Performance-Showdown between Athlon and Pentium III – THG.RU

Athlon at 500, 550, 700 and 750 MHz!

Although I might take the risk of boring you with even more Athlon-stuff, I decided to beef up the information about AMD’s new CPU on this website. It’s most likely that we at Tom’s Hardware are the first to present Athlon at clock speeds of 500, 550, 700 and 750 MHz and thus we didn’t want to waste any more time and decided to post our findings. Most of you might be very surprised that we could run Athlon at all those clock speeds, considering the fact that Athlon is normally not overclockable. Tom’s Hardware Guide found a way around that, but so far I’ve decided to keep the information on how in between the walls of my lab. Let me tell you why, before some of you start getting upset about my decision. Overclocking Athlon is not a ‘Softmenu’ or ‘jumper/dipswitchy’ kind of job. It requires quite a bit of technical knowledge and electronical abilities. I’d say that only people who weren’t afraid of doing the SMP-‘operation’ on Slot1-Celerons could possibly consider facing Athlon-overclocking. It requires cracking open of the Athlon-housing and then the real fun starts with several SMD-components. That’s all I want to say about this topic for the time being and it’s useless to ask me in emails. I will keep the information under disclosure for now, but as soon as I found an easy way for you I’ll finally post it. Please give me a few weeks. Until then I hope AMD won’t strike down on me.

Athlon delivered a perfectly stable performance

It was a very pleasant surprise to find out that our Athlons ran at up to 750 MHz without any additional means of cooling. This shows that the Kryotech-system could possibly run even faster. Athlon ran as stable as a rock, and when I say stable, I really mean it! I am not talking of crazy overclocking and calling a system stable that needs to be rebooted twice daily. Athlon performed really stable and any software I threw against it ran without any crashes for days. The stability-issue brings me to another thing. Many of you wrote me concerned emails, enquiring if Athlon-platforms might have the same compatibility-problems as the Super7-platforms for K6, K6-2 and K6-3. We have four different Athlon-motherboards here in the lab and each platform ran perfectly with any soft and hardware we used. This includes all known graphics cards from our ongoing “33 3D-card” mammoth-test, even the power-hungry 3Dfx Voodoo3 3500, which made several BX-boards cough really hard. This V3-3500 issue is actually not at all 3Dfx’s fault, it’s due to weak power supplies on several, even well known, BX-boards. Brent is currently working on two articles that follow up on the above said.

Pentium III was cranked up to 650 MHz as well …

Talking about Athlon’s scaling alone didn’t seem appealing enough to me, so that I decided to put Intel’s Pentium III against it. It wouldn’t have been fair to run Pentium III at only up to 600 MHz, which is why I also included results of a Pentium III at 650/100 MHz. You can imagine that I was using a Pentium III 600 without multiplier-lock for this. Remembering my comments about failures of PIII 600 CPUs, actually also reported by several other publications in Germany and the UK, should give you an idea how hard it was to run all the benchmarks with an even overclocked PIII 650. Only one PIII 600 in the THG-lab was able to walk through all tests, and I had to use additional fans to keep the PIII 650 alive. This is not how it used to be. I always praised Intel CPUs as highly overclockable, but PIII 600 is different. The PIII 600 at 650 MHz was always just about able to finish the benchmarks, there was no way of running a system stably at this speed for several days.

Gigabyte’s Athlon Platform

I used the Gigabyte GA-7IX motherboard for Athlon, because Brent was using the MSI-board for graphics cards tests in the Kryotech Cool Athlon 800-system. It showed that the GA-7IX is well capable to supply the current needed by an Athlon 750 without any additional cooling, so that my concerns about the missing heat sinks on the GA-7IX have disappeared. Still, the MSI MS-6167 scores slightly higher in graphics-benchmarks, so that my Athlon-results in Quake2 and Expendable could be even a little bit better. You might find that the results scored in this benchmark are a bit higher than in previously published Athlon-articles. This is due to the fact that I tweaked the Gigabyte GA-7IX to the same settings used in the Athlon-motherboard article.

Asus P3B-F – The New Pentium III Reference-Platform

Intel’s Pentium III was tested in the brand-new P3B-F from Asus. This board combines the stability and reliability of a typical Asus-motherboard with the overclocking-friendliness of Abit’s BX6. Asus calls that feature ‘jumperfree’. The P3B-F scores higher in benchmarks than a BX6 and it does not have power supply-issues with a Voodoo3 3500. You will see that several PIII-scores in this review are slightly higher than in the first Athlon-article we posted. This is due to using the P3B-F. Brent is currently working on a BX-motherboard review that will tell you more about the P3B-F.

Some Overclocking Details

I needed to change Athlon’s core voltage a bit to run it at all the different clock speeds. You may think that I’m talking of higher voltages, but as a matter of fact the opposite was the case. Athlon would not run at 500 or 550 MHz unless I lowered the core voltage down to 1.45 V. On the other side, I only had to raise the core voltage from the default 1.6 V to 1.65 V and Athlon worked beautifully at 700 and 750 MHz. 800 MHz were left to the super-cooled Kryotech-Athlon, since I didn’t want to tamper with my Athlon at higher voltages than 1.7 V and this was obviously not enough to run it at 800 MHz without Kryotech’s cooling.

Pentium III used to 2 V for 500 and 550 MHz, the default of 2.05 V for 600 Mhz and for 650 MHz I raised the core-voltage to 2.1 V. Pentium III did definitely not like any higher voltage than that, it responded with definite crashes to 2.2 V.

Both processors are produced in .25µ-technology and so it’s pretty impressive to see the differences. Intel’s Pentium III can barely reach the 650 MHz at 2.1 V, while AMD’s Athlon runs just fine at up to 750 MHz at only 1.65 V.

System Setup

Office Application Benchmark Results

I used BAPCo’s SYSMark98 again, a benchmark that I like better and better the more I run it. Office application performance already used to be the strength of K6-2 and K6-3, but you had to rely on a Super7-platform that wasn’t always stable. At the same time, AMD was finally not able to reach the same high clock speeds as Intel’s Pentium III. These times have changed. Athlon was designed for high clock speeds, leaving Pentium III considerably behind. The Athlon-platforms, as few as there are right now, have proven reliable and stable in our lab-tests, and Athlon’s office application performance continues to be significantly ahead of the Pentium III at the same clock speed.

The scaling of Sysmark98 under Windows98 is pretty much identical as under Windows NT, only that the NT-scores are higher, as we know for a long while. NT is a real operating system while Windows98 is still the toy-OS. This might finally change with Windows2000, thanks almighty Microsoft! Up to 650 MHz both CPUs scale close to perfectly linearly, above that the Athlon scores may be impacted by the hard drive speed, less likely the memory bus speed. It’s interesting to note that the gradient of Athlon and Pentium III is identical. Pentium III at 650 MHz can pass Athlon at 550 but not even reach Athlon 600.

3D-Gaming Benchmark Results

I decided to only use Quake2 and Expendable and I ran both at 640×480 to avoid a major impact of the graphics card on the performance numbers. Quake3 is too hard on the graphics cards and Half-Life’s ‘Smokin’-demo scores unrealistically low results. Quake2 was executed with the 3DNow!-patch and is thus 3DNow!-enhanced, but not ISSE-enhanced. Expendable is one of the few games that are already using ISSE and it has some minor 3DNow!-optimization too. You will be able to see the difference below:

Both CPU’s scale well in Quake2’s crusher-demo, at least up to 650 MHz core clock. Above that, Athlon is again probably impacted by the graphics card, the AGP-bandwidth or the memory bandwidth, so that the graph becomes a bit flatter, although not too much. Pentium III 650 can’t even reach the scores of an Athlon 550 in Quake2.

The story changes with Expendable and the ISSE-usage. You can see that Pentium III is not far behind Athlon, getting pretty close at 650 MHz. A Pentium III 700 could probably reach an Athlon 700 in Expendable, but we shouldn’t forget that Rage did certainly not use the latest 3DNow!-instructions included into Athlon’s instruction set. At the same time it’s important to remember that Expendable was used by Intel’s PR at the Pentium III-launch. It wouldn’t be surprising if Rage put a whole lot of work into the ISSE-enhancements. Still, Expendable shows that the battle between Athlon and Pentium III becomes closer as soon as a game favorites ISSE-enhancements over 3DNow!-enhancements. That’s where Intel will try to intervene, making sure that software uses ISSE and not 3DNow!. It’s the only way to make Pentium III look good or at least reasonable against Athlon.

3D-Rendering Benchmark Results

Again I used 3D StudioMax and rendered the good old ‘ktx-rays.max’ file. We know that Athlon’s floating-point unit is way ahead of Pentium III, but let’s have a look how bad it leaves the Intel-competitor behind.

This time we can look at a very tough picture. Athlon is not only starting at a higher score, its gradient is a lot steeper than Pentium III’s as well. If you have a close look you can see that it will take a Pentium III at 750 MHz to reach the floating-point power of the cheapest Athlon at 500 MHz. If Pentium III would ever want to reach the 3D StudioMax-scores of an Athlon 800, it would take some 1.6 GHz. I guess that hurts.

Summary

In terms of computing power Athlon’s future looks very good. The performance numbers scale beautifully from 500 to 800 MHz core clock and I have no doubt that this will even continue beyond 1 GHz. It is also very impressive as well as surprising to see that current .25µ-Athlons can already run at 750 MHz. This leaves a lot of room for AMD to finalize their .18µ- Athlon-development, which will lead to clock speeds of 1 GHz and beyond.

Now we should hope that finally more motherboard makers will supply Athlon-boards soon, so that those powerful but lonely AMD-chips can finally find a new home in computer systems. It is ridiculous enough to see how much the battleground has moved these days. Not long ago CPU-makers fought with performance-numbers, now they fight in the motherboard-market and in the offices of software developers. Athlon certainly deserves a reasonably sized motherboard market and fairly programmed software. Whoever disagrees with that shoots himself in the foot, because the last that anyone should want is Intel losing competitors. It’s the only thing that Intel should never lose.