By using this site, you agree to our Privacy Policy and our Terms of Use. Close
Intrinsic on 14 March 2019
Edit Thread
Bofferbrauer2 said:

14nm to 7nm doesn't half the size, even though the naming scheme implies it. But those names are decoupled from the actual sized since over 15 years now (since they got smaller than the wavelength of the visible light, which is around 300nm in fact). Vega VII has almost the same amount of transistors as a Vega 64 but is actually only 32% smaller. I calculated the 500mm2 by taking the Vega VII plus 65% of the size of a Zen Cpu, which would be 456mm2. But to reach those 14TFlops, Navi would need quite a few extra functions, so I added 10% to give space for those, which results in 500mm2.

PS4 Pro chip is about the same size as the original PS4 chip, which is about 350mm2. A Vega VII, which is produced in 7nm, is already is 331mm2. How do you want to reach that power in a ~350mm2 package if 95% is already occupied by the GPU - and that GPU isn't even strong enough for the leak. While Navi could be more powerful, it will need more transistors, and thus die space, for that. Getting that much power out of such a small chip is pretty much impossible with the 7nm process.

Vega 64 ad Vega 7 are really bad references to make when talking about next gen consoles. And that is primarily becase they both use HBM RAM. The memory ontrollers for those takeup around 30% of the die space available in those chips!

If you really want to analyze die space usage ou have to do a lot better than that. Take the XB1Xfor instance, it has 12GB of GDDR5 RAM.. its memory controllers fit into a chip thats about 350mm2 along with a CPu and  40CU GPU. Going from 12GB of GDDR5 to 24GB of GDDR6 (just an example) will take the same amount of space with regards to on chip memory controllers. Its been calculated that the Ryzen CPU will take about the sameamount of space on a die that jaguar takes right now. That leaves a gd dalaount of space for the GPU. 

Lets be realistic here..... what are we really asking for? Either a GPU with 64CU running at a clok of around 1500Mhz or one ofaround 72CU running at around 1300Mhz.  Fitting either of that into a die sizeof about 350mm2 to say 380mm2 willnot be a problem. The real issue is how effecient is the architecture, how much power will it draw.... how hot wil it get?