Question Why does Navi require 2x the transistors of Polaris?

[mildew]

RX5700, 36CU's, 10.7B transistors. RX590, 36CU's, 5.7B transistors. I realize that Navi has twice the ROP's but as far as i know they don't use that much die space. So why does Navi require almost double the transistors of Polaris?

 

[digitalgriffin]

AMD Graphics cards have ALWAYS been heavy for the transistors/GFlop performance.

With GCN it was additional circuitry for compute heavy workloads with extra SP's. It's what made it such a great mining card.

Navi this time is focused on gaming performance. It shows in the compute ratings as well. (Preliminary results show OpenCL as either broken driver wise or low on performance) So I'm not sure where the extra transistors are coming from.

That said additional ROPs/CU's do come with a price. And there's additional cache. The more CU/ROPs the more complex the intercommunication. (The complexity is dependent upon design of the cache)

 

[mildew]

Yes, having done some more research myself i see that L1 is increased - but being L1 we will be talking KB's not MB's. I get the feeling a decent portion of the extra transistors are dark/dead silicon to allow higher clock speeds. I just cant see anything else that could account for a doubling in transistor budget.

 

[TJ Hooker]

FYI transistor counts are for the full die, even if parts are disabled for a given GPU. So 10.3B transistors is the count for a 40 CU 5700 XT.

Edit: And it uses the new RDNA architecture, which could use more transistors in general for any particular functional block.

 

[mildew]

Duh, i had completely missed that factor thanks TJ. So that accounts for 11% of the extra transistors.