Silicon Motion already Designing SSD controllers based on PCIe 5.0 - 16 GB/s

Published by

Click here to post a comment for Silicon Motion already Designing SSD controllers based on PCIe 5.0 - 16 GB/s on our message forum
https://forums.guru3d.com/data/avatars/m/232/232349.jpg
Bad ass! Getting closer and closer the eliminating that storage bottleneck in throughput. I hope I get to live to see the day non volatile storage devices exceed the speeds of that of current memory configurations today. Storage like direct memory access, but like 20 terabytes of it!!!
https://forums.guru3d.com/data/avatars/m/282/282473.jpg
there's not gonna be much difference for a home user unless random r/w improve significantly who the hell keeps copying those hundreds of GBs of data every day if those sequential times are so important ?
https://forums.guru3d.com/data/avatars/m/246/246171.jpg
DeskStar:

I hope I get to live to see the day non volatile storage devices exceed the speeds of that of current memory configurations today. Storage like direct memory access, but like 20 terabytes of it!!!
Well, that's kinda already happened with ReRAM, it just hasn't taken off yet since it's hugely complex to adjust an OS to work in such a way. Intel has also made NV DIMMs. Keep in mind too that much of what makes a NVMe drive so fast is its on-board cache. I get the impression we're still a ways away from having storage that's actually fast enough to keep up with speeds like 16GB/s without the cache.
https://forums.guru3d.com/data/avatars/m/258/258688.jpg
I'm sure it's "in the works"...but to me that translates to a long time from now for any commercial availability. First comes the adoption of PCie5 chipsets and motherboards, then come the PCie5 NVMe drives, etc. Intel so far cannot make a full PCie4 system bus, so I think it's going to be awhile. Absolute earliest I see for Zen4 PCie5 debut is 18 months from now--add on another six months for market penetration--then another six months before "real" PCIe5 NVMe devices become available at premium pricing and the bioses for the new breed of PCIe5 CPUs get through their teething stages. (Isn't new tech so much fun? *cough*)
https://forums.guru3d.com/data/avatars/m/56/56686.jpg
will all the extra speed come with even HIGHER temps for storage? they are already pushing GPU temptures with these 4.0 storage and i bet they will increase price of storage per tb too, which is already ridiculously
data/avatar/default/avatar18.webp
DeskStar:

Bad ass! Getting closer and closer the eliminating that storage bottleneck in throughput. I hope I get to live to see the day non volatile storage devices exceed the speeds of that of current memory configurations today. Storage like direct memory access, but like 20 terabytes of it!!!
In a world where all files are massive and sequential, which isn't the real world. If you compare 4KQ1T1 for SATA and PCIe 4.0 you are not seeing anything approaching the same percentage gains you see with sequential speed. The 870 EVO gets about 47MB/S 4KQ1T1 read while the 980 Pro gets about 90MB/S 4KQ1T1 read. That is a 2X multiplier even though PCIe 4.0 M.2 is 13 times faster than SATA 600 in potential. That will grow to 26 times the potential for PCIe 5.0 but small random reads will barely see any change at all. At some point we need something like Optane technology but with more NAND like costs and capacities to come along and address the issue with NAND and small random files.
https://forums.guru3d.com/data/avatars/m/232/232349.jpg
nosirrahx:

In a world where all files are massive and sequential, which isn't the real world. If you compare 4KQ1T1 for SATA and PCIe 4.0 you are not seeing anything approaching the same percentage gains you see with sequential speed. The 870 EVO gets about 47MB/S 4KQ1T1 read while the 980 Pro gets about 90MB/S 4KQ1T1 read. That is a 2X multiplier even though PCIe 4.0 M.2 is 13 times faster than SATA 600 in potential. That will grow to 26 times the potential for PCIe 5.0 but small random reads will barely see any change at all. At some point we need something like Optane technology but with more NAND like costs and capacities to come along and address the issue with NAND and small random files.
In a raid setup I can easily reach 4-600MB/s in small file transfers. Still a ways off, but better to say the least. Just awaiting the days where persistent storage is just as fast as today's memory is all. Technology has been moving at a great pace and I just hope I get to live to see the day where a serious break through is available to every day Joe's like myself. Yeah because I'm done slitting my wrists trying to make the technology come closer to myself early in the game.
data/avatar/default/avatar21.webp
DeskStar:

In a raid setup I can easily reach 4-600MB/s in small file transfers. Still a ways off, but better to say the least.
Under optimal conditions, try intentionally checking worst case. This has been tested to death, NAND is not good at moving small files one at a time. Look at this bench, this is on my quad Optane workstation. You can clearly see the optimal VS worst case numbers, this is roughly 3 times better than NAND: https://i.imgur.com/8sCtdLb.jpg I can pretend that I am getting nearly 3000MB/s in 4KQ1T1, but how often is that actually happening?