AMD said it itself: its RX 7900 XTX will not compete with the RTX 4090. However, we know that on the green side, we will not stop at the 4090 since the AD102 GPU still has some under foot. However, it is possible according to multiple rumors that AMD has a high-end version to compete, we speak of a RX 7950 XTX. The Reds could be looking to fight back by continuing to press where it hurts: high-performance graphics cards with a design that allows for more aggressive pricing. Let’s note here that we are in the midst of speculations and that this brief is a synthesis of them (having taken care to evacuate the most delirious ones).
RX 7950 XTX: Towards the return of the dual GPU?
With the new RDNA3 cards, AMD has put forward products with a very interesting power/price/performance ratio. As we all know, increasing performance on graphics cards means unlocking all the capabilities of a GPU (and therefore sorting and selecting the best ones) but also feeding it a bit more… Which means a clear increase in power consumption. In Nvidia’s case, getting complete (or almost complete) AD102s means selecting the rarest chips at the end of production, i.e. the ones with the least defects. An operation that makes the price of the GPU higher. AMD could adopt a completely different method, taking advantage of its new architecture. The future top-of-the-range reds could be built around 2 Navi 31 chips. A solution that would allow to bypass the yield problems and obviously to contain the prices. Another point, taking advantage of the presence of Infinity cache on both chips, choosing to increase the VRAM might not be necessary. AMD could stay on 24 GB, 12 for each group of chips or choose to share it. Obviously, on paper “simple”, this operation will require other technical choices on the part of AMD, particularly with regard to the width of the bus. But reducing the width of the bus and lowering the frequency of the two chips could however allow AMD to release a very competitive card, cheaper to produce than its competitor and potentially with controlled consumption not generating additional design costs.
A lot of speculation, we can easily agree. But AMD may not have said its last word on the very high end.
