   |
|||
|
|||
|
:
News
:
Archives
|
|
NVIDIA SLI Interview |
|
|
A Few Words from Chris Daniel |
|||
|
|
|||
|
by Josh Walrath |
|||
|
Doesn’t PCI-E provide enough bandwidth and low enough latency to render this somewhat redundant? The SLI connector is critical to achieving the 2X performance gains proven by SLI technology. We looked at implementing SLI without the over-the-top connector, but we decided against it. Without the SLI connector you will have ~1GB/sec more traffic going over the PCI Express Bus, which will slow performance. The faster you render, the more data you will need to transfer over the bus, leading to a self-choking pattern. When Unreal 3 and Quake 4 hit, this issue will only be exacerbated. The SLI connector provides perfect digital compositing of signals, asynchronous with the rendering rate, without consuming bandwidth over the PCI Express Bus. After talking to a couple of motherboard manufacturers, many of them have said the same thing, “Bring me more lanes!” Currently the nForce 4 only supports 20 PCI-E lanes. While the nForce Professional 2200 and 2050 are multi-chip products that provide plenty of lanes when used together, they are not aimed at the desktop space. Is there any technical reason why more than 20 lanes can’t be implemented on a desktop level chipset? There are no technological challenges to implementing more lanes, even though current core logic and motherboard implementations have a limited # of lanes. Is it possible to use SLI with configurations such as 10 or 12 lanes for each video card? The PCI Express standard currently supports x1, x2, x4, x8, x12, and x16 lane widths. Therefore, a 12 lane configuration would be possible with SLI. Do you see any advantages/disadvantages to enabling the use of two fairly separate cards in a multi-card solution? What kind of potential issues would such a pairing entail? NVIDIA currently uses symmetric rendering to ensure maximum scaling. In a scenario in which two different GPUs (different pipes and clock speeds) would be used together, the faster card would be ‘dumbed down’ to the slower card’s configuration. While this may sound like a flexible solution, in a typical case this would lead to little to no performance improvement, as compared to the single GPU performance of the faster card. SLI is being touted as a “buy one now, get a second one much later for cheap” upgrade plan. Let’s say that I buy a SLI motherboard with a 6800 GT today, and in six months I want to buy another one. What are the chances that I can buy an identical model? How do other factors such as BIOS revision affect compatibility and overall performance? With the graphics industry moving forward as fast as it is, and manufacturers updating and upgrading their SKUs, how will these factors affect the true upgradeability of this setup? A common misconception about SLI is that the BIOS revision has to be identical between the cards. This is not true. A manufacturer can rev the BIOS without affecting SLI. However, the two cards do currently need to be identical models from the same manufacturer. With the majority of SKUs lasting on shelves for over a year, often much longer, SLI offers a strong path for upgradeability. We do however suggest that you check with the manufacturer about each particular SKU that you are considering. Does SLI support Windows XP 64-Bit and Quadro drivers? SLI support in these drivers is coming soon.
If you have found this article interesting or a great help, please donate to this site.
Copyright 1999-2005 PenStar Systems, LLC. |
|||
