AMD Confirms HBA High Memory Bandwith for Graphics Cards

Published by

Click here to post a comment for AMD Confirms HBA High Memory Bandwith for Graphics Cards on our message forum
https://forums.guru3d.com/data/avatars/m/47/47197.jpg
That looks very promising. Increased x86 investments, I wonder why they would want to invest in x86-architecture instead of x64 ๐Ÿ˜•
https://forums.guru3d.com/data/avatars/m/225/225084.jpg
Looking at those pictures then i'd have one slight worry about this stacked vram. Getting off future HSF's might be a total nightmare, if the Vram is actually quite a bit higher than the GPU then i can see many a broken GFX card. You might in fact just twist it off with the heatsink ๐Ÿ™‚
https://forums.guru3d.com/data/avatars/m/243/243702.jpg
Looking at those pictures then i'd have one slight worry about this stacked vram. Getting off future HSF's might be a total nightmare, if the Vram is actually quite a bit higher than the GPU then i can see many a broken GFX card. You might in fact just twist it off with the heatsink ๐Ÿ™‚
schema
https://forums.guru3d.com/data/avatars/m/245/245459.jpg
Looking at those pictures then i'd have one slight worry about this stacked vram. Getting off future HSF's might be a total nightmare, if the Vram is actually quite a bit higher than the GPU then i can see many a broken GFX card. You might in fact just twist it off with the heatsink ๐Ÿ™‚
Yeah, I agree, I thought something similar when I saw the diagram of the HBM stacked memory being higher than the GPU core yet being right next to it. To me this would mean that heatsinks will have to be specifically designed to sit flat on the GPU core, and then have a step down to allow the HBM to fit. I think this would be a difficult design to manufacture accurately, and I think we will see inefficiencies with cooling due to heatsinks not mating to the surfaces properly, as well as the increased likelihood of damage being caused to the structure of the HBM when the heatsink is being placed & also when removed. On a side note with HBM, it's just allowing larger bandwidths and throughputs, but AMD will need to manufacture an efficient core in order to take advantage of that extra bandwidth otherwise it's pointless.
https://forums.guru3d.com/data/avatars/m/220/220214.jpg
Waiting for BF5 + DX12 + R395X. That should be a nice step forward.
https://forums.guru3d.com/data/avatars/m/258/258664.jpg
Waiting for BF5 + DX12 + R395X. That should be a nice step forward.
So those are your plans for late 2016? ๐Ÿ˜€
https://forums.guru3d.com/data/avatars/m/88/88351.jpg
Looks good, but no PhysX! :j/k:
https://forums.guru3d.com/data/avatars/m/253/253059.jpg
What percentage of the cards power usage comes from memory vs. the GPU in, say, the 290? Will this shave a few watts or more than that? I wonder how many layers they will have stacked, too. And are the memory and GPU cores fabricated at the process size. I'm assuming the RAM is fabricated elsewhere then fused with the GPU die, or else AMD is also getting into memory fabrication...
https://forums.guru3d.com/data/avatars/m/258/258664.jpg
If I recall correctly, it will have 4 layers at the start, but can go up to 8 layers with HBM gen 1. Might change later. Also, I wonder when they will start to think of putting two or more of those blocks onto a PCB ๐Ÿ™„
data/avatar/default/avatar30.webp
That looks very promising. Increased x86 investments, I wonder why they would want to invest in x86-architecture instead of x64 ๐Ÿ˜•
x86 is x64. x86-32 is what is referred to when people say 32bit x86-64 is what is referred to when people say x64