Skip to main content

Future Promise for Graphics: PCI Express


In consideration of our theoretical forecasts and all the question marks we've raised, it's hard to reach a conclusion. On the one hand, the new bus opens the door for many new possibilities in the world of graphics cards. On the other hand, it will take some time before the casual user will see any advantages in the form of improved performance. 3D games aren't designed to utilize the greater transfer rates yet, and even if they did, the question remains how much of an impact the greater bandwidth of PCI Express will really have. Sure, more and more calculations are being offloaded from the CPU to the graphics processor, but the CPU and memory would still need to (be able to) generate 4 GB per second of data first and then transfer it via the bus. The same holds true for video upstreams - the system memory will be hard-pressed to keep up. Even the newest hard disks can only write an average of 50 MB/s, a data rate that even PCI writes could handle with ease. And as far as real-time video editing is concerned - the CPU will still need to make a few calculations every now and then - and again the memory will probably prove to be the bottleneck. Continuous data transfers at 4 GB/s look set to remain a fantasy for a while to come.

When PCI Express launches in the expected April or May timeframe, NVIDIA will be able to present a complete product family for the new bus, thanks to its HSI bridge chip. While these cards will not yet be able to take advantage of all of the improvements PCI Express offers, they should - theoretically - enjoy a certain (theoretical) performance advantage over existing AGP cards.

However, this will not be enough for the absolute all-out enthusiast / early adopter who will upgrade as soon as the new gear becomes available. Then again, this crowd will probably wait for the next generation chips from NVIDIA and ATi anyway, since they are already just around the corner. NVIDIA's main business for its PCX line will likely be smaller cards. It will also likely be confined to the OEM market. In this performance category (PCX 4300 = GF4 MX; PCX 5300 = GF FX 5200), a faster bus won't yield an advantage anyway. Lastly, it remains to be seen how high a premium NVIDIA will charge for the privilege of being part of the PCI Express hype.

Things won't get interesting until this summer, when ATi and NVIDIA introduce their new processors with native PCI Express support. Then we'll finally be able to gauge how big the PCI Express performance boost really is in practice. Unfortunately, there is the distinct possibility that the AGP versions using bridges will come with lower clockspeeds so as not to get in the way of the PCI Express hype. That will make comparing the two versions of the cards all that much harder.

The outcome, in many ways, is as unpredictable as the weather. PCI Express has a lot of potential, and not just because of its higher bandwidth. Bandwidth is no argument today - as our comparison between APG 4x and 8x prove. Whether graphics cards will benefit from this and how attractive PCI Express is for graphics usage largely depends on the motherboard and software developers. Updating only for the sake of higher bus bandwidth won't be incentive enough in most cases. What's needed are things like multi-slot solutions etc...think SLI!