That should keep cache hit rates high, which means effective bandwidth - even from a 192-bit memory interface - should be much higher than Nvidia's similarly-equipped RTX 3060 12GB.ĪMD didn't reveal all of the specs, but based on that image, it looks like RX 6700 XT / Navi 22 will max out at 96 ROPs (Render Outputs), and the total die size looks to be in the neighborhood of 325mm square, with around 16-17 billion transistors (give or take 10%). The answer appears to be "not very much." The Biggest Navi chip has up to 80 CUs and 128MB of Infinity Cache, so AMD cut the computational resources in half but only lopped off a quarter of the cache. We were very curious about how far AMD would cut down the Infinity Cache from Navi 21.
Except, it still has a honking 96MB 元 Infinity Cache.
Raw theoretical performance sits at 12.4 TFLOPS, and assuming AMD uses 16Gbps GDDR6 again (which is likely), it will have 384GBps of bandwidth. Even with a drop down to 40 CUs (from 60 CUs on the RX 6800), the higher clock speeds should prove relatively potent. The RX 6800 XT and RX 6900 XT both have 2250 MHz game clocks, though in actual benchmarks, we've seen speeds of more than 2500 MHz already - the Game Clock is more of a conservative boost clock. The AMD Radeon RX 6700 XT comes in with the highest GPU clocks we've to date, 2424 MHz.
0 kommentar(er)
