<!–#set var="article_header" value="DDR For Pentium III –
VIA’s Apollo Pro 266 Chipset” –>
Introduction
We have already mentioned it before; DDR-memory is not really anything brand new. It has been used very successfully on high-end 3D graphics cards for a couple of years, but it took until last October, that DDR-memory finally made its way to the system memory banks of motherboards.
Theoretically, DDR-memory is offering double the bandwidth of equally clocked single data rate memory such as the well-known PC133 SDRAM. That is why the performance expectations of systems equipped with DDR-SDRAM main memory were pretty high. Unfortunately the expectations haven’t quite been met.
First There Was DDR For Athlon
While the launch of the first PC-chipset with DDR-memory support is lying back two months now, it is still close to impossible purchasing any motherboards or the actual DDR-memory anywhere. Problems with AMD’s 760 DDR-chipset and the unavailability of Athlon-processors with 133 MHz processor bus clock delayed the actual introduction of DDR-memory to the PC-market.
In late 2000 ALi had released its MaGiK1 chipset, which also offers DDR-memory support, but unlike AMD’s 760 chipset it has not yet been able to show any major performance boost over Athlon platforms that use the well-known PC133 memory.
It seems as if DDR-platforms for Athlon are slowly but surely becoming available now offering you a 0 – 10% performance increase over VIA’s Apollo KT133A PC133 solution, depending on the DDR-memory, the motherboard and the application used.
Intel Preferred RDRAM
The story is a bit different for Intel processors. First of all Intel doesn’t offer any DDR-chipset for any of its processors at all yet. Instead, it started betting on Rambus’ RDRAM more than a year ago. At first Intel had a lot of problems implementing RDRAM in its chipsets. In November 1999 there was the Camino-Gate story, later joined by the MTH-debacle and the drop of Timna. Regardless of those mishaps and failures however, the expensive RDRAM had never been able to show any performance benefit with Intel’s Pentium III processors, which was why most reviewers, including us, considered RDRAM as a major failure. The questionable policy of Rambus Inc. was another good reason to dislike this ‘IP’-company and its RDRAM memory.
Things changed when Intel finally launched its Pentium 4 processor a couple of months ago. This microprocessor was tuned for two things, extremely high clock speed and high memory bandwidth. While Pentium 4’s behavior in today’s benchmarks is rather flaky and unable to convince the experienced ones of us of its claimed superiority, Pentium 4 was finally the first processor that was really able to benefit from high bandwidth memory such as the RDRAM P4-systems come equipped with. In fact, Pentium 4’s anyhow questionable performance in today’s applications suffers badly once its memory bandwidth decreases.
The above graph taken out of the initial Pentium 4 article shows quite clearly how badly Pentium 4 suffers once it’s equipped with PC600 RDRAM, the memory found in the typical Pentium 4 system from Dell or HP today. Office or 3D-gaming performance drops some 3-8%.
Intel’s and Rambus Cover-Up Strategy Continues With Pentium 4
The situation with Pentium 4 finally reveals Intel’s and Rambus master plan behind the forced, painful and obviously pointless introduction of RDRAM in 1999. Intel was fully aware of the fact that Pentium III wasn’t able to benefit from RDRAM one bit, but it wanted to make sure that RDRAM is available in the market once Pentium 4 gets introduced, because this processor would give awful performance results with single data rate SDRAM. At the same time, Intel has still not committed to DDR-memory and some people concluded that this was because only RDRAM could satisfy Pentium 4’s craving for memory bandwidth. It is a matter of fact that WE DON’T KNOW that. Due to Pentium 4’s memory bandwidth dependency, it might be THE microprocessor to show the virtues of DDR-memory, but if Intel would offer Pentium 4 with DDR-memory now it would easily outsell P4-systems with RDRAM (simply because of price), thus jeopardizing Intel’s sweet little deal with Rambus Inc. Therefore, instead of cheering for Rambus’ rebirth, I would suggest to remain as suspicious as possible. Again I ask ‘who would trust Rambus?’ Why is Intel deliberately delaying a DDR-platform for Pentium 4 until the third or maybe even fourth quarter of 2001?
Pentium III And Memory Bandwidth
Lot’s of us were surprised to see Pentium 4’s good results with RDRAM, remembering the sad results Pentium III had produced with the RDRAM-chipsets i820 and i840. The below graphs are taken out of the article about Intel’s i815 chipset from June 2000. They show how much a 1 GHz Pentium III suffers from the RDRAM-platforms i820 and i840.
After all i840 offers the same memory bandwidth as Pentium 4’s i850 chipset, but Pentium III on i840 was not even able to outperform the good old BX-chipset. Remembering the somewhat sobering performance gains scored by Athlon DDR-systems over PC133-systems should already give you a taste for what you can expect from Pentium III plus DDR compared to Pentium III on i815 with PC133 memory. It also gives us an explanation. Pentium III is not designed for high data bandwidth. AMD’s Athlon is already taking very little advantage of the higher memory bandwidth offered by DDR-memory platforms. Pentium III is actually even worse, explaining the bad results of Pentium III on RDRAM-platforms with i820 or i840 chipset. Keeping that in mind might get you prepared enough for our findings with VIA’s new Apollo Pro266 Pentium III-chipset with DDR-support.
VIA’s Apollo Pro 266 Chipset
VIA has not exactly been particularly quick with the release of a DDR-chipset. While at VIA’s booth at Computex 2000 (June 2000) DDR had been all over the place and the officially announcement of VIA’s first DDR-chipset goes back some hefty 4 months, it took until 2001 that VIA was finally able to join the DDR-bandwagon and team up with the other two DDR-chipset providers AMD and ALi.
It is less of a surprise to see VIA launching their Pentium III DDR-solution before the only recently announced Athlon DDR-solution ‘Apollo KT266’. As much as VIA may commit to AMD and its processors, as much it is aware of the fact that the majority of sold microprocessors are still carrying the Intel-logo. VIA wants to make money and so it decided to introduce its DDR-chipset for Pentium III first, just as it supplied its first PC133-chipset for Pentium III and not for Athlon as well.
With the name ‘Apollo Pro 266’ VIA is obviously not trying to give the impression as if this new chipset is particularly different to its predecessor that went by the name of ‘Apollo Pro 133’. In fact, the specs don’t really look so much different. The major improvement is the support of PC266 (or PC2100, which means the same) DDR memory and the support of ATA100. The latter isn’t even that special, because the Apollo Pro 133 chipset was already able to offer this feature once a board would come equipped with the 686B south bridge chip instead of the initial 686A chip, which only supports ATA66.
‘V-Link’
Still VIA has made one major change, which is the interconnection between the ‘north bridge’ and the ‘south bridge’ of the Apollo Pro 266 chipset. I have no idea which wise person once invented the names ‘north’ and ‘south’ bridge, because your motherboard will actually work just fine even if you point the north bridge to the west and the south bridge to the east. Basically it’s a naming that distinguishes the two chips that make up the vast majority of PC chipsets, in which case the ‘north bridge’ is the chip that hosts the interconnect to the processor, the AGP and the memory, while the ‘south bridge’ hosts all the other I/O-connections to the hard drive, the USB, serial and parallel ports as well as AC97 sound and modem stuff.
Traditionally, those two chips are using the PCI-bus as their connection, which means that either of those chips has to share this ‘highway’ with other PCI-devices, such as SCSI-adapters, network cards, sound cards and more. The PCI-bus is 32-bit wide and clocked at 33 MHz, so that the bandwidth that’s shared between all those PCI-devices, including north and south bridge, is 133 MB/s.
The introduction of Intel’s 810 chipset marked some kind of advance over the above-mentioned old design. Intel’s 810 and all other 8xx chipsets are using the lovely-named ‘Intel Hub Architecture’, which only points out the fact that Intel is now connecting ‘north’ and ‘south’ bridge via a dedicated little bus that is 8-bits wide, but double-pumped and clocked at 133 MHz, providing an unshared band width of 266 MB/s between the two chips. The PC-bus is in this case only connected to the ‘south’ bridge, or as Intel names it, the ‘ICH’ = I/O Controller Hub.
What Intel can do VIA finally figures out as well and so Apollo Pro 266 is now the first chipset with a solution that’s virtually identical to Intel’s ‘hub architecture’, but with the name ‘V-Link’ not baptized half as kewl. It’s hard to beat Intel’s marketing department, wouldn’t AMD know that as well?
The average computer user will not notice any difference between the ‘old’ PCI-bus solution and the new ‘hub architecture’ or ‘V-Link’ architecture, but once you remember how high the data bandwidth of PCI devices such as IEEE1394, SCSI-3 or the next ATA-solution can actually get you might feel a bit safer with the new 266 MB/s interconnect.
Test Board
Our Apollo Pro 266 test candidate was Gigabyte’s brand new GA-6RX Socket370 motherboard, which is a very well equipped tweaker platform. Besides Gigabyte’s well-known Dual-BIOS feature for safe FLASH-procedures, 5 PCI-slots, 1 AMR-slot, 4 DDR-DIMM slots, Creative’s CT5880 onboard sound, ATA100-support, Promise’s onboard ATA100 RAID0/1 chip, four USB-ports and more, it also offers all overclocking features from FSB-tweaking to Vcore adjustments either in its BIOS-setup or on the board with dipswitches.
We decided to put it up against a motherboard with i815 chipset, which resembles the most sensible solution for Pentium III ‘Coppermine’ processors to date. Of course our reference board Asus CUSL2 was used for that.
Expectations
Well, of course we would like to see the VIA Apollo Pro 266 board smoking the i815-platform due to its superior memory bandwidth, but we know how little Pentium III was able to benefit from RDRAM and thus we don’t expect much of an advance over Intel’s PC133-memory chipset solution i815. Due to the fact that DDR-SDRAM doesn’t have the high latency penalty of RDRAM, we at least don’t expect a performance drop vs. i815, such as we had seen it with i820 and i840 before.
Benchmark Setup
Hardware | |
CPU | Intel Pentium III Socket370 |
Motherboards | Gigabyte GA-6RX, Rev. 0.2, BIOS 6RX P4 12/27/2000
Asus CUSL2, Rev. 1.02, BIOS 1005 beta 7 |
Memory | 128 MB Micron/Crucial PC2100 222, Setting: CL2, 133 MHz, 4-way Interleave
128 MB Wichmann WorkX PC133 CL2, Settings 2-2-2-5/7 |
Graphics Card | NVIDIA GeForce 2 Ultra (64 MB), Driver 6.67 Win98/2k |
Hard Drive | IBM DTLA-307030 30 GB ATA100 |
Software and Settings | |
Operating Systems | Windows 98 SE, Version 4.10.2222 A Windows 2000, Version 5.00.2195, SP1 |
Screen Refresh Rate and Resolution | 1024x768x16x85 |
DirectX Version | 8.0a |
Quake 3 Arena | Retail VersionSetting Normal, 640x480x16 bit color, no sound |
FlasK Settings | Video Codec: DivX 3.11 alpha, Fast-Motion, keyframe every 10 seconds, compression 100, data rate 910 kbps Audio Codec: audio not processed Video Resolution: 720×480, 29.97 fps, interlaced Resizing: Nearest Neighbor |
Webmark2001 | MSIE 5.01 Timer Resolution 10 ms Java enabled, 5.0 JVM 5.0.0.3310 JavaScript 1.3 Adobe Acrobat Reader 4.05 Cult3D 5.0.1.43 Flash 5.0.30.0 Microsoft Agent 2.00.0.3422 RealPlayer 6.0.9.357 Quicktime 4.1.1 NetMeeting 3.01 Windows Media Player 6.4.09.1109 Windows Media Services 4.1.00.3918 |
ViewPerf | Version 6.1.2 1280x1024x16 |
Office Application Performance
I ran this test about 9 times and the result was that both platforms perform identically. In the disk-intensive office applications that don’t really care too much about memory bandwidth the Apollo Pro 266 is just as fast or as slow as i815.
Internet Performance
BAPCo’s Webmark2000 managed to show at least a tiny little difference between i815 and Apollo Pro 266, but it’s hardly worth talking about.
3D Gaming Performance
The picture is pretty much the same once more. Apollo Pro 266 and i815 are almost on par, with only a tiny lead of the DDR-solution.
MPEG4 Encoding Performance
FlasK MPEG shows that Apollo Pro 266 is able to score even slightly worse than i815 although it has the benefit of the greater memory bandwidth. We know that Athlon was able to take advantage of the higher memory bandwidth of DDR in this test, but Pentium III doesn’t gain the slightest bit.
Professional OpenGL Performance
If you think of the 133 MHz FSB platform article from March 2000, you will remember that SPECviewperf was the only benchmark where i820 and i840 were able to outperform the BX chipset, because those benchmarks are indeed able to benefit from increased memory bandwidth. You can see the same thing here. Apollo Pro 266 and its DDR-memory are indeed performing better than i815 and PC133, but the gain is once more very little.
Conclusion
We shouldn’t be too disappointed by the results we saw, although they show that there is hardly any justification for an upgrade from a PC133-platform to a DDR-platform for Pentium III. In fact, there is hardly any sensible justification for the purchase of a Pentium III processor in the first place anymore, because today AMD’s Athlon processors outperform Pentium III at a significantly lower price point. This means that upgrading a Pentium III system on a budget can only lead to the conclusion to switch over to an AMD-platform.
Pentium III was unable to take advantage of RDRAM and thus it hardly gains any performance with DDR-SDRAM as well. Instead of telling you that it would make sense to rather buy Pentium III plus Apollo Pro 266 and DDR memory once it meets the same price points as i815 and PC133 memory, I’d suggest to forget about getting a new Pentium III processor altogether. If you want excellent performance at a good price it’s close to impossible getting around buying an AMD Athlon system.
I think it’s sweet that VIA has released Apollo Pro 266. It shows VIA’s sportsmanship and its feeling for marketing, trying hard to ride the DDR-pony. By all means of sensibility I cannot see any need for this chipset. If I had a bunch of engineers to occupy, I’d rather try to design a chipset with DDR-memory support for a processor that is actually able to take advantage of this memory. The release of Apollo Pro 266 and all the developing dollars that went into it seem a huge waste. Now VIA will have to start the marketing machine to convince innocent customers of its hidden virtues. For me, Apollo Pro 266 joins the line of other pointless Pentium III chipsets, such as i820. I doubt that Intel made an awful lot of money on i820, but some of those chipsets were indeed sold. Maybe VIA can even sell more Apollo Pro 266 chips, because after all DDR-memory isn’t as ridiculously overpriced as RDRAM was in the early days of i820 and i840. It might still be really cool to own a Pentium III system that is equipped with the funky DDR-SDRAM. Still the fact remains, Pentium III ain’t gonna gain nothin’ from it.
If you are the owner of a Pentium III system with a BX, Apollo Pro 133 or i815 platform then let me assure you that VIA’s Apollo Pro 266 is nothing that should make you get your knickers in a twist. Keep your actual system or upgrade to an Athlon. If DDR, then with some sense please!
Hello VIA! What we really need is a DDR-chipset for Pentium 4! This processor is able to show what power is hidden inside DDR-SDRAM. That’s why Intel and Rambus don’t want a DDR-chipset for this processor for the time being. The only alternative is an AMD processor that will be able to make use of high memory bandwidth. However, it doesn’t seem as if AMD will be able to pull that off for the next few quarters. That’s too bad for the ‘Team-DDR’ and it will make Rambus feel a whole lot better.