Files Shared

Traditional multi-GPU solutions, ones that involve two physical add-in cards, are available from the big two video card developers, NVIDIA and AMD, with each using their own method of implementation. Generally, the two cards split the workload and render frames as a team to provide close to double the performance as a single GPU system.  NVIDIA’s solution is called SLI, which stands for Scalable Link Interface, and AMD’s is called CrossFireX. To use either solution, you need at least two PCI Express x16 slots on your motherboard and a supported northbridge chipset. With this being the case, a multi-GPU setup is something you need to plan for before building a desktop system, it’s not something you can easily fall back on for an upgrade if you don’t already have an SLI/CrossFireX-ready motherboard. For an NVIDIA-based multi-GPU system, you need two cards with identical GPUs. You can use cards from different manufacturers, but if you have a GeForce 9800GT, you need another GeForce 9800GT to go with it. CrossFireX allows for a little flexibility in its pairings, letting you to pair cards that are within the same subseries (e.g. HD 4870 will pair with HD 4850). Unfortunately, these traditional solutions are can be buggy and are not widely supported. That makes traditional multi-GPU solutions an expensive venture for something you might not even be able to fully utilize. Video cards can be expensive enough, buying two or more can be left to the bleeding edge enthusiasts.

Luckily, multi-GPU solutions have evolved to overcome the hindrances of the traditional setup. Each solution offers a Hybrid variant (Hybrid SLI, Hybrid CrossFire) that is designed to incorporate the use of an integrated graphics processor already on the motherboard to handle light-duty graphics tasks along with the much better performance of a GeForce or Radeon GPU. These hybrid solutions are perfect for notebook systems that usually are already using low cost integrated graphics, providing extra graphics processing power when it’s needed. You can find these multi-GPU solutions in place in a number of notebooks already on the market, but, like traditional SLI, these hybrids are not without their drawbacks.

The first drawback is that powering two GPUs at all times when you’re only really using one can drain your battery quickly and the second is that you’ll often need to reboot your system when you enable or disable the extra GPU. This second drawback will probably be overcome by technologies like NVIDIA’s Optimus and Apple’s “automatic graphics switching” found in their current generation of MacBook Pros. These multi-GPU solutions will seamlessly switch between energy-efficient and performance modes without the need to reboot by powering down the unused GPU. This gives you the convenience of a virtually instant performance only when it’s needed, consuming less power, and translating to better battery life for mobile applications. This best-of-both-worlds solution is definitely something I am interested in and I expect to see solutions like this to become widely available in the near future.

If seamless multi-GPU switching doesn’t take over, it will be due to the fact that integrated graphics are gradually being moved from the motherboard’s northbridge chip to the CPU itself. Technically, it can still be a multi-GPU solution, but the hardware is much different and the GPUs aren’t working together. Intel’s Core i3 and i5 CPUs already feature a graphics processor integrated onto the die and AMD is planning on releasing the Fusion series, their incorporation of Radeon technology with AMD CPU microarchitecture. As CPUs have hit a ceiling as far as clock speeds and moved to including more cores, you can expect GPUs to take a similar path. There have already been a few high-end video cards released with two GPUs stuck together, and it’s only a matter of time before they’re small enough to fit many GPU cores alongside the multi-core CPUs of today.