[erlang-questions] Tilera 64-core chip - let's help them help us!
G Bulmer
gbulmer@REDACTED
Thu Sep 6 04:46:20 CEST 2007
Hopefully Tilera's marketing group have a google alert set up, and so
will discover our posts soon.
Maybe Erlang isn't 'on their radar'? That would be a real shame as I
can imagine Erlang+specialised embedded pieces, using their 'native'
technology, may be a great approach to exploiting their architecture.
There may be some enormous benefits to supporting Erlang for them.
So, to help Tilera, could we say what we'd need? Then maybe Tilera
will consider the possible costs and benefits. I'll start:
0. Decent documentation for the architecture. The published NVIDIA
8800 documentation isn't good enough for confident design and
extrapolation from tests; I feel better documentation would have helped.
1. I'd like a free or very cheap (under $50) license for an emulator
and the Tilera software development kit (e.g. like NVIDIA's CUDA).
I'm fine if that software is restricted to use for evaluation. I want
to understand the cost and time to learn how to use Tilera's
technology, and I don't *need* hardware for that. Someone may figure
out, without hardware, whether it's worth porting Erlang to it, and a
free Tilera emulator & SDK would enable more people to participate
and tackle parts of the solution.
2. I could be satisfied with a 'cut-down' version of their PCIe board
(with fewer GigE connections, no 10 GigE, fewer processors, less on-
board memory, maybe running at a lower clock rate, maybe narrower
PCIe) providing I could make reasonable extrapolations from it, so
that I could do realistic de-risking, evaluation and development. A
cut-down board should to be significantly cheaper than a 'real' board
to make it a justifiable alternative because I assume I'd need 'the
real thing' for production development if it all worked out.
I'd offer the same suggestions to Intel; if they want to get their 80-
core widely used, Erlang is the way to go, so support the community !-)
Garry
PS - I'd suggest we pool orders, but that's likely a nightmare to
make work, and it may be too late for (my) Christmas presents ;-)
More information about the erlang-questions
mailing list