Hi all,
Over in the X1 forum, they have a utility & thread to speed up SD card access, and they are benchmarking it with SKTools. It also briefly discussed in our graphics performance thread, but NeoS2007 justly noted it was a bit off-topic, as that thread is actually about OpenGL and D3D (maybe he can change the title)
Anyway, what really bugs me is not that the X1 utility does not work for HD, but that the HD has soo much less RAM (and flash) performance then the X1. Why? Is it not supposed to be nearly identical base machine? Have they used a cheap RAM chip, or is the timing of it wrong. Would that be a bootloader setting?
In fact, I suspect comparing flash speed between X1 and HD is irrelevant if the RAM is so different, you are in fact still measuring RAM.
For example, this is my HD:
(posting from others show it is a fairly normal HD, only a bit high in storage write, maybe because I have a Ultra-card):
Code:
RAM access: 544
Bitmaps: 785
Main write 1776
Main read 6413
Storage write 1303
Storage read 8657
The typical X1:
Code:
RAM access:1519
Draw Bitmaps:797
Main (write):11070
Main (read):17066
Storage (write): 483
Storage (read): 16832
Thats 3 times the RAM speed
And as everything depends on RAM, it does not even matter to mention the 6 times flash write speed
In this thread, I am hoping that people who have significant other findings (you can test this with the trail version of SKTOOL) post so, just to see what really is normal.
Or better yet, if someone smart is able to somehow change the RAM timing, by modding a bootloader or poking some bytes into a chip register somewhere. (I suspect the RAM timing is not a windows setting, but deeper than that)
If you do post your setting, please include if you applied any tweaks, and the brand/type/class of microSD.
Thanks
if you look at the index in sktools benchmark the HD gets given a score of 560 v 472 for the diamond on ram access.
i get on my x1
Integer 329.2257
FP 7.233
Ram 1502
Bitmaps 799
MS Write 8727.27
MS Read 15170.39
SD Write 487
SD read 16168.42
it's curious as I also thought that the two devices were internally very similar, what tweaks are you using?
Do you get different results in integer and floating point, as some people believe their X1s run at 400mhz
Wow thats huge difference in ram speeds between blackstone and hd. But interestingly enough blackstone seems to perform better with in coreplayers benchmark. That maybe suggests that sktools is not accurate, or its buggy etc.
Integer 316.9227
FP 7.294
Ram 461
Bitmaps 769
MS Write 1860.13
MS Read 6849.50
SD Write 461.71
SD read 4740.74
Click to expand...
Click to collapse
Tested using SKTools
Am running SPB Benchmark at the moment to compare
my Touch HD:
Integer 330.3489
FP 7.605
Ram 576
Bitmaps 814
MS Write 1897.47
MS Read 7177.57
SD Write 192.25
SD read 6729.46
what the....?
why am i getting so low result for SD write? i have installes SD tuneUP and still....
noris08 said:
what the....?
why am i getting so low result for SD write? i have installes SD tuneUP and still....
Click to expand...
Click to collapse
Annoying, but I suspect the top 4 benchmarks are much more critical in whether the device feels snappy than the latter 4, and of those, I think the read has more influence than the write. So maybe it makes you feel less bad to know the SD write is probably the least important one?
Also, it may simply be the SD card you have. I just bought an 'ultra', most people have a class II, maybe you have a class I?
yeah, i know that this is the less important but it still bugs me.
that and the huge difference compared to x1
i have the 8gb Transcend microsd card that came within the package. i have no ideea what class it is. how can i find out?
it has a circle with a 6 inside of the circle printed on. that means it's a class 6? then the results i am getting worries me a lot
Since I kind of started this discussion in the other tread mentioned by cybermaus in his initial post, I think, I can put this to rest also. Run the Memory card benchmark in pocket mechanic (Trialversion available here: http://www.wizcode.com/downloads ) and you will see that the read/write performance of the 8GB card that came with the HD is quite remarkable, my results were even better than the ones posted over in the X1 thread here: http://forum.xda-developers.com/showpost.php?p=3154958&postcount=257.
It's a more sophisticated bench for the card performance and we are talking about read performance of 22.5 MB/s unbuffered, 42 MB/s buffered and write performance of 1.80 MB/s unbuffered, 5.33 MB/s buffered, so all is good. The readings in the SK Tools bench seem to be false, I would suggest that this also goes for the ram performance readings. This is also consistent with the observation that the HD is the slightly faster device overall.
razorblader said:
Since I kind of started this discussion in the other tread mentioned by cybermaus in his initial post, I think, I can put this to rest also. Run the Memory card benchmark in pocket mechanic (Trialversion available here: http://www.wizcode.com/downloads ) and you will see that the read/write performance of the 8GB card that came with the HD is quite remarkable, my results were even better than the ones posted over in the X1 thread here: http://forum.xda-developers.com/showpost.php?p=3154958&postcount=257.
It's a more sophisticated bench for the card performance and we are talking about read performance of 22.5 MB/s unbuffered, 42 MB/s buffered and write performance of 1.80 MB/s unbuffered, 5.33 MB/s buffered, so all is good. The readings in the SK Tools bench seem to be false, I would suggest that this also goes for the ram performance readings. This is also consistent with the observation that the HD is the slightly faster device overall.
Click to expand...
Click to collapse
did that, got
read unbuffered 11,86 mb/s
read buffered 32,75 mb/s
write unbuffered 1,47 mb/s
write buffered 3,38 mb/s
Couldn't do the RAM readings as it kept complaining it needed 21,5MB free even tho 70% was free. But indeed quite a difference to SKTools (phew).
rhtizzy said:
did that, got
read unbuffered 11,86 mb/s
read buffered 32,75 mb/s
write unbuffered 1,47 mb/s
write buffered 3,38 mb/s
Couldn't do the RAM readings as it kept complaining it needed 21,5MB free even tho 70% was free. But indeed quite a difference to SKTools (phew).
Click to expand...
Click to collapse
Does the same thing for me on the RAM bench, the low level tests for the the card bench also crash the program on my HD, so we are left with the file read/write tests, but good enough for me at the moment.
would be nice to do that benchmark before and after the install off sd_tuneup
Vinski- said:
Wow thats huge difference in ram speeds between blackstone and hd.
Click to expand...
Click to collapse
Blackstone and HD are names for the same device .
noris08 said:
why am i getting so low result for SD write? i have installes SD tuneUP and still....
Click to expand...
Click to collapse
The "tuneUP" doesn't improve the writing speed, it just increases the reading speed.
johnpatcher said:
The "tuneUP" doesn't improve the writing speed, it just increases the reading speed.
Click to expand...
Click to collapse
i've noticed that 'cause i've made the test with and without tuneUP.
that's is too bad because one may need fast writing speed when taking movies or burst photos
i still can't understand the huge differences compared to X1
btw, i have got better results testing with pocket mechanic. i did it last night and i don't remember the numbers, but, for writing test, there were something like 2.35 mb unbuffered and 4.95 mb buffered.
menwhile i found out that he sd card is class 6.
but i am still bugged by the high numbers X1 scored
johnpatcher said:
Blackstone and HD are names for the same device .
The "tuneUP" doesn't improve the writing speed, it just increases the reading speed.
Click to expand...
Click to collapse
Oh, looks like I mistyped it yesterday, I meant X1 versus blackstone
But still wondering is the benchmark buggy, or why so huge difference. X1 doesnt feel that snappier, and looses Coreplayer benchmark, which uses software decoding in both, meaning pure processing power.
johnpatcher said:
The "tuneUP" doesn't improve the writing speed, it just increases the reading speed.
Click to expand...
Click to collapse
Hi guys,
I found two interesting things when I tested my HD with SKtools.
1) I compared the results of SKTools SD card reading test before I install SD tuneUP and after. It was very suprised for me. The results are following:
Without SD tuneUp: ~5500 KB/sec
With SD tuneUp: ~ 2500 KB/sec
2) I made RAM SKTools read/write tests on HTC Touch Diamond and I was very-very suprised.
HD results:
Main storage (read): 5779 KB/sec
Main storage (write): 1692 KB/sec
Diamond results:
Main storage (read): > 8000 KB/sec
Main storage (write): > 8000 KB/sec
Who can explain that?
razorblader said:
Since I kind of started this discussion in the other tread mentioned by cybermaus in his initial post, I think, I can put this to rest also. Run the Memory card benchmark in pocket mechanic (Trialversion available here: http://www.wizcode.com/downloads ) and you will see that the read/write performance of the 8GB card that came with the HD is quite remarkable, my results were even better than the ones posted over in the X1 thread here: http://forum.xda-developers.com/showpost.php?p=3154958&postcount=257.
It's a more sophisticated bench for the card performance and we are talking about read performance of 22.5 MB/s unbuffered, 42 MB/s buffered and write performance of 1.80 MB/s unbuffered, 5.33 MB/s buffered, so all is good. The readings in the SK Tools bench seem to be false, I would suggest that this also goes for the ram performance readings. This is also consistent with the observation that the HD is the slightly faster device overall.
Click to expand...
Click to collapse
But this, and may posts above, are focussing back to the SD card.
I really think the SD card is far less important than the RAM for general performance of the device. Checking the SD is like wanting a speedy car by checking how fast you can open the booth.
I'd like to indeed think that it may be a artefact in SKTool that causes the difference between HD and X1 RAM benchmarks (500 vs 1500). Especially since people tell me practical benchmarks like Coreplayer tell a different story.
But the difference is too significant to ignore. Does anyone know a different RAM/CPU (not SD) testing program?
so you don' think writing speed of the sdcard is important!
i'll say it again - when you record a movie or burst some photos writing speed will count or not? maybe a drop more then opening the booth?
Actually im thinking RAM is the issue here, the reason being yes the read speeds of the SD card would impact on the performace of loading the data which is why we have buffers, the problem i see is that im not getting buffer underruns so we dont have a shortage of data to play. So we are loading the data in to RAM without much issue.
RAM would also explain any lag issues people might be gettings with the over all performance of the device.
noris08 said:
so you don' think writing speed of the sdcard is important!
i'll say it again - when you record a movie or burst some photos writing speed will count or not? maybe a drop more then opening the booth?
Click to expand...
Click to collapse
I am not saying it is not important. I am saying it is less important than RAM speed:
If you record a video at full VGA resolution, it would mean a 640x480x30x3 or 27 MB/s stream into RAM/CPU/GPU (depending which does the compression). Next, it is compressed (so cycles around between RAM and CPU/GPU a bit) and finally needs to be written to a file in the compressed stream of maybe 1 or 2 Mbps. Which is 0.125 or 0.250 MB/s
If you play a XViD video, the same thing in reverse, except now the RAM bandwidth is even 35 MB/s because we want the full WVGA
So, SD requirements: between 0.1 and 0.3 MB/s. RAM requirements, way above 35 MB/s (the CPU also needs to calculate, not just move data)
Like you, I'd also like a faster SD. I would hope it would speed up loading a new program after selecting it from the launcher. Also, interrupt calls from the SD could cause jerkiness in video. And that one guy with 0.192 MB/s SD defenitely would like a faster SD.
But the fact SD is important, does not take away the fact RAM is more important, which brings us back to the original question: Is the SKTool test flacky, or is X1 significantly faster in RAM?
cybermaus said:
And that one guy with 0.192 MB/s SD defenitely would like a faster SD.
Click to expand...
Click to collapse
that guy happens to be me
i am not arguing about bigger is better (i am speaking about RAM)
but considering the fact that X1 is also made by HTC, and it would be logic that they keep the best part of the cake for themselves (HD) i can only imagine that SE found a better management software of the given hardware.
can this be an explnation for the almost 1 year delay of the X1?
Related
SKTools tests:
Integer;325.4344;Moves/25 usec
Floating point;6.517;MWIPS
RAM access;459;Speed index
Draw bitmaps;570;Speed index
Main storage (write);1503.67;KB/sec
Main storage (read);5120.00;KB/sec
Storage Card (write); 260.04;KB/sec
Storage Card (read);7366.91;KB/sec
Tests Explained:
- Integer: Comparable to other MSM7201a devices and in SKTools listed as one of the fastest.
Only faster devices are Touch HD2 and Samsungs OmniaII, scoreing 813 and 530 respectivly, showing what we already knew, the new generation of ARM's is quite a bit faster.
- Floating Point: Okish with only new CPU's being much faster (HD2 is 5x faster in FPU operations.)
- RAM Access: REDICULOUSLY SLOW, but exactly the same as all other MSM7201a devices. But scoring just around old 200Mhz device speeds. There are PXA2xx based devices with a RAS from 4x that of what our MSM7201a's show to 20x that speed.
- Draw Bitmaps: Okish compared to other MSM devices, but way slower then Intel PXA's and even the HD2 is only 2ce as fast. (PXA's go up to 3x-4x as fast, Omnia 1 and Omnia 2 being nice examples.)
-Main Storage (Write): Here is where it gets interesting and clear something is funky as hell, either the HD has ****ty ROM chips or the interface in Windows Mobile is messed up. There are devices scoring from 2x to 10x the speed the HD does and some of them (notably the Diamond and Sony X1i which are MSM72xx devices) are 10x faster.
-Main Storage (Read): Same as with Write, slow as ****, the Diamond scores 3x the Diamond, same for X1i, with the HD2 surprisingly being able to read its internal storage at 64MB/s! Our HD's around 5MB/s, pitifull, even a Class 2 SD reads 2ce that easily.
- Storage Card (write): I ran this test a billion times and found the score going from 50KB/s to 500KB/s but never higher. There is a serious problem here, I use class 6 SDHC's of several sizes and all of them score very low. The SD interface is slow as hell which I'm certain is a software / driver problem, because I've tested some drivers from other devices, which worked mindbendingly fast.
As you can see, Ram, Internal Storage and SD speed are all lacking extremely.
Compared to other devices (notably, even the Diamond) its quite ridiculous how slow the HD's internal storage really is.
I propose taking a new approach to speeding things up, most other projects so far have been about tweaking graphics drivers, to offload graphics to more optimized routines, streamlining roms, etc etc.
I propose we start looking for ways to get Ram, Internal storage and SD storage speeds up to par with other devices. That is, if its not all limited to hell by substandard hardware. Which would piss me of even more, since the HD is the most expensive phone HTC had at the time and way cheaper devices like the Diamond score much much better.
A few weeks ago, I tested a driver for the SD Interface in Windows Mobile and overal, read and write for the SD were boosted to phenomenal levels, Class 6 cards were actually worth using (while the above results show a class 2 card doesn't even get used right)
Problem was that after a while I was experiencing data corruption on the SD card, mostly after syncing the phone with Outlook to my Exchange account and getting several thousands of small files on there.
I cleaned the card and did the sync again and the same corruption happend.
My tests with big files (MP3's and AVI's) showed no corruption.
So, I'm at least sure that SD speed can be improved quite seriously. And seeing there are devices like the Diamond, which have a lot of the same hardware the HD does, internal storage should be possible to be boosted quite a bit too.
RAM is weird and I don't really see how it can be so bad because of software alone. Did HTC really set us up with ****ty ram chips there? I don't know.
So yeah, anyone with me to explore the storage area to improve the HD's speed?
What SD Card format setup for speed and stability? [Test Updated 8/22 & Conclusion]
See Post 8 and 9 for test data, updates and conclusion.
What setup for the best speed and stability are you using when you format your SD card? What program are you using?
I have tried a number of different configurations and methods including the PC format, the Panasonic and the WinMo format in the phone. I have used about all combinations of cluster size. I have found the Storage Tools program on the phone with the following configuration:
FAT32
512 byte Sector Size
32KB Cluster Size
Backup FAT - NO
seems to give me slightly better results than the second runner up which is WinMo on the phone.
I still am not able to benchmark to the level I could get when my Transcend 8g, class 6 came out of the box. (12-14) Now getting 9-12.
What are you using and what is your best solution?
Does "the weakest link in the chain" apply here?
Does "the weakest link in the chain" apply here? If the SD card has the wrong format will it effect the performance/stability of Android when run off of the card?
Is there anyone who has determined what the optimum SD card format for an HD2 using Android should be? Does it even matter?
If we are running the OS from the SD than the SD becomes the first link in the chain of events. Why no information or knowledge concerning how this first link should be formatted?
Very interested in knowing some exact configurations for the sd formatting..I have done quick format and have done full format and still seem to have problems..I just bought a 4gb class 4 so I am going to see how it works..I am going to try without formatting and see what happens and by then hopefully someone posts the perfect configuration..
I wouldnt mind hearing some oppinions on this too as I am just waiting for my SD backup to complete before fomatting (in the hope of curing some slight stutters in android).
I'd mainly like to know if there is any firm eveidence that formatting on the device (HD2) is better than formatting on PC (Vista in my case).... or is it just best to format on the PC? No difference?
BUMP
Test data
Test data on four different SD cards with four different tests on each coming soon. This will only show how each "specs out" not what is best. It will give me a baseline for further tests. Working on the last card tests now.
intersted in this question lets see what others will say
I always used the Format SD Tool in Windows Mobile and have never faced any problems (hiccups, slow downs) while using Android.
16GB Transcend Class 2 card.
First set of test data here.
All disks contain exact same Android folder on root (Mattc 1.5+overclock)
and same other directories with media etc. from Winmo & Android.
Fresh boots to Andro. no tasks killed, live wallpaper running.
Full blown Miri v18 rom, Cookie and apps. Not a crippled Andro. ROM.
Sandisk 6g class 4 - Formated 512/64 (Format method unknown)
H2testw v1.4 Writing speed Win 7/64: 10.9 MByte/s
H2testw v1.4 Reading speed Win 7/64 19.0 MByte/s
Pocket Mechanic Write WinMo: 9.23 MByte/s
Pocket Mechanic Read WinMo: 20.9 MByte/s
SD Card Speed Tester Write Android: 8 MByte/s
SD Card Speed Tester Read Android: 15 MByte/s
Quadrant score: 1539
Transcend 8 Gig Class 6 - Formated 512/64 (Format method unknown)
H2testw v1.4 Writing speed Win 7/64: 9.17 MByte/s
H2testw v1.4 Reading speed Win 7/64 15.0 MByte/s
Pocket Mechanic Write WinMo: 6.11 MByte/s
Pocket Mechanic Read WinMo: 10.96 MByte/s
SD Card Speed Tester Write Android: 6 MByte/s
SD Card Speed Tester Read Android: 13 MByte/s
Quadrant score: 1522
Original 16G, Class 2 that came with T9193 - Formated 512/64 (Format method unknown)
H2testw v1.4 Writing speed Win 7/64: 5.30 MByte/s
H2testw v1.4 Reading speed Win 7/64 19.0 MByte/s
Pocket Mechanic Write WinMo: 2.14 MByte/s
Pocket Mechanic Read WinMo: 7.98 MByte/s
SD Card Speed Tester Write Android: 4 MByte/s
SD Card Speed Tester Read Android: 14 MByte/s
Quadrant score: 1344
Sandisk 4 G, Class 2 - Formated 512/32 (Format method unknown)
H2testw v1.4 Writing speed Win 7/64: 2.94 MByte/s
H2testw v1.4 Reading speed Win 7/64 10.6 MByte/s
Pocket Mechanic Write WinMo: 3.34 MByte/s
Pocket Mechanic Read WinMo: 7.93 MByte/s
SD Card Speed Tester Write Android: 2 MByte/s
SD Card Speed Tester Read Android: 3 MByte/s
Quadrant score: 1528
Note: H2testw v1.4 is the Gold Standard test for Windows drive speed testing. Google it and download for free. Pocket Mechanic has a 30 day trial. Quadrant and SD Card Speed Tester from market for free. Run each test
3 times in a row. Pocket Mechanic " Card Information" test will tell you what your current card is formated with.
Format Program test:
When I formatted the 16 G original Class 2 with all three programs (Windows, SD Format and WinMo), H2testw v1.4 came up with the same basic read and write speeds for all three at 512/64. The only difference that I see is the WinMo
only makes one copy of the FAT and the others make two. Single FAT is supposed to be faster in the real world according to the tech literature because it only has to write the FAT once. This fact only really shows up when the disk becomes fragmented according to the Xperts.
Methodology: This test was about as scientific as making Kool Aid, but I did start with freshly formated SD cards that I then copied the same exact file system on, so they were about as de-fragmented as possible. At least they were all the same. I had to delete some movies/music etc. from the small card so the SD Card test could run. It needs 300 mb of clean space. Test does not use those files anyway. Testing cards with H2testw v1.4 after format and after files were copied on them gave exactly the same read'write speeds basically.
Conclusion: I will run some more tests (see if 512/64 format helps the 4G Class 2 and format one card with all 3 programs and test all benchmarks), and graph the results before opening my big mouth about what I think all this means.
Data From Tests of multiple format methods on individual cards
set of test data and further reports.
Spread sheet of above results below.
Conclusion thus far:
By label the card that should be the fastest/best is not. When it was fresh out of the box it was the fastest. The Sandisk 6G Class 4 is currently the best in all categories. The answer to this may be in what they were formatted with (program) and/or what parameters were used. All I can conclude thus far is "what you see is not necessarily what you get". I will now test to see if the format programs make a difference.
Update:
New Test Data concerning SD Formatter 3.0 (Beta). I formatted the 4G Class 2 Sandisk with the above program using every parameter it allows. You can only "overwrite erase an SD Card". The Quick and Full formats with erase on and off were tested. Who won? Fooled me. The Quick had faster write/read speeds using post format H2Testw benchmark test as compared to the full (which takes forever to format). It was only a little faster on read and the same on write. This would seem to indicate that you might as well take the easy, and quick, way out! Did not test the larger cards.
Update: (8/19/10):
Tests of Format programs and cluster size on Sandisk 4G, Class 2
This is the smallest card in size and rated class that I have.
It was interesting to see the Quadrant scores come out so close and in the same range as all of the larger/faster cards. This would seem to indicate that the Quadrant test is not dependent on the SD cad for it's scores.
Note: During these tests I noticed that the first run of the SD Card Tester under Android would run extremely slow. Usually giving a speed for write of 0 or 1. I would also get errors and SOD's when I used Windows or WinMo to format the card. I do not know why the Android program runs so slow on the first pass. Possibly some difference in windows card format and Android. I do not know. (This also happened on the upcoming 16G, Class 2 tests.) After the 1st pass the tests were normal. Maybe this initial slow down is the cause of some of the problems ppl have with class 2 cards. It is total speculation, but maybe running some initial write/read tests would help smooth out some of the problems on the smaller cards.
Update: Tested a card with direct copy of backup on fresh format and there is no fragmentation shown by windows. Pocket Mechanic shows 0.01 fragmentation so I think this one can be tossed out the window. As files are added like music etc. this may become a problem. But fragmentation does not have any effect on the tests done here.
Final Update:
And the winner of this adult tricycle race is the original 16 G Class 2 card. I am now trying this card in everyday use to see if the Windows 64 format or the WinMo in the phone format have any different characteristics in use.
If I was going to choose "A" Format Program for "Any" card, I think I would choose WinMo in the phone.
Using programs that only write one FAT, versus the normal 2, might help once the disk get's fragmented, but this is easy to fix by defraging or backup to hard drive, format and recopy. The second FAT is great insurance for file system recovery and this test shows that on a defraged card there is little if any downside in speed.
What we learned: (By this very limited test.)
1. Card Class on the label does not have much to do with true read/write speeds. So you can't buy a card by it's cover and expect price/performance.
2. Card size and Class have little, if any, effect on Quadrant scores.
3. The format program and parameters you use, at least by these results alone, are not going to make a great deal of difference. Some programs/parameters might have a slight advantage. Fragmentation and usage may make some format/parameter methods behave differently in the long run. This was not tested here.
Conclusion: So go, my children and format your cards haphazardly. It probably is not going to make a lot of difference.
Update 8/22: Tested suggested Ubuntu format method and found no advantage as far as speeds or scores. (This suggestion seems to have been based on incorrect knowledge.)
why are you even worried about quadrant scores none of it matters...quadrant gives so many different results ...i dont even know why people are boasting about it .....
Not "worrying" about the scores.
thatruth132 said:
why are you even worried about quadrant scores none of it matters...quadrant gives so many different results ...i dont even know why people are boasting about it .....
Click to expand...
Click to collapse
Not "worrying" about the scores. It is just one means of benchmarking SD cards. A way to compare one to the other. I do not know if the card/format even has an effect on Quadrant scores for sure at this point. Quadrant does give me very consistent scores for each card setup when I consistently take the third run.
I would agree that the Quadrant score race seemed a little like a bunch of adults having a tricycle race around the block.
This test is to try and determine what, if any, the optimum SD Card format perimeters are.
Can we run android on 512mb or 1gb sd card?
Don't know, but
10507 said:
Can we run android on 512mb or 1gb sd card?
Click to expand...
Click to collapse
Don't know, but I would guess not very well. Don't have a card that small. These other cards are just what I had on hand from my cameras and my wife's phone. She hasn't missed it yet?
johnboatcat said:
Not "worrying" about the scores. It is just one means of benchmarking SD cards. A way to compare one to the other. I do not know if the card/format even has an effect on Quadrant scores for sure at this point. Quadrant does give me very consistent scores for each card setup when I consistently take the third run.
I would agree that the Quadrant score race seemed a little like a bunch of adults having a tricycle race around the block.
This test is to try and determine what, if any, the optimum SD Card format perimeters are.
Click to expand...
Click to collapse
i agree ..i have a better understanding of what you are doing
johnboatcat said:
set of test data and further reports.
Spread sheet of above results below.
Conclusion thus far:
By label the card that should be the fastest/best is not. When it was fresh out of the box it was the fastest. The Sandisk 6G Class 4 is currently the best in all categories. The answer to this may be in what they were formatted with (program) and/or what parameters were used. All I can conclude thus far is "what you see is not necessarily what you get". I will now test to see if the format programs make a difference.
Click to expand...
Click to collapse
Kudos for doing these tests and for publishing the results here. Good man. Looking forward to seeing the results of the 3 different format options on the one card. Cheers.
Sent from my HTC HD2 Using the XDA App on ANDROID 2.2
adata 8gig class 6 1 quick format in window no special tool run smooth no problem no hickup, slow down ,not even a single SOD, fast ever since
New Test Data concerning SD Formatter 3.0 (Beta)
New Test Data concerning SD Formatter 3.0 (Beta). I formatted the 4G Class 2 Sandisk with the above program using every parameter it allows. You can only "overwrite erase an SD Card". The Quick and Full formats with erase on and off were tested. Who won? Fooled me. The Quick had faster write/read speeds using post format H2Testw benchmark test as compared to the full (which takes forever to format). It was only a little faster on read and the same on write. This would seem to indicate that you might as well take the easy, and quick, way out! Did not test the larger cards.
Will add this to the data pages.
you should test some none sdhc card, which access time is smaller than sdhc.
I only test what I have on hand
qingcai said:
you should test some none sdhc card, which access time is smaller than sdhc.
Click to expand...
Click to collapse
I only test what I have on hand. Why don't you test the non-HC cards and add to everyone's collective knowledge here. All of the test software is free.
Tests of Format programs and cluster size on Sandisk 4G, Class2
Tests of Format programs and cluster size on Sandisk 4G, Class 2
This is the smallest card in size and rated class that I have.
It was interesting to see the Quadrant scores come out so close and in the same range as all of the larger/faster cards. This would seem to indicate that the Quadrant test is not dependent on the SD card for it's scores.
Note: During these tests I noticed that the first run of the SD Card Tester under Android would run extremely slow. Usually giving a speed for write of 0 or 1. I would also get errors and SOD's when I used Windows or WinMo to format the card. I do not know why the Android program runs so slow on the first pass. Possibly some difference in windows card format and Android. I do not know. (This also happened on the upcoming 16G, Class 2 tests.) After the 1st pass the tests were normal. Maybe this initial slow down is the cause of some of the problems ppl have with class 2 cards. It is total speculation, but maybe running some initial write/read tests would help smooth out some of the problems on the smaller cards.
I will add this to the Data Page 9.
I'm wondering whether there's a way to test the internal NAND speed, then we can compare and balance whether to use extend SD card.
I have a Sandisk Class4 !6G microsd which has 19M/16M read and write, but if the NAND still much faster than this card, im considering to use internal storage only(less mp3 and game instead), its hard to balance.
Sorry to re-awaken an old thread, but since you've had no reply: I had the same question, and found an app in the Market called "Android Hacker's System Tool" that includes a storage speed test (among many other things).
On my Motorola Droid1 running MIUI (not o/c'ed currently), I was shocked to see NAND speeds of only 1.9 MB/s read and 0.21 MB/s write. The OEM 16gb SD card showed a bit over 2 MB/s for both read and write... at least symmetrical, but still very slow. No wonder apps on the SD card seem just as quick as those in NAND!
Any thoughts as to why IO is this poor overall, and why NAND write in particular is so pathetic? The phone feels very smooth right now.
Why do you post questions about your Droid in a Windows Phone forum? Samsung Focus is Windows Phone 7. There is nothing in common with your Droid.
rvonder, interest numbers, thanks for the post. I'm a little surprised at the low numbers as well. Especially the writes. Hopefully we will have a test app for WP7 soon.
Just going by deduction from here. If the phone feels snappy even with those numbers, isn't it safe to assume it doesn't matter ?
rvonder said:
Sorry to re-awaken an old thread, but since you've had no reply: I had the same question, and found an app in the Market called "Android Hacker's System Tool" that includes a storage speed test (among many other things).
On my Motorola Droid1 running MIUI (not o/c'ed currently), I was shocked to see NAND speeds of only 1.9 MB/s read and 0.21 MB/s write. The OEM 16gb SD card showed a bit over 2 MB/s for both read and write... at least symmetrical, but still very slow. No wonder apps on the SD card seem just as quick as those in NAND!
Any thoughts as to why IO is this poor overall, and why NAND write in particular is so pathetic? The phone feels very smooth right now.
Click to expand...
Click to collapse
foxbat121 said:
Why do you post questions about your Droid in a Windows Phone forum? Samsung Focus is Windows Phone 7. There is nothing in common with your Droid.
Click to expand...
Click to collapse
rather than waste anybody's time, please read the question first next time, it has nothing to do specificaly with android but rather about the use of external memory card considering the Focus uses built-in NAND memory that might be much faster than the external memory (or not). Since Windows 7 phone "merges" both memory, it would most likely be at the speed of the slowes of both...
thegarmac said:
rather than waste anybody's time, please read the question first next time, it has nothing to do specificaly with android but rather about the use of external memory card considering the Focus uses built-in NAND memory that might be much faster than the external memory (or not). Since Windows 7 phone "merges" both memory, it would most likely be at the speed of the slowes of both...
Click to expand...
Click to collapse
Did you read his question? He was asking why his Droid phone SD card IO is so slow. First of all the Android app in question does something really stupid. Last time I checked, it gives out really low IO score on my Captivate as well. So, I'd say it is the app that's not really designed well, or could be Android OS doesn't offer a good API for testing the SD performance. None of these are related to how WP7 works. As you mentioned above, WP7 treats memory differently and no one knows exactly how SD performance affects entire system (we knew that most 'faster' SD cards don't work). And we don't have any app to test native IO speeds of the combined memory. So, this is really comparing apples to oranges.
i have bought a new Sandisk Ultra 16GB class 10 card and flash nexushd2 rom in native sd but i notice no different than nand in the speed !!!
i test the card with h2testw_1.4 and it was ok the speed was 10mb R/W,
So Where is the Speed that everyone talk about !!!!
x_max_best said:
i have bought a new Sandisk Ultra 16GB class 10 card and flash nexushd2 rom in native sd but i notice no different than nand in the speed !!!
i test the card with h2testw_1.4 and it was ok the speed was 10mb R/W,
So Where is the Speed that everyone talk about !!!!
Click to expand...
Click to collapse
That's basically because it's a myth .
I'm kidding, I don't know if it is or not, but personally, I didn't notice any speed difference or improvement. I tried different ROMs, they all behaved as if they were on NAND. I have the same card as you do.
Someone told me once: "It's because you have a UHS card, HD2 doesn't support them so the speed is degraded".
I am not sure of that, but there was no apparent improvement over NAND installation.
+1, I never saw any speed difference. My guess is that since it would theoretically be faster there was a placebo effect and people actually felt their phones being faster. The same thing happens with build.prop tweaks and the entropy seed generator, both of which had great comments but no actual effect on my phone.
x_max_best said:
i have bought a new Sandisk Ultra 16GB class 10 card and flash nexushd2 rom in native sd but i notice no different than nand in the speed !!!
i test the card with h2testw_1.4 and it was ok the speed was 10mb R/W,
So Where is the Speed that everyone talk about !!!!
Click to expand...
Click to collapse
You should read first my friend,this way you will save some Money,any Sandisk class 4 will be faster than any class 10 in native sd, simply because with a class 10 card you get high sequential data transfer and very low random data transfer, and that is the secret random access transfer rate.
But the manufacturers won't tell you the random access speeds, you have to test.
Use crystal disk mark to test the sd and check for the last 2 values..
And read..and search..
Sent from my GT-I9300 using xda premium
x_max_best said:
i have bought a new Sandisk Ultra 16GB class 10 card and flash nexushd2 rom in native sd but i notice no different than nand in the speed !!!
i test the card with h2testw_1.4 and it was ok the speed was 10mb R/W,
So Where is the Speed that everyone talk about !!!!
Click to expand...
Click to collapse
I can confirm that the problem is with the Sandisk Ultra SD Card. I had good read but very poor write speeds on a Sandisk UHC which caused all sorts of problems. I changed to a Samsung Class 10 32Gb card and the difference is very noticeable The Sandisk gave a write speed of about 3/4 Mbs and a read speed of about 15 Mbs. The Samsung now gives a write speed of 9.7 and a read speed of 20.15. Both tested with a cache size of 2048Kb.
sandymac said:
I can confirm that the problem is with the Sandisk Ultra SD Card. I had good read but very poor write speeds on a Sandisk UHC which caused all sorts of problems. I changed to a Samsung Class 10 32Gb card and the difference is very noticeable The Sandisk gave a write speed of about 3/4 Mbs and a read speed of about 15 Mbs. The Samsung now gives a write speed of 9.7 and a read speed of 20.15. Both tested with a cache size of 2048Kb.
Click to expand...
Click to collapse
I had the same low write speed in native SD (3 mb), but i flash the rom again in DataonExt mod and the speed of write was 7 mb and i feel some speed up in apps (not in the rom) Rom toolbox collect data faster , I think there is something wroung with Native SD ,the speed of write should be higher than 3mb
Edit: it was a placebo effect , there is no speed up in DataonExt , i test the same app in Nand
x_max_best said:
I had the same low write speed in native SD (3 mb), but i flash the rom again in DataonExt mod and the speed of write was 7 mb and i feel some speed up in apps (not in the rom) Rom toolbox collect data faster , I think there is something wroung with Native SD ,the speed of write should be higher than 3mb
Click to expand...
Click to collapse
Well, if you looked, I am on Native SD and generally manage around the 10Mbs write with 20Mbs read so nothing wrong with Native SD.
as the creators of nativeSD and others have pointed out, it's all down to the small block size random read/write speed, that's what most read/writes are when running an OS, small and random.
While reading (for example) a video, it will read bigger, almost certainly sequential, block sizes.
The industry quoted max speed of a card is almost always based on the largest block size, usually read, always sequential speed, so a class 10 can read large block sizes at 10meg, however, test the card for all different block sizes and you'll see that small random r/w's are wayyyyy slower, massively so, often on the order of a factor of 1000 slower.
In general, the lower class cards actually beat the class 10s, often by a large factor, so the rom running from it feels snappier and more responsive than on a class 10.
Check out the (admittedly over a year old, but still a good example) comparison charts HERE, and note especially the 4th and 5th charts, where the class 4 blows the others down by over a factor of 100
http://www.tomshardware.com/reviews/microsdhc-memory-card-performance,3011-12.html
samsamuel said:
as the creators of nativeSD and others have pointed out, it's all down to the small block size random read/write speed, that's what most read/writes are when running an OS, small and random.
While reading (for example) a video, it will read bigger, almost certainly sequential, block sizes.
The industry quoted max speed of a card is almost always based on the largest block size, usually read, always sequential speed, so a class 10 can read large block sizes at 10meg, however, test the card for all different block sizes and you'll see that small random r/w's are wayyyyy slower, massively so, often on the order of a factor of 1000 slower.
In general, the lower class cards actually beat the class 10s, often by a large factor, so the rom running from it feels snappier and more responsive than on a class 10.
Check out the (admittedly over a year old, but still a good example) comparison charts HERE, and note especially the 4th and 5th charts, where the class 4 blows the others down by over a factor of 100
http://www.tomshardware.com/reviews/microsdhc-memory-card-performance,3011-12.html
Click to expand...
Click to collapse
yugoport said:
You should read first my friend,this way you will save some Money,any Sandisk class 4 will be faster than any class 10 in native sd, simply because with a class 10 card you get high sequential data transfer and very low random data transfer, and that is the secret random access transfer rate.
But the manufacturers won't tell you the random access speeds, you have to test.
Use crystal disk mark to test the sd and check for the last 2 values..
And read..and search..
Sent from my GT-I9300 using xda premium
Click to expand...
Click to collapse
This is crystal disk mark for my SD Card , See it and tell me where exactly the problem !!
x_max_best said:
This is crystal disk mark for my SD Card , See it and tell me where exactly the problem !!
Click to expand...
Click to collapse
Look at my scandisk class 2 32 gb below, in the last field at right is basically more than 30 times faster than yours in the random write..
https://www.dropbox.com/s/bgbm6tcv99a1r7j/scandisk class 2 32 gb.JPG
This is crystal disk mark for my SD Card , See it and tell me where exactly the problem !!
Click to expand...
Click to collapse
The random write speed is to low! Look by "write 4k" and "write 4k/QD32".
Look here, the second ranking is importand for us:
http://forums.whirlpool.net.au/forum-replies.cfm?t=1582172
And here from XDA:
http://forum.xda-developers.com/showthread.php?t=1150369
x_max_best said:
This is crystal disk mark for my SD Card , See it and tell me where exactly the problem !!
Click to expand...
Click to collapse
as already noted 4k and 4kqd32 (qd32 = there are 32 or more write operations in the queue, probably quite common when running an operating system) speeds are atrocious, thats gonna be your real speed, more often than not, somewhere between 38K/sec and 0.6MB/sec,,, a world away from 10MB/sec. Course, oftentimes the system is reading those small blocks, not writing, which is faster so you can probably make a rough estimate of total average r/w at around 2MB/sec, still a long way from 10MB/sec
yugoport said:
Look at my scandisk class 2 32 gb below, in the last field at right is basically more than 30 times faster than yours in the random write..
https://www.dropbox.com/s/bgbm6tcv99a1r7j/scandisk class 2 32 gb.JPG
Click to expand...
Click to collapse
Fietz said:
The random write speed is to low! Look by "write 4k" and "write 4k/QD32".
Look here, the second ranking is importand for us:
http://forums.whirlpool.net.au/forum-replies.cfm?t=1582172
And here from XDA:
http://forum.xda-developers.com/showthread.php?t=1150369
Click to expand...
Click to collapse
Your Conclusion is wroung , i retest my card with CrystalDiskMark and the result was interesting (attachment) ,and all i did to get this result is change the test size to 50 mb , i start to believe that there is noting called fast Native SD , it is just placebo effect.
if you want me to believe u ,provide a video in Youtube for the fast Native SD that u have and let me know what do u mean by fast Native SD
x_max_best said:
Your Conclusion is wroung , i retest my card with CrystalDiskMark and the result was interesting (attachment) ,and all i did to get this result is change the test size to 50 mb , i start to believe that there is noting called fast Native SD , it is just placebo effect.
if you want me to believe u ,provide a video in Youtube for the fast Native SD that u have and let me know what do u mean by fast Native SD
Click to expand...
Click to collapse
I don't need to prove anything..if you think it's not fast or it doesn't suits you use nand.. I have more important things to do..
Sent from my GT-I9300 using xda premium
I have a PNY U3 Turbo Performance 64GB High Speed MicroSDXC Class 10 UHS-I, up to 90MB/sec Flash Card. It seems to get slower with time. I mean it is literally taking me like 60 seconds to delete a few pictures. The card is formatted to portable storage. I only store music and photos on it. Reformatting offered no improvement. It is borderline unusable.
So, I benchmarked and found that my read speed is about 59 mb per sec and my write speed is 4.4 mb per sec. I can live with the read speed but the write speed is horrible. What gives?
How many pictures are you talking about, and how large are they?
128KB clusters?
I found trying to use 4K clusters in exfat was resulting in the same slowness. Reformatted to 128KB clusters and it flys (~20MB/s write ~70MB/read). Sandisk ultra plus 64GB.
I have this issue as well I have one of the fast Samsung cards. First few months worked great super fast but now takes a while to carry over a gig of pics and music. I also notice lag in loading and delteing photos through the phone
These are normal pics 5-6 MB. There has to be something wrong.
Irieone said:
These are normal pics 5-6 MB. There has to be something wrong.
Click to expand...
Click to collapse
If you find any info/solution let us know.
Im thinking about running it through one of the tests that show the read/write speed, if its not up to par on that I will contact samsung see if its covered under their warranty.
Why not opt for a 128GB sd card? My Moto X Pure will be here Thursday. I have a 64GB in my old phone but I may get 128.
Has the OP tried backing up their data and reformatting the card?
gpz1100 said:
Has the OP tried backing up their data and reformatting the card?
Click to expand...
Click to collapse
Yes, I have tried this a few times and it did not help. The write speed of this particular card is somewhere between 5 and 9 mb/sec. I have used multiple benchmarking tools that all confirm the same thing. The sad thing, if you drill down into the specifications for a lot of these new sd cards there is no mention of write- speeds. I asked a question similar to my OP on Reddit and had somebody with the same card echo my issue. Terrible write speeds. I am not in the mood to buy another card with great specs only to find it performs poorly in my phone? There is still a part of me that thinks it's hardware or software related and specific to the phone. I can't quite believe that something advertised as "turbo", UHS-1, Class 10, and 90 mb/sec has an actual write speed of 5 mb/sec. It seems criminal.
^^See my post #3. I've found this card to have very good write speeds on the phone of ~14-16 MB/s, reads around 40-50. On the pc through a usb 3 card reader I can write at the speeds posted above.
But yes, unless you drill down, the marketing and advertising doesn't list write speeds. I can see why. It all depends where you'll be using it. I'll be lucky to see sustained 10MB/s read through my dash cam (not sure what the write speed even is) even through the card is capable of 40MB/s +.
I should say, using the moto x, through MTP, I've seen read speeds upwards of 35-40MB/s, write speeds of about 7-10MB. The card is faster through twrp, backup stats indicate ~14-16MB/s.
http://forum.xda-developers.com/moto-x-style/general/micro-sd-speed-chart-t3196020