Wow. 5% slower than a 1080 at 1080P?
Makes no sense. 5% slower than a 1080 puts it at a bit over 60% faster than the RX 580, but with a 10% slower Boost clock.
If we assume a straight Polaris design that would put the die at 420mm^2. Which means the 500mm^2 die rumor is wrong. To keep all the rumors:
It would have to have a xtor density 20% less than Polaris.
- 500mm^2
- 5% slower than 1080
- 1200mhz clock
So basically they are saying Vega:
Seems like complete BS to me.
- Doesn't clock as high as Polaris let alone Pascal
- Is less die efficient than Polaris let alone Pascal
Vega is a much larger die though so even with architectural enhancements it's tough to expect a 500mm^2+ die to hit the same clocks as a ~232mm^2 die on the same node while keeping efficiency high.
I am looking forward to seeing what AMD does with Vega, though.
For one, you can just add more cores to make GPUs faster, which means every node shrink really is a doubling of performance at the same power. GPU architecture is fundamentally different from CPU. You can't just make that comparison without considering the differences.
IBM, Intel, AMD and others have been working to make CPUs faster since the 1970s, GPUs have been niche for a long period and only recently starting to become really important. They have a lot more potential.
Vega is a much larger die though so even with architectural enhancements it's tough to expect a 500mm^2+ die to hit the same clocks as a ~232mm^2 die on the same node
Exactly. I am hoping that Vega brings with it the next big jump in efficiency that we expected Polaris and 14nm to deliver. Now that we know Polaris was basically a shrunk GCN and Vega being a new architecture specifically designed for 14nm and higher clocks/IPC I think the efficiency will be excellent.I really hope higher "IPC" within a NCU core means higher occupancy within the average wavefront, because that should mean more performance at the same power ( everything being equal).
Vega is a much larger die though so even with architectural enhancements it's tough to expect a 500mm^2+ die to hit the same clocks as a ~232mm^2 die on the same node while keeping efficiency high.
I am looking forward to seeing what AMD does with Vega, though.
Wow. 5% slower than a 1080 at 1080P?
Makes no sense. 5% slower than a 1080 puts it at a bit over 60% faster than the RX 580, but with a 10% slower Boost clock.
If we assume a straight Polaris design that would put the die at 420mm^2. Which means the 500mm^2 die rumor is wrong. To keep all the rumors:
It would have to have a xtor density 20% less than Polaris.
- 500mm^2
- 5% slower than 1080
- 1200mhz clock
So basically they are saying Vega:
Seems like complete BS to me.
- Doesn't clock as high as Polaris let alone Pascal
- Is less die efficient than Polaris let alone Pascal
Or it's not really Vega.Either terrible drivers, or Vega = 1070 performance/Fury X performance.
If that memory clock is right, it's operating on just 358GB/s.
Vega is using a 2048 bit interface. It would be 716GBps if it was a 4096 bit interface.Time Spy page shows DDR rate, so multiply by 2 to get effective rate. So we get 1400MHz, or 716.8Gbps.
If that memory clock is right, it's operating on just 358GB/s.
HBM2. 700MHz (1400MHz effective) and 2048bit bus.That would be about 7.7ghz on a 384-bit bus. Isn't small Vega rumored to have a 384-bit bus?
Well, I sure hope Intel wasn't sitting on a French 16th century door breaching charge. That sounds rather dangerous. Not that I quite understand why it would own one in the first place, though.