@ID_AA_Carmack @TimSweeneyEpic hey guys, do you know how many MAC/s a gpu like e.g. the 1080 could do? Is it an alternative to TI DSPs?
-
-
Replying to @PLT_cheater @ID_AA_Carmack
You can read their spec TFLOPs as single-precision multiply-adds per second, so up to 10.8 trillion.
1 reply 0 retweets 2 likes -
Replying to @TimSweeneyEpic @ID_AA_Carmack
that's 10x cheaper per MAC than TI's top-line TMS320C6678 and 100x than AD's (mainly geared for low end tbh)
1 reply 0 retweets 0 likes -
this begs the question why nvidia doesn't do chips that can be run stand-alone in automation and dsp.
1 reply 0 retweets 0 likes -
Replying to @PLT_cheater @TimSweeneyEpic
those tend to be lower margin parts and often have entrenched codebases, but they will get around to it...
1 reply 0 retweets 2 likes -
Replying to @ID_AA_Carmack @TimSweeneyEpic
I imagine with the advent of self drivers it's high time to have NV chips for embedded/rt. After Volta maybe?
2 replies 0 retweets 0 likes -
I imagine they are holding off for full-speed doubles - am I right to guess Volta will deliver this?
1 reply 0 retweets 0 likes
I assume we'll always see single throughout = 2x double, because a double FPU can be sliced almost for free.
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.