Availability

[W. H. Heydt]

I heard before/around Xmas.

Heh... It is "before Xmas" at present and, if you go by US retail practice it is already "around Xmas". Does make one wonder, though, if the launch will happen on the 1st anniversary of the launch of the Pi Zero, Thanksgiving Day in the US (24 Nov. this year).

[jrychter]

Oh, I do hope that nobody is playing games like that (anniversary launches), especially with the essentially non-consumer product that the Compute Module is.

I am hoping for quick availability, but more importantly, I'm really hoping we will finally see some specs. I don't want to have to wait until actual launch just to get power requirements…

[scotty101]

I don't want to have to wait until actual launch just to get power requirements…

You'll have to given previous history. And it will no doubt you will have to rely on some kind hobbyist measuring the power consumed rather than an official datasheet.

[jamesh]

Oh, I do hope that nobody is playing games like that (anniversary launches), especially with the essentially non-consumer product that the Compute Module is.

I am hoping for quick availability, but more importantly, I'm really hoping we will finally see some specs. I don't want to have to wait until actual launch just to get power requirements…

With regard to power - more than the CM1...

Not sure the RPF have ever released official power figures for anything (it varies considerably with what you do with the device for example), so seems unlikely they will do so for the CM3.

[jrychter]

I realize it sounds like I'm complaining, but I just wanted to explain: "more than the CM1" doesn't cut it, when your PoE-powered device sits right on the edge of two different flyback transformers. Do I overprovision and accept the increased cost and physical dimensions of a larger transformer? Or do I wait, putting the product on hold until CM3 is actually physically available?

The CM is clearly designed for a different market, I wish the RPF also accepted that this market works in a different way.

Anyway, I don't want to sound too negative, it's just that the complete lack of information puts my project in a limbo and it's very annoying.

[jamesh]

I realize it sounds like I'm complaining, but I just wanted to explain: "more than the CM1" doesn't cut it, when your PoE-powered device sits right on the edge of two different flyback transformers. Do I overprovision and accept the increased cost and physical dimensions of a larger transformer? Or do I wait, putting the product on hold until CM3 is actually physically available?

The CM is clearly designed for a different market, I wish the RPF also accepted that this market works in a different way.

Anyway, I don't want to sound too negative, it's just that the complete lack of information puts my project in a limbo and it's very annoying.

To expand on the point in my previous post, do you need all four cores runnings? Will you be using the camera? Will you be using the 3D block? Will you be using the H264 decoder or encoder?

Every one of those affects the power requirements of the device. And in fact, the only difference between CM1 and CM3 in all those respects is the ARM cores requirements. Everything else is the same.

Note if using all of the above at the same time, you will also need a heatsink.

[W. H. Heydt]

I realize it sounds like I'm complaining, but I just wanted to explain: "more than the CM1" doesn't cut it, when your PoE-powered device sits right on the edge of two different flyback transformers. Do I overprovision and accept the increased cost and physical dimensions of a larger transformer? Or do I wait, putting the product on hold until CM3 is actually physically available?

The CM is clearly designed for a different market, I wish the RPF also accepted that this market works in a different way.

Anyway, I don't want to sound too negative, it's just that the complete lack of information puts my project in a limbo and it's very annoying.

Much as i agree with your basic point, that there should be a way to get fundamental specifications enough in advance of availability to design and set up manufacturing for carrier boards, it should still be possible to approximate and upper bound for the current requirements. The recommended value for a Pi3B is 2.5A. Out of that, 1.2A is for the USB ports. The LAN chip current requirements should be discoverable. What is left would be a good value to run the SoC, so 1.2A for the CM3 should be good, and very likely that 1A would suffice.

From the consumer side, a somewhat overspec'd power privision is desirable because that will mean that the power circuitry is running well below its maximum rating and--therefore--will be likely to have a longer mean time between failure.

[jrychter]

To expand on the point in my previous post, do you need all four cores runnings? Will you be using the camera? Will you be using the 3D block? Will you be using the H264 decoder or encoder?

Hmm. I'm not sure if these questions are serious (I use two cameras, encoded to mjpeg, no display whatsoever, so my core requirements are actually low), or if you're just trying to show me that this is a complex issue. I do realize that this isn't simple. But please let me show an example of how NXP gets this right.

Let's look at the datasheet for the KL03Z microcontroller, available at http://cache.freescale.com/files/32bit/ ... M48SF0.pdf — Section 2.2.5 on page 10 lists the "KL03 QFN packages power consumption operating behaviors" — you get typical and max currents for all power modes for the main core. This is followed by the WLCSP package, and then by what they call "peripheral adders", meaning how much current you need to add if you are using a particular peripheral. These tables let me quickly estimate the power requirements for my particular use case, with fairly good precision.

I do realize that the BCM283x is a vastly more complex chip, and the CM adds more complexity on top, but it is entirely possible to create a similar table listing the core(s) power requirements, and peripheral adders.

Every one of those affects the power requirements of the device. And in fact, the only difference between CM1 and CM3 in all those respects is the ARM cores requirements. Everything else is the same.

…and this is actually a very useful bit of information, that I did not have until now!

thanks,
--J.

[jamesh]

Each camera will be using perhaps 200mA, they are surprisingly power hungry. Encoding to mjpeg isn't that bad, probably 50mA? H264 is about 60-75mA So the camera side of things will be pushing 475-500mA. So actually, your core requirements are quite a bit higher than most people with a single camera. No display means no 3D which saves some power, so presumably you just need to run the ethernet and one or more of the ARM cores. ARM core power requirements is the tough one. If you are running them all at full tilt, at 1.2GHz, then they are going to need a hatload of power. But that seems unlikely given the other stuff you mention. ANywhere between 200 and perhaps 1500mA depending on the workload would be my guess (Not found figures for A53's on 40nm silicon).

I'm sure it is completely possible to make a power table with approximate figures (Microcontroller are much easier to document). Whether it's cost effective in time for the RPF to do so, I don't know, and I doubt anyone at Brcm is interested nowadays since they got rid of most of the VC4 team.

I would be inclined to get a Pi3, measure the power consumption at the workload you envisage, then add another 200mA for the second camera.