guesstimation of time needed in token / s for 1 laptop vs 2 laptops in MPI mode. #3161
-
|
guesstimation of time needed in token / s for 1 laptop vs 2 laptops in MPI mode. if 1 laptop is 10t/s, 2 laptops = 18t/s? is this a good guesstimation? the more powerful the gpu card, the more powerful the CPU is needed and more ram etc too. so i was wondering if it's just better to buy multiple devices more maybe at 1/3 of the price than just buy 1 powerful machine that will most probably drop its price soon. 33b model is what i am to run "comfortably" (on laptops) but the gpus for laptops are more expensive than desktop (i dont want the extra heat and power consumption and worst, non portability). what do u guys think? 2 laptops instead of 1? |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments 1 reply
-
|
I don't know the answer to your question but only 10% overhead sounds absolutely amazing. I'd be surprised if the real number wasn't at least 30-40%. |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
@staviq @KerfuffleV2 how to calculate the overhead so as to have the right set up i wonder? it also makes me wonder if it's a viable option for multi datacenter processing at what kind of network overhead. how to calculate it? |
Beta Was this translation helpful? Give feedback.
#3099