I thought it was common knowledge that amd deemed the gains of 3stucking on 79xx GPUs as not worth the cost

Venix:

I thought it was common knowledge that amd deemed the gains of 3stucking on 79xx GPUs as not worth the cost

I don't know anything about that but this only seems to confirm the possibility exists. Which makes sense to me, AMD would likely have designed the chips with the possibility in mind even if they don't end up making use of it.

Exodite:

I don't know anything about that but this only seems to confirm the possibility exists. Which makes sense to me, AMD would likely have designed the chips with the possibility in mind even if they don't end up making use of it.

There where talks that they tried vcache and they deemed the gains do not justify the cost, now why ? Only AMD knows , was that they had to operate on lower clocks so the gains where partly or completely negated ? The gains where trivial ? The yields are bad ? A combination of the above ? Only AMD knows

Venix:

There where talks that they tried vcache and they deemed the gains do not justify the cost, now why ? Only AMD knows , was that they had to operate on lower clocks so the gains where partly or completely negated ? The gains where trivial ? The yields are bad ? A combination of the above ? Only AMD knows

I assume there was a performance gain but not enough to justify the additional price. Keep in mind, as much as AMD would like you to think otherwise, they're not the #1 brand, so it doesn't make sense for them to cut into their profit margins only to still be bested by Nvidia. From what I can tell, the RDNA architecture isn't as starved for memory bandwidth as GCN. When you then account for DirectStorage, higher PCIe bandwidth, and high-speed VRAM becoming more affordable, I could see how V-cache might not be necessary; just simply having bigger caches was enough, apparently.

schmidtbag:

I assume there was a performance gain but not enough to justify the additional price. Keep in mind, as much as AMD would like you to think otherwise, they're not the #1 brand, so it doesn't make sense for them to cut into their profit margins only to still be bested by Nvidia. From what I can tell, the RDNA architecture isn't as starved for memory bandwidth as GCN. When you then account for DirectStorage, higher PCIe bandwidth, and high-speed VRAM becoming more affordable, I could see how V-cache might not be necessary; just simply having bigger caches was enough, apparently.

Cache is still very important see how much more cache the 4xxx cards have in comparison to ampere ! Hell it seems the biggest change is the increased cache at least in the die allocation. AMD already took the cache out on the small chiplets so they do not occupy space on the main die keeping it smaller for better yields and as a result cheaper.

makes sense they would do that, probably not for consumer gpus, the advantage to the chiplet setup, is that they can use them in many different products, I can totally see a CDNA gpu with the same chiplets + 3dvcache.

They have made a lot of changes to this architecture and it is possible that the 3d memory has been left for the next generation.