CPU lane restrictions and Ethernet speed capacities.
CPU lane restrictions and Ethernet speed capacities.
Hello everyone, This is a technical question for an upgrade in specific conditions. I need to update a workstation. It is currently using an Intel 7900x (44 lanes) that can support without problem my PCIe cards at full bandwitdth ( x16, x8, x4, x4 / ie : GPU, 8Gb fibre optic, blackmagic video, mobo chipset ). Looking at my options I found a configuration that is suitable if I give up the fibre channel and use the ethernet port instead to communicate with an Internal Server, but it poses one big question. If all my lanes are occupied while i'm using the 10gb ethernet port that shares its lanes with the other chipset features will I limit my ethernet bandwidth if I use usb or sata in parrallele at the same time : Example : Using Davinci Resolve communicating with a Internal server through the Ethernet port and SSDs through the sata/usb ports. - Intel core Ultra 9 285k (24 lanes, which 4 lanes are dedicated to the mobo chipset) - ASUS PRO art z890 creator wifi . - RTX 5090 (using 16 lanes) - Blackmagic video card (using 4 lanes) If anyone has tested this situation I would much appreciate any feedback. Thanks in advance ! PS : For reference I am currently running at 400-430Mbs down/up on the fibre optic channel to the Internal Server.
With just 430Mbps you won’t face any problems. This speed is far below what most ports support and even a basic PCIe Gen 1x1 connection could manage it more than three times. At 430 MB/s you should be fine. Keep in mind the Z890 board runs on PCIe Gen 5 and Gen 4, which is significantly quicker than the PCIe Gen 3 used by your old X299. This results in a much higher overall PCIe capacity, even though it has fewer lanes. The chipset offers 8 lanes for the CPU, not just 4, and being PCIe Gen 4 gives it the same bandwidth as PCIe Gen 3 x16—about 32 GB/s total, or 16 GB/s in each direction.
The 430Mbps issue stems from the internal server, affecting multiple users in the company. If the system becomes congested, a high bottleneck would be needed to drop below 430Mbps. The chipset allocation seems unusual with eight lanes instead of four; according to the site, it lists two PCIe 5.0 x16 slots and one PCIe 4.0 x16 slot, which adds up to more than expected.
The eight chipset lanes remain hidden from the end user—they function within the internal connections of the CPU-chipset interface. The figures you're observing correspond to the actual physical ports on the board, originating from the CPU or chipset.
If this scenario holds, it would still result in four unused lanes out of the total twenty-four. This raises my question.
The chipset offers more than merely eight connectivity paths. If it lacked this, its purpose would be lost since everything could be routed through the CPU. The key is that the chipset connects directly to the CPU with eight lanes, meaning the combined maximum speed for all connected devices is limited to eight lanes. The remaining PCIe lanes serve other components on the board, such as SATA, USB, Ethernet, and audio interfaces. Having only one dedicated slot doesn't imply these lanes aren't being utilized.
The graphics card links directly to the CPU via dedicated PCIe lanes. The initial M.2 slot is often linked straight to the CPU as well. An 8-lane setup (similar to PCIe 4.0 x8) exists between the CPU and chipset, allowing up to 15 GB/s transfer speed. The chipset adds more PCIe lanes—such as the 890 chipset offering up to 24 lanes. A 10GBps Ethernet port delivers roughly 1.25 GB/s, achievable with two PCIe 3.0 lanes or four PCIe 2.0 lanes. Earlier 10G cards used a PCIe 3.0 x4 configuration because the hardware was built for two RJ45 ports. Even current chips support just two PCIe lanes; PCIe 3.0 backward compatibility is limited, as a single lane can only handle about 900 MB/s, with a theoretical max of 970 MB/s. The RTL8127 chip is among the first to function with a single PCIe 4.0 x1 lane, and it’s worth noting that such devices route data separately from other signals, ensuring smooth performance without interference.