by Anton Shilov
03/11/2006 | 10:28 AM
Memory business is akin to oil business: the price of memory chips usually fluctuates significantly depending on the number of chips in the market. On the side of the memory chip manufacturer there is no significant difference between the chips, but there is a huge differentiation for enthusiasts trying to obtain every additional bit of performance. The memory modules business is quite risky itself: you always have to balance the amount of chips in stock, but keep in mind that the prices may fall and you will have to get rid of chips at no profit. One of the solutions to add stability to financial situation is to add product lines which price does not fluctuate depending on market conditions: for instance, memory modules for servers or enthusiasts.
OCZ Technology has both product lines these days: it has Titanium for mission critical applications and various series for performance-starving overclockers.
This year OCZ showcases the fastest memory modules on the planet: the DDR2 that is capable of functioning at 1100MHz with 2.1V voltage. According to OCZ’s Michael Schuette, who is technology chief at the company, the maximum for DDR2 should be something like 1200MHz under normal conditions and providing stable operation, which means that the new modules are pretty close to the limit.
OCZ will start selling its 512MB and 1GB 1066MHz and 1100MHz memory modules by the end of the month. Pricing is still not confirmed, but we can expect premium products to have premium price.
Nvidia Corp. has been pretty successful with its multi-GPU SLI technology: consumers are aware of it and even despite of the fact that technically it is better to have one powerful graphics card instead of two lower performance, consumers are looking at SLI not only in the high-end and mainstream, but also in the price-conscious segment. The users of the premium components, however, are now ready to pay for graphics even more than before.
In an attempt to address the market of enthusiasts who desire to have absolutely the best hardware, Nvidia has introduced its quad SLI technology that combines four graphics cards to get ultimate performance and image quality. The technology is much more complex compared to the original SLI for two graphics cards in terms of both software and hardware, which is why Nvidia and its partners not only need to develop a driver, but also to design graphics cards with two chips that can work efficiently and reliably.
Here are the main goals that Nvidia had to achieve while developing its quad SLI:
Nvidia’s drivers are renowned for the quality and stability, so, quad SLI should work fine from the software perspectives, except for some possible issues that are subject to all hardware in the beginning of their existence.
A major problem that is associated with quad SLI is development of actual hardware. It is not a secret that modern graphics processors and graphics memory consume a lot of power and dissipate tremendous amounts of heat. So, designing graphics cards with two chips that should work in tandem should be quite a challenge. The situation further worsens with the fact that Nvidia decided to improve its performance scaling with the help of the special PCI Express x48 bridge installed onto graphics cards with two chips to ensure higher performance. The following diagram should give you the idea of working principles of dual-GPU graphics cards with the x48: the chip is needed for maximally efficient communication between chips and the PCI Express bus.
Two GeForce 7900-series graphics chips, a high-speed bridge and memory have extreme thermals, which makes it pretty complex to cool-down and support such products eventually. While the issue has been resolved, Nvidia still decided to wait with the broad availability of dual-chip graphics cards. Initially only large makers of high-performance computers will offer systems with four graphics processors, which is quite right: you can get a Lamborghini only from a special dealer and you cannot match Lamborghini with a tuned Ford.
As a result of such positioning – for extreme enthusiasts at extreme cost – Nvidia even did not allow its partners to display quad SLI. All partners, except one, do not showcase quad SLI systems at the show! The only company who has the right to demo quad SLI is Foxconn, the company behind names like Leadtek and one of the world’s largest maker of mainboards. In fact, Foxconn will be the only manufacturer to produce boards for quad-SLI systems.
What we also do know is that what is displayed by Foxconn is a combination of two GeForce 7900 GTX and GeForce 7900 GT. It is not known how do graphics cards work with each other, however, it is yet another evidence of how complex the quad SLI is.
AMD’s Socket AM2 processors for desktops are all over the show and it is clear that they will, in fact, be available to customers starting June 6th, however, there are no traces of the Socket F (1207) chips that are intended for servers.
In fact, server space is the one that AMD is very successful in right now: it has 12.7% server market share worldwide, 20.5% of server market share in the
AMD claims that Socket F microprocessors intended for infrastructure that supports dual-channel DDR2 memory and quad-core chips will be available in Q3 this year, nevertheless, right now partners of the company do not display solutions for the new CPUs, which may indicate that the chips will be available in late Q3.
All that we managed to picture regarding the Socket F and AMD’s first land grid array chips are the following models, nothing else could be found.
Last CeBIT a lot of graphics cards makers, including Asustek, Gigabyte, Leadtek and, perhaps, some other showcased their graphics cards with two visual processing units based on variety of GeForce 6-series chips. This year the dual-GPU boards are not really highlighted by Nvidia’s partners probably because they are not in very high demand. The only product that, for instance, Asus offers is very high-end dual-GeForce 7800 GT and the company is already discussing the dual-GeForce 7900-series product. It looks like graphics cards featuring two mainstream chips did not receive popularity among users.
This year GeCube decided to give dual-GPU graphics cards a try: the company demonstrates a graphics card with two Radeon X1600 XT chips on it. Currently the company says that the board which concept is called Gemini is still in development and the final version should be available sometime in late April or early May. In the light of the fact that ATI recently dropped the prices on its Radeon X1600-series chips, GeCube will reconsider its intention to sell dual-GPU Radeon X1600 XT product for $399.
Other graphics cards makers are reluctant to develop dual-chip cards, so, GeCube will have a unique position this time. Whether it will provide the company any benefits or not remains to be seen.
It is not a secret that one of the problems that the memory module makers, who overclock their products to the maximum, have is insufficient supply of power to the memory slots. The manufacturers have tried to improve the situation in various ways: OCZ introduced special device that improves the quality of power for memory modules as well as allows to put higher voltage, whereas other memory module vendors worked with companies like Asustek Computer to improve memory power supply circuitries.
Geil went further! It added special power connector to its memory modules to allow them to operate at up to 3.0V voltage. It is unclear for how long memory modules can survive under these circumstances, but it is evident that with such a high overvoltage it is possible to expect maximum overclockability, provided that you have a high-performance power supply.
Currently the external voltage regulator for memory modules is still a concept and the company does not discuss speed-bins of products featuring the technology.
It is interesting to note that OCZ has already attempted to increase voltage for memory modules with its DDR booster, moreover, the company has patent pending on external power supply for memory modules.