how hot will the epiphany chip be?

[piotr5]

I noticed that with x86 you can basically turn off the computer by running some cpu-burn program. somehow the cooling-systems are only designed for ordinary runtime-behaviour like in games or data-cruncing and cannot handle artificially created heavy load. now parallella with its 2w likely wont be as hot as those, but the question of a good (fanless) heatsink might still arise. so if someone has a parallella and a laser-thermometer, maybe with trial and error they could design the "hottest" assembler program, a program that running on all epiphany cores will heat up that chip more than any ordinary program could. or maybe the designers know, what would such a program need to do to quickly produce the highest possible heat? especially for testing the casing such a program would be interesting. so, what do I need for a cpu-burn program on epiphany?

[timpart]

I speculate that keeping the floating point pipeline filled with multiply-adds (with no register dependencies) would consume a lot of power. Do some integer adds at the same time to keep the other half busy. Make sure the instructions are 32 bit so there are lots of loads of instructions from memory

This situation is not very likely in real programs because register dependencies limit how quickly floating point operations can be scheduled. Also loads of data have a delay before it can be used in floating point. Perhaps evaluating a power series with complex numbers would be quite close to peak.

Doing a lot of writes to other cores could keep the intrachip network busy but I imagine this is comparatively low consumer of power. Using the DMA to do this would consume extra power.

Tim

P.S. This post is only meant as a theoretical discussion. Actually doing the above might permanently damage the chip. Any experiments entirely at your own risk.