| drkohler said: These posts get longer and longer to digest with all the quoting. So I ask everyone for just one thing: Some people think that 2.23GHz is a boost clock and the thing actually runs at a lower clock some/most of the time. It is NOT. When Cerny mentioned boost in his talk, he meant it in the engineering sense of the word. What they did when testing the SoC, they started with a low frequency and upped the frequency step by step until the gpu was no longer able to function correctly (My guess is a lot of SoCs bit the dust). This gave them the absolute upper clock limit. Then they did the same thing again up to a point where the thermal/power envelope was reached with whatever cooling solutions were tested. Apparently 2.23GHz is the "sweet spot" for the gpu. (Surprisingly the 3.5GHz for the cpu is already problematic due to a particular 256bit command set that needs large amounts of power.) Stepping up the clock is called boosting the clock in the engineering world. It has nothing to do with "This thing runs at x GHz but we can boost x by y%". |
Well Mark Cerny own words are that variable frequency = continuous boost. If you look at the presentation you even see the graphs in the background doing the boost , UP and DOWN constantly. But the gpu will not be constant at 2.23ghz, it will reach that output if it needs on a given moment, but by underclocking the cpu. This means Gpu clock may be boosting from 1,8 to 2.23ghz. He didn't want to give base clock speed , but i can assure it exists. If we use early leaks, base clock should be around 2ghz (9tflops).
We have to wait and see how this approach is played out. AMD smart Shift technology can indeed be something. It doesn't convince me yet, I see stability problems everywhere. And we have yet to see the cooling solution, it must be a robust one.
