Before the year is out, AMD’s Radeon RX 7000-series graphics cards and the RDNA 3 architecture are expected to go on sale. Although AMD hasn’t yet confirmed the moniker, there’s no reason to think the RX 6000-series model numbers will change. But until and unless we are proven incorrect, we will continue to use RX 7000. If the third time is the charm, AMD may even overtake Nvidia. These new GPUs will compete with the finest graphics cards.
The key technical specs for AMD’s RDNA 3 GPUs are still unknown. Although AMD will use “GPU chiplets” and will transfer at least some of the key building blocks to TSMC’s N5 manufacturing node, the remaining aspects of RDNA 3 and RX 7000 should be considered vague, at best.
Four GPU designs are currently dispersed across a whopping twelve different graphics card models in the RDNA 2 and RX 6000-series GPUs, and that’s not including integrated graphics solutions. There may eventually be a comparable portfolio for the RX 7000, but for now, we only know about the Navi 31, Navi 32, and Navi 33 core GPUs.
The top model could achieve 61.4 teraflops of computing with up to 12,288 GPU shaders performing one FP32 FMA (Fused Multiply Accumulate) operation every cycle, and twice that for FP16 workloads. Even the entry-level CPU would be capable of 38.4 teraflops, which is almost twice as much as what the RX 6900 XT now offers. Which does raise the question of whether these purported specifications are accurate or whether throughput will actually be lower. In the upcoming months, we’ll find out.
In comparison to the RX 6950 XT, total memory bandwidth on the top models should rise by 50%. Even though it could be more, we anticipate AMD to stick with “vanilla” GDDR6, and the availability of 18 Gbps chips appears to be very abundant. The Infinity Cache, whose capacity will typically be smaller than RDNA 2 for most models, will increase real-world memory bandwidth. Higher throughput or other design changes might balance that.
The AMD RX 6900 XT uses 308W of power while producing 130 frames per second at 1080p ultra, 106 frames per second at 1440p ultra, and 63 frames per second at 4K ultra.
While the RX 5700 XT consumes 214W and offers 74 fps, 53 fps, and 29 fps at those same resolutions, the lower-tier RX 6700 XT uses 215W and averages 96 fps, 71 fps, and 39 fps. In comparison to the RX 6950 XT, total memory bandwidth on the top models should rise by 50%. Even though it could be more, we anticipate AMD to stick with “vanilla” GDDR6, and the availability of 18 Gbps chips appears to be very abundant. The Infinity Cache, whose capacity will typically be smaller than RDNA 2 for most models, will increase real-world memory bandwidth. Higher throughput or other design changes might balance that.
There are now a lot of unanswered questions in the table, and we wouldn’t feel too horrible if we put one in almost every cell. At this time, there is no confirmation of names, core counts, or any other details; the only thing we can say with certainty is that AMD will employ TSMC’s N5 node for at least some of the chiplets.
The only definite information we have regarding the release dates of the RDNA 3 and RX 7000-series is what CEO Lisa Su has told us for the most of the past year: RDNA 3 will be released in 2022. But time marches forward, and with the impending release of Zen 4 and the new AM5 platform, certain AMD groups may currently be overworked.