Shared memory is inextricably linked with processor-integrated graphics solutions, and represents a key point in AI work.
A few weeks ago, Intel's CEO admitted that his company had missed the boat on artificial intelligence, leaving rivals such as NVIDIA and, to a lesser extent, AMD to dominate a sector that hardly anyone was betting on just ten years ago. While it is not likely to come close to matching the GeForce developer's success, Intel does have a few cards up its sleeve, for example in agentic AI, an emerging sector in which artificial intelligence must act completely autonomously, without human direction or verification.
Share GPU Memory Override © Intel
While waiting for this sector to develop, Intel still needs to find solutions so that its current products can offer at least a decent alternative. To this end, Intel intends to boost local AI performance on its Core Ultra chips. To achieve this, the graphics solution integrated into these processors needs to be freer to act, less constrained by the technical limitations of components not designed for this kind of task. So, like any integrated graphics solution, the Core Ultra iGPU exploits shared memory - a portion of system memory allocated to the GPU - which is often limited to 10 or 15% of RAM. In the best case, this can be increased to 25%.
With a new update to its Graphics Software, Intel is proposing to go much further by integrating the Shared GPU Memory Override functionality. In a nutshell, this function allows you tooverride the limits mentioned above. In fact, it's possible to achieve a shared memory allocation of 87% of system RAM. Of course, care must be taken to ensure that this setting corresponds to the real needs of the applications used, but Intel points out that on AI tasks currently blocked by the amount of memory available, the gain is real. Get testing...
