The Giga Battle – THG.RU

Introduction

Last week something historical was supposed to happen. For the first time a microprocessor maker launched a processor that runs at 1000 MHz or 1 GHz. Two days later the competitor of this CPU maker and its sworn archenemy did the same with its own CPU, so that last week two Giga Hertz processors were born. Looking at these two happenings a bit closer however, makes this great historical moment more appear like a fancy comedy or soap opera rather than a technological or professional business event.

Winning the Giga Race

No, I don’t want to go on about the fact that only very few people actually need a ‘giga-CPU’ right now, also not about religious issues like “my favorite CPU-maker was faster“. I will not quite yet comment on the fact that some Giga-processors are only available in homeopathic quantities or the ‘great things’ a Giga-processor is able to do for us, I mean meaningful stuff like e.g. running Quake at 14,568,394 fps. What I’d like to say is that the behavior of AMD as well as Intel last week looks as if my good ol’ buddy Kyle, the ‘hardest’ man alive with the largest inferiority complex in the universe, had ran away from his shrink and become CEO of both AMD and Intel for a day.

Isn’t it hard to believe that multi billion Dollar companies such as AMD and Intel are behaving like some little boys, showing off who’s got the biggest and longest, racing the Giga Race? We want to be honest; there was no urgent need for a giga-whopper. It’s only a few months ago when both companies had released their processors at 800 MHz and with either of the two your software was flying already. The logical consequence would have been some kind of orderly release of 850/866 MHz, then 900/933/950 parts and finally we would have been blessed with the big ‘G’. Ok, I admit that AMD is looking a lot better here. The Athlon 850 came out last month and on Monday, March 6, 2000 AMD launched Athlon 900, 950 as well as the Giga-Athlon. Intel stepped from the Pentium III 800, released late December 1999 right to the Giga-Pentium III on March 8, 2000 and completely ‘forgot’ that there should be some actual speeds available in between, like 850, 866, 900, 933 and 950 MHz. It doesn’t really matter, whichever way you look at it; you rather want to smile about the comedian touch than stand in awe for such great achievement. Still there was something really positive and useful in for all of us too. Both CPU-makers dropped their prices significantly and from this point of view we can only hope that there will be ‘2 Giga Hertz Day’ soon as well, making processors even cheaper.

The Giga Specs

I don’t want to bore you with the actual specs of both Giganauts for another time, but a few things need to be mentioned.

Giga Blaster No.1 is AMD’s Giga-Athlon . This processor is based on the Athlon core from last August, but this time it’s produced in 0.18 micron aluminum technology. Still Giga-Athlon requires external second level cache, which is also found inside the ‘SlotA’-cartridge. Athlon’s biggest problem is exactly this external L2-cache, because it only runs at a third of the processor’s internal clock frequency and it’s connected to the core with a 64-bit wide data path. Athlon 800 is running its L2-cache at 0.4 times the core clock, but AMD was seemingly unable to find L2-cache modules that would do the required 400 MHz for Giga-Athlon. Thus the Athlon at 1 GHz does not scale up against its brother with 800 MHz as the core clock would make it appear. This problem won’t be solved until AMD will release its ‘Thunderbird’-core (0.18 micron copper tech), which will have the 2^nd level cache on-die, running at core clock frequency and connected to the core with a 256-bit wide data path. Giga-Athlon’s core voltage was upped to 1.8 V and it requires a Giga-heatsink and Giga-power supply. Still my sample ran Giga-stable in any of my Giga-platforms.

The Giga Specs, Continued

Giga-Pentium III is of course based on Intel’s ‘Coppermine’-core. While Coppermine represents a very advanced version of Intel’s five-year-old Pentium Pro architecture, the 0.18 core of Coppermine is one step ahead of Athlon. Giga-Pentium III’s 256 kB 2^nd level cache is on-die, clocked at core clock and connected with a 256-bit wide low-latency data path. This creates a performance advantage over AMD’s Athlon, even though Athlon’s internal architecture is far advanced over Coppermine’s. Pentium III is also including Intel’s ‘Streaming SIMD Extensions’ SSE, which are used by quite a bit of software nowadays, as we will see later on. Athlon’s ‘Enhanced 3Dnow!’ extensions are able to do the same work as SSE, but AMD’s lack of a compiler for those extensions keeps a lot of software developers from using it as much as AMD would wish.

Giga-Pentium III is also using a higher core voltage than its slower brothers. 1.7 V are required for this Giga-processor. What’s touchy though is the temperature specification of Giga-Coppermine. 60 degrees Celsius is the maximum heat that it is allowed to reach, which is Giga-pathetic compared to the 80 degrees C that any other Coppermine can endure. Thus Giga-PIII needs a Giga-heat sink as well unless you aren’t afraid of a lot of Giga-system crashes on a regular basis. OEM’s are already reporting that the Giga-heat sink requires special assembly lines, but as long as Giga-CuMine is only available in ‘Giga-small quantities’, those assembly lines won’t be too busy I guess.

Giga-Expectations

“If Giga, then which Giga?” is the question here. I am fully aware of the fact that only few of you will be getting and needing any of the two Giga-monsters right now, but I also know that it’s just great to read about the performance Giga-battle, ain’t it?

Giga-2^nd Level Cache

Now what do we know and what can we expect from the Giga-fighters? Well, first of all there is the well-known 2^nd level issue, which I already discussed above. Giga-Coppermine with its on-die L2-cache is obviously scaling better than Giga-Athlon. At 800 MHz both processors were pretty much on par, which is why it’s pretty likely that in most applications Giga-Coppermine will be faster than Giga-Athlon. The story is a bit different for scientific and CAD applications that require a lot of number crunching horse power. Athlon’s FPU is far superior to Coppermine’s, so it will smoke Giga-PIII unless the software makes heavy usage of SSE and none of 3DNow!, something that unfortunately exists.

Giga-Chipsets

The next thing we shouldn’t forget is the platform issue. I know, I’ve been talking a lot about that lately and we’ve got quite a list of articles addressing this issue.

• Showdown at 133 MHz FSB – Part 2, The Real McCoy
• Issues with VIA’s Apollo Pro133A
• Performance Showdown at 133 MHz FSB – The best Platform for Coppermine
• The Impact of the AGP-Speed
• Irongate with Super Bypass vs. VIA Apollo KX133
• NVIDIA’s New NT-Drivers for Athlon on KX133
• New Hope For Athlon – The VIA Apollo KX133 Chipset
• AMD’s Super Bypass – AMD Improves their 750 Chipset
• The RDRAM Avenger – Intel’s i840 Chipset
• Intel i820 Chipset Review

What it boils down to is that Athlon is a bit at disadvantage here too. So far there are only two different platforms available for the fastest AMD-processor. Number 1 is AMD’s own, but slightly outdated 750 ‘Irongate’-chipset, which lacks AGP4X and fast memory, as it’s only using PC100 SDRAM. The other and newer platform is VIA’s Apollo KX133 chipset. It supports PC133 SDRAM and AGP4X and is therefore currently the Athlon-chipset of choice.

There is a lot more platforms available for Intel’s Pentium III. First of all there are the three 800-chipset, i810e, i820 and i840. Let’s forget about the first one for Giga-PIII, since it is a low-cost solution with integrated graphics. Intel’s 820 chipset is supposed to be the mainstream platform, but it’s got this little RDRAM-problem you might have heard of. Intel’s 840 is a much more interesting, but unfortunately even more expensive solution. It is also meant to use the super expensive RDRAM, but it offers two memory channels and thus very high memory performance. Too bad is however that Intel’s own OR840 motherboard with the 840-chipset is not able to run Giga-Coppermine, because even the latest BIOS wouldn’t let any CPU run faster than 800 MHz.

With the Apollo Pro 133A VIA is offering a very nice and attractive alternative. This chipset may not be the right one for professional OpenGL workstations, but people who are playing games or running office applications will be very pleased with its performance. PC133 SDRAM is used by this VIA solution as well, which keeps system costs at a reasonable level.

Last but not least there is Intel’s old 440BX chipset. This platform was never meant for Giga-PIII, but you will see that for nerds there’s no better one available. BX was only designed for 100 MHz FSB and not for CuMine’s 133 MHz FSB. Equipped with PC133 SDRAM it can run at this speed as well, but in this case the AGP is out of spec, running at 89 MHz. Thus I am warning every inexperienced user from combining BX and Coppermine @ 133.

Giga-Availability – Athlon Yes / Pentium III No

The European IT-press is badly upset with Intel. It seems as if not even one Giga-part has made it across the Atlantic since the release of Giga-Pentium III. I am not talking of shipments to OEMs, not even one press sample has reached European ground, showing that Intel hasn’t even got enough Giga-PIIIs to supply the reviewers, who are supposed to write nice articles about it so that the readers can try and get a product that isn’t available. AMD is a lot better at that. The press was equipped with Giga-Athlons already before or at least on the release day. It also seems as if Giga-Athlon OEM-systems are actually shipping already.

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. 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 for 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.

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, which proves once more that you’ve got to be insane buying a system with i820. 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. Coppermine is lagging behind Athlon on all ‘official’ platforms though.

3D Gaming Performance – Expendable

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

3D Gaming Performance – Unreal Tournament

Unreal Tournament favors Coppermine as well and as in all other 3D game benchmarks BX is the fastest, VIA’s Apollo Pro 133A second and i820 the slowest platform for Coppermine. 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 Athlon could leave the Coppermine competition behind if I wouldn’t have been so bold to include Coppermine on 440BX. 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 Athlon and even the old and bold BX-chipset can’t save Giga-Pentium III in this benchmark.

The ray-tracing/radiosity software Lightscape requires quite a bit of FPU-power, which is why Giga-Athlon smokes Giga-Pentium III once more. It also seems to require something that BX isn’t able to supply. That is either high memory bandwidth or, more likely, AGP4X.

Professional OpenGL Performance – SPECviewperf 6.1.1, Continued

ProCDRS running on GeForce with the new 508 driver will become the favorite benchmark for Intel’s platform division. It’s the ONLY benchmark where the RDRAM-chipset i820 is declassing all the competition. VIA’s Apollo Pro 133A looks extremely bad. Athlon would probably score a lot better if it had the memory bandwidth supplied by RDRAM or DDR-SDRAM. ProCDRS needs heavy memory bandwidth and AGP4x.

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.

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

The Giga Performance Battle is ending with interesting results. Intel’s Giga-Pentium III is leading in front of AMD’s Giga-Athlon in the majority of benchmarks. However, Intel will still not like those benchmark numbers too much. It must be horrible for Intel’s platform division to see how badly the old 440BX-chipset destroys i820. Even if we forget about the incredibly high scores of BX for a moment, there’s still the excellent results of VIA’s Apollo Pro 133A platform in all 3D games and office applications. It is certainly a shame that i840 was not included in those benchmark numbers, since it could have saved Intel’s face. The reason for this however is Intel itself. In its paranoia against overclocking Intel’s OR840 motherboard with the i840 chipset comes equipped with a BIOS that wouldn’t allow CPUs to run faster than 800 MHz. Intel could not send me a fix until now and the answer was “Giga Hertz systems are supposed to ship on i820 platforms. So far there is no plan to use Pentium III 1000 on the OR840 motherboard.” Isn’t that a typical Intel-answer? The fastest processor build by Intel is not supposed to run on the fastest platform that’s officially available! Somebody understand Intel’s logic!

AMD has no reason to feel bad though. It’s not only nice to see that Giga-Athlon performs best in the majority of the SPECviewperf and FPU benchmarks. In the rest of the tests Giga-Athlon is certainly not exactly far behind Giga-Pentium III as well. Keep this in mind when you realize that Intel can hardly ship any gigahertz Pentium III processors. While Intel has not even got enough processors in the area between 800 and 1000 MHz to release Pentium III 866 and 933, and Europe is bar of any 1 GHz PIII sample altogether, AMD is equipping the press as well as its OEMs with gigahertz Athlons. If you really think that you need a Giga-System, the decision will be very easy. Athlon Gigas are actually available, Pentium III Gigas are a simple phantom. You cannot buy what isn’t there, can you? Let’s see if Intel has successfully impacted AMD’s business by releasing the Pentium III at 1 GHz prematurely. Intel’s plan might have been to damage AMD’s sales by releasing a phantom product at a very low price. If this phantom product doesn’t soon become reality however, Intel’s plan might backfire. I wonder if Craig Barret (or was it Kyle after all?) will succeed with this boasting policy.

Follow-up by reading the article ‘The Giga-Battle Part 2‘.