Directions for Heat Sink?

dvl36 · Published in 2015-12-8 18:52:28

bronco replied at 2015-12-8 18:07
Did you also gave lima-memtester a try? http://linux-sunxi.org/Orange_Pi_PC#DRAM_clock_speed_limit ...

Not yet. I had tried lima-memtester before, but there was a problem with mali driver or something like that.

bronco · Published in 2015-12-8 19:19:38

Edited by bronco at 2015-12-8 19:22

dvl36 replied at 2015-12-8 18:52
Not yet. I had tried lima-memtester before, but there was a problem with mali driver or something ...

Same here. Now I'm really concerned since the preliminary test results doesn't look that promising. And bit flips are really bad leading to all sorts of trouble, data corruption and crashes (we normally attribute to 'software' but it might be also wrong hardware initialisation).

This new approach https://github.com/ssvb/lima-mem ... ange-pi-pc-fel-test comes with everything already prepared. You just need a x86 box where you can connect your OPi to using FEL/USB (and sunxi-tools -- unfortunately all my 'physical' PCs run OS X and I'm still unable to 'make sunxi-fel' -- no time to dig deeper at the moment)

EDIT: Just forgot that I should be able to use any of the other SBCs lying around here and running linux to do the test...

BTW: Did you altered the dvfs table inside script.bin or tested with the defaults (only 1.3V and 1.5V)?

bronco · Published in 2015-12-8 19:42:43

Edited by bronco at 2015-12-8 19:55

Got it running on a Banana Pi M3 after compiling the most recent version of sunxi-tools (the one included in most distro images is too old):

root@bananapi:~/fel-boot-lima-memtester-on-orange-pi-pc# lsusb
Bus 003 Device 001: ID 1d6b:0001 Linux Foundation 1.1 root hub
Bus 002 Device 001: ID 1d6b:0001 Linux Foundation 1.1 root hub
Bus 001 Device 005: ID 1f3a:efe8 Onda (unverified) V972 tablet in flashing mode
Bus 001 Device 003: ID 05e3:0718 Genesys Logic, Inc. IDE/SATA Adapter
Bus 001 Device 002: ID 1a40:0101 Terminus Technology Inc. Hub
Bus 001 Device 001: ID 1d6b:0002 Linux Foundation 2.0 root hub
root@bananapi:~/fel-boot-lima-memtester-on-orange-pi-pc# ./fel-boot-lima-memtester-on-orange-pi-pc 480
Stack pointers: sp_irq=0x00002000, sp=0x00005E08
MMU is not enabled by BROM
Generating the new MMU translation table at 0x00044000
=> Executing the SPL... done.
Setting write-combine mapping for DRAM.
Setting cached mapping for BROM.
Writing back the MMU translation table.
Enabling I-cache, MMU and branch prediction... done.
Writing image "U-Boot 2016.01-rc1-00441-g75507d", 321065 bytes @ 0x4A000000.
0.2 kB written in 0.0 sec (speed: 69.6 kB/s)
Passing boot info via sunxi SPL: script address = 0x43100000
36.0 kB written in 0.0 sec (speed: 737.2 kB/s)
7048.2 kB written in 7.6 sec (speed: 922.7 kB/s)
Starting U-Boot (0x4A000000).
Now the Orange Pi PC board should boot and show a spinning cube on
a HDMI monitor. If the cube stops spinning or does not show up in
the first place, then the selected DRAM clock speed is unreliable.
Also one of the LEDs should be blinking. If this LED stops blinking,
then the test has failed. When the second LED lights up while the
first one still keeps blinking, then enough time has passed already
and we can consider the test a success. The LEDs allow to use this
test even when no monitor is connected.
You still can keep the test running for a longer period (for example
overnight) for extra confidence. Also please consider to have some
extra safety headroom. For example, if the test seems to run fine at
'X MHz', but fails at 'X + 24MHz', then it may be reasonable to use
'X - 24MHz' setup in the end. Good luck.

Copy code

First test with 672 MHz immediately crashed

600 MHz crashed after 2 minutes, now I let 552 MHz try... also crashed after 2 minutes. Woohoo

528 MHz also broken, now trying 480 MHz...

fritz · Published in 2015-12-8 19:50:32

mine passed with 600Mhz.

dvl36 · Published in 2015-12-8 20:40:38

Edited by dvl36 at 2015-12-8 20:57

bronco replied at 2015-12-8 19:19
Same here. Now I'm really concerned since the preliminary test results doesn't look that promising. ...

Checked. At frequencies 504Mhz and above LED stop blinking. At 480Mhz heatsink temperature (at bottom) only 40-41C.
Previously I tested with four linpacks (4000) without any problems. (Two days running)
So it seems that something wrong with this test.

ADD: Default voltages of loborises Debian Wheezy image. Should be 1.5v.

bronco · Published in 2015-12-8 21:20:06

dvl36 replied at 2015-12-8 20:40
Checked. At frequencies 504Mhz and above LED stop blinking. At 480Mhz heatsink temperature (at bott ...

I don't know whether heatsink temperature is relevant here at all. As far as I understand the lima-memtester is a tool that has to run under worst case conditions putting as much stress on the SoC as possible (therefore the spinning cube rendered with Mali) to get a clue when bitflips happen (might be related to undervoltage of DRAM -- this has nothing to with the Vcore voltage of the CPU cores). I don't know how that translates to real-world situations (or running only cpu intensive benchmarks as you did)

And obviously DRAM initialisation/settings differ between what we use now and what the most recent u-boot does:

https://github.com/loboris/Orang ... onfig.fex#L179-L203 vs.

https://u-boot.googlesource.com/ ... nxi/dram_sun8i_h3.h
https://u-boot.googlesource.com/ ... angepi_pc_defconfig

BTW: My 1st OPi PC didn't survive even 480 MHz. Currently testing 456 MHz

bronco · Published in 2015-12-8 21:49:38

fritz replied at 2015-12-8 19:50
mine passed with 600Mhz.

Did you also tested with ssvb's new fel version using his kernel fork and recent u-boot or when running loboris' 3.4.39 kernel?

fritz · Published in 2015-12-8 22:39:20

I don' t use the kernel by @loboris.
Test was done with u-boot binary and uImage provided by @ssvb on opi plus.
Restested with 600 Mhz and this time it failed at loop 1/8:

Bit Flip : testing 111READ FAILURE: 0xffffdfff != 0xffffffff at offset 0x0054a294 (bitflip)

bronco · Published in 2015-12-9 17:33:21

fritz replied at 2015-12-8 22:39
I don' t use the kernel by @loboris.
Test was done with u-boot binary and uImage provided by @ssvb o ...

Seems to be related to DRAM initialisation in the early boot stage: https://groups.google.com/d/msg/ ... 4k8nIE/PKkl7cwUAQAJ

dvl36 · Published in 2015-12-9 18:19:25

Edited by dvl36 at 2015-12-9 18:29

bronco replied at 2015-12-8 21:20
I don't know whether heatsink temperature is relevant here at all. As far as I understand the lima ...

@ssvb at linux-sunxi Google groups:
-----
Currently U-Boot v2016.01-rc2 fails the lima-memtester check unless
the DRAM clock speed is reduced to 456 MHz on my Orange Pi PC. But
if I use the boot0 bootloader to boot the same kernel image
(uImage-3.4-sun8i-h3-lima-memtester), then the test works fine.
This means that there must be still something wrong with the H3
DRAM init code in U-Boot, or maybe some clocks are wrong
------

How we can test it using boot0 bootloader?