Performance Showdown at 133 MHz FSB — The best Platform for Coppermine — THG.RU

Introduction

Who wouldn’t remember October last year, when Intel released its successor of the Pentium III with the ‘Katmai’ core. The introduction of the new Pentium III with the ‘Coppermine’ core was essential for Intel, because this new processor was Intel’s only weapon against AMD’s new Athlon processor, the clear performance leader at that time. Athlon had been so much faster and even cheaper than ‘Katmai’, that Intel had been starting to lose face. ‘Coppermine’ a processor core based on ‘Katmai’s’ core, but improved with full speed and on-die 2^nd level cache that was connected to the computing units with a data path 4 times as wide as Katmai’s, turned out to be significantly faster than its predecessor at the same clock speed and it could at least catch up with Athlon, although not quite overtake it. From that point of view everything should have been fine for Intel. However, a wrong decision from the past had caught up with mighty Chipzilla. Rambus started to wreck havoc with Intel’s future platform plans.

Has the Force Left Intel?

One of the new features of ‘Coppermine’ was its support for 133 MHz front side bus speed. Higher FSB’s are always welcome and so everybody should have been happy with it, if, yes, if there had been an Intel platform to support it. The Force hadn’t really been with Intel last year, which is something that I appreciate, since the dark side of the Force had hit me in April 1999 as well. I had been badly betrayed by the greed of some so-called ‘friends’, but Intel had been betrayed by its own greed, which makes the story a bit more enjoyable. Quite a while back Intel had decided that Rambus is the memory of the future. The reason behind this decision was not necessarily the superiority of the Rambus-design, but the fact that Intel could buy a significant amount of shares of the sweet little company back then, planning to finally rule the memory market, once it had taught the world that there was only one way in the IT-business and that was Intel’s way.

Intel’s ‘Caminogate’ i820

Intel’s new chipset, that was supposed to support the ‘Coppermine’-processor, was first known under the code name ‘Camino’. It was supposed to be the first PC-chipset to support Rambus or RDRAM. RDRAM was and still is not exactly popular in the industry, because Rambus Inc. owns the design and so each memory maker producing RDRAM has to pay royalties to Rambus Inc. Intel didn’t bother about that, since history had proven that whatever Intel proclaimed as the future direction was finally accepted by the world. Unfortunately however, the Force had left the designers of Camino. The effect was rather devastating. ‘Camino’, today known as ‘Intel 820‘ chipset, turned out to have a major bug in the very last minute before its planned release. This had been relayed to the poor motherboard makers so late, that some of them couldn’t even stop their marketing campaigns for i820 anymore. Intel also decided against any further delay of ‘Coppermine’, so that this new processor was released without an Intel-platform for it. What happened after that is history and known to all of us. Finally i820 was released, but castrated to only 2 RIMM slots for the super expensive RDRAM. The Intel820 chipset may perform well, but nobody can afford RDRAM. Thus people considered Intel’s half-baked ‘MTH’-solution that supports PC100 SDRAM. It translates the parallel CPU protocol in the serial RDRAM protocol and then back into the parallel SDRAM protocol, which results in mediocre performance. The next thing that came as a shock was when Intel admitted that i820 plus SDRAM is having problems as well. ECC SDRAM leads to crashes, which is as paradox as it comes, but who would wonder about that anymore? We’ve learned to expect the craziest stuff from Intel today.

VIA’s Apollo Pro 133A, VT82C694X

VIA, who made it into the league of Intel’s arch enemies recently, had come out with a chipset that offered 133 MHz FSB (inofficially!) and PC133 SDRAM support already in June 1999. By that time the performance of the Apollo Pro Plus (VT82C693) platforms was rather mediocre though. Then VIA saw its big chance when ‘Caminogate’ (copyright Mike Magee, The Register) took place. At the time of the Coppermine-launch VIA’s Apollo Pro 133 (VT82C693A) was the only performance chipset that supported the 133 MHz FSB and thus the ‘real’ Coppermine. Nobody was crazy enough to plug this expensive new and fast Intel processor into an i810e-platform, although i810e was at least able to support 133 MHz FSB. The integrated 3D-deccelerator and only PC100 SDRAM support hinder its performance and make it rather unattractive. Anyway, VIA’s new Apollo Pro 133A (VT82C694X) chipset now offers full Coppermine support, AGP4x, PC133 SDRAM support instead of the super expensive RDRAM and it comes without any major bugs. Therefore it is the born competitor against Intel’s cursed i820 chipset. It took a long while though, until VIA moved from Apollo Pro Plus over Apollo Pro 133 to Apollo Pro 133A. The latter (VT82C694X) finally performs well enough to be considered a performance threat to i820, as Asus’ new P3V4X-platform with this new chipset is able to prove impressively.

Quick Comparison

It’s pretty obvious that motherboards with VIA’s Apollo Pro 133A (VT82C694) chipset are clearly the cheapest solution for Coppermine. If the performance is right, VIA’s Coppermine-solution is the clear winner in this competition.

The Good, the Bad and the Ugly — Who Will Get the Coffin Full of Gold?

To make the story easy and enjoyable, each representative comes from the same motherboard maker. I chose Asus because I always get their products on time, even early products work reliably and I get the best support. Curtain up, let’s have a look at our contestants!

The Good

You know how things are, ‘The Good’ could not possibly have been one of the Intel representatives, it’s the representative from VIA, the Asus P3V4X with Apollo Pro 133A chipset. The motherboard looks as solid as it actually is, it sports 6 PCI slots, one ISA for the oldies, one AGP, four DIMM slots, two USB-hubs and thus four USB-ports and the Asus ‘JumperFree’ hardware setup, which can also be disabled, so that the adjustments can be made via dip switches.

The Bad

The P3C-L with RDRAM is playing ‘The Bad’ in this contest. Why? Well, it’s using the bad bad i820 chipset AND the bad bad bad expensive RDRAM on top of it. The Bad is equipped with 5 PCI-slots, one ISA, one AGP, one AMR, it sports onboard LAN with Intel’s i82559 and the two RIMM-slots that Intel allows.

The Ugly

What happened here? Well, you’ve got to admit it. This one IS ugly indeed. What we see is again the P3C-L, but this time equipped with Asus’ DIMM Riser II. This huge piece of plastic hosts Intel’s beloved ‘MTH’-chip, ‘the memory translator hub’.

The DIMM-RiserII can be plugged into a RIMM-slot and then you can put PC100 SDRAM into the riser-card. It might look ugly, but it’s the same thing as an onboard ‘MTH’-chip. This way, you could possibly get the P3C, buy the Riser II and SDRAM, thus saving the money you would have wasted on RDRAM and once RDRAM has dropped in price, some time in 2246, you can take the RiserII off and plug in your new RDRAM.

We want to learn now, which platform is the best for a nice and shiny Pentium III 800 or Pentium III 866. Should it be the expensive i820-plus-RDRAM solution? Is it i820-plus-SDRAM, cheaper, but still good? Or might it be the outsider from Taiwan? An inexpensive motherboard with PC133 SDRAM? It’s lean but powerful, two Asian virtues!

The Testbed

First of all I would like to apologize that we did not use Windows NT. This OS has just become obsolete for reviewers, although I appreciate that many are still using it. Windows2000 on the other hand is very young and the support for it isn’t quite that mature yet. Thus I decided to focus on Windows98-testing, which should anyway give reliable performance data that can be transferred over to WinNT and Win2k.

Our test-suite consisted of the new BAPCo Sysmark2000 for office application performance evaluation, the two OpenGL-games Quake 2 and Quake 3 Arena, two 3D-games running under Direct3D, Expendable and Unreal Tournament and finally some OpenGL workstation stuff tested with SPECviewperf. The test 3D-card was a GeForce DDR reference board as usual.

Benchmark Results — Office Applications — Sysmark2000

Finally VIA’s Apollo Pro 133A with PC133 SDRAM has reached the office performance of Intel’s 820 with PC800 RDRAM. This is rather impressive and shows what good of an alternative the VIA chipset really is. Intel 820 plus SDRAM lags behind with a rather serious distance. The results plus the price point clearly favor the Apollo Pro 133A chipset, which is as fast as i820 w/RDRAM, but vastly cheaper.

Benchmark Results — 3D Games — Quake 3 Arena

Intel’s 820 w/RDRAM is the leader of the pack, but VIA’s Apollo Pro 133A is close behind it. i820 w/SDRAM is far behind the two and shows once more that 820 plus SDRAM is no real alternative.

Benchmark Results — 3D Games — Quake 2

Benchmarking with Quake 2 shows the same results as with Quake 3 Arena. ‘Caminogate’ with the expensive RDRAM is leading; Apollo Pro 133A becomes close second and i820 w/SDRAM lags behind considerably.

Benchmark Results — 3D Games — Unreal Tournament

Unreal Tournament is not quite as particular about the memory performance as the Id-games are and the frame rates are significantly lower as well. Still the rating hasn’t changed, Apollo Pro 133A stays close behind i820 w/RDRAM, i820 with the MTH comes last.

Benchmark Results — 3D Games — Expendable

Expendable repeats the above results once more. All in all VIA’s Apollo Pro 133A in the P3V4X-motherboard from Asus is an extremely attractive alternative to i820, even for 3D-gamers.

Benchmark Results — Professional 3D — SPECviewperf 6.1.1

The order of precedence hasn’t changed a bit, but i820 w/RDRAM is further ahead of Apollo Pro 133A than in the 3D-games. Professional graphics applications do make a difference between the 1.6 GB/s memory bandwidth of i820 w/RDRAM and the 1 GB/s of Apollo Pro 133A. However, if you want to find the very fastest Coppermine-platform for professional workstation software, you might want to go for Intel’s i840 chipset, which offers a memory bandwidth of no less than 3.2 GB/s.

Conclusion

First things first. Intel’s i820 chipset with the ‘elitist’, super expensive RDRAM is still the overall fastest mainstream chipset for Coppermine. The only chipset that is faster comes at an even higher price point and is Intel’s workstation chipset by the name of ‘i840’. Both chipsets definitely require RDRAM to perform well though.

People who think that i820 with PC100 SDRAM is worth its money must be a lot closer to the ‘lunatic fringe’ than any PC-hobbyist. Maybe Mr. Dell should consider that before offering Dell-systems with this crazy constellation. Firstly i820-motherboards are rather expensive anyhow. Then you need to pay extra for the ‘MTH’ ‘memory translator hub’ chip that enables i820 to use the cheaper SDRAM instead of RDRAM. Now you can add PC100 SDRAM at a reasonable price, but look at the above results to see what horrible performance you get in return!

The CLEAR winner in this competition is VIA’s VT82C694X or ‘Apollo Pro 133A’. This chipset offers almost the same performance as i820 with RDRAM and it scores even identically to the expensive competition from Intel under office applications. For a most significantly lower price you get a product that scores slightly worse than i820/RDRAM in 3D-games and some 10% worse in professional OpenGL applications. For a normal consumer there’s no alternative to it, for a workstation guy the story might look a bit different. I was honestly impressed with VIA once more. After the good Apollo KX133 for AMD’s Athlon, which only had to beat AMD’s aging 750-chipset, VIA has now shown that it can take on Intel just as well. Congratulations!

My last words are for system integrators and OEMs that offer i820 plus SDRAM. Please do us all a favor and get lost with this sorry combination! Your customers deserve better.

Follow-up by reading the articles ‘Issues with VIA’s Apollo Pro133A‘ and ‘Showdown at 133 MHz FSB — Part2, The Real McCoy‘.