How Many Graphics Cards [GPUs] Can You Use In A PC Without Bottlenecking?

Ever wondered how many GPUs you can use without bottlenecking? If so, you’re in the right place, because that’s exactly the question we’re looking to tackle today.

I’ll walk you through all the core information you need to know about the performance compromises (and benefits!) of a modern multi-GPU setup.

Ideally, by the end of this article, you’ll have a solid idea of how many GPUs you can work into your ideal PC build and maybe even which ones to use for specific workloads.

I’ll also take time to provide a few additional resources and answers for other multi-GPU setup-related questions before I wrap things up.

So, let’s get into it!

A Brief on Multi-GPU Setups

First, let’s take a moment to describe the multi-GPU setup and its basic requirements. Your average multi-GPU setup consists of the following:

• Two or more discrete Graphics Cards, sometimes connected with a physical “bridge” of some kind to improve multi-GPU performance (ie, an SLI or NVLink bridge from Nvidia— AMD CrossFire doesn’t need a physical bridge)
• A full E-ATX, XL-ATX, ATX or (rarely) a Micro ATX motherboard to make room for 2+ GPUs. The single PCIe x16 slot limitation of a Mini ITX motherboard makes it impossible to run even a dual-GPU setup in an SFF PC.
• A high-end consumer CPU for editing and gaming workloads, OR a proper HEDT CPU for full-PCIe-bandwidth multi-GPU renders.
• Multiple full-speed PCIe x16 slots for the best results!

How GPU Bottlenecking Works

A GPU bottleneck can be defined as any scenario where your graphics performance is being hindered by another component within your PC. This isn’t the same as simply having weak graphics hardware, which would instead result in a CPU bottleneck if you paired a weak GPU with a powerful CPU.

So, what are the most common causes of GPU bottlenecking? I’ve broken them down into a list below:

• Limited PCI Express Bandwidth — An older PCIe standard or insufficient PCIe lanes can throttle GPU performance in bandwidth-sensitive workloads.
• Slow Storage — Slow storage can hinder a GPU’s ability to render and process large files, ie for project renders or even modern games built for SSDs.
• Slow CPU — A slow CPU can hinder a GPU’s raw horsepower, but especially in games or other real-time workloads where a lot of CPU processing is necessary for the GPU to be able to do its work at all.
• (For Projects) Slow or Limited RAM — RAM usually doesn’t have a heavy impact on graphics performance, but if you’re relying on GPU acceleration in rendering and editing projects, you’re also still very reliant on your existing desktop RAM to work back and forth with your GPU and its VRAM. Especially if your VRAM is insufficient for your particular project and your RAM is being used as well.
• (For Games) Poor or No Multi-GPU Optimization — Unfortunately, despite how popular multi-GPU setups used to be for enthusiasts and gamers in general, most titles simply aren’t well-equipped to handle multi-GPU setups, if at all. Certain titles that support multi-GPU operation may still fail to scale performance as expected, as well. Only a few particularly well-optimized titles will get a 2x, 3x etc performance boost according to the number of GPUs you use.

How Many GPUs Can You Use Without Bottlenecking Them?

So, how many GPUs can you use without bottlenecking, in general? This isn’t a question that can be answered for every system and every use case universally.

However, we can address one of the most likely sources of bottlenecking in a multi-GPU setup: limited PCI Express lanes.

There are a set amount of PCI Express Lanes available to your GPUs depending on your processor, chipset, and choice of motherboard.

For the best results in a multi-GPU setup, you’ll want to confirm that your motherboard properly supports running at least two of its PCI Express x16 slots at full speed, simultaneously.

Many cheaper boards aren’t capable of this and instead need to throttle all but one full-size, full-speed PCI Express x16 slot.

HEDT motherboards are also of particularly high utility in multi-GPU setups, since they come with far more available PCI Express Lanes than consumer desktop series motherboards.

Of course, not every GPU needs the entire bandwidth of a PCIe x16 slot. Some GPUs don’t saturate x8 or even x4 lanes and can comfortably run with this limited amount of lanes without performance impact.

How Many GPUs Can You Use Without Bottlenecking In Games?

Unless you’re playing one of the very few games that are actually well-optimized for multi-GPU support, the answer to this one is, unfortunately, going to boil down to “one of them”.

Even in titles that support multi-GPU properly, it’s fairly rare for multi-GPU support to provide a proportional performance boost-per-graphics card as a gamer would reasonably expect from slapping two or more graphics cards into their PC build.

If gaming is your PC’s primary focus, be mindful: multi-GPU support should not be your priority, at all. Not enough developers prioritize the feature for you to be prioritizing it, and for that matter, the budget GPU market tends to steadily improve over time, especially the used market.

Stick to single GPU builds whenever possible, and consider selling your GPU when it comes time to upgrade instead of trying to slap in a duplicate for extra performance.

How Many GPUs Can You Use Without Bottlenecking In Editing and Rendering?

So, how many GPUs can you use in a professional rendering PC without encountering bottlenecking?

Fortunately, this is actually pretty much the best-case scenario for a multi-GPU setup, provided your motherboard has enough PCI Express Lanes to keep up with your workflow!

For example, if you’re doing video editing and compositing on a multi-GPU setup, a high-end mainstream motherboard with 2+ full-speed PCI Express x16 slots should do the trick, and handily avoid any GPU bottlenecking as long as the rest of your system can keep up.

When you move up to proper 3D animation and rendering, you’ll find yourself saturating the bandwidth available to your GPUs much more frequently than if you were just gaming or video editing, though.

For multi-GPU 3D rendering, HEDT platforms can look even more appealing for their PCIe Lane advantage, especially if you’re hoping to use a lot of NVMe storage and/or turn around regular previews of your work.

Of course all of this requires software or engine support, so multi-GPU scaling will depend heavily on whether your particular software supports it or not, and to which extent.

Conclusion

So, let’s put some concrete numbers to the excursion above. How many GPUs can you use in a PC without them bottlenecking?

In a best-case scenario, where your CPU is up to snuff, your software supports full multi-GPU scaling, your case is big enough, your cooling is top-notch, your GPUs are purpose made for multi-GPU setups (blower, dual-slot) and your PSU is kick-ass, it all boils down to your motherboard, how many x16 PCIe-slots it has, how many PCIe lanes, and how those slots are spaced (unless you use riser cables).

So with all of the above excluded as limiting factors, here’s how many GPUs you can put inside your PC and use to its full potential without bottlenecking (or almost no bottlenecking, near perfect scaling):

• Mainstream Motherboard suited for a mainstream Ryzen or Intel Core CPU: 1 – 2 GPUs. Some outliers allow 3 GPUs
• HEDT / Workstation Motherboards: 4 GPUs, some outliers let you add up to 7 GPUs, but you’ll need riser cables to make them fit
• Server Motherboards: Servers optimized for GPUs can fit 8 GPUs or more.

As mentioned, though, you’ll need very specific software that is specifically optimized and developed for scaling across multiple GPUs. GPU render engines such as V-Ray, Octane, Redshift or Blender’s Cycles come to mind, or Machine learning software.

FAQ

What Kind of GPUs Are Best In a Multi-GPU Setup?

Undoubtedly, blower-style graphics cards are the best choice for a multi-GPU setup.

Blower-style graphics cards are built to immediately exhaust hot air outside of your PC as quickly as possible, making them far easier to stack on each other without risking excess heat spreading to other graphics cards or the rest of your PC.

Is NVLink, SLI, or CrossFire Better?

AMD’s CrossFire and Nvidia’s SLI are both essentially-identical technologies for putting multiple GPUs to use. CrossFire doesn’t even need a physical connector to work, unlike Nvidia SLI, though they are both basically the same.

Nvidia’s newer NVLink standard actually does offer improvements of note over SLI and CrossFire, though, especially for professionally-oriented applications.

You can try Alex’s Full Guide on NVLink vs SLI if you want more detailed performance numbers on this, but the general takeaway is if you’re going to do a multi-GPU setup for rendering, using Nvidia NVLink is probably the best route available to you.

Can You Mix GPU Brands In a Multi-GPU Setup?

Yes…technically.

Alex has written a detailed run-down on what you can expect from a multi-GPU setup where you’re utilizing GPUs from different brands in the same build.

The rest of what’s been stated in this article still applies, though, and while mixing GPU brands technically works in some use cases, it certainly isn’t being optimized around.

Plus, the best multi-GPU solutions like NVLink will be unavailable to you if you’re mixing AMD and Nvidia in the same PC.

Over to You

And that’s all, for now!

I hope that this article helped clarify how many GPUs you can use without bottlenecking, whether you’re an aspiring creative with room for an extra graphics card or a gamer on the hunt for more performance.

Feel free to leave a comment in the comments section below and let us know if you have any other questions related to GPU technology! Me or another member of the CGDirector Team will be happy to assist you.

You can also try out our Forums if you want to socialize with other members of our community, or want to use the longer-form format to share your latest project or build.

Until then or until next time, happy building! And remember: if you’re doing three or more GPUs, you may as well go straight to an HEDT platform over mainstream CPUs and motherboards. The high-end of mainstream is still going to start struggling past two GPUs if you’re saturating their bandwidth.