Introduction
I’ve just returned from Taipei and would now like to supply you with all the information I found out in the numerous meetings with Taiwanese hardware manufacturers. Visiting Computex in Taipei each year has nowadays almost become a traditional event for Tom’s Hardware Guide and it has always been an extremely informative as well as enjoyable experience. The Chinese hospitality plus the lovely Asian charm is something I would never want to miss.
Taipei at daytime
Taipei in the evening
Thus I would like to thank everyone who helped making my stay in Taipei as enjoyable and successful as it was. I would like to particularly thank Annie Chao and Michael Deprenda, who were responsible for the scheduling of my appointments and who took care of me for leisure purposes (as few as there were) as well.
Michael and Annie from Tom’s Hardware Guide Taiwan.
Platform News A – Intel 820 or ‘Camino’
There is a lot of stuff going on in the PC-platform area right now. Intel’s 820, also known under the name ‘Camino’, is slowly but surely coming along. I was able to collect a lot of interesting data about Camino from almost each motherboard maker. Most motherboard manufacturers seem to still have problems running the RDRAM at 400/800 MHz. Thus there were a lot of complaints about Camino in terms of stability. The high frequencies of Camino’s RDRAM-interface seem to be a serious design hurdle for many motherboard makers. Very few motherboard makers could proudly say that their Camino-board was running stable at 400/800 MHz, and most complained that so far they couldn’t find any performance advantage over BX. Everyone was complaining about the high price of RDRAM, which is why many expect Camino to become a failure. Intel has changed the design guide once again only a few days ago, coming up with a new solution to support SDRAM. Most motherboard makers understand that the old solution, which uses a riser card known as a RIMM/DIMM-converter, will yield an expensive and extremely low performing solution. Intel is now adding an extra MTH for SDRAM support into the Camino design guide, which is supposed to provide two DIMM (for SDRAM) and two RIMM (for RDRAM) slots per board. At this time it is not yet clear if the MTH will support PC133 or only PC100 SDRAM.
All of these issues are showing that there’s quite a bit of chaos going on about Camino. It’s either not running stable, or it’s not performing great. Additionally to that, the motherboard makers have to change their Camino board design for the fourth time now, if they want to include the MTH for SDRAM support. I personally don’t doubt that Camino will provide a stable platform once it’s released in September. The memory issue of Camino is highly annoying and I agree with 99% of the industry, despising the closed RDRAM-design and the resulting royalty-payments to Rambus and Intel. However, expectations of an extremely high RDRAM-price might be wrong, and I doubt that Camino will be any less successful than the previous BX-chipset, should the platform prove affordable. BX also offered only little performance advantage over its predecessor 440LX. Only for the record, Camino will also support ATA-66 and most importantly AGP 4x, so it’s not quite as crappy as some may think.
Platform News B – AMD K7
Information about AMD’s K7 and especially K7-platforms was not quite as easy to come by, but that’s not due to any non-disclosure issues, but rather due to the way AMD is planning to release K7 and its platform this month. The majority of you may expect that there will be a wide variety of K7-platforms available at the launch date of K7. This is what we are used to from Intel. When Intel releases a new CPU plus a new platform, you can buy this very CPU as well as the according motherboard all over the place, the latest at launch day. The story has always been a bit different with AMD. When AMD releases a CPU, it’s usually only available to OEMs at launch date and even for those OEMs there are only small numbers available initially. The retail market has to wait for quite some time, particularly the fastest AMD CPUs are only available in very small quantities. With K7 the story is becoming even more complicated.
Once K7 is released you will again have to face the fact that you won’t find a whole lot of K7-processors in the retail market, since the majority will ship to OEMs and turn up in your shop inside complete systems. For the few K7-CPUs from retail, you will need a motherboard though, and this is where AMD’s second problem kicks in. AMD’s new K7-chipset is available in pretty small quantities. This is why AMD chose only a few motherboard makers for the initial K7-platform launch. Asus, Gigabyte, FIC and MSI are the only motherboard makers that are able to supply K7-platforms at K7-launch. All the other board makers will have to wait until a third party chipset maker (e.g. VIA, ALi, SiS) will supply them with a K7-chipset, which is not supposed to be before September 1999. One of the ‘left-out’ board makers, a very well known one actually, seems to be particularly upset with AMD and told me that ” for us K7 is still vaporware …”
So far about logistical problems of K7, let’s talk about the performance. AMD thanked me for not trying to get a pre-release K7 from some motherboard vendor or anybody else, because the performance of those early samples doesn’t seem to have much to do with the latest version of K7. What I can tell you right now, before I will have to go under NDA for the evaluation system is that: K7 Revision C seems to be a clear winner. Someone who is supposed to know told me the following, “I’ll give you 100 bucks for each benchmark or application that runs faster on a PIII than on a K7 at the same clock speed.” I heard very similar comments from motherboard vendors at Computex. K7 will most likely be released as 500, 550 and 600 MHz version and that should be enough to give it a clear lead for the time being, until Intel launches its first Coppermine PIII processors. After that it will depend on who is going to make the clock speed race, Coppermine or K7. It will be a close race, where Intel has got a huge advantage when it comes to availability though. So far, K7 seems to have a slight edge in integer performance, floating point performance seems to be some 20-30 % higher than PIII’s though. This will turn around the past, when Intel was the FPU-leader in the x86-world. The K7-platform based on the AMD-chipset uses a 100 MHz front side bus, working at ‘2x’. This means that data can be delivered at the rising and the falling edge of the clock, making the FSB perform as if it runs at 200 MHz. The current memory seems to be PC100 SDRAM only, but that shouldn’t bother us too much now. Memory speed is still hardly an issue when it comes to overall performance, only some more exotic software can really take advantage of a higher memory bandwidth, the CPU-performance is much more important. The AMD-chipset seems to also lack AGP 4x and ATA-66 support. Both will probably included into the third party vendor chipsets though and thus be available once Camino ships and proper AGP 4x-products become available. All in all it doesn’t look bad at all for AMD and K7, let’s only hope that AMD will be able to deliver product.
Platform News C – PC133
VIA is the first chipset maker to make a chipset available that takes advantage of the new PC133 SDRAM. However, before I’ll get into this I’d like to make a comment why the above picture proves VIA’s courage. Right now, PC133 will not be available to anyone officially, because no chipset maker supplying a Slot1-chipset with PC133-support is allowed to advertise this product. The reason is quite simple. If VIA or ALi want to supply their version of a Slot1-chipset, they’ve got to use the Slot1-bus protocol for the communication between an Intel Slot1-CPU and the chipset. Intel licensed this bus protocol, known as `GTL+’, a long time ago. Thus the chipset vendor has to ask Intel for permission to use this protocol and has to pay a royalty to Intel for using GTL+ as well. Intel doesn’t like if anyone runs its current Slot1-CPUs at a front side bus of 133 MHz, which would be pretty easy with a PC133-chipset, without taking the risk of overclocking the AGP and thus jeopardizing the life of the graphics card. Therefore Intel told VIA as well as ALi that they won’t get the license to use their bus protocol and other patents, unless they’ll wait with any official PC133 announcement until Intel released its new CPUs (Coppermine) running at 133 MHz FSB officially. ALi was very quiet about their PC133, VIA didn’t seem to be too bothered though, as you can see when you look at this beautiful booth announcement at Computex. Still VIA does only ship the ‘693’-north bridge right now, which only supports 100 MHz FSB officially. The ‘693(A)’ will be the official PC133 chipset, although it’s exactly the same as 693. Of course you’ll be able to run 693 at 133 MHz FSB in any motherboard that uses it, but that’s only the motherboard makers ‘fault’, VIA advises against it officially, although with very little effort of course.
As already said, ALi tried to keep their PC133 chipset a lot more quiet, nevertheless they’ve got one as well, as you can see when you look at the reference micro-ATX board above. The ‘name’ of this new chipset is ‘Aladdin Pro IV’ or simply ‘M1641/M1535’. Whilst VIA is fighting PC133 performance issues with its 693-chipset, as you already know from my article ‘PC133-A First Look’, will we have to wait for ALi’s Aladdin Pro IV until August 1999. Let’s hope that its performance is going to be better than what we’ve seen of PC133 so far.
Platform News D – Problems with TNT2 and ALi’s Aladdin V Socket 7 Chipset Finally Sorted Out?
So far Nvidia’s TNT2 3D chip doesn’t run on any Socket7 board that’s based on the Aladdin V chipset. I spoke to Ali about this issue and they admit that it is a bug in their chipset, colliding with a special AGP-transfer Nvidia is using in the TNT2-drivers. The problem was finally located last Friday and now ALi and Nvidia have been working on it. I haven’t had the chance to see if the problem has been sorted out with the latest driver release from Nvidia (0188), but both companies seem confident that the problem will finally be solved very soon.
Platform News E – Dual Celeron PPGA Platforms
At Computex99 two motherboard makers offered a really nice solution for people who want to take advantage of the fact that Intel’s Celeron processors for Socket370 are actually able to run in dual-CPU configuration by using a small trick. The still existing pin for ‘GTL+ Input’ (pin AN15) needs to be connected to ‘GTL+ I/O’ and both need to be pulled up to 1.5V. This sounds too technical for most of you, but Abit and QDI came up with a solution that doesn’t require any drilling or soldering. As a matter of fact you can just plug in two Celeron CPU in two Socket370s and you can take advantage of SMP. Quake3 will support SMP (multi processing), so that the two solutions may not be that uninteresting to many of you after all.
This is the BP6, Abit’s solution for a dual-Celeron platform. It’s a lot cheaper than the usage of two special Slot1/Socket370-converter boards with a dual-CPU BX-motherboard, and it’s a lot easier to use as well. However, this board is a bad investment once Intel removes the dual-CPU ability from Celeron for S370 completely. The question is if Intel can’t just stop bonding AN15 when packaging the chip, or if AN15 is required for other operations as well. Anyway, once Intel gets really annoyed about people using dual-Celeron systems, they will make sure that Celeron doesn’t do SMP anymore for good and then Abit will have serious problems selling the BP6.
QDI’s solution is called ‘TwinMagic’ and seems even cooler to me. So far you cannot purchase the dual Socket370 card for Slot1 on its own, you have to get the BX-motherboard with it. This makes sense though, because not every single BX-motherboard supports dual-CPU operations, dual operation requires a special ASIC on the board.
We will see which solution proves to be better. Abit has the advantage of providing less sensitive connectors, the CPUs are directly plugged into the board. On the other hand, the QDI-solution is less wasteful. The single BX-motherboard can still be used for any current Intel P6-CPU, either by directly plugging it in to the Slot1 or via a converter card. If Intel removes the AN15-bonding wire from the Celeron PPGA CPUs it doesn’t become quite as pointless as the Abit BP6. Anyway, I’d like to congratulate each of the two board makers for their innovative idea, which won’t make them a lot of friends within Intel though.
Platform News F – SiS 630 Chipset, the Intel 810-Killer?
SiS is lately not playing a major role in the performance platform area anymore, instead of this they are concentrating on integration. The SiS620 has already been able to show that SiS is able to produce an interesting integrated chipset solution for the low cost market, and now the SiS630-chipset is supposed to be a very strong competitor to Intel’s 810 chipset. You will be surprised about the features of SiS630 once you had a look at the block diagram below. The additional remarks are mine, sorry if they disturb you or if you can’t read them.
After the presentation at the SiS-booth, I could only ask if this thing is also able to make coffee, since it seems capable to do almost anything else. The integrated graphic is based on the 128-bit and partly dual-pipelined architecture of SiS300. SiS620 can take full advantage of the 128-bit graphics memory interface when you add up to 8 MB frame buffer cache onboard. Then this FBC-memory holds the upper half of the frame buffer, the lower half is found within main memory. Thus the graphics controller can work with up to 16 MB frame buffer, which enables highest 2D resolutions and color depths and allows almost the same in 3D. The memory interface is ready to run with PC133-memory and the FSB of the CPU can also be 66/100/133 MHz. The display output can be upgraded to a second CRT/flat panel/TV via analog or digital interface by adding a SiS301 chip to the board. SiS620 supports 4 PCI bus masters and 4 USB connectors, which equals two hubs, not only one. The keyboard controller is integrated and so is a full 10/100 Mbps Ethernet controller, a Home PNA controller, an AC97 interface, a hardware wave table and more ….
I guess the only thing that chipset is really missing is a built in CPU. Combined with that it would indeed be the first system on a chip. Anyway, depending on the promising looking 3D-performance of SiS630, it could be a very interesting competitor to Intel’s 810 and 810e chipset indeed.
Epilogue
Well, the article has become really really long indeed, although I only reported facts as briefly as possible. This is why I will post a second part by tomorrow, which will still include some platform news, but also news about 3D and some more …
Wait until tomorrow to find out what Asus Vice President of R&D/Graphics Division H.C. Hung, the father of V3800, had to tell Tom.