Platform Conference 2002 February 04, 2002 Tuan 'Solace' Nguyen

Introduction

Page:: ( 1 / 8 )

Kicking off 2002

It’s not one of the largest trade shows around, but it’s definitely one of the most informative. Platform Conference, a show where vendors come to exchange information, continues to garner more audience each time it occurs, and always with more interesting information. Like its name suggests, Platform Conference focuses on low-level computing platforms. This is where chipsets, memory, and board manufacturers come to meet. The main sponsors of the show are AMD, VIA, Micron, Infineon, NVIDIA and the HyperTransport Consortium. From that list, you can tell that the primary focus of the conference is either on memory technology and/or AMD technology.

As we went around and browsed through the two-floor conference, we noticed that a few motherboard companies that attended the last PC were noticeably missing from the show. GigaByte, who attended the last conference with a slew of boards to show, was nowhere to be seen this time around and we didn’t see companies like Iwill or Epox anywhere on the show floor.

Other exhibitors included companies like Philips Semiconductor, who produces ICs and timing modules for various board makers. Tundra Semiconductor, who flew in from Montreal, Canada, talked about its bridging technology and the ability to support PCI-X on existing platforms. While there were interesting technologies here and there, the majority of talks were based around memory, chipsets and processor architectures – specifically 64-bit.

Spy Game

While it initially began as a show dedicated to supporting a VIA/AMD platform, there are a lot more interesting technologies that are beginning to surface at Platform that can be applied to many other platforms as well. Because of this, Intel always sends secret agents over to Platform for product snooping. Determined to find out if VIA has any ulterior motives, Intel engineers went around talking about products that don’t actually exist just to see if anyone will accidentally leak out confidential information. While standing at the booth of one board manufacturer, an Intel representative with his badge flipped over to hide his identification asked “are you guys showing the VIA dual P4 solution?”

Rest assured all parties were left baffled by the question. Of course, there aren’t any dual Pentium 4 solutions out there and there likely won’t be. Intel has specific markets and already holds an iron grip on the server market with its Xeon processor so bringing the P4 into the foray only confuses and takes away its own market share. Besides, we’re sure chipset manufacturers will want to avoid getting into huge legal issues with Intel.

Relaxing and kicking butt

If you want to have lots of fun, shows like E3, CeBit and Comdex are probably the ones to attend. Platform is a show reserved for press and vendors. You won’t see the typical end-user at Platform and thus most of the discussion that occurs there remains rather business oriented. Even still, Platform attendees still know how to have some fun!

NVIDIA is sure to always have a cool booth, and this time it was of course showing off Xbox, with the party-favorite DOA3. This game naturally gathers an audience quickly when being played on a 50” wide screen plasma display and before anyone knew it, there was an Intel/AMD melee.

AMD won 5 to 0.

SIDEBAR: This year’s first Platform Conference was noticeably smaller than last year due to the economic downturn.

AMD Hammer

Page:: ( 2 / 8 )

Intel can’t touch this

One of the main topics at Platform this time was AMD’s Hammer core, an 8th generation processor core with two cores, one handling 32-bit instructions and the other one handling 64-bit instructions. What’s unique about AMD’s Hammer core compared to Intel’s IA-64 architecture is that it is able to run legacy 32-bit programs without the use of emulation. We all know that emulation is always slower than hardware processing, so AMD’s two-core design for Hammer will greatly help the move from 32-bit software to 64-bit.

Ever since its conception some years ago, Intel’s IA-64 architecture has been all but widespread. Because of its sharp abandonment for 32-bit support, IA-64 was too much, too soon. Currently, IA-64 is featured in the Itanium processor and is effective in a very small sector of the processor market distribution. Most workstations and servers are currently populated by Intel’s more popular Xeon processor and other processors like Alphas, already at 64-bit.

Understanding the situation that Intel is in, AMD immediately sought to take full advantage of it and thus we have Hammer. But Hammer isn’t just about a hybrid 64/32-bit architecture, it also features a full-fledged DDR memory controller right on the die.

No more north bridges?

When you purchase a motherboard, you’re likely to be concerned about which chipset it uses, how it performs, etc. One of the most important things is what type of memory it uses. This is in fact determined by the north bridge, usually the larger chip in a chipset. The northbridge is large because of two reasons: it’s has to connect with both the processor and the south bridge and it has a memory controller to interface with the system memory.

AMD suggests that processors are currently held back by memory bottlenecks, and in a nutshell that is indeed the case. Remember back when AMD was still using L2 cache exterior to the processor core and how the Athlon’s performance really bit the dust? When AMD moved the L2 cache on die, performance shot through the roof and took back Intel’s brief performance lead with the Coppermine core. AMD’s idea with the memory controller is the same; bring it onto the core.

The Hammer core will support all current DDR technologies but will also support future technologies via new chips. This almost removes the need to upgrade your motherboard in some sense. Featuring a crossbar memory architecture, expect to see Hammer taking full advantage of high speed DDR memory well beyond DDR333. Unbuffered and Registered modules are supported as well as an 8 or 16-byte interface for support of up to 8 DIMM slots.

Before you get worried about VIA, SiS and ALi disappearing, they still have a huge role in chipsets. The memory controller is only part of the northbridge and there’s still the AGP controller as well as the interconnecting bus between it and the south bridge. Northbridges will still be integral to the entire chipset but they just won’t be the link between the processor and memory anymore.

Too legit to quit

After a lot of thought and consideration, Intel is currently deciding on a similar 64-bit/32-bit hybrid design like Hammer. However, the implications of such a quick design turnaround may be tough because Intel is already so far into its production schedule. A processor usually takes a few years of design before it even gets a chance at fabrication and a 64-bit/32-bit solution from Intel may not be viable when it does reach stores, if ever.


SIDEBAR: NVIDIA didn’t show up with booze like it did at the previous Platform Conference. Too bad – NVIDIA knows how to host a party.

Memory Technology

Page:: ( 3 / 8 )

SyncFlash

Being that this is Platform Conference, it’s only natural to have a slew of memory companies at the show. Despite the surprising nature of an appearance by Rambus, it’s good to see diversity, as it’s usually all about DDR. In truth, Rambus does have really good technology; it’s just too bad using its technology costs an arm and a leg. Even so, DDR333 was the talk of the show this time around and while it is indeed fast, it’s not as cool as Micron’s SyncFlash memory.

Micron is developing and perfecting SDRAM (DDR soon to come) modules based on a type of Flash RAM with a standard SDRAM interface. Because it’s not volatile, Flash memory can be used to store information and have that information be retained even when power is switched off. Combine this attribute in standard system memory and you have very fast solid-state storage

Not one to leave without a show, Micron had a demo system with a stick of 32MB SyncFlash RAM in a standard VIA chipset motherboard using an existing chipset with Windows installed right into RAM. There were no hard drives, no CD or floppy drives – everything was executing right off of system memory. Micron provided a reset button on the system to toy around with just for you to see how fast the system boots into Windows. We think it was around 4 or 5 seconds before the system reached the desktop from BIOS load.

The benefits of SyncFlash are obvious and there are many situations where SyncFlash would prove to be very useful. With SyncFlash, you can install programs into system memory and because programs are executed in RAM, there’s no load time and programs simply execute where they’re stored. SyncFlash modules can be installed along with regular modules so no changes are needed to the motherboard physically. Micron says that only minor BIOS tweaks are needed to support SyncFlash.

There are indeed downsides to SyncFlash like its write speed. Since Flash memory isn’t very fast at writing data, you’re likely not going to want to store vast amounts of data onto it; you’ll still need to rely on conventional memory for the short term. SyncFlash is currently limited in capacity by price. It’s much more expensive to produce a 128MB SyncFlash module than it is to produce a 128MB PC2100 module.

Rambus’ future on the PC

Intel, while still supporting RDRAM in Xeon systems, looks like it will begin to let DDR RAM phase out RDRAM on desktop systems. This is a big blowback for Rambus since the desktop market generally speeds adoption of new hardware faster than other markets. DDR, now at 333MHz, is catching up to PC800’s bandwidth. Rambus did demo some PC1060 RDRAM over in its booth but they seemed to be high grade overclocked RDRAM rather than a true fabrication. In comparison, Corsair is now selling what it calls XMS memory, which are DDR modules that use very fast Micron chips which allows FSB overclocking to reach 166MHz.

[image]

Storage and Networks

Page:: ( 4 / 8 )

ATA133, Serial ATA and Big Drives

The only hard drive manufacturer that was at Platform was Maxtor, and with good reason. If you don’t already know, Maxtor acquired Quantum a few months ago and all of Quantum’s products have moved over to the Maxtor brand name leaving DAT and DLT devices in the Quantum name. Looking back at history, Quantum’s timeline has been good overall with a few bumps and bruises. Initially, Quantum produced very high quality hard drives and when it then proceeded to expand its business to backup storage, some of its desktop drives deteriorated in quality. We remember the first time we saw a Quantum Big Foot drive and it wasn’t the prettiest piece of equipment.

Quantum did pioneer a lot of technologies in the IDE field and many of the faster UltraATA interfaces that we all use today stemmed from Quantum roots. In the light of that, Maxtor is now pushing ATA133 as the next step for hard drive speed. However, ATA133 may see some trouble becoming widespread due to lack of support from Intel and the only ATA133 drives available right now are ones from Maxtor.

Serial ATA is on everyone’s waiting list lately as it’s proving to be a promising hard drive interface technology. While we like the fact that Serial ATA drives use very small gauge cables compared to current IDE cables, because it is a point-to-point technology, only one drive can be used per cable per channel. While cables can reach up to a meter long, the fact that you have to have four connectors on the motherboard to use four drives is a little daunting. It’s good that Serial ATA connectors are small for standard drives but if you want Serial ATA drives with hot-swapping capabilities, the resulting connector is fat and almost as wide as a 68-pin SCSI connector.

Shown alongside Serial ATA is Maxtor’s Big Drive technology, where they will address the 137GB limit that exists on current ATA technology. Maxtor claims that drives will be capable of addressing 100,000 times more data than today’s largest ATA hard drives but we doubt drives that large will be spinning anytime soon (and we’re ready to insert foot firmly into mouth).

Gig-E Ethernet

If you thought gaming over a 100Mbit LAN was fast, wait until things start rolling out with gigabit Ethernet. Gig-E has already been around for quite some time but it’s just only recently that it’s beginning to make suggestions of appearing on desktops. Like the name suggests, gig-E is ten times faster than current Fast Ethernet (100Mbit). Available in both copper and optical interfaces, gig-E is the next step in networking.

We saw a few motherboards from various manufacturers that already have gig-E built into them but we’ll reserve commenting on which companies and boards specifically supports gig-E as these boards were shown “under the table”. Come IDF time, we’ll reveal more about these boards and their specific features. But from what we saw, they are the cream of the crop in terms of features.


SIDEBAR: An optical OC192 internet backbone connection has enough bandwidth to send 9.6Gb of data per second. That’s 6400 times faster than a T1 line.

Bus and Bridges

Page:: ( 5 / 8 )

Old and aging

While getting on to the topic of busses, let’s think about the previous page for a second. Maxtor is pushing ATA133, and gig-E is bound to explode onto the desktop anytime, which leaves us wondering where all these high speed devices will converge. Current PCI slots are run at 33MHz and are capable of moving data at 133MB/sec. However, with ATA133, a few bursts of data will instantly saturate the bus, leaving other streams of data clogged in the pipeline. Add in gig-E and we already see the need for higher bandwidth PCI solutions, or get rid of PCI altogether.

PCI does have other flavors to choose from, most noticeably 64-bit PCI as well as PCI-X which runs at a top speed of 133MHz with a bus width of 64-bit. These beefier versions of PCI will need to start appearing on motherboards if we’re going to move to faster storage and faster networking. Besides the hard drive, the current 32-bit PCI bus is probably the other most-limiting factor in any system.

Although we talked about Intel’s 3GIO technology before, it didn’t make an obvious appearance at Platform and instead remained to a few discrete conversations between hardware observers. 3GIO looks to be a very promising bus technology and the benefit is that it’s already compatible with current PCI and HyperTransport technologies. For the near future however, 64-bit PCI and PCI-X still remain the bus of choice for most chipset and board manufacturers.

Bridging the gap

It’s possible for manufacturers to switch to an all 64-bit PCI solution for the desktop right now but the costs would be brutal for the consumer. One of the other problems is that there aren’t any current desktop-level chipsets that support 64-bit PCI slots. Manufacturers wanting to dish motherboards featuring PCI-64 will have to use a supporting chipset. However, it’s not always possible to be totally backwards compatible but with the help of some bridging technology from Tundra Semiconductor, we may be able to see some desktop boards with 64-bit PCI or even PCI-X.

Tundra’s bus bridges allow 64-bit PCI only systems to be compatible with 32-bit PCI cards by using a bridge chip. The same can be said about 32-bit PCI only systems that want to support the use of 64-bit PCI or PCI-X cards. With the bridge chip, a PCI-X card for example, can be used in an existing 32-bit PCI solution without fuss. However, since the main bus is still only 32-bit PCI, the PCI-X card will effectively be only a 32-bit PCI card. But it’s not all bad. For solutions like gig-E that’s only featured on 64-bit PCI cards, having bridges on a motherboard can greatly reduce the costs for someone looking for an inexpensive PCI-64 or PCI-X capable solution.

SIDEBAR: Tuan used to live in Toronto, Canada, home to ATI.

Chipset

Page:: ( 6 / 8 )

VIA

VIA didn’t expose too many details on its next generation chipset for the Athlon platform, but we do know it as KT333. From the documents we’ve read, KT333 will fully support DDR333 memory. Also supported is ATA133 as well as upcoming Athlon XP processors. For the past while now, VIA’s consecutive releases of the KT series has primarily been of speed and memory support but no real changes to the chipset’s architecture. Besides retaining the KT tag, VIA plans to introduce a faster system interconnect between the north and south bridge on KT333. It seems the existing V-Link technology is no longer providing enough bandwidth to satisfy VIA. This change can only mean good things as we’re curious to know if VIA plans to support 64-bit PCI in its chipsets. Provided that KT333 will have a superior chipset interconnect to VIA’s previous solutions, we wouldn’t be surprised if 64-bit PCI shows up on VIA’s roadmaps.

KT333 will be in direct competition with SiS 7xx chipset which will also be a DDR333 chipset for the Athlon. If SiS keeps up the good work it has done for the 645 chipset, we’ll be talking a lot more about them in the near future. As of currently there’s not much talk about what SiS is doing with its 7 series chipset but it’s safe to say that it will be a very competitive solution. We’re hoping SiS will attend future Platform events since it is a big supporter of DDR.

On the Pentium 4 side of things, VIA plans to launch P4X333 sometime late this year with support for an ultra fast version of V-Link, 533MHz FSB, and AGP8X (also called AGP3.0) support.

ServerWorks

ServerWorks joined the party too, revealing information on its upcoming Grand Champion LE and GE chipsets. Both chipsets support DDR RAM and are similar in all features except the number of processors they can use.

GC-LE specifications

• Dual processor support
  
• 400MHz/500MHz FSB support
  
• DDR200/266
  
• 16GB memory capacity
  
• Up to 2 PCI-X segments

• Quad processor support
  
• 24/32GB memory capacity

Motherboards

Page:: ( 7 / 8 )

AMD 760MPX troubles

There were three board manufacturers at Platform displaying dual Athlon, 760MPX motherboards: Asus, MSI and Tyan. All three companies unfortunately suffer from the same problem – lack of onboard USB support. Right at the end of production schedules for these companies, AMD halts the assembly line and declares that USB support in its 760MPX chipset is shot. This is definitely a show stopper and doesn’t look good for a company like AMD that’s trying to make a positive impression for server chipsets.

All three manufacturers we spoke to are including PCI USB cards with their boards but Tyan is the only company including a USB1.1 card while the others are including 2.0 cards. AMD assures that a new stepping for its 760MPX will arrive within a few weeks but this also means trouble for board manufacturers who want to sell boards now. Customers will feel ripped off and cheated and it simply looks bad for everyone when the emails start flowing in with complaints.

From our brief time with the boards at Platform, all three companies had impressive solutions. However, one manufacturer has a few extra toys to play with but we’ll keep quiet on 760MPX for our upcoming roundup.

VIA’s Eden platform

There were other boards to look at but most of them have been on the market for some time now and just didn’t seem that interesting anymore. What did catch our eyes was VIA’s new Eden system, designed to be extremely compact and fuel efficient. Using a low power VIA chipset and C3 processor, Eden is able to run in a tight form factor (it looked like FlexATX) without requiring an active heatsink. Our take on Eden is that it won’t take off too far off the ground before falling back down. There have been concept systems that strive for small box sizes but in the end they don’t do much but remain novelty items. VIA’s C3 processor isn’t going to score points in the latest Onion benchmarks either. Eden will likely be reduced to small tablet devices like Transmeta’s Crusoe technology.

Taking a tour

Not much else was interesting as we walked around from booth to booth. Most of the motherboards on display were ones that we’ve already talked and heard all too much about already. Taking a walk around from booth to booth we noticed that some motherboard manufacturers were using the same computer systems, with the same cases and even the same mods. We quickly found out that these systems were built by Corsair – yes, the memory company.

Called Cobra, Corsair is building barebones systems that feature very high overclocking as well as an aluminum case decked with a laser-cut Corsair logo on the side panel and a blue cold cathode blue tube on the inside. There are two different Cobra models differentiated by video card (the lower model doesn’t come with one) and processor. Included in the high-end model is an Asus V8200 Deluxe GeForce3 Titanium 500 that features support for video in as well as 3D shutter glasses. We’ll be taking a look at what Corsair has done with its Cobra to differentiate itself from rest of the crowd.


SIDEBAR: Speaking of NEC, Tuan still plays on his TurboGrafx16 – still considered by many to be one of the best systems. At the time when Nintendo had its Nintendo Entertainment System, NEC’s TurboGrafx16 was already at Super NES levels.

Conclusion

Page:: ( 8 / 8 )

The end of another show

Overall, Platform was a decent show with a few interesting technology tracks to attend and listen to but it doesn’t offer much compared to other shows like IDF (Intel Developer Forum), CeBit or Comdex. The difference between Platform and the others is target audience, and Platform remains a show for vendors to meet with vendors.

We managed to take a look at some promising technology like Micron’s SyncFlash and AMD’s Hammer architecture. All the other technologies revolving around AMD and their support for Hammer will likely provide a solid platform for AMD to launch at the end of this year or early next. Hopefully AMD will get its production fixed so that issues like 760MPX USB won’t appear in Hammer or any other AMD product. It would be a shame if there were memory controller bugs inside Hammer when it debuts.

DDR memory continues to climb the performance ladder steadily with the release of DDR333. The next logical step is DDR400 but we won’t be seeing that until AMD announces new Athlon cores, which are due out in a few months. Whether they will be SMP capable or not remains to be seen but AMD will likely phase out Palomino by the end of the year so expect to see DDR400 sometime in early 2003.

Alongside chipsets and memory technology, storage technology is also advancing steadily. Maxtor is pushing hard on its Big Drives and Fast Drives technology but the lack of ATA133 from Intel does hurt. AMD hasn’t announced products that support ATA133 yet either and we haven’t seen products that do on its roadmaps. Hopefully Maxtor is willing to make ATA133 more open to public adoption than keeping it for itself.

Things are only beginning to get interesting and will really heat up when IDF hits. There are a lot of high-end technologies trickling down the pipe onto the desktop level that were only previously seen on servers. Thankfully the trend has been the same for some time. With the rapid growth in technology, we’re seeing newer technology arrive faster and faster each year as we advance on a curve rather than a straight line. While there are a few technologies that we’re still not quite so sure of, they’re all worth keeping an eye on this year as the computer industry begins to make its recovery. If Platform is a small indication of things to come, we can’t wait for IDF.


SIDEBAR: What technologies are you looking forward to most? Will your next system be running PCI-X, SerialATA, SyncFlash, Hammer, or all of the above?

Let us know in our News Comments!