Published May 25, 2026, 9:00 AM EDT
Ever since a kid, Shaheer has always been intrigued by tech and how its components work, and has always geeked out over new PC component releases. His enthusiasm for tech, gaming, and composing eventually led him to become a writer, a career path he hadn’t quite planned or thought of, but it proved to be a natural fit. Nevertheless, Shaheer has been writing for more than 4 years, having penned more than 2000 articles now across multiple accredited publications.
Shaheer mostly covers PC components, including in-depth coverage of each component, but his speciality subsumes within hardware of all sorts.
Even if modern motherboards come equipped with several M.2 slots, most people are unaware that populating all of them incurs a significant performance penalty. Your CPU has limited lanes, and sharing those between the GPU, NVMe, and other ports means it can only transfer so much data at once. As such, installing too many NVMe drives will cause the lanes to split — ultimately impacting your PC’s performance.
Related
Your CPU only has so many PCIe lanes
Most slots route through the chipset instead
Intel CPUs 12th-14th gen expose 20 lanes directly to the CPU, of which 16x is utilized for the GPU and 4x for the secondary PCI Express slot or the M.2 slot closest to the CPU. These direct lanes run at their full speed, directly connected to the CPU. However, additional ports, including remaining M.2 slots, USB, Ethernet, SATA, etc., are routed through the motherboard’s chipset.
Even if your motherboard’s chipset supports 16 lanes, the link to the CPU might only be four lanes wide on platforms like the Z490 or Z590. This is why any data directed through the ports through this chipset will be slow, and the same extra step also introduces latency.
So, if you have any additional NVMe drives connected through the chipset, they will run at a much slower speed than intended, since the latency and lane bottleneck will limit their performance. This is also why certain SATA ports are disabled if a certain M.2 port is populated, because all these ports share the same route via the chipset.
Certain motherboards link more than 1 M.2 slots directly to the CPU, in which case a bottleneck would occur at the CPU link, and your GPU and NVMe drives would then contend for bandwidth.
The chipset is where things slow down
The bottleneck becomes a real issue under load. When more devices compete for the bandwidth in an already congested lane, all hardware routed through the chipset will be slowed. For GPUs, the performance loss is minimal — usually 0–5%. However, with NVMe drives connected via the chipset, the effect is more pronounced, and read and write speeds are noticeably reduced.
Fortunately, newer platforms are less affected by this issue. Newer GPUs, like the Nvidia RTX 50 Series, support PCIe 5.0, with speeds double those of PCIe 4.0, reaching ~ 4GB/s per lane, for a total of 64GB/s at 16x. The flagship RTX 5090 features 32GB, so even halving the GPU lanes from 16x to 8x still delivers 32GB/s bandwidth, freeing up lanes for additional NVMe drives or PCIe devices.
AMD CPU owners are better off when it comes to lane splitting, since the Ryzen 7000/9000 Series CPUs support 24 lanes directly to the CPU, so more devices are given access to full-speed connections.
This only affects certain workloads
Casual gamers and users can most probably ignore this issue
Even if your hardware is maxing out the required lanes on your motherboard, you won’t experience bandwidth limitations unless they’re maxed out all the time. As such, lane limitations aren’t a concern for everyone, but only for a specific user base.
Now, for someone who handles heavy workloads like video editing with multiple drives, the system would read from one drive and write to the other, saturating the available bandwidth.
If you’re gaming, however, lane splitting won’t really matter much, since you’ll only be stressing your GPU and the drive you’re pulling the game from. The rest of the drives aren’t being stressed, so lane splitting won’t hurt, since they’d be sitting idle and wouldn’t be competing for bandwidth.
How to check if your drives are being bottlenecked
Use GPU-Z and CrystalDiskMark to check
There are multiple ways to check if your motherboard is allocating bandwidth to your on-board devices, slowing them down. GPU-Z can display what speed your GPU is running at under the bandwidth section. If the speed shown in GPU-Z is running lower than the rated spec, the GPU may be running in a shared configuration.
You can also use CrystalDiskMark to benchmark your NVMe drives. If the read/write speeds are slower than the advertised speeds, the drive may be bottlenecked via the chipset. You can then check your motherboard’s manual to see if that specific drive is being routed through the chipset or directly to the CPU. I recommend installing the fastest NVMe in a slot connected directly to the CPU.
Smart drive placement solves most of the problem
The solution to this problem isn’t to avoid multiple NVMe drives in the first place; it’s about being methodical with the placement. As I mentioned earlier, if your workload doesn’t stress all your hardware at once, you probably won’t notice the bottleneck much. Also, it’s essential to place the NVMe drives so that the ones you want to run at full speed go in the right slots by checking your motherboard’s manual.
GPU-Z
OS
Windows
Developer
TechPowerUp
Pricing model
Free
Leave a Reply