Theoretical Performance

Without discussing actual performance figures (those will be published in part 2 of this article) we can make some assumptions about the performance of VC-SDRAM.

In the event that a single memory master is requesting data from main memory, VC-SDRAM could potentially be slower than regular SDRAM since the memory master now has to go through a Virtual Channel to access the memory it would normally have direct access to. This performance hit should be negligible because of the high-speed nature of the Virtual Channels.

For regular business applications the performance improvement should be negligible.

The benefit of VCM architecture, specifically VC-SDRAM over regular SDRAM, will be seen when multiple memory masters are attempting to concurrently access the system memory. In this case overall system performance will be increased, and depending on how intense the memory accesses are the performance increase could be significant.

An application where a performance increase would be seen is in a 3D gaming environment where the CPU, AGP card, and sound card are all trying to access system memory at the same time. Applications that are memory streaming intensive or AGP texturing intensive will experience a performance benefit courtesy of VC-SDRAM.

Implementing VC-SDRAM

Unlike DDR-SDRAM, VC-SDRAM is pin compatible with SDRAM and is completely transparent to software. It does require chipset support which is provided for by both the Apollo Pro 133 and the Apollo Pro 133A, which is why VIA is pushing the standard. Micron has already begun shipping systems based on the Apollo Pro 133A using VC-SDRAM and they are currently the only major manufacturer to be shipping the new Pentium III 600B since both Dell and Gateway are waiting on Intel's i820 release before they can ship a 133MHz FSB system.

It is currently available in frequencies up to 143MHz which means it can be used in PC133 compliant systems, such as those built around the Apollo Pro 133/A. According to NEC there is no additional cost of ownership for VC-SDRAM, but at the same time this means that you can't move your VC-SDRAM with you if you decide to migrate to a platform that doesn't support VC-SDRAM. Also the availability of VC-SDRAM is currently limited, but it is in existence, so depending on the demand it may or may not grow to be a more popular solution.

Final Words

Tomorrow, we'll be taking a look at the performance of the Apollo Pro 133A with VC-SDRAM in various real world applications to see if the performance improvement is actually there. VC-SDRAM + Apollo Pro 133A could become a viable alternative to Intel's RDRAM + i820 move, and a much cheaper one.

Initial claims state that going down the RDRAM path adds up to $300 to the overall cost of a system. Keep in mind that these claims are coming from Micron, a company that is supporting their decision to go for VC-SDRAM + the Apollo Pro 133A, but one thing is for sure -- RDRAM won't be cheap once it hits.

Unless DDR-SDRAM becomes a quick reality, RDRAM will probably be the trend Intel pushes the market towards and, although it won't be cheap at first, it will most definitely become more affordable. But what solution do you have between the time that it isn't affordable and the time that it becomes so? The Apollo Pro 133A is an option. However, wait until tomorrow before deciding on whether or not the real world performance supports or refutes that argument.

Memory Masters (cont)
Comments Locked

0 Comments

View All Comments

Log in

Don't have an account? Sign up now