Programming? Don’t forget also the other hardware of your system 🙂
PCI-e 5.0 has increased by a factor of 4 compared to version 3.0, while RAM has increased by a factor of 2 since DDR4-3200 to DDR5-6400. The bandwidth of the dual-channel DDR5-6400 system is approximately 90 GB/s. Even if you can make an array of 4×10 = 40GB/s, just storage and retrieval is really a workload of 2 x 40 = 80GB/s to/from RAM. This means that the memory bus has already become saturated. The CPU will then have to process that data “quickly”, which is a difficult task (40 GB/s over 16 cores = 2.5 GB/s per core)
It would be great if you could move data directly to GPUs, but on consumer boards it also means you have to split 16x lanes into 2x 8x lanes for SSDs and GPUs. If the chipset and the SSD’s CPU slot are already PCI-e 5.0, you’ll get the same amount of bandwidth (although you may be hitting other I/O ports).
Unfortunately, HEDT (or: AMD) rigs are still a bit behind in the PIC-e 5.0 area, but then you won’t have the hassle of splitting lanes. It is also possible that you have 4, 6 or 8 channels of memory at your disposal, so don’t saturate your working memory with a few SSDs.
Either a pretty cool device, but I don’t know if you should put this in the consumer panel. However, the possibility is always there, and even with an x8 connection to the CPU, all the same SSD will still be used fairly smoothly (7.5GB/s max?).
