GDDR5X vs HBM2 which is better?

That_Tech_Guy_Again

Distinguished
Aug 27, 2016
201
0
18,690
So i was having a look at different gtx 1080's and i saw the news that amd vega should be coming out soon using HBM2 instead of gtx 1080 GDDR5X memory type. So does anyone know which memory type is better? and why?
 
Solution
Hbm is not faster speed, it's slower as pointed out in the link. Also the size is much less of a concern. Performance is what everyone cares about most. Being smaller does not create more room for more cores. That is a limitation of the heat produced, else we'd have bigger dies being used. Gddr is not even large. Look at a pic of a gpu pcb.

Speed and bus width is less important since bandwidth is the end of the equation and is what matters. Max capacity is higher than what's needed even for top cards. We're going to see gddr6 and hbm3 before more is needed but hbm is the future as is any 3d tech. It just allows for better performance and lower power. The only disadvantage is that it's new. This means higher cost, new designs vs gddr...

FnaticMeister

Respectable
Aug 9, 2016
312
0
1,860
First of all HBM is considered to be better. This is because HBM can be compressed into a WAY smaller size than GDDR5X has to take up to match the vRAM capacity of HBM. HBM can also run at higher speeds than GDDR5X, which is a advantage for all.

Another thing you have to take into consideration: Speeds for games usually barely matters on vRAM speeds. Most of the performance comes from the cores and what speeds those are running at and how many there are. But also notice that by using HBM, you use less space in the graphics card for memory, which opens up the card for more cores and more efficient cooling with all the space, etc.

If you're asking the broad question: Which is better, HBM2 or GDDR5X, and not the difference in performance between the new AMD and nvidia card, then yes, HBM is better.

 

That_Tech_Guy_Again

Distinguished
Aug 27, 2016
201
0
18,690




I probably should have included this before, but anyways, here is a link with the different specs for each type listed. http://graphicscardhub.com/gddr5-vs-gddr5x-vs-hbm-vs-hbm2/


I am most interested in the differences between

maximum capacity
maximum speed
bus width

and what they translate into in terms of performance.

Also in a broader context, is there any differences in terms of which is better for any particular circumstances, which one of them might do better than the other?
 

FnaticMeister

Respectable
Aug 9, 2016
312
0
1,860


Obviously, HBM has a WAY smaller size, which I already mentioned, which will give the card optimizing HBM to have more memory while taking out space, which companies could add more cores and better cooling. So, HBM basically has all of the advantages here.

I don't know much about the maximum speeds that can be occurred between each memory type, but I believe GDDR5X will give a little better speed, but that won't matter in almost every situation you run into.

HBM wins in the bus width section.

So, HBM has a lot more advantages to its much larger counterpart of GDDR5X.
 

That_Tech_Guy_Again

Distinguished
Aug 27, 2016
201
0
18,690



I understand everything you have said. I just have 1 more question. I heard someone, somewhere say that 1 Gbps of HMB2 is as fast as about 8Gbps of GDDR5X. So if the maximum speed of HBM2 is 2 Gbps then that would equal about 16 Gbps of GDDR5X.

Can you confirm whether this is true or not?
 
Hbm is not faster speed, it's slower as pointed out in the link. Also the size is much less of a concern. Performance is what everyone cares about most. Being smaller does not create more room for more cores. That is a limitation of the heat produced, else we'd have bigger dies being used. Gddr is not even large. Look at a pic of a gpu pcb.

Speed and bus width is less important since bandwidth is the end of the equation and is what matters. Max capacity is higher than what's needed even for top cards. We're going to see gddr6 and hbm3 before more is needed but hbm is the future as is any 3d tech. It just allows for better performance and lower power. The only disadvantage is that it's new. This means higher cost, new designs vs gddr implementations and low supply.

In terms of performance, as pointed out, the gpu matters more than vram. This will always be the case. As long as vram has enough bandwidth as the gpu itself needs then it's all good. A good example was fiji and hbm. Even with higher vram bandwidth, it doesn't make up for the gpu so the 980ti beat the fury x.
 
Solution

TRENDING THREADS