Matmul-16 example gets stuck

[Calle]

[Calle]

And if I run 32x32 matrices I get errors in every element every second row (row numbers 1, 3, 5, etc).

[cklin]

[ticso]

Unfortunately I have to say that one of my two boards I'd received today show the same problem.
Of 10 tests about half matmul-16 went wrong.
The board with failure is SN 0003573
The good board is SN 0003957
I directly attached the heatsink and added a big fan on first powerup.
Both Zynq and Epiphany feel cool and never run hot for me.
Powersource is a workbench supply set to 5V/3A, but could do 5A as well.
I used the same setup for testing both boards - same supply cable, same supply, same SD card, same ethernet cable, ...
I didn't attach anything to USB or HDMI and just ssh to the board.
The good board take a few mA more power - currently 1.010 to 1.020 A idle.
Would have to reattach the failing board for current rate on that, but it was slightly below 1A.
The working board gets 4.69V because of cable loss - would have to measure the other one, but should be a bit higher because of lower current.
Volatage drop a few more mV during process run.
I could do retest the faulty board with raised voltage to compensate cable loss if required.
What is the upper supply limit?

[ticso]

Btw: It would have been good to ship the boards with ESD shielding.
This pink foam is plastic which doesn't build charge by itself, but it is not shielding.

[ticso]

Voltage doesn't seem to influence and temperature variation also seems to make no difference.

[ticso]

I noticed a difference in marking of the Epiphany.
The good one is marked in the lowest line as 428 and on the problematic board is marked 5494.

[miro]

Hi, I'm unfortunately also affected by the Matmul-16 and ERROR: C_epiphany is different from C_host !!!.
I ran is about 20 times and I get about 50% correctness. Does someone has a code to identify the faulty address?
(Mine is 0002571, epiphany 2646)

[timpart]

[aolofsson]

miro, cklin:
We are working on a fix that could possibly fix your Epiphany issue. (slightly higher voltage) Can you hold on to the faulty board for two more weeks? If we haven't released a fix by June 1st, then please send the problematic boards back to us using the rma process.(and we'll send you new boards).
Thanks,
Andreas