Does it need to be faster than a 1080 custom ? Perhaps custom vega's will be fast enough for that. Also in terms of noise I would wait for the launch to judge how noisy it gets. Price and performance are both important and if they can come in under the 1080 price with 1080 performance it will make noise.
1080 performance , similar power usage and low price would be a great trifecta for me. I would jump from my 290
Linking their article ...
Probably geometry considering the OpenCL Catmul results. With low levels of tessellation and clocks normalized, the Vega was almost exactly half the performance of Fiji. Those results increased 3x or 1.2x respectively for Catmul5. So it could be half the geometry pipelines disabled, clockspeeds way off their design targets, or simply horrible performance with simple geometry. Not like an engineering sample needs fully enabled for testing.
That result would be extremely odd, as Vega has all the Polaris geometry pipeline improvements, 2x raw geometry throughput compared to Polaris, and improved geometry load balancing. Half performance per clock vs Fiji would mean that it behaves 4x worse in geometry bound scenario than raw specs would let us believe.
687F:C1 vs Fury
OpenCL doesn't have any access to hardware tessellator or GPU geometry pipelines. OpenCL = pure compute kernels, just like CUDA. If this is pure OpenCL benchmark, then it has nothing to do with hardware tessellation or fixed function geometry performance.
There is a relationship between TDP-TFLOPs-frequency though that cannot be broken that much, Nano is a classic example of this with the rigid power management to stay close to 175W and its impact on core frequency with real world compared to spec rating TFLOPs relative to Fury X as I outlined in my previous post.
Radeon RX 480 STRIX is clocked at 1310 MHz (base) / 1330 MHz (boost). It runs at pretty stable clocks (at least in our game). Stock Radeon 580 is running at 1340 MHz boost clock. AMD marketing has already clearly said that Vega NCU was designed to allow higher clock rate. This has been mentioned by many architecture previews. I'd say 1500 MHz is plausible. But I'd guess it is the boost clock rate, not the base clock rate. It is still a huge improvement over the Fury X 1050 MHz max clock rate.
Remember the official rating with TDP from AMD and going from a 380 to 580 gave 29% clock gains as I laid out in earlier post.
And that just reimphasis my point, the reference spec boost figure TDP/TFLOPs was 5.8, in fact with the increased clocks for 580 AMD on their site now says "up to 6.2 TFLOPs" and has a TDP/TBP 185W while custom designs are hitting 200-230W.
A number of things.