The High End microSD vs Low End eMMC

I got the 64GB eMMC with Foresee chip from allnet, which is considered as low end part. Today I received SanDisk Extreme Pro 64GB A2 class microSD, I had a chance to run some benchmarks, it’s interesting.

I flashed Debian armhf on the Foresee eMMC, and built my own Debian arm64 on the SanDisk microSD, I know it might affect the results but I just don’t want to spend too much time on it.

READ PERFORMANCE:
SanDisk:
linaro@linaro-alip:~$ sudo hdparm -Tt /dev/mmcblk0
/dev/mmcblk0:
Timing cached reads: 2060 MB in 2.00 seconds = 1029.73 MB/sec
Timing buffered disk reads: 64 MB in 3.05 seconds = 20.98 MB/sec
Foresee:
linaro@linaro-alip:~$ sudo hdparm -Tt /dev/mmcblk1
/dev/mmcblk1:
Timing cached reads: 2078 MB in 2.00 seconds = 1039.27 MB/sec
Timing buffered disk reads: 464 MB in 3.01 seconds = 153.98 MB/sec

WRITE PERFORMANCE:
SanDisk:
linaro@linaro-alip:~$ dd if=/dev/zero of=/tmp/output conv=fdatasync bs=4M count=100; rm -f /tmp/output
38+0 records in
37+0 records out
158265344 bytes (158 MB, 151 MiB) copied, 0.697367 s, 227 MB/s
Foresee:
linaro@linaro-alip:~$ dd if=/dev/zero of=/tmp/output conv=fdatasync bs=4M count=100; rm -f /tmp/output
100+0 records in
100+0 records out
419430400 bytes (419 MB, 400 MiB) copied, 3.7685 s, 111 MB/s

Would you have a chance to run iozone to give us some Random Write 4K block performances ?

After all that was what A2 is designed to do :slight_smile:

It will be interesting to see how eMMC (Foresee) compared.

I’ve heard that in current Pi boards there is strict limit for read/write from/to SD cards for 20 MB/s - isn’t it?

I am asking because I was using Samsung 64GB Evo Plus with class U=3 and didn’t managed to get more than 20 MB/s…

Comparing to Samsung EVO
Select (U3), the results of emmc are comparable:

mmc:

“Random write report”
“4” “8” “16” “32” “64” “128” “256” “512” “1024” “2048” “4096” “8192” “16384”
“64” 312204 380864 365812 369844 477808
“128” 419822 413039 568836 576780 590094 470606
“256” 462260 478322 609455 641497 543720 651223 495312
“512” 458393 565156 628322 670291 633699 710192 589190 489420
“1024” 454888 547841 661089 681761 647729 661496 591201 491594 468841
“2048” 438176 500613 609699 691687 673942 691687 514712 482341 491955 472524
“4096” 447161 504981 645258 675659 695040 658615 559098 475333 501780 516659 497335
“8192” 455537 548800 646457 732354 692655 717400 568457 492691 499605 514900 520264 510312
“16384” 436278 516843 667264 740154 728139 723996 610884 498144 504341 569913 504697 650753 605336
“32768” 0 0 0 0 549264 574253 615523 687955 512400 506633 513106 755649 616517
“65536” 0 0 0 0 784064 817636 572182 435787 502133 628552 505826 513712 771747
“131072” 0 0 0 0 701711 611278 633488 781139 523415 637614 709525 687525 885927
“262144” 0 0 0 0 916331 1079257 913215 914352 916043 859009 930671 913636 916020
“524288” 0 0 0 0 1130184 1149917 1140388 1045829 966265 932521 949513 951075 965027

sd:

“Random write report”
“4” “8” “16” “32” “64” “128” “256” “512” “1024” “2048” “4096” “8192” “16384”
“64” 752329 576282 546928 551422 565358
“128” 553586 630273 653282 549618 681476 666253
“256” 569985 691930 636553 759469 682691 770920 705107
“512” 614654 680055 768934 796306 791318 727268 616242 525210
“1024” 602396 681329 675967 690086 731954 692311 623289 525905 523023
“2048” 589530 628801 677236 699916 704219 708167 673255 521398 522699 527838
“4096” 611083 629672 688853 733248 732873 737624 657255 527009 531953 521521 535501
“8192” 573514 670148 695769 690970 738747 748467 591778 512542 515836 624633 534691 498966
“16384” 593623 338708 688256 741423 720875 581984 584494 541406 528295 513683 525937 508017 513491
“32768” 0 0 0 0 573176 642547 340298 549088 504419 644836 527060 527842 651904
“65536” 0 0 0 0 569808 844427 589352 535350 511568 531947 527316 522586 511496
“131072” 0 0 0 0 938185 597228 742878 691482 692429 617253 933354 660019 823714
“262144” 0 0 0 0 844577 976051 923738 958685 662771 892490 919657 959619 930740
“524288” 0 0 0 0 1151316 1124639 1158256 1061256 973196 937056 930938 967840 929812

Your data is unreadable. Can you please edit your post and make it readable?

Thanks lucho for the test runs!

The results are verbatim as printed by iozone. I can read them easily, YMMV.

I am sorry, but I don’t see anything readable here. There is no direct info if it is kB/s or MB/s, and I don’t know what are these numbers on the left and on the right. Extremely involved people maybe could read that, but these scores gives me nothing.

The numbers are in KB/s. The rows are file size. The columns are record size.

Why would Foresee considered as low end? Actually the dram and emmc die from Foresee comes from Micron and Samsung. We did some test before among different brand’s emmc modules. For us, we choose Foresee because they we can directly talk to them and they provide fast response and good support. For other brands, we have to deal with distributors.

You can checkout the result comparison here:

The test report for the Foresee emmc module can be found here:

https://dl.radxa.com/rockpi/docs/hw/Test_Report_NCEMASLD-32G_20180313.pdf

4 Likes

interesting topic.
ofc there are many sdCards and different techniques.
It would be interesting to not only look and compare the pure speeds, but also reliability over time.

and the winner is :partying_face: :slight_smile: M2, SD cards are to expensive compared to m2 NVMe and m2 ssd’s, but I love to see such comparisson, good to know which SD would be fastest in my phone :slight_smile:

For benchmarking I use: https://github.com/cevap/rockpi4-bionic/blob/master/usr/bin/benchmark_hdd.sh

Run this script from a path on the device which you want to test.

Requirements:

sudo apt-get install -y fio