Intel’s Cascade Lake-X CPU coupled with X299 chipset motherboard gives me 48 PCIe lanes straight to the CPU (out of that EVGA X299 Dark supports “only” 44 but that has nothing to do with how much platform can do and everything to do with it being made before 48 lane CPUs existed).
Video card gets its own 16 lanes to CPU. I have attached boot drive to U.2 port that has its own 4 CPU lanes. I have populated both M.2 NVMe slots with drives (one for data I am working on that needs fastest possible performance and other as fast “scratch” volume) and each one gets its own 4 CPU lanes, still leaving me 16 lanes available.
Last, but not least, thanks to latest (1.23) BIOS update mb now supports bifurcation of PCIe slots and their use with aftermarket PCIe storage cards which allows me to use card populated with up to four NVMe drives, Windows will see them all and each one will get its own 4 CPU lanes.
Why this excites me? It is not (just) the obvious part of being able to have plenty of simultaneous I/O without bottlenecking that excites me about it. It enabling use of NVMe drives on PCIe card as “SSD” tier in Windows Storage Spaces is what really thrills me. Now I can put that tier in front of HDD tier to end up with “hybrid drive” that gives me best of both worlds: huge capacity per $ of HDD with blazing performance of NVMe for files on it that are accessed most (Windows behind the screen maintaining “heat map”) and all without having to deal with everything use of RAID cards or VROC chip would result in. Screenshots below show 18TB volume I got by combining 2TB NVMe tier with 16TB HDD tier and benchmark that proves SDD tier is automatically used. Yes, figures are bit lower than they would be for individual NVMe drive but that is because drives in each tier are mirrored (I have used two NVMe and two HDD drives).
Icing on the cake is that by using “just” two NVMe drives on PCIe card I still have 8 PCIe CPU lanes left unused for future expansion in the case I end up needing 4K capture card or whatever.
In other words: For some uses clock speed or number of cores is all that matters. In my case expandability/layout is all that matters, if I can’t have layout I need then how much faster some other CPU might be is completely irrelevant to me.
