[SemiAccurate] Intel Details Knights Corner Architecture At Long Last

gorobei

Diamond Member
Jan 7, 2007
4,110
1,613
136
if the performance is anywhere near a first gen tesla/fermi card, this could be very bad for nvidia's hpc aspirations. native x86 and tcp/ip would reduce the gpgpu customers to a very narrow market of those who explicitly need the peak performance of a gpu vs a linux farm on a chip like KC.
 

BrightCandle

Diamond Member
Mar 15, 2007
4,762
0
76
As a developer I prefer the Intel approach so far. I would still rather have the os be aware of the different performance cores and allow me the programmer to hint where I want it run but a cluster on a card is a better step than open cl which frankly is awful.