GodisanAtheist
Diamond Member
- Nov 16, 2006
- 7,750
- 8,765
- 136
Frankly I hope AMD can keep RDNA high end parts with a GCD shrink by using both monolithic and RDNA3 style chiplets.
A lot of the promise of "chiplets", even the approach RDNA3 uses, is that you can upgrade just the GCD while keeping the same fanout/interposer/memory subsystem and realize extra performance thanks to the additional compute resources and clockspeed without a ton of additional resources/using volume pricing for those existing parts.
Maybe that's a pivot ( vs full blown chiplet arrays that got pushed to RDNA5) AMD is trying to make at the moment, hence all the hush hush and lack of credible leaks around RDNA4.
A lot of the promise of "chiplets", even the approach RDNA3 uses, is that you can upgrade just the GCD while keeping the same fanout/interposer/memory subsystem and realize extra performance thanks to the additional compute resources and clockspeed without a ton of additional resources/using volume pricing for those existing parts.
Maybe that's a pivot ( vs full blown chiplet arrays that got pushed to RDNA5) AMD is trying to make at the moment, hence all the hush hush and lack of credible leaks around RDNA4.