3D DRAM?

[8l]

[Wolf29]

I was wondering about RAM.
I am planning to build out a web server on this board, and apache can be very piggy about ram. I am not as concerned with the number of processors since the epiphany-16 will probably handle the stress. It looks like RAM is the next bottleneck. Am I just wrong? Should I wait patiently for my board to arrive?

Wolf

[hamster]

You might have a wrong idea of what the board is designed for / capable of.

The Parallella isn't a 16 or 64 core wide SMP system - it's a two core ARM system with something akin to a 16 or 64 floating point co-processors. Each of these co-processors are a bit like a GPU shader, able to run a small program. This makes your question a bit like saying "I've put a high end GPU into my PC, will it make my web server super fast, or will I have a RAM bottleneck?"

So the answer is "No" - this board will serve pages like any other ARM board, that is unless you are using the Epiphany to accelerate the processing of data that you will be serving.

[Wolf29]

Thanks, hamster,

I may have been a bit confused.

wolf

[piotr5]

now that this technology is ages old, anyone has some news on that, in relation to parallella? since parallella needs a redesign anyway, could this technology be used there? or in the new redesigned epiphany4 chip?

as for an epiphany5, I was thinking, what about a 3d-mesh network, with north,south,east,west, and up and down? maybe a pin-out allowing for 6 elink connections? 12GB/s, 6 in and 6 out, that sounds quite interesting. well, maybe not for 64 cores, but 4096 cores based on a 16x16x16 grid sounds much more managable than a 64x64 network. maybe even give fpga access to 3 elink-connections instead of 1...

of course my ideas wont work: heat-management in 3d is much more complicated. and there simply is not enough space for 150% pins (so I guess epiphany and fpga need to talk spi to eachother). but at least I can dream.