Would we not benefit from 8Gb RAM in the same way that RPi has, or is that too simplistic a view, particularly given we’re highly unlikely to use the Pyra as a server...?Agreed. There is no practical point to going to 8GB RAM. But, beyond marketing specs, there wasn't much practical point to going to 4GB RAM either. I don't think I have even crossed 1GB RAM in use on my prototype.
There is always a clamoring for new/better/more though. People have been suggesting other SoC options fairly consistently - whether the suggestions are even remotely practical for this type of device is another conversation entirely. Just making a first pass at SoCs for finding something that fits in our heat envelope, physically fits on the board, has the needed peripheral support, isn't too expensive, is available to the project, has long term product availability, and at least doubles the 3D graphics performance - that weeds things out pretty quick.
Sharing the same SoC as other devices would indeed be handy too. The few options I've seen that meet all that were actually Mediatek based. Unfortunately many in the GNU/Linux community would burn me at the stake for even suggesting such a thing due to Mediatek having done some sin or another years ago. So, that adds another wrinkle - Linux community acceptance.
Ok so the only way 8Gb would ever be useful to us is with a 64-bit capable CPU with lower thermal constraints?The RPi 4 has more CPU power, and less in the way of thermal constraints. It is also 64-bit, which means that one process could easily access all 8GB if needed, whereas on the 32-bit OMAP5 in general each process is limited to 3GB each.
Actually, the amount of address space dedicated to the kernel is adjustable. As each process has its own set of virtual memory configuration, the remaining reserved memory space is only limited by the amount of shared memory you actually need to interact with other processes or the kernel itself. The only reason why such a huge chunk of address space is being allocated to the kernel itself is performance: if everything the kernel needs uses the same virtual memory address mappings you do not need to reprogram their entries within the MMU configuration when switching to other tasks, which also affects other related things like the TLB.whereas on the 32-bit OMAP5 in general each process is limited to 3GB each.
Yes, only 22 2GB orders.
We probably will NOT create them, instead, these will be updated for free to 4GB versions.
Unless you REALLY want a 2GB one, as we have some prototype PCBs lying around
Agreed. There is no practical point to going to 8GB RAM. But, beyond marketing specs, there wasn't much practical point to going to 4GB RAM either. I don't think I have even crossed 1GB RAM in use on my prototype.
There is always a clamoring for new/better/more though.
The OMAP5 is actually running LPAE which is an emulated 64bit mode. It provides extended address space for the extra RAM. This is what allows us to use 4GB. I don’t know if it can support up to 8GB though I haven’t looked into that.Ok so the only way 8Gb would ever be useful to us is with a 64-bit capable CPU with lower thermal constraints?
I wasn't sure, but when I considered the way the shop works, I thought it through and the above is where I ended up.
Yes, only 22 2GB orders.
We probably will NOT create them, instead, these will be updated for free to 4GB versions.
Just wondering, would 8Gb RAM not serve useful for doing CAD work, video editing etc. on the Pyra?
Ah yes, how could I forget the vm’s and containers!There are a thousand things I can use 8GB of memory for. My desktop has 64GB and hosts multiple virtual machines and containers, runs Firefox on dozens of bloated websites (thanks Bootstrap, React, etc.). It's helpful with video editing, software compiling, and allowing for remote user sessions.
If I were doing mobile app development still, the emulators would eat RAM for breakfast.