AMD Radeon RX 6000 series: Up to 80 CUs with a 256-bit memory interface, 128 MB of Infinity Cache, a 300 W TDP and PlayStation 5 GPU clock speeds
Details about Radeon RX 6000 cards have started appearing online, just a few days after AMD announced the series. So far, AMD has confirmed the reference design of at least one card, along with a launch date of October 28. The RDNA 2 architecture will be at the heart of the Radeon RX 6000 series too, even if its dedicated webpage does not reflect that yet. New information published by RedGamingTech (RGT) sheds some light on AMD's upcoming cards, though.
According to RGT, the Radeon RX 6000 series will offer up to 80 Compute Units (CUs), presumably on the triple-fan design that has leaked twice this week. AMD will also lower-end SKUs equipped with 64 CUs, which could feature in the dual-fan card that has leaked too. Additionally, the Radeon RX 6000 series will operate on a 256-bit memory interface, the same as AMD uses on cards like the RX 5700 XT. Moreover, RGT alleges that the Radeon RX 6000 series will have up to 128 MB of Infinity Cache. This could be separate to the GPU cache, but RGT's sources are divided on this one.
Apparently, GPU clocks will reach 2.2 GHz or above, which would put them at PlayStation 5 levels. Sony's next-generation console has a 36 CU GPU clocked at 2.23 GHz, for reference. High-end Radeon RX 6000 series cards will be able to reach 2.2 GHz and beyond thanks to their 300 W TDPs. However, RGT notes that the dual 8-pin power connectors shown in leaked photos will allow cards like the RX 6900 XT to exceed that power threshold through overclocking. 2.3 GHz could be an architectural limit of RDNA 2 though, RGT notes.
There are no details on VRAM volumes yet. GDDR6X is unlikely to feature on the Radeon RX 6000 series though as Micron developed it in collaboration with NVIDIA. Clock speeds of 2.2 GHz would be way beyond what NVIDIA's RTX 30 series offers, but AMD would need to take a huge step forward to reach RTX 3080 or RTX 3090 performance levels.