NicoD Posted September 30, 2019 Share Posted September 30, 2019 Hi all. I've made a new video where I test all storage devices for the NanoPi M4. 8GB, 16GB and 32GB eMMC modules, NVMe drive, sd-card and SSD over USB3. Here's the video. Here's my gathered data. Storage speeds -------------- 8GB eMMC module read : 160 MB/s write : 11 MB/s 16GB eMMC module read : 175 MB/s write : 50 MB/s 32GB eMMC module read : 173 MB/s write : 111.6 MB/s 256GB NVMe read : 697.7 MB/s write : 401 MB/s (630MB/s for 2.5GB, then 300 MB/s) M4 sd-card read : 70 MB/s write : 20 MB/s M4 swap read : 1.9GB/s 0.01 msec access time Greetings, NicoD Link to comment Share on other sites More sharing options...
qstaq Posted September 30, 2019 Share Posted September 30, 2019 Interesting stats Surprised that a 256GB NVMe drive only has 2.5GB or 1% of its capacity as fast flash and then the rest is QLC I presume? What model is it? Seems a bit of a con to have an NVMe drive that throttles down to half SATA speeds once the MLC buffer is full Link to comment Share on other sites More sharing options...
NicoD Posted September 30, 2019 Author Share Posted September 30, 2019 33 minutes ago, qstaq said: Surprised that a 256GB NVMe drive only has 2.5GB or 1% of its capacity as fast flash and then the rest is QLC I presume? What model is it? Seems a bit of a con to have an NVMe drive that throttles down to half SATA speeds once the MLC buffer is full I can't read the model number now. It's a no-brand. I've gotten them for free, so a con it's not I know samsumg EVO's are a lot faster. But they also cost a lot more money. Cheers Link to comment Share on other sites More sharing options...
qstaq Posted September 30, 2019 Share Posted September 30, 2019 Oh well, that's just a bonus then Just out of interest, did you write test the zram performance? And what compression algorithm are you using? I've had some strange performance issues with zram on Arm v Power7/8 v Intel Arm seems to be the only modern platform where LZO and LZ4 are not hardware accelerated. I don't know if this is a limitation of Arm's NEON vector extensions or if nobody got round to implementing vector instructions in the Arm ports Link to comment Share on other sites More sharing options...
NicoD Posted September 30, 2019 Author Share Posted September 30, 2019 23 minutes ago, qstaq said: Just out of interest, did you write test the zram performance? And what compression algorithm are you using? I've had some strange performance issues with zram on Arm v Power7/8 v Intel Yes I did. 1.9GB/s for the M4 its zram. The Odroid N2 gets 2.4GB/s. I'd like to test the M4V2. I'd need to get Armbian running on it. I think the friendlyElec images run the lpddr4 at the same clock as the lpddr3. There's even a little less performance. So I want to know why and how this is.It's the default settings of zram-config. I think that's LZ0. Not sure. Cheers Link to comment Share on other sites More sharing options...
qstaq Posted September 30, 2019 Share Posted September 30, 2019 1.9G for read and write? In which case it must using NEON instructions now. That's good Link to comment Share on other sites More sharing options...
mar0ni Posted January 20, 2020 Share Posted January 20, 2020 At the weekend, I tested the 64GB eMMC module and here are the results, just for the records 64GB eMMC module read : 180 MB/s write : 165 MB/s ... and also a Samsung 970 Evo Plus (500GB) with an amazing write speed of 730 MB/s 500GB 970 Evo Plus read : 736 MB/s write : 730 MB/s 1 Link to comment Share on other sites More sharing options...
![m][sko](https://forum.armbian.com/uploads/n2a/n2a-180904eb9612afa23a4c3358385bd86f384613e3.avatars_letters_gd.png){kind=avatar}
Recommended Posts