The concerns regarding bandwidth sharing between the PCIe x16 slot and the primary M.2 slot aren't new — PC builders have been worried about PCIe lane sharing for at least 10 years, if not more. When reading up online to get to the bottom of this issue, I found too many conflicting reports and opinions. People just can't seem to agree on whether an NVMe SSD really "slows down" a graphics card or not.

While there's enough documentation from CPU and motherboard manufacturers to make this a non-issue, it turns out the information overload from the community is enough to overwhelm anyone, especially new PC builders. I'll break it down in the simplest terms, so you know what to look for when buying a new motherboard.

Intel and AMD behave differently

It has to do with the different approaches to PCIe 5.0

So, let's begin with how your processor communicates with components like graphics cards, NVMe SSDs, and other less resource-intensive hardware. Every motherboard has a fixed number of PCIe lanes distributed between those connected directly to the CPU and those to the chipset.

Depending on which PCIe, M.2, and other slots you use to connect your components (and in what combination), you could be utilizing either the lanes directly going to the CPU or the chipset. Ideally, you'd want to use those that have a direct connection to the CPU to minimize any latency. Hence, it's always advisable to connect high-speed components like graphics cards and fast NVMe drives to the slots having the most direct-to-CPU PCIe lanes.

These slots are almost always the PCIe x16 (usually the topmost) slot for the graphics card and the primary M.2 (usually labeled M.2_1) slot for your fastest NVMe SSD. One might install the graphics card and the primary SSD to these two slots respectively and call it a day, but things get a bit tricky depending on your processor as well as motherboard; bear with me.

Both Intel and AMD CPUs have a total of 28 PCIe lanes coming from the CPU, but their makeup is different. Intel's 13th Gen and 14th Gen CPUs feature 16 PCIe 5.0 lanes shared between the primary PCIe and M.2 slots, and 8 PCIe 4.0 lanes for the chipset. AMD's Ryzen 7000 CPUs, on the other hand, have 20 PCIe 5.0 lanes shared between the primary PCIe and M.2 slots, and 8 PCIe 5.0 lanes for the chipset and other connections.

Your NVMe SSD could halve your GPU bandwidth

It still depends... on your motherboard chipset

With an Intel CPU, installing an NVMe SSD (any generation) in the primary M.2 slot can reduce the number of PCIe lanes for your graphics card by half. So, instead of your GPU using all the 16 available PCIe 5.0 lanes in the case of an unpopulated M.2_1 slot, it'll now be getting only 8 lanes, and 4 lanes will be used by the SSD. Due to the nature of the bifurcation of PCIe lanes, a 12x/4x split isn't possible.

However, this scenario only occurs on Z790 motherboards with a PCIe 5.0 M.2 slot, and not all Intel motherboards. You'll need to check the motherboard manual for the specific model you're planning to buy, so you're not caught off-guard. Even on affected motherboards, using a secondary M.2 slot that uses PCIe 4.0 instead can circumvent this problem.

And, as you would expect, AMD processors allow 16 and 4 PCIe lanes to be used, respectively, by your graphics card and primary SSD — no 8x/4x split needed. That said, only the X670E chipset will guarantee PCIe 5.0 lanes for your GPU; other chipsets can still have only PCIe 4.0 support for the graphics card. AMD's Ryzen 9000 processors also continue this support, but this time on both X870 and X870E chipsets.

So, in conclusion, AMD processors, at least in theory, are better for users who want to use a high-end GPU and PCIe 5.0 SSD together without sacrificing bandwidth for either one.

However, if you only have a PCIe 4.0 SSD, you might skip the primary PCIe 5.0 M.2 slot entirely on your Intel motherboard (and AMD also, of course), and use one of the PCIe 4.0 slots that are linked directly to the CPU. This way, you don't need to worry about any lane sharing between your SSD and GPU. The motherboard manual can help you find the right slot, and also appropriately guide your purchase decision.

Why you will not notice the difference anyway

The performance reduction is negligible

With all that out of the way, the difference in GPU performance between 16 and 8 PCIe lanes will not be noticeable to most users. Firstly, there are no existing graphics cards that support PCIe 5.0, so your GPU is running on PCIe 4.0 anyway, even in a PCIe 5.0 slot. Second, even if you add an SSD in the primary M.2 slot, leaving 8 instead of 16 lanes for your graphics card, the drop in gaming performance is negligible — around 1–3%.

Graphics cards are still a long way from saturating the full PCIe 4.0 x16 connection, so reducing it to x8 doesn't impact your performance as much as you might think. Only if you're running workstation-grade systems will you probably need the most bandwidth, but in that case, you probably need workstation CPUs such as AMD's Threadripper chips and others.

PCIe 5.0, at least for GPUs, is still a fairly new standard, so we're still years away from GPUs needing the full PCIe 5.0 bandwidth. In fact, they've still not come close to saturating the full bandwidth of PCIe 4.0.