Phison announces new USB SSD controller

[1_rick]

https://www.anandtech.com/show/16425/phison-at-ces-2021-new-usb-and-nvme-ssd-controllers

Not a USB-SATA or USB-NVMe bridge, an actual direct-to-flash controller.

"For portable SSDs, Phison is introducing the U17 and U18 controllers. The U17 uses a USB 3.2 Gen 2x1 (10Gb/s) host interface and a two-channel NAND interface running at up to 1200 MT/s. The U18 doubles these: USB 3.2 Gen2x2 (20Gb/s) and a four-channel NAND interface. The performance specs may look lackluster compared to Phison's NVMe SSD controllers, but they are pretty close to saturating what their respective USB host interfaces can handle, and performance will be competitive with NVMe+USB bridge based portable SSDs. However, the U17 and U18 will have a significant power efficiency advantage, lower cost and smaller PCB footprint than existing portable SSDs. Phison will also be providing TCG Opal encryption support on the U18, enabling a level of security they say is impossible to achieve with NVMe+USB bridge solutions."

 

[Deleted member 89018]

Neat!

 

[KazeoHin]

I was pretty underwhelmed when I dropped a basic 1500MBps NVM-E drive into a Simplecom NVM-E USB3 enclosure and was only getting around ~300MBPs.

I would not mind having something that could saturate the 5Gbps (~600MBps) bus.

 

[Format _C:]

As long as they don't do a Western Digital like some of their external drives no SATA/USB board the USB was the same board as the drives control board.

 

[UnknownSouljer]

As long as they don't do a Western Digital like some of their external drives no SATA/USB board the USB was the same board as the drives control board.

I think you misunderstand the point of this controller because that's exactly what this is.
It will be just flash memory attatched to this USB controller. As mentioned in the article the great advantage of doing things this way is saturation of USBs limited bandwidth (as opposed to overkill - and having a controller that is optimized), cost, and size. This is basically a net win. It'll be great when we can get flash storage that's 10Gb/s, 2-4TB in size, and the size of a current USB-stick (it doesn't actually mention total size, but we can dream) while costing a fraction as much as current solutions. If you don't get why this is amazingly beneficial despite not being able to be shucked, then I don't know what to tell you. I have use cases already that could benefit from this.

 

[Deleted member 89018]

I’m thinking this would be great for an ssd for game consoles.

 

[Nside]

Eventually I hope USB will be seen as the redundant and limited protocol that it is, much like SATA, and it takes a backseat to PCIe. Since PCIe will by-far be the fastest connection to the CPU for the foreseeable future, everything performance dependent should just use that, and a standardized external connector should be made mainstream.

Not saying USB should die, but it has become more convoluted than anything with "Universal" in its name should ever be. As a basic connection to a PC for stuff like a mouse and keyboard, it's fine. But for anything that is performance related, the average joe has to go to Wikipedia to try to figure out what the differences are between the multitude of color-coded ports, and then to the user-manual of their devices to make sure they're actually capable.

Ever since the revised numbering scheme of USB was adopted, I've seen it as nothing more than trying to be the jack of all trades from Laptop-charging to external data storage, and it just can't be good at all these things, not universally... Even with the expensive type-C ports you see on laptops and various mobile devices that do seem to handle most of what USB is capable of, that's only 1/3 of the equation, Can the cable handle what you are trying to do? Can the device on the other end of the cable handle what you are trying to do? etc... It's too convoluted, and flies in the face of what USB was originally meant to be... Let USB connect input devices like M/KB, even basic thumb-drives. But give those people who want actual standardized performance without a bunch of head-scratching, research, and trial-&-error an option to just tie into the PCIe bus externally.

 

[bonehead123]

Well, I've neva seen the point using an external enclosure that has a USB bridge chip, and sticking an nvme drive in it, which will only give you normal USB 2/3 performance anyways....

Perhaps this product will finally provide a solution to that issue

 

[Format _C:]

I think you misunderstand the point of this controller because that's exactly what this is.
It will be just flash memory attatched to this USB controller. As mentioned in the article the great advantage of doing things this way is saturation of USBs limited bandwidth (as opposed to overkill - and having a controller that is optimized), cost, and size. This is basically a net win. It'll be great when we can get flash storage that's 10Gb/s, 2-4TB in size, and the size of a current USB-stick (it doesn't actually mention total size, but we can dream) while costing a fraction as much as current solutions. If you don't get why this is amazingly beneficial despite not being able to be shucked, then I don't know what to tell you. I have use cases already that could benefit from this.

Sorry I was attempting to be sarcastic as I had a friend of mine damage the poor designed crap know as Micro USB 3.0 connector so my attempt to get his files back were to take the drive out of it's casing and put it into another case but there was no way as the USB board was also the drives control board! luckily he had backups (not all people do this! I do for my important stuff but movies and music I can't afford more 14 TB drives so important stuff it is pictures and documents)

 

[UnknownSouljer]

Eventually I hope USB will be seen as the redundant and limited protocol that it is, much like SATA, and it takes a backseat to PCIe. Since PCIe will by-far be the fastest connection to the CPU for the foreseeable future, everything performance dependent should just use that, and a standardized external connector should be made mainstream.

SATA/SAS won't die until rotational HD's die. And the only way that is going to happen is if/when flash memory becomes cheaper, more dense, and more plentiful than magnetic media. I don't think that there will be a cost effect 10TB NVME flash drive anytime soon, let alone close to the current top 18TB rotational drives.
The other major issue of course is PCI-E lanes. Right now an SAS controller can easily slap in 24+ drives into a single PCIE slot. Trying to have 24 flash drives go into that many lanes is a lot more difficult. This is especially true as expansion goes up. A single server with SAS expansions could be in the 100s of drives. There simply aren't enough PCIE lanes supported on any board to move all of that to NVME/PCIE flash storage (even if flash storage was cheap).
The only way around that of course would be more servers. So the cost of storage goes up not only because flash media is more expensive, but also because more servers are necessary for the same amount of drives. Those are important and significant hurdles.

This just goes into the idea and concept that what is best isn't necessarily what is the fastest. There is still a lot of use for these "old protocols". It's the same reason why someone might want a Golf TDI over a Ferrari F458. Sure one is faster, more powerful, handles better, and in theory gets you from A to B faster, but it also costs 5x as much and burns through way more gas and is wildly less practical for most people trying to move more than 2 people with any amount of cargo. I don't foresee SATA/SAS dying any time soon.

Not saying USB should die, but it has become more convoluted than anything with "Universal" in its name should ever be. As a basic connection to a PC for stuff like a mouse and keyboard, it's fine. But for anything that is performance related, the average joe has to go to Wikipedia to try to figure out what the differences are between the multitude of color-coded ports, and then to the user-manual of their devices to make sure they're actually capable.

Ever since the revised numbering scheme of USB was adopted, I've seen it as nothing more than trying to be the jack of all trades from Laptop-charging to external data storage, and it just can't be good at all these things, not universally... Even with the expensive type-C ports you see on laptops and various mobile devices that do seem to handle most of what USB is capable of, that's only 1/3 of the equation, Can the cable handle what you are trying to do? Can the device on the other end of the cable handle what you are trying to do? etc... It's too convoluted, and flies in the face of what USB was originally meant to be... Let USB connect input devices like M/KB, even basic thumb-drives. But give those people who want actual standardized performance without a bunch of head-scratching, research, and trial-&-error an option to just tie into the PCIe bus externally.

In theory it's on its way out. USB4 has effectively/finally merged TB3 and USB3 into the same spec. TB is essentially PCIE. Apple is way ahead of the curve on this as essentially they've been using TB3/USB3.1 connectors for the past 4.5 years. PC vendors have been incredibly slow to adopt Thunderbolt in any capacity. In fact if I wanted a Z370/Z390 board as an example, I can only think of one that had access to a single Thunderbolt port, which was the Gigabyte Designare. PC side is doing a little bit better on laptops, Razor and Lenovo as an example have both put ONE(!) TB port on most of their laptops. Again, now that USB4 is a thing, it's more likely that PC manufacturers will start making the ports more plentiful because saying you support USB4 will become part of the spec war, where it wasn't that way before.

 

[RanceJustice]

USB4 is still more complex than I'd like looking at its Wiki page on the spec but its a major step forward. If every new chipset can just target USB4 40gbps with all "optional" features, then we'll be in business - compatible with Thunderbolt 4, highest bandwidth and backward compatibility regardless, all power delivery and display options (including Display Port 2.0! , and USB-C connector. I'd love a future where that was the norm on the next gen of motherboards and those with older hardware but PCI-E slots/lanes available could buy comprehensive PCI-E > USB4 40gbps (ideally multi-port) USB-C cards.

 

[Nside]

SATA/SAS won't die until rotational HD's die. And the only way that is going to happen is if/when flash memory becomes cheaper, more dense, and more plentiful than magnetic media. I don't think that there will be a cost effect 10TB NVME flash drive anytime soon, let alone close to the current top 18TB rotational drives.
The other major issue of course is PCI-E lanes. Right now an SAS controller can easily slap in 24+ drives into a single PCIE slot. Trying to have 24 flash drives go into that many lanes is a lot more difficult. This is especially true as expansion goes up. A single server with SAS expansions could be in the 100s of drives. There simply aren't enough PCIE lanes supported on any board to move all of that to NVME/PCIE flash storage (even if flash storage was cheap).
The only way around that of course would be more servers. So the cost of storage goes up not only because flash media is more expensive, but also because more servers are necessary for the same amount of drives. Those are important and significant hurdles.

This just goes into the idea and concept that what is best isn't necessarily what is the fastest. There is still a lot of use for these "old protocols". It's the same reason why someone might want a Golf TDI over a Ferrari F458. Sure one is faster, more powerful, handles better, and in theory gets you from A to B faster, but it also costs 5x as much and burns through way more gas and is wildly less practical for most people trying to move more than 2 people with any amount of cargo. I don't foresee SATA/SAS dying any time soon.


In theory it's on its way out. USB4 has effectively/finally merged TB3 and USB3 into the same spec. TB is essentially PCIE. Apple is way ahead of the curve on this as essentially they've been using TB3/USB3.1 connectors for the past 4.5 years. PC vendors have been incredibly slow to adopt Thunderbolt in any capacity. In fact if I wanted a Z370/Z390 board as an example, I can only think of one that had access to a single Thunderbolt port, which was the Gigabyte Designare. PC side is doing a little bit better on laptops, Razor and Lenovo as an example have both put ONE(!) TB port on most of their laptops. Again, now that USB4 is a thing, it's more likely that PC manufacturers will start making the ports more plentiful because saying you support USB4 will become part of the spec war, where it wasn't that way before.

Not only did I NOT say SATA was dying, I specifically said "I'm not saying USB should die" in reference to what it should become. SATA will never see another significant update, and it's because of PCIe being so much better at handling throughput, that's my point. USB is too convoluted to keep trying to be everything. SATA couldn't compete with throughput, neither can USB.

As for limited lanes, I also said "eventually" ... Processors can easily have more lanes, just look at Epyc or Xeon. Most USB bandwidth has to use shared PCIe lanes anyway. There's only a handful of the ports on the back of any given motherboard that go directly to the CPU, the rest go through the chipset. If you devoted less (or none) space dedicated to USB on the I/O portion of the processor, you could dedicate that space to PCIe lanes instead.

I know it's in some people's nature to be contrarian, but at least address the things I said, not something I didn't say.