Fastest Micro SD card for Pi 4? Does it support XC U3?

[morphious66]

Hello community, I'm Ready to order a Pi 4 and want to order the fastest Mcro SD card currently available.

Does PI 4 support the new Class 10 XC I U3 micro SD cards?

Will it take advantage of the speed increase over the U1?

Want to order the best / fastest micro SD. Looking at the https://www.samsung.com/us/computing/me ... e128ga-am/

-D

[lb]

What matters most is the random access performance (IOPS). So get an A1 or A2 class card if you want good performance. Apart from that, little else matters as only up to DDR50 transfer mode is supported. The Samsung card you posted does not have an A1 or A2 certification, so I wouldn't recommend it.

Note: A2 cards require special support by the host controller for best performance. I am not sure if that is implemented for the new SD controller in the Pi 4. So A1 may be the safer (and cheaper) choice.

[morphious66]

ok, understand - Yes this spec is Much better / faster then the Samsung card I had above.

- DDR50(max bandwidth: 50MB/s)

- A2 category: Random Read: 4000 IOPS - Random Write: 2000 IOPS

Looks like the best that is currently available is the Sandisk https://www.sandisk.com/home/memory-car ... me-microsd & on Amazon https://www.amazon.com/SanDisk-Extreme- ... 2313&psc=1

"Note: A2 cards require special support by the host controller for best performance. I am not sure if that is implemented for the new SD controller in the Pi 4. So A1 may be the safer (and cheaper) choice."

Would Love to know if it is, Admin or anyone else might know if this is the case?

[morphious66]

I do plan to Overclock this card as I previously did with my Pi 3 - I was about to double the bus speed to 100mghz and get 31.66 MB/sec Read speeds. From this thread - https://www.raspberrypi.org/forums/view ... 3&t=140461

I hope that I can do the same with the Pi 4 and get in the 75 - 100 MB/sec range.


Can some one with a PI 4 go to the command line interface and run the command to test the MHZ speed on the MicroSD bus

Mod: removed insecure link

Then can you place the results here?

[lb]

No, you cannot overclock the SDHC interface on the Pi 4 (at least not easily). It's also not recommended or required.

[morphious66]

No, you cannot overclock the SDHC interface on the Pi 4 (at least not easily). It's also not recommended or required.

I was told that before....... didn't stop me then from doing it with zero issues to date,. Outcome was a Much faster responding Pi 3.

My previous results:


Mod: removed insecure link
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 2227 100 2227 0 0 1296 0 0:00:01 0:00:01 --:--:-- 1297

Raspberry Pi Dramble microSD benchmarks
microSD clock: 100.000 MHz

Running hdparm test...

/dev/mmcblk0:
Timing buffered disk reads: 96 MB in 3.03 seconds = 31.66 MB/sec

Running dd test...

51200+0 records in
51200+0 records out
419430400 bytes (419 MB) copied, 69.3182 s, 6.1 MB/s

Running iozone test...
Iozone: Performance Test of File I/O
Version $Revision: 3.434 $
Compiled for 32 bit mode.
Build: linux-arm

Run began: Tue Jul 2 09:01:45 2019

Include fsync in write timing
O_DIRECT feature enabled
Auto Mode
File size set to 102400 kB
Record Size 4 kB
Command line used: ./iozone -e -I -a -s 100M -r 4k -i 0 -i 1 -i 2
Output is in kBytes/sec
Time Resolution = 0.000001 seconds.
Processor cache size set to 1024 kBytes.
Processor cache line size set to 32 bytes.
File stride size set to 17 * record size.
random random bkwd record stride
kB reclen write rewrite read reread read write read rewrite read fwrite frewrite fread freread
102400 4 1392 1822 6237 6287 5652 984

iozone test complete.

microSD card benchmark complete!

[alphixt]

Commenting here to throw in what I've attempted to gather and the responses I've received. Mostly due to trying to obtain relatively the same information a few days ago both from a independent post and responding in the mega thread quoting someone with similar questions.

My thread:
https://www.raspberrypi.org/forums/view ... p?t=243615

Mega Post... post:
https://www.raspberrypi.org/forums/view ... 2#p1486832


Good luck, it feels as if it is some sort of secret that we cannot know! (Jokes) not really, it feels that way.

[Gomoto]

Perhaps this is also an option to consider:

The fast sd card is 19 bucks. You get SSD drives new for around 15 bucks. An SATA USB adapter for around 8 bucks. Boot from sd card and put your root filesystem on ssd. If you do not mind the space. Performance is eye opening. Just did this on my new Pi 4 with an unused 64GB ssd I had lying around. I am amazed by the results.

[Andyroo]

Do not run A2 cards on a Pi.

The Pi does not have the correct hardware to use the A2 config and it’s a waste of money. Reports on the forum starting e these result in slower than A1 speeds

[morphious66]

Commenting here to throw in what I've attempted to gather and the responses I've received. Mostly due to trying to obtain relatively the same information a few days ago both from a independent post and responding in the mega thread quoting someone with similar questions.

My thread:
https://www.raspberrypi.org/forums/view ... p?t=243615

Mega Post... post:
https://www.raspberrypi.org/forums/view ... 2#p1486832


Good luck, it feels as if it is some sort of secret that we cannot know! (Jokes) not really, it feels that way.

I agree with you, I just had my links removed that test the speed, saying its "unsecure"

I came to the same conclusion: Does A2 work now, will it work in a future firmware version or will it work past DDR50 with a Overclock method.

[morphious66]

Do not run A2 cards on a Pi.

The Pi does not have the correct hardware to use the A2 config and it’s a waste of money. Reports on the forum starting e these result in slower than A1 speeds

Can you share the links and or the test results showing this please?

[morphious66]

Perhaps this is also an option to consider:

The fats sd card is 19 bucks. You get SSD drives new for around 15 bucks. An SATA USB adapter for around 8 bucks. Boot from sd card and put your root filesystem on ssd. If you do not mind the space. Performance is eye opening. Just did this on my new Pi 4 with an unused 64GB ssd I had lying around. I am amazed by the results.

Sorry for my use this has to be all inclusive and I cannot have a 3rd party device hanging off the side of it.

[alphixt]

Perhaps this is also an option to consider:

The fats sd card is 19 bucks. You get SSD drives new for around 15 bucks. An SATA USB adapter for around 8 bucks. Boot from sd card and put your root filesystem on ssd. If you do not mind the space. Performance is eye opening. Just did this on my new Pi 4 with an unused 64GB ssd I had lying around. I am amazed by the results.

Sorry for my use this has to be all inclusive and I cannot have a 3rd party device hanging off the side of it.

Same use case as me then. Trying to optimize and get some official word as to exactly what is and is not supported in regards to "reduced the bottleneck". Also as to what the future firmware could hold (ie, A2 support). But, getting this information seems harder than it should be.

I just want information on the SD Card slot, that's all.

[Moonmarch]

Here is what I saw on the Canakit website, the RPI4 complete starter kit includes the Sandisk 32gb SDHC SD card class 10 that contains NOOBS, here is a link to the product page:

https://www.canakit.com/raspberry-pi-4- ... r-kit.html

Don't know if this helps, here is what I can tell you about SD cards, search the internet for SD Card read/write speed benchmarks, usually the more expensive card will have faster read/write speeds, for the RPI computers you can expect performance improvements with faster SD cards, the only way to know the actual read/write speed is to run benchmark programs comparing different SD cards, and SD cards are not very expensive when the cards are on sale.

If you do plan to buy a brand new SD card you will need to install the operating system yourself, here is a link on how to write the OS image to the SD card:
https://elinux.org/RPi_Easy_SD_Card_Set ... 6_10_Hints

[morphious66]

I'm going to order and try the 64gig Sandick A2 U3 V30 card from https://www.amazon.com/SanDisk-Extreme- ... 313&psc=1

Will post up test results once complete.

[morphious66]

Also looking into an ultra small Samsung MUF-128AB option https://www.samsung.com/us/computing/me ... -256ab-am/ reviews do not mention of any heating issues and its ultra small as well. and may be small enough to hide of on the side with up to 300mbs reads. ( 80mbs writes from one persons benchmark. )

[morphious66]

ok, here are the results with the new Rpi4 with a 64gig Sandick A2 U3 V30 micro SD card

RPi3 - overclocked to 100mghz
Timing buffered disk reads: 96 MB in 3.03 seconds = 31.66 MB/sec
419430400 bytes (419 MB) copied, 69.3182 s, 6.1 MB/s

RPi4 - 50mgzh stock clock
Timing buffered disk reads: 128 MB in 3.00 seconds = 42.62 MB/sec
419430400 bytes (419 MB, 400 MiB) copied, 11.6795 s, 35.9 MB/s

10.96 Mb/sec increase in upload speed
29.8 Mb/sec increase in download speed

Looking forward to seeing what this will do when we can figure out how to overclock the bus.

--------------------------------------------

here is the full run-

Running dd test...

51200+0 records in
51200+0 records out
419430400 bytes (419 MB) copied, 69.3182 s, 6.1 MB/s

/dev/mmcblk0:
HDIO_DRIVE_CMD(identify) failed: Invalid argument
Timing buffered disk reads: 128 MB in 3.00 seconds = 42.62 MB/sec

Running dd test...

51200+0 records in
51200+0 records out
419430400 bytes (419 MB, 400 MiB) copied, 11.6795 s, 35.9 MB/s

Running iozone test...
Iozone: Performance Test of File I/O
Version $Revision: 3.434 $
Compiled for 32 bit mode.
Build: linux-arm

Contributors:William Norcott, Don Capps, Isom Crawford, Kirby Collins
Al Slater, Scott Rhine, Mike Wisner, Ken Goss
Steve Landherr, Brad Smith, Mark Kelly, Dr. Alain CYR,
Randy Dunlap, Mark Montague, Dan Million, Gavin Brebner,
Jean-Marc Zucconi, Jeff Blomberg, Benny Halevy, Dave Boone,
Erik Habbinga, Kris Strecker, Walter Wong, Joshua Root,
Fabrice Bacchella, Zhenghua Xue, Qin Li, Darren Sawyer,
Vangel Bojaxhi, Ben England, Vikentsi Lapa,
Alexey Skidanov.

Run began: Tue Jul 30 14:29:56 2019

Include fsync in write timing
O_DIRECT feature enabled
Auto Mode
File size set to 102400 kB
Record Size 4 kB
Command line used: ./iozone -e -I -a -s 100M -r 4k -i 0 -i 1 -i 2
Output is in kBytes/sec
Time Resolution = 0.000001 seconds.
Processor cache size set to 1024 kBytes.
Processor cache line size set to 32 bytes.
File stride size set to 17 * record size.
random random bkwd record stride
kB reclen write rewrite read reread read write read rewrite read fwrite frewrite fread freread
102400 4 2465 2354 8663 8665 7376 3343

iozone test complete.

microSD card benchmark complete!

[RetroBoyAdvance]

I wouldn't read too much into speed labels, and try to find reviews that actually put cards to the test with a benchmark, usually CrystalDiskMark. Of course then you won't know exactly how the Pi will handle it, but if the performance is not there in the first place it's no use trying. When the 4K write manages to hit 3MB/s and not fall below 1MB/s that should be a good basis to start with, indicating it to be decent enough for the limited performance of the Pi. That is where is the difference is at for the OS's responsiveness and boot times.

Note: A2 cards require special support by the host controller for best performance. I am not sure if that is implemented for the new SD controller in the Pi 4. So A1 may be the safer (and cheaper) choice.

I don't think that's true. A1 / A2 is just a new label to give you an idea about the minimum random read/write performance of the flash memory used in the card. Beyond that it's a matter of the device's controller being fast enough to make use of it. Apparently it is twice as fast now.

Not like the UHS spec, where UHS-II does require additional pins to function (does not have them), or it may perform worse than a UHS-I variant. That's the only thing that requires hardware changes, while SDHC/SDXC is mostly down to the operating system. Anything else is just labels.

Not to mention the actual A1 spec required IOPS are not actually very good. The only thing that's guaranteed with A1 is that it won't be as good as an A2 spec card, because they would have put that on there. Chances are that quite a few decent cards from before the Application spec existed are A2 already, just not labeled as such.

It might also be a card that is marketed differently. The Samsung EVO Plus (or EVO+) is a card that (according to reviews) performs very well, and even better than the Pro in some cases for our purpose, but neither have an Application spec label (yet).

[trejan]

I don't think that's true. A1 or A2 is just a new label that gives you an idea about the minimum random read/write performance to be expected of the flash memory used in the card. Beyond that it's just a matter of the device's controller being fast enough to make use of it.

No. There is a difference for A2 cards. The controller and card both need A2 support as they added some extra commands. The developer version of the page you linked mentions it but your consumer page doesn't.

The Application Performance Class 2 (A2) is defined by SD Physical 6.0 specification. It makes SD memory card much higher performance than A1 performance by using functions of Command Queuing and Cache.

A2 performance is available only the combination of A2 supported host and A2 supported card.

https://www.sdcard.org/press/thoughtlea ... Class.html explains what the two commands do.

Command Queue

The new CQ mechanism allows the SD memory card to accept several commands in a series (without their associated data) and execute them (with the data) whenever the memory card is ready. It contributes mainly to random read performance.

During the data transfer, additional commands may be sent to the card as long as the maximum number of queued commands does not exceed the maximum queue depth supported by the card (the SD standard allows queue depth of min 2, max 32).

With CQ, advanced information on intended commands is provided to the card. The card may manage and optimize its internal operations to prepare for the various commands in advance. Multiple tasks can be handled at one time in arbitrary order. New information on next commands may be sent to the card during current execution and during data transfer.

Cache function

In order to overcome the relatively limited write speed operation of flash memory, the Cache function allows the card to accumulate the data accepted by the host in a high-speed memory (e.g., RAM or SLC flash)) first, release the busy line and perform the actual write to the non-volatile slower memory (e.g., TLC NAND Flash) in the background or upon flush command. The card may cache the host data during write and read operation. Cache size is card-implementation specific; flushing of contents stored in cache is done in less than one second. It is supported by OSs today for embedded memory devices and is assumed to be easy to implement for cards.

Data loss is allowed if the data is not moved to non-volatile storage. In order to minimize data loss, the flush cache feature shall be supported and implemented by host before power-down operation. The data is not guaranteed as long as the flush operation is not completed. Cache contributes mainly to random write performance. Cache flushing is triggered by host followed by up to one second of the card’s busy time.

Tests of A1 and A2 cards in a RPi currently show that A2 cards are optimised for use with those two extra commands.

[RetroBoyAdvance]

I don't think that's true. A1 or A2 is just a new label that gives you an idea about the minimum random read/write performance to be expected of the flash memory used in the card. Beyond that it's just a matter of the device's controller being fast enough to make use of it.

No. There is a difference for A2 cards. The controller and card both need A2 support as they added some extra commands. The developer version of the page you linked mentions it but your consumer page doesn't.

Ah yes, i guess you could say memory controller requirements are hardware requirements. If it is indeed the controller. However, one would think it has to be higher spec memory first, which is faster in any case, removing at least one bottleneck. Then these new functions would aim to optimize and ensure the host delivers as advertised, which may be more than a Pi could handle anyway. Or are there actually tests (preferably on the Pi itself) showing that the situation is similar to using UHS-II cards in a USH-I only device?

[RetroBoyAdvance]

ok, here are the results with the new Rpi4 with a 64gig Sandick A2 U3 V30 micro SD card

That's very interesting. Could you decipher the results there for me?
How many IOPS or 4K random did you get on the Pi 4 without overclocking.

And please stop saying Sandick.

[trejan]

Ah yes, i guess you could say memory controller requirements are hardware requirements. If it is indeed the controller. However, one would think it has to be higher spec memory first, which is faster in any case, removing at least one bottleneck.

Yes. It is a software difference instead of an electrical difference but you do need support for it.

If it is indeed the controller. However, one would think it has to be higher spec memory first, which is faster in any case, removing at least one bottleneck.

The rated specifications for A2 cards are assuming you have those extra commands. If you don't then they can actually perform worse. Some tests at https://www.jeffgeerling.com/blog/2019/ ... spberry-pi Read the followup post as well.

There is currently no reason to pay extra for an A2 card since Linux doesn't support those extra commands and an A1 card performs better than an A2 card anyway. Once Linux has support then it'll be time to reevaluate the situation.

[RetroBoyAdvance]

If it is indeed the controller. However, one would think it has to be higher spec memory first, which is faster in any case, removing at least one bottleneck.

The rated specifications for A2 cards are assuming you have those extra commands. If you don't then they can actually perform worse. Some tests at https://www.jeffgeerling.com/blog/2019/ ... spberry-pi Read the followup post as well.

There is currently no reason to pay extra for an A2 card since Linux doesn't support those extra commands and an A1 card performs better than an A2 card anyway. Once Linux has support then it'll be time to reevaluate the situation.

If this is the test that you've based these conclusions on then i'm afraid i don't see A2 as the defining culprit here. All these cards being tested are different cards, that includes the Extreme, and Extreme Pro cards being compared. While the 'regular' Extreme card may have the A2 label, it is not a Extreme Pro, which is also available with updated A2 label. There is even a Extreme Plus in between those. So these are just cards being compared - that have little to do with the Application spec.

While the first test does have two Extreme cards being compared, they are different capacity for starters, which again makes them perform differently. More tellingly is the lack of A1 label, which probably means it is an older model. So who knows what's changed in the type of memory used, shifting throughout the lineup. Production processes in the world of memory change all the time even without changing labels.

The conclusion one can take out of this is that A2 does not 'automatically' make it a top performer, but maybe we knew that.

Note that the cards being tested are all decent, so none of these should perform badly at least. So yeah, back on point. Don't read too much into these classifications.

[trejan]

Yes. You're going to have trouble finding an exact like for like comparison except for the class. There have been other people testing these cards on this forum with similar conclusions but you'll need to dig around to find it as the search tool isn't the greatest. For what has been shown, buying an A2 card for usage on a RPi is currently a waste of money as an A1 card performs as good if not better for much less money. When Linux gets full support for the A2 commands then it'll be time to rerun the tests.

[gnarl]

So far the fastest sdcard i've found for my raspberry pi 4 is the Sandisk Extreme Plus (32gb). I've compared it to a Lexar 300x (64gb) and a Kingston U1 (32gb). Even just starting up the pi the first time, the speed difference was quite noticeable. This sdcard has much better write speeds and rand 4k read and write speeds. They all seem about the same in sustained read speeds.

Here's the benchmarks, averaged out from 4 runs:
hdparm read: 43.82MB/sec (best: 43.86)
dd write: 40.1MB/sec (best: 40.2)
rand 4k read: 9621 (best: 10,334)
rand 4k write: 4512 (best: 4561)

Compare this to the other 2, which performed identically pretty much:
hdparm read: 43.40 (kingston) 43.18 (lexar)
dd write: 16.1 (kingston) 17.9 (lexar)
rand 4k read: 6220 (kingston) 6528 (lexar)
rand 4k write: 1220 (kingston) 1260 (lexar)

hth.