Cool! Relevance of HBM for consumer products decreases day by day.

Cool! Relevance of HBM for consumer products decreases day by day.

This ^

So...GDDR5x and GDDR6 will perform the same....? ....

Cool! Relevance of HBM for consumer products decreases day by day.

This ^

Considering that the speeds of GDDR5x and GDDR6 can't even remotely touch HBM2, let alone HBM3, not sure what you two are smoking.

Cool! Relevance of HBM for consumer products decreases day by day.

Try fitting GDDR/X on an APU. 🙂

Considering that the speeds of GDDR5x and GDDR6 can't even remotely touch HBM2, let alone HBM3, not sure what you two are smoking.

Can't remotely touch? But they already do. AMD's Vega with HBM2 will have max 512GB/s (launch with even less - 409.6GB/s) memory bandwidth. 1080 TI with GDDR5X already has 484GB/s. New GDDR5X and GDDR6 will surpass that. You need to "count in" the whole "package" - number of chips, memory bus width, etc. So, as I said - for consumer products HBM looks less and less viable.

Try fitting GDDR/X on an APU. 🙂

You can try fitting HBM to APU and see its cost quadrupple. Not going to happen any time soon.

Considering that the speeds of GDDR5x and GDDR6 can't even remotely touch HBM2, let alone HBM3, not sure what you two are smoking.

HBM have also issues due to their technology that limit the impact of all the goods it can bring (heat, controler, ...). HBM is inovative i agree but those GDDR mem is a good solution (even a better solution?).

You can try fitting HBM to APU and see its cost quadrupple. Not going to happen any time soon.

it is already planned by AMD... but i agree about the cost

Try fitting GDDR/X on an APU. 🙂

APUs are not high-end, they are low-end to mid-range at best, they will not get an expensive high-end memory like HBM. People buying high-end don't use integrated GPUs, so who would waste much money on this?

@Aura89 and you really think any consumer would care, if the product costs a lot more? cost for 8gb is about 160$, so i know it will mute the price advantage AMD had in the past (over Nv).

Can't remotely touch? But they already do. AMD's Vega with HBM2 will have max 512GB/s (launch with even less - 409.6GB/s) memory bandwidth. 1080 TI with GDDR5X already has 484GB/s.

You're forgetting that only accounts for 8GB of HBM2 memory. Try 16GB, which considering we're already doing 11-12GB of GDDR5x memory isn't too far off. As i said, GDDR5x/GDDR6 doesn't even touch HBM2/3, especially with how little of an improvement GDDR6 is apparently going to give us.

@Aura89 and you really think any consumer would care, if the product costs a lot more? cost for 8gb is about 160$, so i know it will mute the price advantage AMD had in the past (over Nv).

No reliable information exists for costs of HBM2. Yes, i saw where it states $160, but it does not state if thats how much of a premium it puts on it, or is the cost. Considering the cost of HBM1 was massively lower then that, $160 does not make sense. There's no point speculating about an irrelevant thing like costs when we don't even have a clue as to what the actual cost is. But when did this become a conversation about cost? And when has cost ever been a determining factor if a better technology gets used? If that was the case, DVD and blu-ray would never have gotten off the ground since at their initial release price and for a few years after it aswell, the costs for the drives, let alone the movies, were astronomical compared to the predecessors price at the same time. http://cdn.overclock.net/a/ae/aeac88da_Sys-plus-1.png

@Aura89 and you really think any consumer would care, if the product costs a lot more? cost for 8gb is about 160$, so i know it will mute the price advantage AMD had in the past (over Nv).

and in some country AMD is more expensive than NV since more than 3 years now (sadly Ryzen is expensive too there 🙁 ).

http://cdn.overclock.net/a/ae/aeac88da_Sys-plus-1.png

i understand what you want to say... but this isn't a good exemple the R9 290X were performing better in 80% of the case than Fury and cost less money... :)

APUs are not high-end, they are low-end to mid-range at best, they will not get an expensive high-end memory like HBM. People buying high-end don't use integrated GPUs, so who would waste much money on this?

The Pro level "Opteron" APU's will be the ones with HBM. They won't be cheap, but they will be fast. They have GDDR/X tech for the mid to low end, they have just bet on HBM for high end graphics cards, and it hasn't worked out due to fab problems, and yields. I'm not going to tell a company to stop innovating. Without innovation, we get companies like intel that just sit around and scratch their arse.

i understand what you want to say... but this isn't a good exemple the R9 290X were performing better in 80% of the case than Fury and cost less money... :)

290x were performing better then Fury? In what exactly?

Can't remotely touch? But they already do. AMD's Vega with HBM2 will have max 512GB/s (launch with even less - 409.6GB/s) memory bandwidth. 1080 TI with GDDR5X already has 484GB/s. New GDDR5X and GDDR6 will surpass that. You need to "count in" the whole "package" - number of chips, memory bus width, etc. So, as I said - for consumer products HBM looks less and less viable. You can try fitting HBM to APU and see its cost quadrupple. Not going to happen any time soon.

I'm absolutely certain that NVIDIA is using HBM in all of its really high end offerings (not what we peasants can get), because it makes no difference.

i understand what you want to say... but this isn't a good exemple the R9 290X were performing better in 80% of the case than Fury and cost less money... :)

The 290x, Hawaii in general, was actually always slower than Fiji. I don't know where you dreamed that.

@Aura89 Ur not even comparing the same thing. One is a consumer product that almost everyone will need that doesn't have a console and wants to watch movies from a disc. What consumer that is not gaming/en-/decoding, will buy a dedicated gpu for their computer? Not even talking about higher end cards, as i doubt the will put HBM on a 530.. and seeing the cost was about 40$ for HBM1 and added 120$ to the card, 160 doesnt seem too far off. Just because something is better as some previous/existing tech, doesn't mean its something that will replace everything used before. E.g. thats why we dont have nuclear powered planes or cars. ;-)