SD Association Unveils microSD Express Format That Promises Transfer Speeds of Up To 985 MB/s (engadget.com) 72
The SD Association has unveiled microSD Express, a new format that will bring speeds of up to 985 MB/s to the tiny memory cards used in smartphones and other devices. From a report: Like SD Express, it exploits the NVMe 1.3 and PCIe 3.1 interfaces used in PCs to power high-speed SSDs. The tech is incorporated onto the second row of microSD pins, so the cards will work faster in next-gen devices while maintaining backward compatibility with current microSD tech. PCIe 3.1 allows for low power sub-states, so the cards will not only offer much (much) higher transfer speeds, but consume less power than regular microSD cards. It'll also open up features like bus mastering, which lets memory cards communicate with other components without going through the CPU first.
Countdown to... (Score:5, Insightful)
... bus mastering being used in an Intel processor exploit in 10, 9, 8 ...
Re:Countdown to... (Score:5, Insightful)
This was my first thought as well. It seems to me extending to the PCIe bus to all kinds of untrusted hot plugged devices has bad idea written all over it. USB 2.x we "permissive" enough in terms of memory access.
Re: (Score:3)
Well, those that prefer speed over security will at least not have security. The funny thing is that these morons are the ones to complain loudest when they get hit because of their own stupidity...
Re:Countdown to... (Score:4, Interesting)
We already did, twice. First was ExpressCard, which is a card version of a x1 PCIe slot. Second time was Thunderbolt, which I believe the current iteration is up to x4 PCIe.
And yes, I believe there are Thunderbolt RAM attacks though because of the IO controller, it's somewhat mitigated.
Re: (Score:2)
I don't know if there were actual exploits, but FireWire absolutely had direct memory access.
As bad ideas go, minimally bad (Score:2)
I am personally OK with the tradeoff of much better local performance, for an increased security risk around physical presence.
After all, a hardware maker can do things to make sure ports are disconnected when systems are locked, or in the most drastic cases you can physically render external ports inoperable.
Re: (Score:2)
One, if the hardware maker is not the OS maker, they can only provide hooks and leave it up to the OS. Two, I don't want all my storage unmounting every time I walk away from my desk.
Should be possible to do something (Score:2)
One, if the hardware maker is not the OS maker, they can only provide hooks and leave it up to the OS.
For desktops that is probably true, although they could have a "lock system" button... you could put it right next to the "Turbo" button some systems used to offer. :-).
For laptops though, the hardware maker could easily have some kind of physical interlock that disabled anything but power (or even that) to outside ports until the case was opened. The problem there of course, is people that want to run la
Re: (Score:2)
if there's already a storage device mounted, there's very little reason to unmount if the system is locked. It's already been mounted and had a chance to do it's bad deeds if it's bad.
He's talking about the idea of not enumerating newly plugged devices if the system is locked, which is 100% in the OS ballpark. I don't want the hardware maker involved in that decision at all, or else the hardware starts doing shit the OS doesn't know about, and a patch to the OS can't fix. That's where bad security proble
Re: (Score:2)
It seems to me extending to the PCIe bus to all kinds of untrusted hot plugged devices has bad idea written all over it.
Do you bolt your TV down to your house, or do you lock the front door? I agree that there are security implications with exposing a bus like this, but none that can't be managed externally.
But in reality how is this any worse at all than Thunderbolt or ExpressCard?
Re: (Score:2)
... bus mastering being used in an Intel processor exploit in 10, 9, 8 ...
If you buy a modern system without a IOMMU, you deserve exactly what you get. We figured out that this was a problem back in the firewire days, and first servers and then desktops sprouted IOMMUs to solve it.
Re: (Score:2)
Don't worry Intel is immune. They don't provide enough PCIe lanes to attach useful peripherals.
Re: (Score:2)
For this reason, Windows now has IOMMU virtualization enabled to prevent DMA attacks (starting with Windows 10 RS4/1803/April 2018 Update): https://twitter.com/AmarSaar/status/985618204184768513 [twitter.com]
In conjunction, tianocore also has IOMMU based DMA protection for 2 years now: https://github.com/tianocore/edk2/tree/master/IntelSiliconPkg/Feature/VTd [github.com]. So even if the OS isn't up yet DMA attacks are still locked out.
Assuming you are running a recent OS and firmware, this is now a non-issue.
Re: (Score:1)
W.C. who? Fields? I never heard him say that!
Re: (Score:2)
It was said in a water closet.
Re: (Score:1)
the philosophy of failure, the creed of ignorance, and the gospel of envy
Also known as MAGA?
MB/s not Mb/s (Score:5, Informative)
microSD Express format supports up to 985 MB/s not 985 Mb/s.
MB/s is megabytes (1,000,000 bytes) per second.
Mb/s is megabits (1,000,000 bits) per second.
References:
https://www.sdcard.org/press/T... [sdcard.org]
Re: (Score:1)
Re: (Score:1)
Some people are more comfortable with that than others.
Just why is 10 Gbps still so expensive? (Score:2)
I mean, what is has it been -- 5 years at least that 10GBase-T has been out, and it's still expensive?
Are network gear makers just charging a premium because there's still a lot of business/enterprise upgrading to be done or because they don't have anything else "premium" to offer once 1 Gbit becomes as obsolete as 100 Mbit?
Or is it just the industry not bothering to mass produce it because 1 Gbps is like 640k, it ought to be enough?
Re: (Score:2)
Because nobody's buying it. Economies of scale don't kick in until you have scale. And not just a bunch of R&D costs.
Re: (Score:2)
Chicken and egg, then? Odd that 90% of the desktop use cases didn't need 1 Gbps, either, but the economies of scale made sense to bake it into the chipsets anyway which drove down costs.
The parts costs for 10Gb must be high enough that razor-margin industries just don't feel pressed to step it up vs. paid-for commodity 1 Gbit parts.
Re: (Score:2)
The big problem with 10gbps is that it's too fast to handle with more than a few feet of copper, and fiber is a BITCH to terminate.
The other problem with 10gbps is that it's simultaneously too fast AND too slow.
From the perspective of a single endpoint device (like a computer), 10gbps is absurdly fast for Ethernet... but the two things you might actually WANT to wrap up and transmit over Ethernet -- HDMI and multi-lane PCI Express data -- blow past 10gbps and keep going without looking back.
So... as a trans
Re: (Score:3)
The big problem with 10gbps is that it's too fast to handle with more than a few feet of copper, and fiber is a BITCH to terminate.
Maybe its misinformation? The spec lists it at 100 meters over Cat-6a and 50 meters over Cat-6. I've had zero problems with it using Cat-6 at ordinary data center lengths up to about 10 meters.
It's advantage in many environments isn't just the added speed -- it adds a ton of performance for even rotational media SAN, but the fact that its running on a 10x faster clock, cutting latency times as well, even for applications which aren't exceeding 1 Gbit throughput limitations. As for too slow, well, you ca
Re: (Score:2)
When most non-business network traffic is WAN-link limited by a shitty ISP that can't even do gigabit, 95% of non-business customers don't need more than gigabit.
There's still a lot of rotating-rust drives out there that can barely saturate 100Mbps. 10GbE and higher are still the domain of the server room, so volumes are low; thus price is high.
Re: (Score:2)
Technically a megabit is 1,048,576 bits but for marketing use they like to use 10 to the 6th power to represent a megabit (megabyte would be 1,048,576 bytes). I have no idea if they are using the true megabit in this case or the marketing megabit though.
Re: (Score:2)
Sorry I forgot they call those Mebibits now.
Re: (Score:2)
Networking speeds have always been measured in base 10.
limits (Score:5, Interesting)
Re: (Score:2)
Why is MicroSD limited to a single chip? Is this a physicals space limitation?
Re: (Score:3)
There are some cards with more than a single flash chip inside them, but they are always full size sdcards.
Here's an example.
https://upload.wikimedia.org/w... [wikimedia.org]
Most will only be a single chip though.
The big bottleneck is the controller itself, which manipulates the flash. SDCard uses a serial protocol, not a parallel data IO direct to the flash chip. The flash chip could be hella fast, but if there is a cheap and slow controller driving it.. That's like putting an SSD on a SATA I interface.
Re:limits (Score:5, Interesting)
Until they begin making (micro)SD cards out of 3DxPoint, ReRam, Phase Changing RAM or Mermistors...
You see, the (micro)SD format is not tied to Flash, therefore, the need to future-proof the bus...
Re: (Score:2)
It's hilarious that you think any iPhone ever will be able to interact with external storage of any kind.
Re: (Score:2)
It's hilarious that's you haven't seen the iXpand drives on the flash drive shelf at Walmart, where normal non-techies buy them for everyday use.
Re: (Score:2)
Ok, so the "of any kind" was inaccurate. However, having that thing hanging off your lightning port isn't exactly convenient if you were to use it for every-day storage like you can with an SD card equipped device.
Not that Google has been doing Android any favors with how Android 8 handles SD cards, but you know what I mean.
Hang on, this test Android device on my desk has told me that external storage has stopped responding again...
Re:limits (Score:4, Informative)
Ignore the OP. He's buying cheap Chinese shit. Most reputable cards will happily max out the SD card's current bus for a sustained write across their entire capacity and despite his assertion that NAND is the limiting factor to 7MB/s you'll find most SSDs have either 2 or 4 NAND chips on them and happily crank out several gigabytes per second of data.
Re: (Score:2)
This F-Stoppers video shows some 1.5GB transfers with times on various cards and readers. 88MB/sec write on a 95MB/sec claimed write speed.
https://youtu.be/ZlWhvc-UCOA?t=500
Re: (Score:2)
NVMe on a single chip shitNAND? lol. this is pure marketing bullshit.
Hmmm I have only 2 chips on my NVMe SSD and I can do 3400MB/s so as usual there is more to technology than counting the number of black things on the bigger thing that is plugged into the other thing.
And I'm genuinely surprised. I think I have a 256GB SD card with a 256GB controller buffer on it. Who knew. After all I can sustain 88MB/s on my SSD card just fine when copying massive amounts of data to and from it. I guess all those people who actually record 4K footage also have those magical SD cards you've
IOPS (Score:1)
It would be interesting to know the IOPS of the new cards, SD card latency is traditionally a fraction of embedded flash.
Re: (Score:2)
I don't know about these new cards but "A1" and "A2" cards have higher IOPS than others. SanDisk Extreme microSD Cards with A2 [anandtech.com] IOPS are at least 4000 read and 2000 write.
Bus mastering microsd? (Score:1)
Where nothing can possiblye go wrong!
Re: (Score:2)
"Possibly" go wrong.
Huh. That's the first thing that's ever gone wrong.
Re: (Score:1)
wear leveling is indeed part of the solution.
The other, is assuring that write operations are efficient. If you write 1MB of data as a barrage of 512 byte sectors (such as with some variants of NTFS), you will burn the card up very quickly, because the controller inside the card natively writes a much larger chunk than 512 bytes. (often closer to 2 or 4 MB, depending on the card!)
Peppering the drive with shitloads of 512 byte writes causes the card to overwrite 2 to 4 MB of flash cells EACH TIME YOU WRITE.
T
High-Endurance SD cards. (Score:2)
I want an SD card that lasts long and fails gracefully
Go for a "High Endurance" card. They're designed and marketed for use in dash-cams and other loop-recording devices, where constant heavy writing is expected to occur.
Transcend was formerly the gold standard for these, although it now looks like Samsung is the current top performer.
Larger cards (Score:1)
Am I the only one who would want larger cards in exchange for similar speeds to desktop storage? Something about the size of Compact Flash, but with the ability to work as similar to an actual SSD found on a desktop or laptop. 985 Mb/s is quite slow in comparison to the 500 MB/s we have with desktop storage media. Maybe the power requirements are too high or there are other reasons it won't work. I'd love to be able to take a standard M.2 drive and stick in my phone or camera, even if we have to increase
Re: (Score:1)
Not sure if the article was updated, but it's 985 MB/s, so almost double the 500 MB/s you listed for desktop storage. So the small size should be fine for speed.
Re:Larger cards (Score:4, Informative)
The speed is actually 985 MB/s not 985 Mb/s. The article was wrong. See https://www.sdcard.org/press/T... [sdcard.org]
985MB/s Sustainable Into or Out of the Card? (Score:3)
When you see "985MB/s transfer speeds", I suspect that you're assuming that the card can read and write data at this speed all day long.
But, I suspect that there are limits in terms of writing and accessing data. I'm sure burst speeds of 985MB/s is possible (with longer read bursts than write) but the overall/average speed will probably be 20-50MB/s, which is still very good, but not what you're being lead to believe.
Re: (Score:3)
Re: (Score:3)
Interestingly the heat throttling of modern NVMe SSDs has nothing to do with the memory and everything to do with the controller. I would highly recommend a heatsink, and when you get the heatsink you want to apply the pad so it touches only the controller. NAND works better when it's hot which is why it has a minimum temperature rating.