The Giga-Battle Part 2 – THG.RU

Introduction

Last week we brought you the first evaluation of the two Giga-processors from AMD and Intel, using a wide selection of platforms, but unfortunately Intel’s latest flagship chipset, the i840, had to be left out. The latest official BIOS of Intel’s i840-motherboard ‘OR840’ wouldn’t let any processor run faster than 800 MHz, which to my understanding must be another proof for Intel’s paranoia. Obviously Intel wanted to make sure that nobody is able to run processors in the OR840, that could be overclocked to a higher frequency than Intel’s fastest processor at the date of the BIOS release. It seems as if the group of BIOS engineers for the OR840 has still not heard of the Giga-Hertz Pentium III, which is why this new processor isn’t supported yet. It could also be however, that this group of BIOS engineers is well aware of the non-existent supply of Giga-Pentium IIIs, which is why they can’t be bothered to enable OR840 for this processor.

I published this information along with Intel’s peculiar statement, saying that systems with Giga-Pentium III are not supposed to ship with Intel’s fastest official platform and finally Intel’s helpful and knowledgeable German PR-team was able to supply me with an OR840 BIOS that has no ‘high speed lock’. Thus I was finally able to run Giga-Pentium III on an i840-platform as well.

Intel 840 or ‘Carmel’

If you should not be too familiar with i840, I suggest you to read my article about this RDRAM-chipset by the name of ‘Intel’s 840 chipset – The RDRAM Avenger‘. However, I will try and summarize the important facts of i840 right here:

• Code name ‘Carmel’, bigger and stronger brother of ‘Camino’ or i820
• Official workstation and entry-server chipset
• Designed for RDRAM (Rambus) support
• Two Rambus channels, resulting in double the memory bandwidth and half the latency (through interleaving) of i820
• Bandwidth 3.2 GB/s with PC800 RDRAM, 2 GB/s with PC600 RDRAM, no support of PC700 RDRAM
• Support of PC100 SDRAM with additional MTH-chip, i840 can support up to 4 MTHs vs. only 2 MTHs supported by i820
• Recently published bug with ECC SDRAM, leading the majority of Taiwanese motherboard makers to drop plans to build i840-motherboards
• AGP4x – support
• ATA66 – support

You can see that i840 is a lot more powerful than i820, but it requires at least two instead of only one RDRAM-modules, to populate each Rambus-channel. Intel’s OR840-motherboard is a dual processor motherboard with integrated LAN and I guess you can imagine that it does not sport any overclocking features whatsoever.

ECC SDRAM Bug Turns Out to be i840’s Killer

The recently released bug, saying that i840 is supposed to fail with ECC-SDRAM, was almost a killer for this chipset. Due to the high prices of RDRAM and due to the dual-channel architecture, i840 was an attractive solution for an SDRAM platform with 133 MHz FSB-support, using onboard MTHs. i840 was supposed to be used as workstation/server platform, where large amounts of ECC-memory are rather common. After the bug-release i840 ceased to be an interesting platform for systems using ECC-memory, unless the user accepted the high price of ECC-RDRAM. The result was that Taiwanese motherboard makers dropped their i840-plans, because there’s only a minority of customers who have enough money to buy large amounts (1-2 GB) of RDRAM for their servers.

Show Me an Intel Platform for Pentium III Xeon Processors at 133 MHz FSB!

You may not be aware of it, but the recent i840 ECC fiasco puts Intel in an even more ridiculous situation when it comes down to platforms for their latest processors. I wonder if you can answer this question. “In which workstation or server platform can you stick a Pentium III Xeon 800EB?” Well, unless you find a Slot2-motherboard with i840 (which doesn’t exist to my knowledge), you are pretty much left out in the rain. Intel has currently no server chipset with 133 MHz FSB-support and there are no Slot2-motherboards for 133 MHz FSB Xeon processors from Intel! Intel’s 840 chipset was never designed for the high-end server, but only for workstations and entry-level servers. Now since i840 fails to support ECC-SDRAM there is no alternative server chipset from Intel that would support Intel’s fastest Xeon-processors. Is that crazy or what? The only alternative is ServerWorks‘ ServerSet III chipset, a server/workstation chipset that comes in three different flavors, supporting 133 MHz FSB, PC133 SDRAM, but no RDRAM.

Giga-Pentium III and Intel 840 – A Good Solution?

Well, the answer to that question will end up in an ‘if – then’ answer.

• If you don’t mind about the high price of RDRAM, then YES it is a solution that’s way ahead of i820 as well as any Giga-Athlon-platform.
• If you don’t want to run a BX-platform out of spec at 133 MHz FSB, then i840 is the fastest platform for Giga-PIII.
• If you don’t want to use RDRAM, then you better forget about i840.
• If you want to run a Giga-Server, you better look at the ServerSet III HE chipset from ServerWorks.
• If you simply want Giga-performance at an affordable price, you might want to either look at Giga-Pentium III on VIA’s Apollo Pro 133A chipset or at Giga-Athlon on VIA’s Apollo KX133 chipset instead.

i840’s dual Rambus-channel architecture offers a much better performance than i820. We will see in the benchmark results, if i840 is able to outperform 440BX at 133 MHz FSB as well. The results from my latest platform article make that more than questionable though.

The Giga Test Systems

NVIDIA’s New Driver 5.08

In this test we used NVIDIA’s new pre-release driver rev. 5.08, as available at www.reactorcritical.com. This driver vastly enhances GeForce’s OpenGL-performance on all platforms and it removes the ‘fast writes‘ bug of the driver rev. 3.68. At the same time the 5.08-driver reduces Direct3D-performance though. I still decided to use it, because it shows a lot better performance on non-Intel platforms than previous ones. The frame rates of Quake 3 Arena, Quake 2 and SPECviewperf are up to 5% higher than before; the frame rates in Expendable and Unreal Tournament D3D are reduced however.

Office Application Performance under Windows98SE

BAPCo’s Sysmark2000 has become the de-facto standard for office application benchmarking. You can see it, Giga-Athlon does not score as high as Giga-Pentium III, but it’s not too far behind. Athlon on KX133 is slightly faster than on Irongate with enabled SuperBypass. The Giga-Pentium III results are more interesting. They show once more that nothing can beat the good old Intel 440BX chipset even with Giga-Coppermine. VIA’s Apollo Pro 133A is just as fast as Intel’s i820 using the fast PC800 RDRAM, but both are far behind BX. i840 is rather close to BX, but it can’t quite reach it. It’s remarkable however, that i840 with PC600 RDRAM is still scoring better than i820 or Apollo Pro 133A.

3D Gaming Performance – Quake 3 Arena

NVIDIA’s new 5.08-driver produces excellent results for each system. Giga-Coppermine on 440BX is incredibly fast and way ahead of any competition. Surprisingly the new driver makes VIA’s Apollo Pro 133A score better than Intel’s 820 chipset, and even i840 can not quite reach it. Athlon lags behind Pentium III, but with frame rates of more than 130 fps it is still looking very good.

3D Gaming Performance – Quake 2

Quake 2 with its special 3DNow!-support would almost have been won by Giga-Athlon, if there wasn’t this crazy old 440BX chipset. Giga-Pentium III is lagging behind Athlon on all ‘official’ platforms, but i840 makes quite a good figure here.

3D Gaming Performance – Expendable

Expendable is clearly ruled by Giga-Coppermine, which again performs best on the ‘inofficial’ BX-platform, closely followed by i840 and then VIA’s Apollo Pro 133A. Giga-Athlon on KX133 is at least as fast as Giga-PIII on i820. I guess that’s surprising too.

3D Gaming Performance – Unreal Tournament

Unreal Tournament favors Giga-Coppermine as well and as in all other 3D game benchmarks BX is the fastest, i840 is second, then there comes VIA’s Apollo Pro 133A with i820 bringing up the rear. NVIDIA’s 508-driver has changed the picture quite significantly.

Professional OpenGL Performance – SPECviewperf 6.1.1

Advanced Visualizer is clearly limited by GeForce’s fill rate. Only Irongate with its missing AGP4X support is not quite reaching the ceiling.

The story is different with the CAD-program ‘Design Review’. Here Giga-Athlon is pretty much on par with Giga-Coppermine on i840 or BX. SPECviewperf is the one benchmark set where BX doesn’t look quite as good anymore and where i840 is really able to kick ass. You can see once more that VIA’s Apollo Pro 133A is not good for professional workstation software. It scores by far worst.

Professional OpenGL Performance – SPECviewperf 6.1.1, Continued

IBM’s Data Explorer favors Giga-Athlon or Giga-Pentium III on i840, which is now winning this competition as well.

The ray-tracing/radiosity software Lightscape requires quite a bit of FPU-power, but i840s superior memory performance can make up for the inferior FPU of Giga-Pentium. Giga-Athlon won this competition before i840 entered the scene, but now it shows quite clearly that Giga-Athlon is in the urgent need of a high-end platform that doesn’t cripple its performance anymore.

Professional OpenGL Performance – SPECviewperf 6.1.1, Continued

It is pretty obvious that there is something wrong here. I cannot really complain about NVIDIA’s 5.08-driver, because it is pre-release, but you can see that it has some issues here. It cannot possibly be that i840 performs worse than i820 in this benchmark. Giga-Athlon and especially Giga-PIII on the Apollo Pro 133A-platform are also scoring by for too bad. Hence I suggest you forget those numbers until the driver has been fixed. I only included them to be complete.

Floating Point Calculation Performance

3D Studio Max Release 1 may not be the latest version anymore, but that has one big advantage. This early version is still fully usable and it is neither using SSE nor 3DNow!. Therefore it is able to show the pure FPU-performance of a processor. The rendering performance in this benchmark does not depend on the platform, because the benchmark is purely running in the processor’s L1-cache. That is why I didn’t make any difference between the different platforms for each processor, even i840 cannot make a difference here.

Athlon’s pure FPU performance is some 33% percent higher than Coppermine’s. This might not be important for the average user, but whoever is using scientific software will certainly appreciate Athlon’s big performance advantage.

Conclusion

There you have it! With the inclusion of Intel’s 840 chipset in the Giga-evaluation, Giga-Pentium III is able to win each real world benchmark in this set. Only Giga-Athlon’s pure FPU-performance is on top of Giga-Coppermine’s.

I mentioned it already in the first part of the Giga-project, most of you will read these numbers like reading the performance numbers of the latest Porsche 911 Turbo (my favorite car). You are impressed, maybe shaking your head, but only very few of you will buy it right now.

As for the winner of this competition, the Giga-Pentium III, you are in a strange situation. First of all the product doesn’t seem to be available. World wide only a handful of reviewers has done testing with it. European reviewers are ranting and raving about Intel’s inability to supply review samples. Actual Giga-system are hardly in sight anywhere and I doubt that there will be any boxed Giga-Pentium III in retail shops for quite some time to come. So what does that leave us at? A Phantom-product has won the competition. It’s as if an Athlete broke the latest 100m-race record by running on the moon. Even if Giga-Pentium III was available though, you would get the best performance out of an officially ‘outdated’ 440BX-platform, or out of an expensive i840-system, which currently doesn’t support Giga-Pentium III as well.

Giga-Athlon may look like the loser in the benchmark results, but it might turn out as the actual winner. Even though the supply of Giga-Athlon is short as well, there is still the actual chance to purchase an Athlon Giga-system right now. This product is no phantom, and that makes quite a difference. AMD should also realize that Giga-Athlon lost this competition against Giga-Pentium III for two reasons:

• Giga-Athlon should have been released with the ‘Thunderbird’-core, which doesn’t suffer from a slow 1/3-speed second level cache. Pushing Copper-Thunderbird in the Dresden-fab and holding off this crazy Giga-release by a bit would have made sure that Giga-Athlon is able to look a significant lot better against Giga-Coppermine. I am sure that Intel wouldn’t have released Giga-Pentium III as prematurely as they did as well, so AMD could still have won the race.
• It’s about time for some Giga-platforms for Giga-Athlon. VIA’s Apollo KX133 may be a step into the right direction, but it’s hardly good enough against Intel’s chipsets. Dual-processor support, dual SDRAM channels, DDR-support, …. is what Athlon will require to look good against its archenemy Coppermine and particularly against the upcoming Willamette.