Whatever happened to motherboards with support for multiple GPUs?

Contrary to popular belief, motherboards with dual GPU support never went away, and chances are, the motherboard in your PC actually supports dual GPUs right now. However, you can’t just put two GPUs in your PC and call it a day, as Nvidia’s SLI (Scalable Link Interface) and AMD’s CrossFire are no longer supported. As a result, you can put two GPUs in your PC, but it won’t really help you in gaming.

What happened to SLI and CrossFire?

Big cards and complicated software

We’ve already talked about how 3dfx started the dual GPU craze with the Voodoo2. If you put two of those in a machine, they share processing-related information with each other. This was done with a cable that connected both cards, and they would render the horizontal pixel lines in an alternating sequence, essentially increasing the graphics processing capabilities of the system.

Nvidia later bought 3dfx, debuting SLI with its GeForce cards in 2004 and building more complex algorithms to better utilize the power of two graphics cards. Split-frame rendering (SFR) and Alternate-frame rendering (AFR) were two such algorithms, where SFR divided each frame horizontally and sent each section to a different GPU. AFR gave better performance by forcing the GPUs to take turns rendering the frames, but it could cause microstuttering.

However, between the rising TDP of GPUs and the growing physical size of GPUs, you could barely fit multiple GPUs in a regular chassis. Plus, both SLI and CrossFire were difficult for developers to optimize for, with it also being a taxing process for both Nvidia and AMD to release driver updates that would take into account multiple GPU setups. Eventually, they became enthusiast-only features, and disappeared entirely a few years later.

Why use a dual GPU setup nowadays, and what do you need for it?

It’s all about PCI lanes and productivity workloads

Source: Wikipedia

If you’re considering a dual GPU setup still, then there are a few things to consider. First and foremost, it won’t help you in gaming; at least not directly. Because cards won’t work together, you could put two GPUs in a machine and delegate individual workloads to them. For example, you could game on one and render video on another, and that would work just fine and is the most practical usage for a setup like that.

However, you also need to consider whether your motherboard will support it. While multi GPU setups were common enough in the past for motherboard manufacturers to at least consider supporting it, but now it’s not really a consideration. High-end chipsets like Intel’s Z790 and AMD’s X870E will work, but you’ll be running the first GPU with 16 PCIe lanes in the Z790 and four PCIe lanes for the second GPU, and each GPU will get eight PCIe lanes a piece in the X870E.

This is also dependent on the motherboard’s lane allocation, but that’s typically how it, as motherboard OEMs will prioritize single GPU configurations. If you plan on building a multi GPU machine, you’ll need to research carefully. Four PCIe lanes is enough for bandwidth for a lot of tasks like video rendering, but you’ll also need to research the workloads you intend on putting on your secondary GPU to ensure that you get what you need.

Why use a dual GPU setup nowadays, and what do you need for it?

It’s all about PCI lanes and productivity workloads

If you’re considering a dual GPU setup still, then there are a few things to consider. First and foremost, it won’t help you in gaming; at least not directly. Because cards won’t work together, you could put two GPUs in a machine and delegate individual workloads to them. For example, you could game on one and render video on another. That would work just fine, and some kind of setup that requires dedicated workloads on specific GPUs is the most practical usage for a setup like that.

However, you also need to consider whether your motherboard will support it. While multi GPU setups were common enough in the past for motherboard manufacturers to at least consider supporting it, but now it’s not really a consideration. High-end chipsets like Intel’s Z790 and AMD’s X870E will work, but you’ll be running the first GPU with 16 PCIe lanes in the Z790 and four PCIe lanes for the second GPU, and each GPU will get eight PCIe lanes a piece in the X870E.

This is also dependent on the motherboard’s lane allocation, but that’s typically how it, as motherboard OEMs will prioritize single GPU configurations. If you plan on building a multi GPU machine, you’ll need to research carefully. Four PCIe lanes is enough for bandwidth for a lot of tasks like video rendering, but you’ll also need to research the workloads you intend on putting on your secondary GPU to ensure that you get what you need.

Plus, your CPU needs to support it, too. We’re already seeing problems with PCIe lanes and PCIe 5.0 when it comes to both GPUs and NVMe drives. And, finally, there’s the issue of getting a power supply that can provide sufficient power to both of your GPUs and your high-end CPU that has enough PCIe lanes to support running everything. Some motherboard chipsets will even have specifically allocated PCIe lanes for specific M2 slots.

In short, motherboards supporting dual (or more) GPUs never really went away, but the use cases definitely did. Nowadays, you don’t need a system with multiple GPUs unless you’re doing some really heavy lifting on your machine, and the cost of running that kind of build can be incredibly prohibitive. Home lab users who want to run hardcore AI workloads or game on separate VMs might also have a use for multi GPU setups like these.

In short, you need a lot of power, a motherboard that supports it, a good enough CPU, and you need to be careful about your lane allocation. For most people it simply isn’t worth it to try, but enthusiasts and productivity workloads can see big benefits when done right.