enjoy wrote: ↑
Sun Jul 28, 2019 7:33 pm
jamesh wrote: ↑
Wed Jun 26, 2019 7:42 am
Gavinmc42 wrote: ↑
Wed Jun 26, 2019 5:47 am
I am thinking multiple version of Computes.
CM4 and CM4+ which has the PCI-e broken out.
Those USB/Lan parts can be on the motherboard, two types of motherboards?
Third party MB's, 6 UART versions?
Wonder if BCM can make enough 2711's?
We don't have the staffing levels to do so many variants.
We are still thinking about how, if we were to go ahead, the CM4 would work, as as stated above, there are not enough pins on the SODIMM connector to expose everything.
CM4 would be such a great product. Have you guys thought about SMARC standard? It has 314 pins and lots of I/Os.
SMARC is too big in many applications and migration of products from CM3 (SODIMM200) to SMARC would be impossible. Therefore I would propose instead to migrate to 260 contact SODIMM socket like https://www.te.com/usa-en/product-2309413-1.html
. This way you could have all interfaces broken out and fit it mechanically in the same space as the CM3. Of course, the downside is you will have to re-route the base PCB, but you have to do that for the SMARC case as well.
But if you change the pin-out intelligently, so that the legacy signals remain more or less at the same mechanical position on the SODIMM, the effort for re-routing could be minimized to a matter of hours. You could add a pair of contacts every 6 ... 7 contact pairs. At one position you would have to add two pairs (for PCIe).
If it were my choice, this is how I would do it. I agree that there should by a backward compatible CM4 variant in SODIMM200 form factor, without PCIe.
To sum up, my proposal would be:
- CM4 in SODIMM200, backward compatible to CM3(+)
- CM4 Lite in SODIMM200, backward compatible to CM3(+) Lite
- CM4-260 in SODIMM260, with PCIe and other additional interfaces broken out to the SODIMM260, eMMC on module
- CM4-260 Lite in SODIMM260, with PCIe and other additional interfaces broken out to the SODIMM260, eMMC off module
We have an industrial application in mind and need PCIe as a fast and low-latency data link to a FPGA. I would really like to see that CM4 sticks to the principles of CM3: Means, only have the minimal set of components on the module, that keeps the module low-cost and highly flexible at the same time. Functions like USB3 controller or Ethernet PHY don't belong on the module but on the base board.