Today marks a new day for the Radeon division of AMD as the Silicon giant has finally shifted from the GCN architecture. After a series of leaks and rumors regarding benchmarks and updated pricing, the Navi based AMD Radeon RX 5700 and 5700XT are in the market. AMD decreased the price of these graphics cards by $50 across the board, making them more attractive for the consumers, especially those consumers who want to jump into 1440p gaming at a reasonably lower cost.
These graphics cards are based on the new RDNA architecture, which is a hybrid of the upcoming Navi architecture and the old GCN architecture. The “intermediate” architecture allows AMD to carry the benefits of the GCN architecture over to the Navi while at the same time allowing it to be scalable for the most part. It is the first desktop grade GPU architecture that can be scaled down for the use in the mobile SoCs.
While there are many pros of the new architecture, the initial testing reveals a major drawback that many benchmarkers will not find particularly pleasing. AMD has removed CrossFire support from the new Navi GPUs. CrossFire is AMD’s equivalent of Nvidia’s NVLink. It allows users to scale the power of two or three similar GPUs (depending upon the type) up to almost double or triple respectively. The method does not scale the power of the cores; instead, it allows the programs (read games) to use more memory bandwidth.
These GPUs support the new PCIe 4.0 interface which means one x8 slot of the PCIe slot will be similar to one x16 slot of the previous PCIe 3.0 slot. If you insert two RX 5700 or 5700XTs in your beefy motherboard, it will not give you the option to implement CrossFire. AMD’s reasoning for the lack of support seems unsatisfactory as the company is relying on the developers to implement the use of dual GPUs in the games. The statement that AMD gave in response to Techpowerup‘s question is as follows.
“Radeon RX 5700 Series GPU’s support CrossFire in ‘Explicit’ multi-GPU mode when running a DX12 or Vulkan game that supports multiple GPU’s. The older ‘implicit’ mode used by legacy DX9/11/OpenGL titles is not supported.”
On the other hand, Nvidia is increasing the implementation of the NVLink interface. With the introduction of Turing architecture, they announced that two GPUs linked with the new NVLink interface would be able to enjoy precisely doubled VRAM space and bandwidth.
With both companies going opposite ways, it is in the hand of the developers to either implement the use of dual GPUs in games or refrain from it. If we look at the current market structure, the devs will most probably side with Nvidia as it might not require a lot of work on the devs part and secondly, dual GPUs won’t be the norm anytime soon.