The Meteor Lake architecture is about to debut on the market and although we have already seen many of its benefits, there was another one left to know, which Intel has taken care of putting on the table. And AMD’s Ryzen AI has been a resounding blow in AI within a red CPU compared to what was seen in Intel, which had been left behind. Well then, Meteor Lake will compete head to head with Phoenix Pointand given what has been seen, I may overcome it thanks to the fact that Core Ultra will have GNA 3.5.
Intel can’t stay behind NVIDIA and AMD in GPUs and CPUs. Therefore, his preferred toolkit called OpenVINO has obtained an important update to version 2023.1, which mentions the improvements to the tools, and also the three new features that are included in reference to Intel Gaussian & Neural 3.5.
Intel Gaussian & Neural Accelerator, a coprocessor that will be replaced
The blues have been with their Gaussian and Neural accelerator for some years now, but for those who do not know what we are talking about specifically, we will give you a brief summary to understand its importance within the AI sector.
Intel Gaussian & Neural Accelerator (GNA) it’s a low-power neural coprocessor for inference. When power and performance are critical, this neural and Gaussian accelerator provides always-on, energy-efficient computing power and support to the P-Core and E-Core.
GNA is designed to offer computing power in AI voice and audio applicationssuch as neural noise cancellation, and at the same time Frees up CPU resources for overall system performance and responsiveness.
Understanding GNA as such, it must be emphasized that Intel has a specific library that provides an API for the inference of this coprocessor. This library is integrated within the commented OpenVINO and therefore, there is a specific driver that is updateable and essential for GNA to work well.
Hardware improvements for Intel GNA 3.5 focused on Meteor Lake before the VPU arrives
As we have mentioned before, the new update of OpenVINO 2023.1 It brings three specific improvements within what GNA is, and in them we have the key that interests us:
- Support was introduced for GNA 3.5the new version of GNA HW included in the Intel Core Ultra (Codename Meteor Lake).
- The C++ and Python automatic speech recognition samples are deprecated and will be removed with 2024.0.
- Optimized data preprocessing using AVX instructions in the CPU has Improved runtime performance.
- Introduced support for automatic layout conversion, allowing more layout models to be supported. TensorFlow ready to use.
So if GNA 3.5 debuts on Meteor Lake, what’s the real problem if what you’re going to do is pit Ryzen AI against Phoenix Point? Well, Intel is going to say goodbye to this coprocessor:
Intel’s GNA is being discontinued and Intel Core Ultra (known as Meteor Lake) It will be the last generation of hardware to include it. For this reason, OpenVINO 2023.2 will also be the last version to support the GNA plugin. Consider the new Visual Processing Unit (VPU) from Intel as a low-power solution to offload neural network calculations for processors that offer this technology.
Well yes, Intel is going to load up on GNA in one fell swoop and is going to bet everything on its VPU Movidiusso AMD better have something in store with Ryzen AI V2 or similar, because Intel’s intentions are clear and they are betting everything on a single color, which, as we already discussed at the time, looks really good.