On 06/20/2018 08:36 AM, Michael Di Domenico wrote:
On Tue, Jun 19, 2018 at 10:46 PM, Stu Midgley <sdm...@gmail.com> wrote:
Where Intel made the mistake was to assume they could shift people from
GPUs. People who have spent years writing and optimising won't shift
easily... cause they have to go through that whole process again. Getting
people back to x86 once they have shifted is a long long term goal.
this is where my environment fell over with respect to KNC and KNL.
Once the dev's got their hands on the cards they realized they had to
refactor the code away from gpu's. and the few that did, didn't get
much performance benefit over just using a gpu which they already
knew.
intel could have cornered the market from gpu's if they had released
KNL first and not been so late to the market. once CUDA took hold,
game over.
Well, there was the Intel Larrabee project, which was killed before it
ever went into production, which set Intel back some.
https://en.wikipedia.org/wiki/Larrabee_(microarchitecture)
And I'm sure there was some arrogance in this mix, also. I bet Intel
figured they could be late to the game, and still win, because they're
Intel, or would just use the same tactics that they used to crush AMD
once the Nehalem processor came out. They were late that party, too, but
still won.
Prentice
_______________________________________________
Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf