Shehriazad
Senior member
- Nov 3, 2014
- 555
- 2
- 46
Hm... I'm not sure if I can agree completely. It's kind of like driving a V10 car around in normal city traffic. In most cars, you're still powering all 10 cylinders (unless the car has the ability to shutdown cylinders, which some have), which aren't really necessary. So, how does this relate to GPUs? If you're using a nVidia card and are trying to overclock, that memory is still powered regardless of whether you're using it. The video cards are also limited to a certain power threshold, which means that memory eats into your available power and potentially limits your OC.
Which is where HBM comes into play. I'm not entirely sure about the numbers anymore...but I think HBM was somewhere around 40% to 50% less power consumption than GDDR5 at the same amount.
That frees up quite a bit of power...now imagine a Titan X with 12GB of HBM gen 1 instead of GDDR5 .
But then again...Nvidia is waiting for Gen 2...maybe that consumes even less power...not sure...some people on this forum still have all the data flying around, I'm sure.
But 1st Gen HBM GPUs will be available this year and 2nd gen next year. And since 4GB is pretty much going to be the minimum, anyway as no one will put HBM on slow GPUs, 8GB will probably be somewhat of a standard in 2016...now this isn't official...that's just my gut feeling.
Last edited: