CPU breakthrough for machine learning and AI domain

As a well known fact, for machine learning, an artificial intelligence (AI) GPU or graphical processing unit does matter to accelerate at least 5x for whatever tasks involved. It is due to the fact that there are a lot of execution units (or cores) involved.

However for the coming new CPU, it may not be that anymore. Latest breakthrough and new algorithm proposed by computer scientists from Rice University, claims to make the CPU 15 times faster than some leading-edge GPUs. In theory, it is much easier to achieve high performance using compute GPUs as most algorithms are based on matrix multiplications.  If that research findings were relevant, then it would change the current architecture, and favour “unified system on a chip (SOC)”. Whatever current market dedicated machine learning or AI dedicated core may or may not really be that relevant, or we expect for the price to decline. It has the same impact on the GPU market as well, if GPU does not matter and it can be handled by the CPU, massive amounts of GPU being used or planned to use for the machine learning and AI may be in transit with better, faster and more cost efficient solutions.

CPUs continue to evolve, in particular for ARM based chip, unified and system on a chip, power the device for whole day longs and power efficiency, and keep revolutionary breakthroughs by stacking the architecture to become more powerful chip then previous generation of product.

Data science and machine learning in the past, when trying to elicit performance of training models at scale and experience processing lag, may eventually become the history, as datasets get larger. Make use of GPU for image and video processing on the fly is not secret for years, even typically image and video processing software will leverage it for the same purpose. The computation power has been heavily used by those who perform bitcoin mining for years.

So, you will not find surprising, a lot of business applications are GPU processing power intensive for the multi core and mathematical computation, not really just for the graphic processing purpose for the 3d rendering and visualisation.

Just at the beginning of 2022, all the CPU houses already published their new line of CPU and data to show how they outperform the Apple M1 Max. 

E-SPIN Group in the enterprise ICT solution supply, consultancy, project management, training and maintenance for corporation and government agencies did business across the region and via the channel. Feel free to contact E-SPIN for your project requirement and inquiry.

Other post you may interest in:

1. Machine learning what it is & why it matters
2. How Artificial Intelligence (AI) deliver real value for companies
3. System-on-chip (SoC) from evolutionary to revolutionary
4. 5 Common ML Challenges Data Scientists Face
5. Advantages and Challenges of Deep Learning