As per German outlet, Igor’s Lab, AMD’s much-anticipated Big Navi graphics card based on the RDNA 2 uarch will be on par with NVIDIA’s newly launched Ampere flagship, the RTX 3080, but slower than the RTX 3090. Furthermore, considering that the Navi 2x GPUs will be fabbed on TSMC’s 7nm node which is much more efficient than Samsung’s 8nm process, Big Navi will have a slightly lower power consumption than the RTX 3080: 300W vs 320W.
Igor wasn’t very clear with the details and said that at 300W, Big Navi catches up with the RTX 3080 (320W) and at 275W, it performs between the RTX 3070 and 3080. As of now, the AIB partners haven’t received received the specifications of the new GPUs, nor the chips themselves. As such, we’ll only be getting the reference cards this year, with the third-party cards launching around Christmas at the earliest.
Update: Here’s the link to the video. Keep in mind it’s in German.
AMD has already confirmed that it’ll be ditching the blower-style heatsink in favor of a more effective and attractive design for the reference cards so this shouldn’t be much of an issue for early adopters.
Let’s have a look at the rumored specs and the block diagrams:
As you can see, the full Big Navi die will feature a whole 80 compute units or 40 WGPs, resulting in a total core count of 5,120 cores. NVIDIA’s GA102 has a lot more raw shaders at 8,704, but half of them are dedicated FP32 cores while the other half can either do FP32 or INT32 operations. AMD’s SIMDs, on the other hand, can do both, albeit only one per cycle. Furthermore, Ampere takes two cycles to complete executing a warp while RDNA 1 and 2 take only one cycle. As such, the core counts can’t be directly compared, but I’d put the RTX 3080 and Big Navi on par on this front.
Then, there’s the matter of the BVH and intersection units. We don’t know much about how AMD’s units handle it but we do know how NVIDIA does it. The 2nd Gen RTcores are supposed to be 2x faster than Turing. One oddity with AMD’s units is that they are linked to the texture units and both share the same resources. As such, the effective capabilities of the ray-tracing cores will be slower. However, as already explained we have no idea how powerful AMD’s ray-tracing hardware is.
There are two other things. AMD’s Big Navi has the same number rasterizers as the RTX 3080 but the render backends or ROPs are half as much. While this necessarily won’t be a bottleneck considering that both architectures are quite different from one another, it gives NVIDIA’s GPU an edge in terms of the last part of the fixed-function graphics pipeline. This means faster triangle culling, pixel blending, stencil testing, triangle re-ordering, etc.
Last, but not least, we have the memory. NVIDIA’s higher-end Ampere lineup uses GDDR6X memory and from what we know, it’ll be exclusive to GeForce cards for the time being. Therefore, Navi 2x will continue to use GDDR6. However, the bus width is likely to be wider.
While, all these things might put Big Navi in a bad light, keep in mind that the 2nd Gen RDNA cards will leverage TSMC’s 7nm node which is much more efficient that what NVIDIA is using. That translates into much higher clocks, better power efficiency and higher transistor density.
Overall, I believe Big Navi has a really good chance against the GeForce RTX 3080. That’s doubly true considering AMD’s habit of aggressively pricing their products. If the RX 6900XT is priced $100 lower than the RTX 3080 and performs roughly the same, then I’d have no issue recommending it over the latter.