JEMC said:
Slimebeast said:
But this is horrible. 5.5 TFLOPs is not even much faster than a GTX980 (5.3TFLOPS) and significantly less than GTX 980Ti at 6.5 TFLOPs. So not that much faster than a GTX 980 and yet it's rated at 150W? AMD has talked so much about 2.5x performance per Watt , and considering a GTX 980 draws only 165W, a Polaris 10 shouldn't draw more than 100W.
I really really hope this rumour is false or else AMD is toast.
|
That's why the FLOPS figure is never an accurate measure of performance. And I don't know if some of those numbers are right either...
A few days ago, Pemalite wrote this on the "GTX 1080 unveiled; 9 teraflops" thread:
Case in point: Radeon 5870.
Even though the Radeon 5870 has 2.72 Teraflops of performance, majority of games it will lose against the Radeon 6950 at 2.253 Teraflops and the Radeon 6970 at 2.703 Teraflops (Usually by a healthy margin). And will even lose against the Radeon 7850 at 1.761 Teraflops, Radeon R7 265 at 1,843 Teraflops and so on.
The Radeon 270X will more than double the Radeon 5870's performance despite having 2.688 Teraflops verses the Radeon 5870's 2.72 Teraflops.
Graphics is more than just single precision floating point.
And if that's not enough, there's this: R9 290X = 5.6TFLOPS vs GTX 980Ti = 5.6TFLOPS. Or the GTX 970 = 3.4TFLOPS vs GTX 780Ti = 5.3TFLOPS. TechPowerUp has a database where you can see the specs of any chip, including its GFLOPS figures. You may find some surprises.
In summary: FLOPS doesn't translate into real world performance, just like synthetic benchmarks. It's worth keeping that in mind.
|
There's a strong correlation though, even if it can be off by 20% or so.
5.5TFLOPS rated at 150W has no chance of being even close to Raja Koduri's claim "2.5x performance per Watt".
Our only chance is for the rumour to be totally wrong.
