| Frant wrote: |
| Ankh wrote: | | StrEagle wrote: | it's from the previous PC along with a SSD, legacy files on them and so on..
I have 2x NVMEs  |
Nice  i've got 3 myself. Best buy ever (apart from upgrading to SSD the first time) |
Three? You have a board with three M.2 ports or do you have a PCIe slot-in card NVMe/M.2 card?
I've got one NVMe (1TB 980 PCIe). I love it. The difference isn't as mindblowing as when I got my first SSD though, it's more of an evolutionary step. I did believe that the step from ~520GB/s to ~3500GB/s would be massive when it came to performance but I realised that it's not just about raw speed that the device can deliver, it's the PC that does a ton of other things while reading/writing to it that adds a considerable amount of overhead. The current NVMe tech is far beyond the average PC tech. The PCIe/NVMe subsystem (transfer protocols, CPU PCIe <-> memory controller/performance, RAM size, CPU and RAM performance etc.) must evolve to take advantage of the performance that modern NVMe technology offers.
They've introduced a first step in that direction with DirectStorage which, if I remember correctly, use the Compute cores on your GPU and VRAM to decompress the packed data files lightning fast for a particular game instead of using the CPU and DRAM which is a much slower process. As it decompresses the data which is usually read sequentially it can transfer it to DRAM at the same time which is where the performance improvement comes from. It requires a 30x0 series card (  ) and games that include support for the DirectStorage API.
While it's being "ported" to Windows 10, it's a limited version and you're getting the full boost only in Windows 11 since they've apparently created a new storage subsystem or whatever it was. It's limited to games at the moment and thus doesn't accelerate general computing where random reads/writes are the norm which is much slower than sequential reads/writes. That's one area where they really need to improve. I'm not certain how current SSD-drive controllers work these days but I presume they're evolved versions of the ones I remember from my OCZ SSD days. They need to improve the random read/write performance by increasing the number of "cells" that the controller can adress simultaneously (ie. accelerate the requests) as well as design non-volatile chips that can be read from more addresses as fast as possible without the latency that comes from requesting data from random adresses on any particular non-volatile chip. |
1 on the motherboard and 2 on PCI-E
Edit: I agree with the step up to SSD btw, it was amazing compared to the good ol days!
shitloads of new stuff in my pc. Cant keep track of it all.
