Tx2 read/write performance drop

TX2 read/write performance on L4T R32.3.1(jp43) is lower than on L4T R32.2.1 (jp3.3) as nvpmodel was set to 0.
Is it normal? any reason?

Plus, as Tx2 nvpmodel is set 0, the read/write performance is lower than as nvpmodel is set to 3.
Is it normal? any reason?

Here are the test results.
r_w_results|690x406

Thank you,

Hi HuiW,

Please share the dd command you are using.

Hi WayneWWW,

Here are the command.
write: dd if=/dev/zero of=1GB bs=1024 count=1048576 conv=fdatasync
read: dd if=1GB bs=1024 | dd of=/dev/zero

Thank you,

Hi HuiW,

Are you enable jetson_clocks?

sudo nvpmodel -m 0
sudo jetson_clocks

Hi carolyuu,

Yes.

also tested nvpmodel with 3 and new L4T performance also lower then previous L4T.

Thank you,

Hi HuiW,

We can’t reproduce your issue.
For our internal testing, we’re using “IOZone” to do read/write performance test.
We also tried using your DD command, but nvp mode 0 result is better than mode 3.
Could you also try using IOZone tool?

Hi carolyuu,

Thank you for your prompt support.
Would you kindly share the results you got? (nvpmodel 0)
Which L4T did you test?

Would you also share the IOZone command you used too?

Thank you,

Hi HuiW,

We’re test with r32.4.2 (JetPack-4.4 DP release).

Share the steps for IOZone:
sudo apt-get update sudo apt-get install iozone3

==Sequential read==
$ iozone -ecI -+n -L64 -S32 -s64m -r512k -i0 -i1 -l8 -u8 -m -t8 -F /mnt/file1 /mnt/file2 /mnt/file3 /mnt/file4 /mnt/file5 /mnt/file6 /mnt/file7 /mnt/file8

==Sequential write==
$ iozone -ecI -+n -L64 -S32 -s64m -r512k -i0 -i1 -l8 -u8 -m -t8 -F /mnt/file1 /mnt/file2 /mnt/file3 /mnt/file4 /mnt/file5 /mnt/file6 /mnt/file7 /mnt/file8

Hi carolyuu,

Could you test L4T R32.3.1(jp43)?

Yes. As your tested,
We found r32.4.2 (JetPack-4.4 DP release) did not have the drop issue as we tested.

Thank you,

Hi HuiW,

The JetPack-4.3 result the same.
The mode 0 performance result is better than mode 3.
Are you still see the issue on JP-4.3?

Hi carolyuu,

Would you share the DD test result?

Is TX2 read/write performance on L4T R32.3.1(jp43) higher or lower than on L4T R32.2.1 (jp3.3) as nvpmodel was set to 0?

We have two users got the drop issues on JP43.
Is there any idea about the different?

Thank you,

Thank you,

Hi HuiW,

I would suggest you and your customer could try to use iozone. DD is not like a formal tool to benchmark the I/O performance. You could see even run dd on host could have varied value from low to high.

BTW, jp3.3 is rel-28.x but not rel-32.2.1. What release are you comparing exactly?

Hi WayneWWW,

Sorry for the typo.

It’s L4T R28.2.1(jp3.3).

TX2 read/write performance on L4T R32.3.1(jp43) is lower than on L4T R28.2.1 (jp3.3) as nvpmodel was set to 0.

Thank you,

Hi HuiW,

Please share your TX2 module serial number.
Thanks!

Hi Carolyuu,

My Tx2 module (with jp43) SN is 0325117093125.
Tx2 module (with jp44) SN is 0322017038178

Thank you,

Hi HuiW,

Are they B02, C03 or D01 modules?

Hi WayneWWW,

Sorry. I don’t know how to check it.
Would you share a way to get the information?

Thank you,

Hi HuiW,

Do you check the SN from the sticker of your module or from the software?

Are you able to see something like " 699-83310-1000-D00" on your sticker?

from Tx2 module

My Tx2 module (with jp43) SN is 0325117093125–699-83310-1000-B02 C
Tx2 module (with jp44) SN is 0322017038178 — 699-83310-1000-B02 C

Thank you,