Amd files a new patent to get closer to the performance of nvidia gpu

AMD has not given up on the fight for graphics processor supremacy, and has recently filed several architecture patents for its upcoming post-Navi graphics cards. The objective? Of course closing the performance gap with Nvidia and revamping the old CGN architecture.

Has AMD ever been ahead of Nvidia?

If you're the world's second-largest manufacturer of high-performance desktop graphics cards, the least you can do is try to move forward and gain ground on your first-place rival. This is precisely what AMD has been planning for a long time, but has never achieved in the graphics world.

This is why AMD has filed new patents to implement a new architecture on its post-Navi graphics cards, the current and newly released 7nm technology.

As in any great company, there are always lights and shadows. In this case, the light is undoubtedly represented by Zen technology, with which AMD has managed to position itself after many years at par or even surpass Intel in innovation. Its new Threadrippers are worth seeing, and the best thing is that they are much cheaper than the Intel ones.

But on the other side we have the shadow, one that has stretched from the time AMD bought ATI to cover the desktop graphics market under the AMD Radeon name. We will agree that AMD has good graphics and cheaper than Nvidia, but they have never managed to beat this one in terms of pure performance. Cards like the recent RX Vega 64 put AMD on par with the performance of an Nvidia RTX 2070, a graphics card that is by no means the top of the brand range. And now with its new Radeon VII and its innovative 7nm, which no one has yet been able to reach, it has equaled in performance a not inconsiderable RTX 2080, or at least so say the benchmarks of the brand and the independent ones. But of course, there are still quite a few well above this one, like the 2080 Ti and the incredible Titan RTX.

Another field where AMD cannot compete is in real-time ray tracing. Nvidia took this new technology out of “the sleeve” with its new Turing range that further opened the gap against AMD, with graphics cards that far exceed the needs of this generation's games. How could I then do an AMD flip? Well, they won't even know it, but at least they will try.

A new architecture to improve graphics performance and power

This attempt consists of a patent application named “ Stream Processor with high bandwidth and low power vector regitry file ” which is the next in the list of a previous patent with an almost equally tedious name “ Super single instruction multiple data (Super- SIMD) for graphics processing unit (GPU) computing ”.

What these patents let us know, is that a new architecture is being cooked that seeks to eliminate the GCN (Graphics Core Next) architecture that has been in force since 2011.

In the image we can see, as far as we can understand, that AMD intends to implement a series of high-bandwidth flow processors with its own cache memory, buffers and instruction queue. Something similar to what Nvidia has implemented in Turing, doubling the cache capacity of its CUDAs. And it is that, sharing resources facilitates construction, but also generates large bottlenecks in the performance of electronic components. We do not know if in this way they will be able to provide their graphics with ray traces in real time, but at least we will have a possible renewal, which is already underway.

Do you think AMD will be able to match Nvidia in performance one day? We trust that yes, but Nvidia will not be less crossed, either, and the advantage they have today is undeniable.

Hothardware font