Security experts at the University of Virginia's School of Engineering and Applied Sciences said that it is possible to retrieve instructions from the so-called micro-op cache. In such a cache, which is available in both Intel and AMD processors (Intel processors have had this cache since 2011), the simplest instructions for the branch prediction are stored. Researchers have identified mechanisms that can extract this data and use it to obtain confidential information.

The problem is that the level of the microinstruction cache is much lower than the level at which vulnerabilities like Specter work. Therefore, the Specter Vulnerability Protection does not work against the new vulnerability. Moreover, since the new vulnerability attacks a more primitive layer in processor architectures, an attempt to weaken or close it will seriously affect the performance of both Intel and AMD processors.

The researchers reported their discovery to all interested processor architectures. But they themselves doubt that such a vulnerability can be closed with patches without seriously affecting the performance of computing systems.

It all look for me as staged things.

AMD also can be hitting performance wall and new generations can be now focused on "more security".