Samsung Unveils First UFS Removable Flash Cards, Up To 530 Mbps And 40K IOPS

[Guest]

Samsung unveiled its new UFS devices, which are designed to offer a blazing 530 MB/s of throughput and 40,000 random write IOPS for mobile devices.

Samsung Unveils First UFS Removable Flash Cards, Up To 530 Mbps And 40K IOPS : Read more

 

[User360]

What devices support it/ will support it?

 

[SimonSM24]

will Samsung S7 be compatable with those cards?

 

[3ogdy]

will Samsung S7 be compatable with those cards?

It's likely that it will. It WILL NOT be compatable, though, that's for sure. It would probably be compatible

 

[okenny]

Title should read MB/s not Mbps!

 

[PaulAlcorn]

Title should read MB/s not Mbps!

Good catch, fixed!

 

[paladinnz]

will Samsung S7 be compatable with those cards?

No, sadly not compatible, the S7 supports MicroSD, not UFS.

 

[rigurat]

Oh no, not another memory card standard coming in. It's like 2001 all over again.

 

[razor512]

It will need a new card slot, and thus for it to catch on, the slot must be standardized, with everyone being allowed to use the same slot without having to be worried about being used.

 

[dosmastrify]

how long before somebody tries to RAID 0 a bunch of them

 

[PaulAlcorn]

how long before somebody tries to RAID 0 a bunch of them

What a wonderful idea!

 

[bit_user]

how long before somebody tries to RAID 0 a bunch of them

What a wonderful idea!

I used to think about running RAID-5 of DVD-R's and USB flash drives, years ago, but this is a head scratcher.

These seem mostly targeted at tablets, phones, and laptops, that would have only one slot. But that's no different from DVD-R's.

Where I get stuck is, assuming you're not RAID'ing them just for capacity (in which case there are cheaper and simpler options), what would it take actually to get a performance boost by doing it? Would your SoC or PCH have any bottlenecks that limit the combined speed of the connected readers? On Haswell systems, I know the PCH would bottleneck even a RAID-0 of two SATA 3 SSDs. How much software overhead does reading/writing over USB 3.0/3.1 usually incur, and how much CPU headroom would be needed for the software RAID?

Anyway, I think we can say that RAIDing these for capacity is probably not a good move. I think it'd be more cost-effective either to simply use a bigger USB 3 flash drive or to put a SATA SSD in a USB 3 enclosure. For performance, I wonder if there aren't already USB 3.1 enclosures for NVMe drives, which would at least exceed the sequential throughput of a single UFS drive (while USB would cut down on random performance, any likely UFS RAID configuration would probably also rely on USB 3).

 

[bit_user]

Oh no, not another memory card standard coming in. It's like 2001 all over again.

At least it's a big enough improvement to justify the trouble.

 

[PaulAlcorn]

how long before somebody tries to RAID 0 a bunch of them

What a wonderful idea!

I used to think about running RAID-5 of DVD-R's and USB flash drives, years ago, but this is a head scratcher.

These seem mostly targeted at tablets, phones, and laptops, that would have only one slot. But that's no different from DVD-R's.

Where I get stuck is, assuming you're not RAID'ing them just for capacity (in which case there are cheaper and simpler options), what would it take actually to get a performance boost by doing it? Would your SoC or PCH have any bottlenecks that limit the combined speed of the connected readers? On Haswell systems, I know the PCH would bottleneck even a RAID-0 of two SATA 3 SSDs. How much software overhead does reading/writing over USB 3.0/3.1 usually incur, and how much CPU headroom would be needed for the software RAID?

Anyway, I think we can say that RAIDing these for capacity is probably not a good move. I think it'd be more cost-effective either to simply use a bigger USB 3 flash drive or to put a SATA SSD in a USB 3 enclosure. For performance, I wonder if there aren't already USB 3.1 enclosures for NVMe drives, which would at least exceed the sequential throughput of a single UFS drive (while USB would cut down on random performance, any likely UFS RAID configuration would probably also rely on USB 3).

I mentioned it in jest, though it would be interesting. You are right, card readers would likely encounter some serious roadblocks, which could be sidestepped by using a PCIe-based adaptor. CPU overhead for software RAID is a pain, but we have been able to surpass 3 million IOPS and 21 GB/s of throughput with PCIe-based storage, so that should be sufficient. http://www.tomshardware.com/reviews/intel-xeon-e5-2600-v4-broadwell-ep,4514-4.html

I totally agree that a large USB, or just SSD, is the best option. There is no sense in taking the risks associated with RAID 0, and RAID 5/6 is just a waste of capacity. R5/6 also incurs significantly more CPU overhead in any type of software implementation.

 

[bit_user]

I totally agree that a large USB, or just SSD, is the best option. There is no sense in taking the risks associated with RAID 0, and RAID 5/6 is just a waste of capacity. R5/6 also incurs significantly more CPU overhead in any type of software implementation.

In defense of DVD-R RAID-5's, there once was an undeniable cost advantage over just about anything else. I wonder whether video editors or anyone else ever actually did that.

 

[PaulAlcorn]

I totally agree that a large USB, or just SSD, is the best option. There is no sense in taking the risks associated with RAID 0, and RAID 5/6 is just a waste of capacity. R5/6 also incurs significantly more CPU overhead in any type of software implementation.

In defense of DVD-R RAID-5's, there once was an undeniable cost advantage over just about anything else. I wonder whether video editors or anyone else ever actually did that.

Honestly, I did not consider optical, and there is merit to that. If we examine Facebook and Panasonic's Blu-Ray initiatives they really deliver the best archival costs, or at least the promise of the best cost.

Here is a news post I wrote about that a while ago. Facebook actually has an entire datacenter dedicated solely to Blu-Ray archival.

http://www.tomsitpro.com/articles/panasonic-bluray-storage-archive-data,1-3097.html

 

[bit_user]

here is a news post I wrote about that a while ago. Facebook actually has an entire datacenter dedicated solely to Blu-Ray archival.

http://www.tomsitpro.com/articles/panasonic-bluray-storage-archive-data,1-3097.html

Wow, thanks!

I don't know anyone else who has a Blu-ray burner, but I still use them to back up important stuff and offload some media that I don't want to delete. I burned some files on one for a friend, but it turns out he hasn't even a reader, in his PC.

I burned over 1 TB worth of DVD-Rs, back in the day. Always used top quality burners and media, for best longevity.

 

[dosmastr]

how long before somebody tries to RAID 0 a bunch of them

What a wonderful idea!

I used to think about running RAID-5 of DVD-R's and USB flash drives, years ago, but this is a head scratcher.

These seem mostly targeted at tablets, phones, and laptops, that would have only one slot. But that's no different from DVD-R's.

Where I get stuck is, assuming you're not RAID'ing them just for capacity (in which case there are cheaper and simpler options), what would it take actually to get a performance boost by doing it? Would your SoC or PCH have any bottlenecks that limit the combined speed of the connected readers? On Haswell systems, I know the PCH would bottleneck even a RAID-0 of two SATA 3 SSDs. How much software overhead does reading/writing over USB 3.0/3.1 usually incur, and how much CPU headroom would be needed for the software RAID?

Anyway, I think we can say that RAIDing these for capacity is probably not a good move. I think it'd be more cost-effective either to simply use a bigger USB 3 flash drive or to put a SATA SSD in a USB 3 enclosure. For performance, I wonder if there aren't already USB 3.1 enclosures for NVMe drives, which would at least exceed the sequential throughput of a single UFS drive (while USB would cut down on random performance, any likely UFS RAID configuration would probably also rely on USB 3).

I mentioned it in jest, though it would be interesting. You are right, card readers would likely encounter some serious roadblocks, which could be sidestepped by using a PCIe-based adaptor. CPU overhead for software RAID is a pain, but we have been able to surpass 3 million IOPS and 21 GB/s of throughput with PCIe-based storage, so that should be sufficient. http://www.tomshardware.com/reviews/intel-xeon-e5-2600-v4-broadwell-ep,4514-4.html

I totally agree that a large USB, or just SSD, is the best option. There is no sense in taking the risks associated with RAID 0, and RAID 5/6 is just a waste of capacity. R5/6 also incurs significantly more CPU overhead in any type of software implementation.

I was not kidding, I remember people on here doing RAID 5 on usb flash drives before ssd's were a thing. Highly doubt it was worth the trouble, (maybe for iops?) but while I won't do it, I'm sure someone else will.

 

[dosmastr]

But I think it's a pretty funny idea.

 

[PaulAlcorn]

how long before somebody tries to RAID 0 a bunch of them

What a wonderful idea!

I used to think about running RAID-5 of DVD-R's and USB flash drives, years ago, but this is a head scratcher.

These seem mostly targeted at tablets, phones, and laptops, that would have only one slot. But that's no different from DVD-R's.

Where I get stuck is, assuming you're not RAID'ing them just for capacity (in which case there are cheaper and simpler options), what would it take actually to get a performance boost by doing it? Would your SoC or PCH have any bottlenecks that limit the combined speed of the connected readers? On Haswell systems, I know the PCH would bottleneck even a RAID-0 of two SATA 3 SSDs. How much software overhead does reading/writing over USB 3.0/3.1 usually incur, and how much CPU headroom would be needed for the software RAID?

Anyway, I think we can say that RAIDing these for capacity is probably not a good move. I think it'd be more cost-effective either to simply use a bigger USB 3 flash drive or to put a SATA SSD in a USB 3 enclosure. For performance, I wonder if there aren't already USB 3.1 enclosures for NVMe drives, which would at least exceed the sequential throughput of a single UFS drive (while USB would cut down on random performance, any likely UFS RAID configuration would probably also rely on USB 3).

I mentioned it in jest, though it would be interesting. You are right, card readers would likely encounter some serious roadblocks, which could be sidestepped by using a PCIe-based adaptor. CPU overhead for software RAID is a pain, but we have been able to surpass 3 million IOPS and 21 GB/s of throughput with PCIe-based storage, so that should be sufficient. http://www.tomshardware.com/reviews/intel-xeon-e5-2600-v4-broadwell-ep,4514-4.html

I totally agree that a large USB, or just SSD, is the best option. There is no sense in taking the risks associated with RAID 0, and RAID 5/6 is just a waste of capacity. R5/6 also incurs significantly more CPU overhead in any type of software implementation.

I was not kidding, I remember people on here doing RAID 5 on usb flash drives before ssd's were a thing. Highly doubt it was worth the trouble, (maybe for iops?) but while I won't do it, I'm sure someone else will.

Oh, someone will do it. It doesn't matter if it makes sense or not, I would still like to see it as well. There is always a cool factor to things like that.

 

[bit_user]

I was not kidding, I remember people on here doing RAID 5 on usb flash drives before ssd's were a thing. Highly doubt it was worth the trouble, (maybe for iops?) but while I won't do it, I'm sure someone else will.

Oh, someone will do it. It doesn't matter if it makes sense or not, I would still like to see it as well. There is always a cool factor to things like that.

Imagine using RAID-5 or RAID-6 to distribute sensitive data across different hiding places, or as a resiliency mechanism for sending data over the postal mail. There could be practical uses for RAIDing removable media other than simply increasing capacity or performance.

 

[JTWrenn]

Wait...a brand new interconnect...and it only supports the speeds we can make right now? That seems silly. Why not make something with a little bit of....oh....ahhh I see what you did there....capitalism at it's finest.

 

[bit_user]

Wait...a brand new interconnect...and it only supports the speeds we can make right now? That seems silly. Why not make something with a little bit of....oh....ahhh I see what you did there....capitalism at it's finest.

Geez, cynical much?

This is for low-power, mobile applications. There might be good reasons not to make it faster, at least for the time being. And perhaps they can scale it up in a backward-compatible way, if/when needed.

IMO, the protocol enhancements mentioned in the article are reason enough to justify a new standard, not to mention the fact that it's 2-4x the speed of UHS-II.