Computer

Intel removes a feature from its CPUs that improves gaming performance

Apparently Intel forgot to remove AVX-512 instructions from its desktop CPUs and now has plans to remove them altogether. And no, we are not talking about a simple blockage, but rather that they are not found in any of its cores from the Core 12 ranges onwards. How do they intend to do it?

Rumors say that Intel did not plan to launch the Xeon Sapphire Rapids for workstations and that it was an afterthought. The objective was that the high-end Intel Core 12 were the ones oriented to them, but in the end those of Pat Gelsinger decided to back down. The problem? The AVX-512 to be eliminated was already in the Golden Cove cores and there was no time for changes.

Intel is going to completely eliminate AVX-512 in its Core 12

Intel Core 12 Overclock

The Golden Cove core or P-Core of the twelfth generation is common in the Alder Lake and Sapphire Rapids architectures, however, Intel is not interested in one of its capabilities, such as the execution of AVX-512 instructions, remaining only for its CPUs. for workstations or servers. Technically, it makes sense to use the same core to avoid duplicating resources unnecessarily, but this means that certain capabilities reach certain CPU ranges that are not in Intel’s interest.

Let’s not forget that Intel has never officially mentioned the AVX-512 instructions in relation to its Intel Core 12. What’s more, you need to disable the E-Cores through the BIOS in order to activate these instructions. A feature that was included on Z690 chipset motherboards and later removed by Intel on 600 series motherboards via a BIOS update. That’s why we shouldn’t be surprised by Intel’s plan to remove AVX-512 entirely from cores. P-Core or Golden Cove.

That is, Intel has developed a version of the performance core of its most powerful desktop CPUs at the moment that cannot execute AVX-512 instructions in any way. Which will make the first batch of Alder Lake-S architecture processors more valuable for those using applications optimized for Intel’s 512-bit SIMD instructions.

Intel intends to promote AMX instructions

AMX

Intel’s move to remove AVX-512 instructions is in response to the fact that they have added their new AMX instructions. Which are optimized not for vector calculus but for matrix calculus that is widely used in deep learning applications. That is to say, they are the equivalent of the Tensor units of NVIDIA GPUs and AMD CDNAs, but in a CPU. These new instructions for the moment have not been adopted by AMD at the desktop level, although the veteran ones AVX-512 as of Zen 4. So it is clear that Intel intends with it that the new versions of the applications are optimized for the new type of unit and instructions that they have exclusively to have an advantage over its rival.

We don’t expect Intel to bring back support for their 512-bit long instructions in Raptor Lake, given their policy of removing any remnants of them from their desktop and laptop lines. In any case, it is a question of controversial instructions due to the fact that they require a lot of energy to work and are a problem for the paradigm of increasing performance with respect to consumption in each generation.

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *