Blazing Fast NVMEs and Direct Storage API for PCs *spawn*

There's a handful of various M2 to SATA Adapters and SATA to USB 3.x Enclosures, each with their own specificity of support and speeds. I think you'd really have to do some research to find one based on current NVME, let alone trying to get NVME Gen4.
 
Just remembered something... would one possibility be that a large chunk is transfered and then decompressed on the fly by a compute shader that does virtual texturing.
 
There's a handful of various M2 to SATA Adapters and SATA to USB 3.x Enclosures, each with their own specificity of support and speeds. I think you'd really have to do some research to find one based on current NVME, let alone trying to get NVME Gen4.


No I was just wondering what the protocol for usb storage was and if it had revisions? And if includes things like trim/ncq or whatever makes nvme special.
 
I think what makes mvme special is it uses pci-e lanes and after the memory bus it's the fastest connection we have on pc
edit: brainwave, make a drive that connects via a dimm slot
 
I think what makes mvme special is it uses pci-e lanes and after the memory bus it's the fastest connection we have on pc
edit: brainwave, make a drive that connects via a dimm slot


Close enough :eek:

Combo%2BSDIMM%2Bw%2BCFast_lo.jpg
 
I think what makes mvme special is it uses pci-e lanes and after the memory bus it's the fastest connection we have on pc
edit: brainwave, make a drive that connects via a dimm slot

Some of Intel's 3DXPoint products are in DIMM form factors, but of course it's very expensive.
 
Both CUDA and Tensor cores sit within the SM so it still could be either.

Minuscule... what, at 7GB/s throughput from the SSD? If so that sounds impressive. Can't wait to see some real numbers around this.
In another post I made on this forum in which I was actually referencing this tweet, I took this to assume that it was using the shader cores.. but I guess that's not necessarily true. The implication is that it's RTX only, which could imply that it's actually using the tensor cores in some capacity.

Would be nice to have clarification. Maybe that's what makes the performance hit so miniscule/negligible?
 
PS5 decompression hardware lists 22 GB/s max with 8-9 GB/s typical.
PS5 10.28 TF AMD GPU hardware using first version does 60-120 GB/s. Radgames developer was hoping for more improvements in later versions of their GPU decompression.

Napkin math of around 6 to 12 GB/s per TF on AMD GPU. So maybe 2TF to hit 12 to 24 GB/s?
 
Any takers on when first biggr game seriously utilizing directstorage will come out? My bet would be ue5 engine and some time 2022. Demos are pretty much guaranteed 2021 as ue5 early version becomes available. Might be a bit of a drag though if one has to use some beta windows version to run ue5.
 
Any takers on when first biggr game seriously utilizing directstorage will come out? My bet would be ue5 engine and some time 2022. Demos are pretty much guaranteed 2021 as ue5 early version becomes available. Might be a bit of a drag though if one has to use some beta windows version to run ue5.
First party MS games should likely have that type of feature set requirements around 2022 or so
 
Back
Top