1

(1 replies, posted in Hardware)

Hi dpapathanasiou,

Have you planned some cut-outs above and around the CPU heatsink?

2

(3 replies, posted in Hardware)

Hi roheve,

roheve wrote:

While looking for more info on power consumption and heat generation (and cooling) for the iMX6 SoC, I found an application note AN4579 (PDF) describing some scenarios. The note also describes some interfaces and methods for the SoC to reduce power consumption (or better, to find a balance between performance and functionality and power usage). This is an interesting read (and I an curious if the some of it is in the Linux kernel for the Novena already).

Can't comment from 1st-person PoV as I haven't observed this, but a friend of mine who has Novena stated last week that the CPU is throttling when being heated enough, so for me this means that the thermal management is already working.

roheve wrote:

It is not helpful that GPU hardware acceleration is not yet operational (so the CPU has to work more now).

CPU doesn't come even close to the GPU hardware in terms of performance. Also, the GPU hardware is already supported for years by the FSL kernels, so if you're fine with some binary blobs clogging around, you can build/play with it.

roheve wrote:

I noticed that even when just running the debian (animated) screen-saver, both the SoC and the DRAM will be 'hot' (not to hot to touch, but, you know, 'warm').

Well, the simple truth is that any high performance SoC nowadays will heat considerably. If it doesn't heat, then either it's throttling, or it's not fast enough.

About the DDR memory - the module consumes about 800mA @ 1.5V, which makes 1.2W. If you want lower dissipation, you'll have to sacrifice performance. For example going to 400 MHz DDR3 clock will reduce the dissipated power to 1W, which is 33% reduction. Also, you can save power by disabling some of the CPU cores, limit their maximum clock frequency, change the scaling governor policy or implement your custom control policy. Also, disabling the unneeded SoC modules and gating their clocks will definitely help reduce the consumption further. My experience when ported U-Boot for a customer imx6 board was that having enabled only the few needed clocks made the IC just warm, and having all the CCM clocks enabled made it very hot (couldn't hold my finger on the IC).

Regards.

3

(3 replies, posted in Hardware)

As long as your LCD panel resolution is no bigger than 1280x800 or 1366x768 @ 60fps, the answer is "yes".

4

(12 replies, posted in Firmware)

Hi Damien,

Please check in Novena's board config file for an option called

#define CONFIG_TFTP_BLOCKSIZE 4096

Either remove it (to use the default value), or change it to 1468.

U-Boot's tftp code is somewhat flaky and OpenBSD's tftp server doesn't like that at all (printing obscure messages). Please share if this fixed the issue for you.

Regards.

5

(4 replies, posted in Hardware)

Short answer: Novena is maxed out, don't waste your time.

Looong answer: imx6 can't support more than 3.75 GiB of memory, due to hardware limitations (DDR3 address space mapped at physical address 0x10000000). The CPU doesn't support PAE so no chance to support more than 32-bit address space, and also some of the system memory will be reserved for the kernel (including a big 256-320 MiB chunk for CMA), so the available memory for userspace applications is expected to be about 3-3.25 GiB. Which is actually more memory than my old ThinkPads had.

While Novena uses standard SO-DIMM DDR3 modules, the maximum supported data rate by the imx6 memory controller is 1066 MT/s. Also, different SO-DIMM modules have different memory organizations and timings, which are stored in the SPD memory, and Novena should read this data and properly calibrate the DDR3 controller for the memory. The factory U-Boot supports only 1 fixed configuration, which was designed specifically for the shipped 4 GiB modules, but there's also a Novena git repo which supports dynamic boot-time calibration. Or said in other words - if you're using the factory U-Boot shipped with Novena, there's a non-zero chance that the board will not boot with module other than the original.

Hope this helps. Regards.

spanner888 wrote:

Now got asssesment for $205.44    =  GST 156.59 + processing fee 48.85    ($US1195 -> 1434.95 ... even though paid nearly twelve months back when closer $A to parity with $US!)

Yep, that's unfortunate, but happened everywhere. The currency exchange rate is calculated at the moment of importing the goods in your country, not the moment of paying the invoice sad.

7

(17 replies, posted in Hardware)

Hi roheve,

Thanks for verifying this. It's good to know that it's not a widespread issue.

Regards.

8

(17 replies, posted in Hardware)

Here's the new discussion related to power optimizations:

http://www.kosagi.com/forums/viewtopic.php?id=175

9

(3 replies, posted in Hardware)

This interesting discussion was started by dbtayl. The purpose is to measure the current consumption of different Novena subsystems and find how to minimise it, while still using the full Novena functionality.

LCD consumption:

dbtayl wrote:

It looks like bunnie/xobs are saying 4.8W max is more typical: http://www.kosagi.com/w/index.php?title … _Main_Page (under "eDP Adapter" section, 1.3W for panel, 3.5W for backlight at max brightness)

No idea what's causing it, but I'm seeing something like 800 mA current from my 3S battery pack (as reported by the Senoko) when idle with moderate backlight, which seems high to me. That's something like 10W total- on par with my Core2Duo laptop idling. And said laptop has a 25W TDP CPU and an SSD that the Novena isn't running. My recollection is that the IMX6Q is something like 3-5W peak, and that's when REALLY stressing it (all CPU cores, plus the GPU).

Maybe this will be solved (or at least helped?) with the upcoming 3.19 kernels? I'm just not sure where the extra power is going. At least the CPU is clocking itself down.

Novena board consumption:

picmaster wrote:

Here's a list of voltages I tried, and the measured current consumption (only serial port console login, no ethernet):

7.2V: 594mA, 4.2W (fully discharged 2S Li pack)
8.4V: 520mA, 4.3W (fully charged 2S Li pack)

10.75V: 422mA, 4.5W (fully discharged SLA battery)
10.8V: 426mA, 4.6W (fully discharged 3S Li pack)
12.6V: 378mA, 4.7W (fully charged 3S Li pack, fully charged SLA battery)

14.4V: 344mA, 4.9W (fully discharged 4S Li pack)
16.8V: 308mA, 5.1W (fully charged 4S Li pack)

The measurements have been recorded from the power supply front panel (Instek PSP-2010), and are not of ultimate precision, but can serve as good starting point.

dbtayl wrote:

Those values are without the LCD?

I did some tests with the LCD. All of these measurements are from the Senoko 'stats' output. My setup is running from batteries (3S pack), so voltages vary slightly. I've got the desktop Novena unit configured, except:

-No SSD
-No speakers attached
-No GPBB or SDR module

The wireless module was removed for some tests, as noted below. The only external device plugged in is a ThinkPad USB keyboard.


With everything on, the backlight 3 steps up from the bottom, I measure ~660 mA @ 12.156 V

Same as above, but without WiFi (card physically removed), I get ~610 mA @ 12.162 V

Turning the backlight to barely on, WiFi removed, and disabling ethernet (ifconfig eth[01] down) gives ~570 mA @ 12.136 V


So by my rough calculations, the WiFi card is 0.6 W, the two ethernet interfaces are 0.35 W combined, and the backlight is something like 0.05 - 0.1 W per step.

That's getting closer to what I'd expect to read. A little more is probably acounted for by the LVDS-eDP controller. Maybe the audio CODEC and FPGA count for a little more, I don't know.


EDIT: The numbers reported fluctuate a bit; I've tried to pick out where they seem to settle, but take it with a grain of salt.

xobs wrote:

You can save a bit of power by turning the Ethernet ports off (ifconfig eth0 down; ifconfig eth1 down).  Just running the PHYs can take a few tens of mA.

10

(17 replies, posted in Hardware)

Hi dbtayl,

dbtayl wrote:

Those values are without the LCD?

That's correct.

Would you mind if I split this discussion in a separate thread, dedicated to the power consumption topic and possible optimizations?

Regards.

11

(8 replies, posted in Hardware)

Hi ballanux,

It looks nice. Did you had issues with the flat cable just below the SSD? Mine is bent upwards, just where the SSD sits.

12

(17 replies, posted in Hardware)

I've put my Novena desktop back together. It works fine without issues on 7.2V, including the LCD, which was my primary motivation for this research.

I'm happy now big_smile. The mistery is resolved!

13

(17 replies, posted in Hardware)

Just "repaired" R105 (Senoko passthrough) by adding a 13K resistor in parallel, so the combined resistance is 10.183K, close enough to design value.

The results - Novena now boots perfectly at 7.2V, together with Senoko passthrough!

14

(17 replies, posted in Hardware)

Hot news.

I checked the voltages in the SMPS regulators, and the first check revealed problem - U11N was receiving a sawtooth wave (swinging between 4-7V), instead of DC. It looked like the DC input voltage was sagging when U11N wanted to draw current and then turned off.

I verified the sawtooth wave on Q100 drain (Senoko passthrough main MOSFET switch), and the issue was still there. The voltage on the gate was flat DC, no sawtooth. Then I understood something is very wrong - the Vgs on Q100 is 1.75V, while it should have been 5V (1/2 of the 10V input voltage, divided by R104 & R105). Which in turn means that either Q101 is defective, or R104/R105 are defective or wrong values.

It turned out that R105 has wrong value - 47K instead of 10K. I verified the Novena ECO list, but there's no such change of R105, so I assume it's a factory defect. Well, 1.75V are below the maximum Vgs (th) voltage of 3V, so it's understandable why it can't deliver the inrush current of the main SMPS regulator.

Well, back to the soldering iron...

15

(2 replies, posted in Hardware)

Hi dbtayl,

Unfortunately I also can't find this datasheet. Here's what I can suggest for measuring the current you're interested in:

Boot your Novena and while the eDP + LCD are working, carefully attach amper-meter across the MOSFET switch Q10L, this will give you the combined currents of IT6251 and LCD. Since you now have the LCD datasheet, there's already an estimation for the current consumption @3.3V, so you can calculate the difference and have a rough value of the IT6251 current.

Q10L is on the PCB bottom, so you'll need to dismount the PCB and prepare in advance the multimeter clips to reduce the chance of frying the board while it's alive (either SMD pin clips or soldered wires will do the work).

Regards.

16

(17 replies, posted in Hardware)

The next experiment - attached Senoko passthrough to Novena and powered the boards via connector JP200. Novena doesn't boot below 13V.

There's something very strange. When I turn-on the PSU at 9V and start increasing the voltage in 1V-steps, I hear tiny buzzing noise coming from the boards at 10V, 11V, 12V and the noise disappears at 13V, when Novena boots. The buzzing noise probably comes from some of the inductors.

Wow - we have progress. I hope the issue is something dummy with the Senoko and not with the LCD.

17

(17 replies, posted in Hardware)

I did the following experiment - took the Novena main PCB out of the desktop enclosure, removed the Senoko passtrough board and all peripherals (USB extension, GPBB, SATA Wifi, eDP). Then I powered the Novena directly via connector JP10N. Here's a list of voltages I tried, and the measured current consumption (only serial port console login, no ethernet):

7.2V: 594mA, 4.2W (fully discharged 2S Li pack)
8.4V: 520mA, 4.3W (fully charged 2S Li pack)

10.75V: 422mA, 4.5W (fully discharged SLA battery)
10.8V: 426mA, 4.6W (fully discharged 3S Li pack)
12.6V: 378mA, 4.7W (fully charged 3S Li pack, fully charged SLA battery)

14.4V: 344mA, 4.9W (fully discharged 4S Li pack)
16.8V: 308mA, 5.1W (fully charged 4S Li pack)

The measurements have been recorded from the power supply front panel (Instek PSP-2010), and are not of ultimate precision, but can serve as good starting point.

The result of this experiment is - the Novena PCB works OK through the full voltage range I expected, which is very nice.

18

(17 replies, posted in Hardware)

Hi dbtayl,

Thanks for the useful info. I can confirm that my display model is the same (Innolux N133HSE-EA1). The display is specified to work in the range 5-21V, and 12V is showed as typical voltage. Max power consumption is documented as 5.15W, which is 430mA consumption @ 12V, which seems quite reasonable.

The only thing which looks suspicious is the 1.5A inrush current, which is specified without a voltage.

19

(17 replies, posted in Hardware)

Hi xobs,

Do you have the datasheet of the display? It would be great to see the exact numbers for LCD power supply range and required current. My hope is that the desktop and laptop models are using the same LCD panel, so the panel power requirements should be reasonable to cover the laptop use case.

I'm interested in running the Novena desktop on 12V SLA battery, so my work doesn't get interuppted by power outages (not so frequent in the city, much more frequent in the summer house). I would love to experiment how well the Novena works on Lithium pack, but this would be more for satisfying my curiosity than for practical purposes.

Regards.

20

(17 replies, posted in Hardware)

I'm looking at what is the operational voltage range of the desktop board. It works with supplied 18V adapter, which is fine, but how low it can go?

Tried 12V with a programmable power supply I have, but it didn't started. At 13V the Novena started, but had display turns off at regular intervals (blinks). At 13.5V the display and the board are working OK.

My analysis of the passthrough board shows that it has only local 3.3V VREG for the Senoko chip, and the power supply is passed through Q100 power switch directly to the desktop board. So no regulation/limitation at least on the Senoko side.

Next, the desktop board is powered via J10N and goes to BATT_PWR power rail. There are 2 consumers of this rail - the main 5V regulator (U11N) and the LCD backlight.

So probably the senoko and desktop boards themselved should be able to work properly to quite low voltages, for example 8.4V as provided by 2S Lithium cells.

The main question is - what is the working voltage range of the LCD backlight?

Thanks in advance. Regards.

21

(24 replies, posted in Hardware)

Hi ballanux,

My LCD bracket doesn't fit properly, as the holes are offset by at least 1mm each. I managed to hold it with several of the screws, but the bracket is bent, having 8-10mm distance in the center of the LCD panel.

I'll do some machining in the next day or two, to resolve this mechanical issue. Unfortunately I'm not sure whether/how it will help with the backlight bleeding.

Regards.

22

(1 replies, posted in Hardware)

Hi robryk,

To answer your questions:

1. Is there any reverse polarity protection?
Well, yes, but it doesn't work by default and there are quirks. If you power the board via JP200, reversing the power supply polarity should fry the Senoko (Novena and all attached peripherals should be protected by Q100). One possible measure which can be taken is to short P102, so D105 will be forward-biased when the power supply is reversed, thus *attempting* to protect the boards. Although the diode has low forward-bias voltage (0.49V), the maximum continuos current is only 4A, which can be less than the current capability of an external/laboratory power supply, and if the diode failure mode is to open when overloaded, this is bad news for the silicon on-board.

2. What's the use for D105?
D105 can be used for 2 things - to power the board via P102 and having a limited reverse-polarity protection, or to short P102 and do what I described earlier. Let me explain why powering the board via P102 again allows only a limited protection. When D105 is reverse-biased, the reverse current is around 5-6uA, which means that although not instantly, C117 will be charged to an increasing potential, which sooner or later will be enough to conduct current through parts of the circuit which are not designed to work in such mode.

What would be the best protection?
If you don't mind loosing 0.8-1W on D105, you can power the board via P102 and implement small fix - add one more reverse-oriented power Schottky diode between D105 cathode and ground. In the case when power supply is reversed, the reverse current of D105 won't do any harm, as the other diode will start conducting at 0.3-0.35V and will protect the silicon.

Even better approach would be to use schematic equivalent to Q100 - P-channel MOSFET with grounded gate, which conducts only when input voltage is positive. Just like active MOSFET rectifier.

Hope this helps. Regards.

23

(9 replies, posted in Software)

Well, you can start digging into Yocto (branches dizzy & master), understand how the Vivante components are working there and port them to your Linux distro. Or wait for someone else to do it.

I've used the GPU & VPU with Yocto on custom imx6 board, and they seem to work.

24

(21 replies, posted in Chat/off-topic)

Today just shared with a friend of mine (also a Novena-fan), that according to Murphy's law, I'll be the precisely last guy who will receive his Novena(s). Ehh, fate...

The most positive news is that at least I now know that my package is in the bureaucrats... I mean, Customs' hands, so hope to have it here soon.

25

(12 replies, posted in Hardware)

Great info, thanks for sharing!