Overclock.net banner
1 - 6 of 6 Posts

113802

· Banned
Joined
·
8,594 Posts
Discussion starter · #1 ·
Source: Microsoft's DirectStorage API to Support PCIe 3.0 NVMe SSDs, DirectX 12 GPUs

Microsoft yet has to officially disclose the DirectStorage hardware requirements for PCs. However, a software developer who saw an up-to-date Microsoft DirectStorage presentation and even shared some slides from it said in a Reddit post that the new API will be supported by all DirectX 12-compatible GPUs and SSDs featuring a PCIe 3.0 interface and supporting NVMe. Unfortunately, it is unclear whether all versions of the NVMe protocol will be supported, or whether SSDs will need to support any other capabilities (like a minimum speed requirement.
 
It starting to seem like some of the in development features are going to converge and create a "whole that's greater than the sum of it's parts". Maybe these Meh things will make sense once I see all of it being utilized together.
 
It starting to seem like some of the in development features are going to converge and create a "whole that's greater than the sum of it's parts". Maybe these Meh things will make sense once I see all of it being utilized together.
DX12 Ultimate in a nutshell.
 
What seems really odd about this article is the (Existing storage stack) and (Optimized storage stack) labels on the first step.

Why the **** wouldn't we always use the optimized storage stack?

Great, the GPU can decompress its assets keeping load off the CPU but the assets still have to be copied off the SSD into system memory. If they can make that step faster for streaming GPU assets why can't they make if faster for everything?

We really need this optimized storage stack. I don't even care about games or DX12 or any of it, just fix the storage stack in Windows. :p

Edit:
It is very nice that it works with PCIe 3.0 SSDs too, it will be interesting to see how my Intel 905p Optane drives compare to my Samsung 980 Pro.
 
Great, the GPU can decompress its assets keeping load off the CPU but the assets still have to be copied off the SSD into system memory. If they can make that step faster for streaming GPU assets why can't they make if faster for everything?
My understanding of DMA is that it allows PCIe devices to share data (notably from non-volatile storage and RAM into your GPU) without first loading the assets/instructions into RAM. This saves a lot of overhead processor, chipset, and memory controller, not to mention it makes use of all that juicy PCIe4 16x bandwidth.

It’s not that it makes anything operate faster at a hardware level. Instead, it dramatically improves performance for the end user by bypassing a ton of unnecessary “computational bureaucracy”. Instead of having everything pass through CPU/RAM at least once, (probably many times since there’s so many steps involved with the most trivial I/O operations), the GPU can now make its own I/O requests and write the results into VRAM.
 
1 - 6 of 6 Posts