   |
|||
|
|||
|
:
News
: Archives
|
|
NVIDIA's SLI |
|
|
Not Cheap... Not Slow |
|||
|
|
|||
|
by Josh Walrath |
|||
|
Split Frame Rendering is in many ways more interesting, but possibly more problematic than AFR. Each frame is split between the two cards, and the driver and hardware in the nForce 4 SLI chipset calculates the workload for each card and splits the screen as needed. For titles that show more shader bound performance in the lower half of the scene (as most scenes in 3D apps show only sky at the top), the driver will split the frame so each card has as close of a rendering workload as possible. Titles such as Far Cry utilize SFR, and when the SLI load graph is enabled the user often sees a 40/60 split between the bottom and top respectively. The advantage to this is that loads are more evenly balanced than AFR in certain situations, but the main disadvantage is in geometry scaling. Again, for each scene, each card is given the same geometry data, and that data is only utilized in the area that each card renders. So, while there is a small geometry boost overall, it is not nearly as large as with AFR.
The penultimate SLI setup? Note that this is a four card SLI rig, but the Voodoo 2's don't count for much anymore. Note the over-the-top connectors though. Currently there are no plans to offer a “digital scanline” type output, which would mirror what 3dfx did with their Voodoo 5 series of cards. I am unsure if it can be supported by the hardware, but with so many 3dfx engineers now working at NVIDIA, it wouldn’t surprise me if it supported such an output. Still, between SFR and AFR/AFR2, there might not really be much of a need for such an algorithm. Getting SLI There are currently two ways of getting a SLI enabled computer. The first is to buy it from a manufacturer like Alienware, Falcon Northwest, or Monarch. This will guarantee that the user will have a computer that was designed from the ground up to run SLI without a hitch, as well as getting their respective warranties. The only problem with this is that the user will have fewer customization options for their system. The second, and more interesting way, is to spec out the parts and build a system from scratch. This allows the user to customize their system to their heart’s content. It also allows them a lot more flexibility in terms of overall price. Pre-built computers regularly have a higher overall price than a home built system using similar parts. The main thing to remember here is that a home built PC will have to be supported by the person who built it, and returns and problems are often easier for the end user to handle when using a pre-built system. Building From Scratch My initial impressions of building my own SLI machine was that it was easy, and I didn’t know what NVIDIA was so worried about. This of course came back to bite me. Choosing components should not be done on price alone! Especially when dealing with SLI!
Note the 2nd GT board is actually touching the fan on the MSI K8N board. Luckily the fan is not affected by the card sitting on top of the housing. I was sent two 6800 GT’s directly from NVIDIA, but I purchased a MSI K8N Neo 4 Platinum/SLI and an Antec NeoPower 480. I thought that though NVIDIA states that at least a 500 watt power supply should be used, Antec had a history of overbuilding their supplies, so I wasn’t too worried. I used a full tower Antec case with plenty of cooling, and I had the AMD Athlon 64 3800+ Venice and 1 GB of Corsair XMS 3200 XL. I thought I was set! I was wrong. With the two 6800 GT’s pulling more than 80 watts a piece, and eating up slightly less than 30 amps, it was pushing the Antec 480 watt supply more than the part was designed for. Add to that the rest of the system, it is no wonder that issues started to pop up. The first three weeks were uneventful in testing, and I started to get a great handle on the setup and its performance characteristics. Then I started to have trouble. Applications that used to not crash suddenly became unstable in SLI mode. I then noticed that the temperatures on both of the cards started to approach the mid-90’s (C) when running a 3D application. At first I thought it might be too hot in the lab, but that was not the case. I then thought that the thermal interfaces between the chips and the heatsinks somehow had broken down, so I replaced this with Arctic Silver. I continued to get higher temperatures, and then the artifacts started to show up. The boards were not even overclocked, and the temperatures were much higher than I had seen with 6800 GT’s. When I first received these cards they were in the mid-70’s when running 3D apps. The artifacts were caused by a heat buildup in the memory, and while the applications didn’t crash, they were unplayable. I contacted NVIDIA about this, and once I started talking to Nick Stam, we finally figured out that my power supply could be the source of the troubles. If the PS is not supplying the correct amount of power, or is being pushed too hard, it can have a very negative effect on the hardware being used. The main two effects that the dying power supply had on the system was that of a thermal runaway with both cards, and actually burning out the memory on one of them. I underestimated the importance of sticking with the specifications that NVIDIA gave for SLI enabled power supplies, and for the average user it would have been an expensive lesson to learn.
If you have found this article interesting or a great help, please donate to this site.
Copyright 1999-2005 PenStar Systems, LLC. |
|||
