TECHNOLOGY - Thanks for the memory

It seems that so far, no one has taken much notice of Intel's new Ram and memory developments, but all that is about to change.

If Intel gets its way, dealers could see a run on workstations andRam and memory developments, but all that is about to change. servers in 1999 as users upgrade processors from Pentium II to the third generation P7. But Intel is also pushing a memory development that should not be forgotten.

In 1999, Intel will ask users to celebrate the end of the 20th century by digging into their pockets to upgrade workstations and servers to the latest microprocessor - the P7. But, while users are used to changes in processors, this time the upgrade will come with a double punch. Intel is using this 'next generation microprocessor' to work alongside a new type of Ram it is backing. Planning has been as public as such technology-led issues can be, but with Ram about as sexy as a cross between Friday the 13th's Jason and TFI Friday's Chris Evans, no one has paid much notice.

As Sander Siezen, Packard Bell product manager, says it has been hard to get the industry to recognise the significance the Ram changes will have. But when it means there could be demand to scrap servers or change motherboards, dealers and users should stand up and take notice. Siezen says increased demand for 3D graphics, multi-media and ATM style networking has put pressure on systems.

So far, the industry's response has been to increase the amount of processing clock speed. The problem is that while processor clock speeds have doubled every 18 months, Ram performance hasn't kept pace and has become the principal bottleneck to system performance.

This was recognised more than two years ago when Pentium PC's memory subsystems were mostly alike: fast-page-mode dynamic Ram (DRam) for main memory static Ram (SRam) for the secondary cache. Speed seekers opted for extended data out (EDO) DRam and pipelined-burst SRam. That all changed when Dell and other vendors moved to even quicker synchronous DRam (SDRam) system memory to cope with the increased demands of a multimedia PC.

Compared with EDO DRam, SDRam was capable of transmitting data on every clock cycle, conferring a peak system performance improvement of 22 per cent. But that increase was insufficient as SDRam was only capable of clock speeds of 100MHz and with processor speeds expected to rocket even further, the techies were burning the midnight oil for a way of getting the Ram to catch up.

Siezen says there are several technologies people are investigating at the moment. Some wanted to push SDRam further, as lab speeds of up to 500MHz had been recorded. Others favoured technology based around embedded DRam. However, Siezen says that while some of these chips might find their way on to shelves, it will be a short-term option as Intel want the Direct Rambus RD-Ram, which goes eight times faster, to match the P7 processor which will be on the market by then.

'Intel is the powerbroker in this situation. It was up to Intel to decide what will be used next because it will have to fit with its motherboards and processors,' he says. Intel's aim is to stabilise the Ram market for five years after Direct Rambus is introduced. This is because changing the architecture to deal with different types of Ram is often more difficult than increasing processor speed.

An Intel representative says the company has taken the existing technology as far as it can. Technology has to move on. If there was no demand for the product we would not produce it, he adds.

Intel has written the PC SDRam specifications for 66 and 100 MHz SDRam, and 100MHz SDRam products will be launched within the next couple of months with high-end machines running them in the fourth quarter. However, Intel will dump the technology once Direct Rambus chips are up and running.

It does not look like SDRam will be saved by Rambus falling behind on its development timetable or not finding backing from the manufacturers.

So far, Intel and Rambus have been on schedule and have a significant number of pledges to make the Ram from manufacturers.

'We set this schedule in 1996 with Intel and our DRam partners,' says Allen Roberts, vice president and general manager of Rambus' Memory and Technology division. 'We've been working in parallel with some of our DRam partners for almost two years to reach this point. During that time, they have been progressing on their proprietary Direct RDRam core designs and we have been releasing elements of our interface design as they have been completed. We're proud to announce we've met this critical program milestone and the Direct Rambus program remains on track' says Roberts.

In September 1997, 18 companies representing the leaders in system-memory implementation products - including memory modules, connectors, clock chips and test systems - announced their intention to support Direct Rambus technology.

They were Fujitsu, Hitachi, Hyundai Electronics Industries, Kingston Technology, LG Semiconductor, Mitsubishi Electric, NEC, Samsung, Smart Modular Technologies, Texas Instruments, Berg Electronics, Cypress Semiconductor, Integrated Circuit Systems, IC Works, International Microcircuits, Advantest, and Teradyne products.

The companies are planning the Ram to be put into PCs, multimedia and graphics, communications systems, digital television and consumer electronics memory.

Hewlett Packard announced in July 1997 its intention to ship semiconductor automatic test equipment capable of testing Direct Rambus devices. Late last year, Rambus announced details of the interface required to run the chip allowing manufacturers to produce products capable of taking the new chips. This was important as it enabled manufacturers to start designing specs around the new Ram.

In February, Rambus handed over the Direct RDRam implementation package, which helps DRam makers integrate the standard Direct Rambus interface design with their proprietary memory cores. The package includes complete schematics, a transistor netlist, layout, behavioural models, process-independent timing analysis, package information, test criteria and test vectors for the interface design as well as a core-interface specification.

Among the list of licensees released by Rambus was Fujitsu, Hitachi, Hyundai Electronics, IBM, LG Semiconductor, Micron Technology, Mitsubishi Electric, NEC, Oki, Samsung, Siemens AG, Texas Instruments, and Toshiba.

This was the last significant link in the development process and at the time, industry watchers believed the most aggressive Direct RDRam partners could have the first silicon for evaluation within a few months.

'We were pleased to see Rambus deliver the Direct RDRam interface design on schedule,' says Peter MacWilliams, Intel Fellow and director of Platform Architecture. 'This is an important milestone to support the transition from today's PC100 SDRam to Direct RDRam beginning in 1999, reinforcing that Intel's main memory roadmap is right on track.'

Ironically, he used the word transition while PC100 SDRam had yet to make its impact on the market.

But the Intel/Rambus strategy has won praise from industry watchers.

'By releasing a standard design simultaneously to all 13 DRam companies, Direct RDRams from suppliers will be similar, eliminating compatibility problems,' says Steve Cullen, senior DRam analyst at InStat. 'This strategy goes a long way to smoothing the transition to this new technology.'

LG Semicon has already released a 64Mbit chip as a second-generation RDRam with a data rate of 700Mbps, about four times higher than that of high-speed SDRam.

The company says the new chip incorporates a new design technology which will be adopted for third-generation RDRam, called Direct Rambus DRam.

LG Semicon said the 64-bit RDRam chip will help it advance the development of Direct Rambus DRam by about a year as it has already secured the key chip design technology.

LG's new chip will initially be used as a graphic memory for workstations, servers, digital TV sets and ATM exchanges which require high-capacity graphic memory chips. It has recently begun mass production of 18Mbit RDRam, seeking to capture a substantial share of the graphic chip segment which accounts for about 25 per cent of the entire memory market.

The new 64Mbit chip can store four times more data than the 18Mbit product.

However, this product would only be a stage one of what Intel and Rambus want, as the Direct Rambus interface can be supported with 64Mbit, 256Mbit and 1Gbit DRam densities. But dealers selling the new product might find selling the new concept less popular than Intel and Rambus would like.

'When the product finally hits the shelf, dealers might find themselves with an unpopular sale on their hands,' warns Siezen.

'Some will have just invested in Pentium II, or PC100 upgrades and will be annoyed at having to scrap that spend money on something new that soon,' he says. While many may be happy with the additional power, particularly if they are installing the latest graphics or video conferencing packages, the cost in the additional stages might be a little high, he continues.

'However, there is no doubt that demand for what this product can do will percolate down from the high-end server market to the workstations quite quickly,' he says.

One distributor, Flashpoint, says the Intel/Rambus strategy was just another way Intel used to maintain its control over the PC market.

Sukh Rayat, Flashpoint's managing director says: 'If competitors are having to constantly adapt to Intel's latest developments they are not having a chance of being a serious threat to them.' Dealers, particularly in the Far East, were losing money because Intel's new developments and upgrades were making their catalogues fall out of date while still sitting on the shelves, he adds.

However, Rayat warns that the P7/Rambus project, following so hard on the heels of the P100 SDRam upgrade, could be the last straw for users who are getting fed up with buying equipment that is going to be out of date within six months. Rayat says: 'People are going to stop buying PCs if the market is not stable. There was a demand for machines to go faster, we have that and users are not using the processing power they have.'

Ideal Hardware product manager Ed Bateman says even if companies like Mitsubishi were planning releases along those lines and time scales, there are other technologies, and it could be that in the end, Intel may not decide Rambus will be the solution that it will go for, although he has not spoken to it recently about that.

He also warns that the industry often moved fast, particularly if a company like Intel, which had a reputation for on-time delivery, was backing a change.

He did not think that when Rambus finally arrived dealers would have difficulty convincing users that they should have it.'The big corporates are less worried about hardware. I think they will be more concerned that their existing software will run on the new Ram,' Bateman says.

However, if Rayat is right, many will use 1999 to dump PCs and go for NCs as a way of stabilising the development cycle.

The Rambus technique

The Direct Rambus RDRam spec was developed by a US company called Rambus with Intel's backing. Standard Direct RDRam devices can achieve 1.6Gbps peak transfer rates from a single chip. But for highest-performing memory systems, 64-bits, they can produce more than 6.4Gbps of performance. A single Direct RDRam device permits four interleaved memory operations and several Direct RDRam devices can support up to eight simultaneous operations.

Direct DRam uses conventional DRam cores, fabrication techniques and packaging.

In addition, Direct Rambus memory modules, called RIMM modules, use standard industry assemblies similar dual in-line memory modules. For PC main memory, a Direct Rambus memory system fits within the same physical, power and thermal profiles of a similarly configured 100MHz synchronous DRam memory system - only it is a lot faster.