As a leading global digital services and solutions provider, Presidio is passionate about exploring the latest emerging technologies, not only to understand the gains to be made but also to consider how they can be utilized to benefit each customer’s use cases.
For instance, Presidio works closely with our key technology ally, Intel, to continually explore the capabilities of their latest innovations and determine how they can help customers overcome specific challenges – improving performance, security, sustainability and other goals, all while optimizing ROI and TCO.
Recently, Presidio and Intel put the 4th Generation Intel® Xeon® Scalable processor to the test in the rapidly growing area of machine learning (ML) with a specific use case in mind: facial recognition. Our objective was to determine how it would perform against the common choice for powering artificial intelligence (AI) and ML workloads, the GPU.
Our initial testing showed that 4th Gen Intel Xeon Scalable processors offer a viable alternative to GPUs and address many of the challenges that GPUs can pose when running these demanding workloads. Rather than default to traditional GPUs, consider the advantages that could be yours by choosing 4th Gen Intel Xeon Scalable processors instead.
The capabilities and challenges with GPUs for ML
The advancing growth of AI across industries has placed increasing demands on cloud and data center infrastructure to handle challenging new workloads to support model training, fine-tuning, and inferencing. For years, the popular choice for powering AI and ML applications has been the GPU, valued for its critical role in machine learning and parallel processing. This makes GPUs ideal for the matrix calculations and neural network training that are central to ML applications.
But, despite the advantages of their performance for ML, GPUs also present several challenges: they are costly, require significant energy consumption, and can be hard to obtain in certain regions due to frequent shortages in the supply chain. These factors have led Presidio and others to seek a more cost-effective and energy-efficient alternative to GPUs in ML – such as 4th Gen Intel Xeon Scalable processors.
Advantages of 4th Gen Intel Xeon Scalable processors for ML
Since the first generation launched in 2017, Intel Xeon Scalable processors have been the CPU of choice for data centers, enterprises and cloud service providers due to their reliable performance and scalability. In addition to those benefits, we’ve now found that the latest 4th Gen Intel Xeon Scalable processors also compete favorably against GPUs for machine learning due to their:
- Parallel processing capabilities – When it comes to multi-threaded ML workloads that involve data preprocessing, model inferencing, and postprocessing, these processors rival GPUs as an efficient and reliable performer.
- Energy efficiency – 4th Gen Intel Xeon Scalable processors are not only designed for high performance, they are also optimized for energy efficiency, requiring less electricity to operate. As a result, they deliver significant energy savings over time compared to more power-hungry GPUs.
- Cost savings – In comparison, high-end GPUs are significantly more expensive than 4th Gen Intel Xeon Scalable processors. Instead of GPUs, these processors enable you to perform ML workloads on potentially fewer servers with a more affordable option, thereby reducing overall hardware costs.
Another key advantage that 4th Gen Intel Xeon Scalable processors bring to ML is their versatility and ability to perform inferencing – a critical element of artificial intelligence and machine learning. Recognizing these potential ML capabilities and benefits, Presidio and Intel teamed up to conduct an initial test for a specific workload to measure the impact from utilizing these latest generation processors in our ongoing quest to find new alternatives to address customer challenges.
Initial facial recognition test reveals a viable GPU alternative
For every customer, Presidio strives to find the best technology option to fit their specific use cases, challenges, and business needs; often that means comparing different technologies side by side.
Recently, a Presidio customer who uses facial recognition to support public safety needed a better solution that would meet or exceed the desired threshold for effectiveness. Presidio and Intel saw this use case as an opportunity to explore the capabilities and advantages of 4th Gen Intel Xeon Scalable processors versus traditional GPUs.
Requirements for effective facial recognition
The challenge in successful facial recognition involves the need to process a series of video frames in near-real time, analyzing facial features to make a positive identification in one second or less, with minimal latency; otherwise, a moment’s delay could degrade the usefulness of the I.D.
Previously, the customer had been using a solution that was slow to return those required actionable identifications. Presidio and Intel conducted initial tests in which two systems – one powered by a 4th Gen Intel Xeon Scalable processor and the other running a GPU – were tasked with recognizing one face in a video stream within one second.
The results were impressive:
- Comparable video frame processing performance within the desired threshold between the GPU and the 4th Gen Intel Xeon Scalable processor.
- Performant hardware at one-third the price when using the 4th Gen Intel Xeon Scalable processor to achieve the same result as the GPU.
- Up to one-third less energy required by the Intel processor versus the high-consuming GPU.
It’s important to note that this benchmarking was performed without any special optimization. Presidio believes that a contributing factor to the success of the Intel-powered option was its faster, more powerful cores and support for DDR5 for improved memory speeds.
The bottom line: 4th Gen Intel Xeon Scalable processors exceeded expectations as a viable and cost-effective alternative to traditional GPUs for ML video processing.
Presidio and Intel continue to explore use case benefits with its latest processor
As this initial test demonstrated, GPUs need not be the automatic go-to for machine learning when 4th Gen Intel Xeon Scalable processors can often provide comparable performance for many AI use cases at significant cost savings.
Presidio and Intel anticipate future use case testing which will only further improve performance by incorporating the advantages of built-in AI acceleration features like Intel® Advanced Matrix Extensions (Intel® AMX) and Intel® Advanced Vector Extensions 512 (Intel® AVX-512).
As you embrace AI and ML in your own operations, it’s important to explore all options, question old assumptions, and seek experienced help from a trusted technology advisor. Contact Presidio for assistance in making an optimized platform decision based on Intel® technology to incorporate artificial intelligence and machine learning into your organization.
