At a game developer conference in San Francisco this year, Microsoft launched a week lectures on modern graphics techniques with the introduction of the next ‘big thing’ by the next ‘big thing’: Co -operative vector. But all the promises offered by the new feature, it would be nice to see that we are playing before making a big difference in your sports. At least not because a GPU vendor has only joined the Matrix Core gang.
If you are wondering what is exactly what the co -operative vector is, you are in good company, because Despite their announcement at the beginning of the year, Microsoft has done a beautiful job in describing what they are, how they work, and what you can really do with them. Fortunately, Sean Hargirvas, head of Microsoft’s Directorate X Development, presented an excellent review of the feature in his graphics API. He is a Windows boy, but it is also worthwhile that he will also come.
Co -operative is the name of the vector An additional API for Directox 12 FamilyThis will allow programmers to directly use the GPU’s matrix or tensor core, without using the vendor’s specific API. For example, in the case of NVIDIA’s GeForce RTX cards, the tanker core automatically handles any FP16 calculation but otherwise, you need to use the choice of CUDA to program them. DX12 Co -Co -API gets around this problem, as it will help AMD, Intel, Nvidia, and even Qualcomm.
You can argue that Coopic is really a group of extra HLSL (high level sheder language) instructions that are specific to matrix vector operations, which is designed to accelerate the tensor/matrix cover. But one of the best ways to see the API is to bring the world of ‘small’ into shaders.
With Microsoft, representatives of AMD, Intel, and NVIDIA talked about what they were developing with Kopok. In the case of the latter, we have already heard about them: RTX structure compression, RTX neural material, and RTX neural radiation cache. Hargirvas described the current developments with Kopak as his ‘low execution fruit’, but these are also very early days that can be done with the new API. The common idea, though, is to use a small nerve network, which is stored on the GPU, which estimate that otherwise, shader warning or data will be very expensive.
Like Nvidia, Intel is also working on its nervous structure compression algorithm, though it is technically based on something Yubisoft developed. Intel’s system is slightly different from NVIDIA, but the purpose of both is the same final result. A metal surface, which is abrasive, erased and marked with dirt, will be offered using numerous, high resolution textures and in terms of visual quality and compression ratio, the standard method of compressing it is not perfect.
Nerve compression works best on many layered materials, and both Intel and Nodia run a pair of demo that offers the main VRAM load. However, the technique is a small fine of performance and it is not really usable on the majority of the structure load. So if you were hoping that the nervous structure compression would eliminate the need for many GBRMs, you would be out of luck.
It was the only thing in which Intel showed in the conversation and the AMD had nothing to say. Its speakers, who are going through a different different thing: the nerve lighting system used for its recent toy shop demo, yes, which left our Andy a bit pressed. Rosec, somewhat quietly pointed out that Kopuk has currently been cooperated only on his latest RNDA 4 Power Redon RX 9070 graphics cards, though AMD is looking to see if older Radon can support it.
There was no Qualcomm mark in the GDC 2025, but Microsoft said it was working to support the API for Snapdragon X chips.
But when I really think that Coopquake is great, it is also in the long postponement and a part of it, it is an AMD mistake to take such a long -standing matrix units to its users’ GPUs. I am also disappointed that compared to NVIDIA, how small AMD and Intel had to show in GDC.
Given that Intel’s Arc GPU has always separated the Matrix Core, you think it will have to go a bit behind the scenes. Perhaps this happens and we can see something more comprehensive in the near future and at least AI-Bosted in AID has the entire demo of rendering.
For this moment, this is NVIDIA, which is once again-telling what the future of real-time rendering looks like. Developers will be able to get their hands soon on Microsoft’s preview version of Copok EP, but initially, there is no plan for a full retail release by the end of the year. Let’s hope that Coopquake’s optic in sports is faster than direct storage, yes?
