Combine PCIe connections for RAID 0 setup (reverse bifurcation)
Combine PCIe connections for RAID 0 setup (reverse bifurcation)
Think about merging two PCIe X8 ports to power your Asus Hyper M.2 card with all four M.2 drives in RAID 0. Your board offers two x16 and two x8 slots (ASUS Prime). You might be able to route hardware from the x16 into two x8 connectors, effectively reversing the setup. You’d likely need to rewire one x8 port to access the final pins in the x16 socket, and possibly use a female-to-female cable to get a proper female connector for the card. This approach probably won’t work for GPUs, but since your board supports RAID across any drives on an AMD chipset, it’s worth checking if there are compatible solutions.
Which specific board are we talking about? If it actually has two X8s and can split them (which seems unexpected), you might consider using two Hyper MX2s or something similar, with two drives per unit.
Asus Prime x299 deluxe II is powered by an Xeon 2255 processor, equipped with 192GB RAM (supports quad channel for 256GB), a Dual 2080ti chipset. It features two 16GB PCIe slots and two M.2 ports. The layout arrangement is complex—starting with x1, then x16, followed by x8, then x16 again, and so on. This configuration means the top x8 slot is occupied by the first GPU, requiring a riser cable to reach it. Using only the lower x8 slot leaves just two M.2 bays available for RAID. I opted for four M.2 drives to boost transfer speeds and storage capacity, currently holding around 5TB across three HDDs. My goal is to merge all this into one drive for better performance and longevity, which would last several years. I acknowledge the system's age but feel updating isn't worth the trade-off in RAM—switching to a single GPU would create a PCIe bottleneck, capping its performance at roughly half of the dual 2080s today. The new 14th-gen CPU’s specs don’t justify the price drop from 196GB to 128GB in quad channel, especially since RAM capacity has increased significantly.
AMD is also worth considering. You can install 192GB on consumer systems using 4x48GB modules. Your connection speed might be slightly reduced, but it should still achieve 80–90GB/s (close to the upper limit of your current setup with quad channel). What kind of tasks are you running? Are you certain the workload is that demanding for PCIe transfers? You could opt for two 4TB NVMe drives or just one 8TB unit. There’s plenty of storage and reduced risk with a 4x RAID configuration. Are you already utilizing all the available NVMe ports? It looks like you have three, two from the chipset and one from the CPU lanes. Your explanation seems a bit unclear, so I reviewed the manual. It appears you’re referring to GPUs in slots 1 and 3, with slot 5 limited to x8 only. To add another GPU there, you’d need a riser. Long shot, but could you split slot 5 into four x2 configurations? Then you could run four NVMes at half speed each. However, the information you gave contradicts what I saw: only two x16 and one x8 slot are visible. Another choice might be to abandon the hyper idea and purchase a card with a real PCIe switch, allowing four or more drives on a single x8 port without issues.
It's the X299 Deluxe model, not the Mark II. The "B" M.2 slot is actually an X16 port but only contains half the connectors like the lower X8 slots. As discussed before, the Hyper M.2 card works on AMD systems with X8, X8, X4, X4, X4, or X4 configurations using bifurcation. If I set up the riser correctly, merging the two X8 slots would mean splitting the Hyper M.2 in half and placing each half into an X8 slot. The VROC then merges all drives back into a single array, utilizing CPU lanes instead of the chipset.