Personal computing discussed
Moderators: renee, morphine, SecretSquirrel
DancinJack wrote:I didn't see anyone post die size yet?
It appears the die (Vega only, no HBM) is roughly 25.90mm x 19.80mm = 512.82mm-ish. (almost cut off a sig fig)
willyolioleo wrote:I'm willing to believe the Fiji drivers theory. Because there is way more going on than just a bad launch. A simple die shrink of Fiji would get better results than this. Why the hell wouldn't AMD just release a 14nm Fiji and be done with it, if these were the numbers they were looking at?
And that's not even counting the new features like the new rasterizer, memory controller, or tile based renderer.
The way it's performing looks like none of the new features are being used, and on top of that, even the raw compute power isn't being used properly either.
If the drivers are basically working like Fiji, but the underlying architecture is different (like how the pixel engine is connected to the L2 cache now) it would basically explain why it's even worse than Fiji.
Otherwise I don't see how they could have managed to screw up THIS badly.
Vhalidictes wrote:willyolioleo wrote:I'm willing to believe the Fiji drivers theory. Because there is way more going on than just a bad launch. A simple die shrink of Fiji would get better results than this. Why the hell wouldn't AMD just release a 14nm Fiji and be done with it, if these were the numbers they were looking at?
And that's not even counting the new features like the new rasterizer, memory controller, or tile based renderer.
The way it's performing looks like none of the new features are being used, and on top of that, even the raw compute power isn't being used properly either.
If the drivers are basically working like Fiji, but the underlying architecture is different (like how the pixel engine is connected to the L2 cache now) it would basically explain why it's even worse than Fiji.
Otherwise I don't see how they could have managed to screw up THIS badly.
It does indeed look like either the drivers are a total mess or they really did screw up this badly.
Just because they're using a new architecture, doesn't mean that the new architecture is any good! I agree a die shrink Fiji could have been a better performer, but a Fiji die-shrink wouldn't have beaten the 1080 Ti either.
chuckula wrote:Vhalidictes wrote:About the whole "It's just Fiji drivers" argument, my question is: If this really is a new architecture then why/how could it even run these programs at all simply using the old driver from a different architecture in the first place? Wouldn't we be talking about a failure to even run properly instead of be quibbling about performance numbers?
chuckula wrote:Vhalidictes wrote:willyolioleo wrote:I'm willing to believe the Fiji drivers theory. Because there is way more going on than just a bad launch. A simple die shrink of Fiji would get better results than this. Why the hell wouldn't AMD just release a 14nm Fiji and be done with it, if these were the numbers they were looking at?
And that's not even counting the new features like the new rasterizer, memory controller, or tile based renderer.
The way it's performing looks like none of the new features are being used, and on top of that, even the raw compute power isn't being used properly either.
If the drivers are basically working like Fiji, but the underlying architecture is different (like how the pixel engine is connected to the L2 cache now) it would basically explain why it's even worse than Fiji.
Otherwise I don't see how they could have managed to screw up THIS badly.
It does indeed look like either the drivers are a total mess or they really did screw up this badly.
Just because they're using a new architecture, doesn't mean that the new architecture is any good! I agree a die shrink Fiji could have been a better performer, but a Fiji die-shrink wouldn't have beaten the 1080 Ti either.
About the whole "It's just Fiji drivers" argument, my question is: If this really is a new architecture then why/how could it even run these programs at all simply using the old driver from a different architecture in the first place? Wouldn't we be talking about a failure to even run properly instead of be quibbling about performance numbers?
Aranarth wrote:Correct me if I'm wrong:
The Vega FE is a professional level card and in games has performance around or a small notch above a 1070.
The RX Vega is a gamer card aimed at gaming and it has performance around or a small notch above a 1080.
Currently there is no "1080TI killer" being talked about yet in the AMD camp.
Is this correct?
