chart above, we found something rather peculiar with the results and so, we decided to plot the performance of each processor with respect to their clock frequency (or model number).


    The chart above gives us a better view of how each processor scales. While the Intel Pentium 4 enjoys a pretty linear scaling curve, the Athlon XP seems to be struggling. The trouble spot seems to stem from the Athlon XP 3000+ processor where AMD has made available two different chips based on the same rating. One of which was for newer 400MHz FSB capable motherboards while the other model was once the flagship of the 333MHz FSB series. Obviously, something didn't quite match up here as the Athlon XP 3000+ CPU seems to be running faster at 400MHz FSB although its clock speed is 66.66MHz slower than the 333MHz FSB model. AMD's 'True Performance Initiative' is really in question here as it's not delivering the kind of performance promised to end users. Was this a mistake right from the beginning? Critics tend to believe so and their beliefs would be further enhanced from this set of results.

    If you dissect the results above further by calculating the slope of each segment on the graph (differentiation of each point on the x-axis - remember calculus?), we came up with the plot below.


    Now, this might seem to be too much to follow, so we'll try to make it simple here. Each data point in this chart represents the amount of performance you get with every 1MHz increase in clock frequency. First, let's look at the data points for Pentium 4. For every MHz increase in clock frequency from a 2.4C to a 2.6C processor, you get an average of about 0.029% increase of performance. While this number may seem small, multiply that by 200MHz and you'll get about 5.8%. So, what can we conclude here? For Pentium 4, the increase of performance for every MHz is pretty linear and they are between 0.026% to 0.028%. To us, that's very acceptable and it's very important that users know for sure that performance scales with clock frequency linearly.

    Next, look at the data points for Athlon XP. While performance increase per MHz were pretty linear up till 2800+, everything starts to go out of hand when AMD began introducing the 3000+ model. It was extremely disappointing to know that you actually gain very little if you moved from a 2800+ processor to a 3000+ (333MHz FSB) processor. This is not surprising as the clock frequency difference between them were only 66MHz while between them was a wide 200+ model difference. We think AMD should really stop making the Athlon 3000+ (333MHz FSB) version since it's totally not worth investing in one - considering how much performance you actually gain from it. Instead, they should replace it with the 400MHz FSB version as it presented a more reasonable performance scaling although it's still abnormally low.

    Now that you know how to interpret the chart above, do you really need us to tell you how rotten is the Athlon XP 3200+? Seriously, AMD's True Performance Initiative looked good at the beginning but when they started compromising clock frequency with FSB speed, we think the model numbers are doing them more harm than good.

    Let's look at the next page as we discuss more about price and performance.
  • <<Prev | Page 24 of 26 | Next>>