Silicon Motion Launches New PCIe 4.0 NVMe 1.4 Controllers Capable of Extreme Speeds

[Peter_Brosdahl]

Image: Silicon Motion



As PCIe 4.0 adoption continues to grow in the PC enthusiast community, we are beginning to see more and more NVMe SSDs hit the scene. Although most games do not fully benefit from their increased speeds, they can still provide other benefits for other applications. An integral part of the evolution of drive speeds is their controllers. New offerings from Silicon Motion are pushing the speed envelope even further.



Featuring 3D TLC and QLC NAND, they cover all the angles, from performance to value. At the top of the stack, we have the SM2264. This premium NVMe controller has blistering-fast 7,400/6,800 MB/s read/write speeds and 8-channel memory. IOPS are...

Continue reading...

 

[Grimlakin]

This is interesting, will be nice to see what Samsung does with these or how they compete.

 

[Elf_Boy]

Cant wait for when a Dev learns how to use the speed on game load.

 

[Burticus]

 

[Grimlakin]

Cant wait for when a Dev learns how to use the speed on game load.

Honestly when we get direct calls to the storage via a video card we SHOULD be able to load the digital assets to the card fast enough to fill it in less than 3 seconds with the speeds obtainable on PCIE 4.0x4. And still have enough bandwidth for the OS. The real problem is getting the data to properly stream over multiple channels.

Right now the speeds we are seeing are tests done in ideal situations.

People talk about sub ms seek times. To get that truly you need a profile of what is called regularly from storage. And you need a controller with enough memory space to meet that need. That's the real reason big enterprise storage is so **** fast. It has learned through AI like algorithms what data is accessed frequentrly, REALLY FREQUENTLY and less often. With memory becoming progressively cheaper if someone came out with a intelligent controller that could take a DDR4 memory stick I would be right there ordering a 16 gig stick to stick in it and having my NVME card being ran off of that.

The problem is to run the algorithm correctly it's like an additional computer to add to your setup and only for I/O performance.

The more cost effective option would be to have a CPU heavy I/O application handle that role and have a carved out memory space to host it. Only time shall tell.