IMX6 Crash when do GPU stability test

[david5icp]

I do GPU test on our board, It's base on SABRESD Board.But we change ddr3 to lpddr2. And using 32 bit single channel mode with 2 chip selects, 512B pre channel. We can pass ddr_stress_tester more then 2 hours  (ver2.52).

And we also pass classic tset (without gpu) more then 4 hours. But when we test with GPU, the system will hang in 5 minutes (always less then 2 minutes), and without any logs. 

We check all power domains are normally... How do i to verify to debug ? Thanks!

BR.

David  

[igorpadykov]

Hi David

crash can be caused by gpu bugs, so may be recommended to use
latest L4.1.15 were they were fixed.
http://www.nxp.com/products/microcontrollers-and-processors/arm-processors/i.mx-applications-process...
Another cause may be power ripples, note ddr tester does not stress well
power supplies. For example one can try the following experiments to determine
if it is a hardware PCB issue. Hand solder 22uF capacitors directly under center of chip.  
Solder on top of the 0402 chips.
Have software Increase voltages of ARM_CAP, SOC_CAP, PU_CAP.

Best regards
igor
-----------------------------------------------------------------------------------------------------------------------
Note: If this post answers your question, please click the Correct Answer button. Thank you!
-----------------------------------------------------------------------------------------------------------------------

[david5icp]

Hi igor,

 Thanks for your reply so quickly... All power domains have more than 22uF on CPU buttom side. And i also adjust ARM_CAP, SOC_CAP, PU_CAP to 1.3v, but still hang. The ripple of these power are less 30mV

 I also do a test, to keep CPU clock rate on 996Mhz and set GPU3D clock rate to 3/64, it will pass the test more than 2 hours.Does it mean HW issue?

 Thanks!

BR

David

[michaelguntli]

Does the system completly freeze until the watchdog reset occurs? I assume it's another problem because the arbitration within the memory controller?
LPDDR2 has much less bus bandwith, thus causing problems when the GPU3D gets "too hungry".
https://community.nxp.com/message/935193?commentID=935193#comment-935321