Do I Need 2 CPUs?

Are you looking into an HPC system for your workloads whether it be a workstation or server? Should you spring for a Dual CPU configuration? We want to flush out any details on considering dual CPUs and highlight what are the advantages and shortcomings of a Dual CPU system versus a Single CPU.

What comes with a Dual CPU system?

Dual CPU systems require dedicated motherboards that enable the installation of two CPUs. Yes, the CPUs must be identical, otherwise, your system will not boot. Please do not mix and match processors or you risk damage to your system.

With a dual CPU system, the motherboard offers twice the memory slots, two independent sets of cache, and more PCIe lanes connected for more graphics cards or other expansion cards, as well as supporting more storage.

The point of a dual CPU setup is not only to increase the number of ports and I/O but also to stack additional computing cores for better parallel performance and improved multitasking like opening 10 Chrome tabs or exporting a video while you edit another.

Dual CPU systems usually have better performance than single CPU motherboards sharing memory resources and synchronizing to process workloads in parallel by splitting the task between the two.These extra components effectively double the performance of a workload. So why is everyone not running a dual CPU configuration?

Does Your Workload Require 2 CPUs?

Many common workloads that day-to-day people do, like gaming, surfing the web, streaming movies, or graphic design and content creation often don’t need anything more than a single CPU. Most casual workloads sometimes cannot take advantage of the second CPU like gaming. Your second CPU stands idle, consuming wattage and generating heat without doing any work not to mention the extra cost incurred. Compute intensive tasks can get by with a modern, top of the line, single socket CPU configuration since the incorporation of higher core counts and higher thread count have become the norm.

Other than the system being too fast for normal workloads, if your workload does reach high throughput, bear in mind dual CPU servers have additional latency due to NUMA (non-uniform memory access) where the two CPUs have to communicate to share resources. NUMA isn't the bottleneck though; the inherent design of incorporating a large number of cores. NUMA is actually quite efficient in solving this issue when sharing CPU1's share of the workloard with CPU2's. However, it is best to avoid NUMA if you can, to mitigate the latency by partitioning workloads between the two CPUs. NUMA isn't a drawback or disadvantage, just a precaution.

When are Dual CPU Systems Used?

As of 2022, present-day CPUs are miles ahead of what they used to be; high core counts, multi-thread capabilities, and above-average performance; dual CPUs are rarely at your desk.

Dual CPU systems can be beneficial for streaming. By splitting the task of gameplay and streaming between the two CPUs, gamers/streamers can avoid stuttering and CPU bottlenecks and provide a higher quality stream. However, instead of investing in a dual CPU motherboard, streamers often opt to build two separate single CPU systems for added flexibility for hardware such as dedicated GPUs and storage for stream recordings.

Dual CPU systems are reserved for HPC servers designed for the smallest footprint. A large data center infrastructure would house thousands of servers executing high-performance tasks. Deploying 5,000 machines as opposed to 10,000 can save on rack space and time spent servicing these machines.

Some use cases for dual CPU systems include scientific simulation and calculations, rendering farms, complex deep learning AI training, or deploying dense storage. The advantage of a dual CPU system is performance density. Multi-threaded CPU-intensive applications can be executed in a breeze, even handling multiple high-capacity databases due to their sheer power. The 64-core 128-thread AMD EPYC is a multi-threaded performance king; with 2 of them in one system, their computing power is no joke.

Looking to speedup your workload and increase your throughput?
Exxact offers NVIDIA GPU Servers for Deep Learning and AI with dual AMD EPYC CPUs or dual Intel Xeon Scalable CPUs configurations.

Furthermore, data centers optimize these CPUs to perform a set of tasks at extremely high speeds with no CPU utilization wasted. For a system that fits in a single rack at the data center, nothing beats a dual CPU system that can be virtualized into two systems. These dual CPU systems effectively have two enterprise-grade single CPU systems populating only a single rack space in the data center. These infrastructures crunch huge amounts of data and need any last drop of extreme performance they can get within a small footprint.

If you are not ready to build a server infrastructure or your workload doesn't require one, a Dual CPU workstation can be a good start. With the increased PCIe slots, dual CPU systems enable multiple GPUs useful for small-scale operation such as AI training or building complex algorithms. By increasing your CPU count, you can double your GPU count cutting the time spent training. They can also be utilized in workstaion simulation softwares such as AMBER and Ansys, and rendering applications sucha s RELION where huge servers are not needed but instead just need a single dual CPU workstation enable more storage and more compute to assist with parallel applications and.

How to Optimize Dual CPU System Workloads

The big deciding factor when deciding on dual CPU or single CPU solutions is the workload. While Dual CPU systems can be beneficial for parallel computing (calculations performed simultaneously) like rendering, they can cause latency issues when executing tasks serially (calculations performed subsequently). While a Cinebench test (parallel) can show double the performance, a 7zip test (serial) won’t be able to take advantage of dual CPU and in some faults, be slower.

The best way to optimize dual CPUs is to make sure datasets/workloads are separate. Something like this might be training a deep learning network for Natural Language Processing where analyzing large datasets split into two. Or set aside multiple sets of calculations for your scientific research or cryo-EM 2D classification and 3D reconstruction. Almost any HPC application where parallel computing can be sped up dramatically.

Deploy a virtual machine using a hypervisor like VirtualBox or RedHat to make two separate systems. Run the two datasets separately on the two VMs. Because VM1 and VM2 are independent machines operating like a single CPU system, these CPUs can essentially work in parallel while executing serial tasks, doubling the speed if we only had one CPU. Now you can run twice the number of unrelated serial tasks at the same time.

So, do you need Dual CPUs?

With current-generation CPUs holding so much power, there isn’t a substantial need for a dual CPU system for mainstream consumer usage. AI training and deep learning workloads can benefit from parallel computing through GPUs. However, if your workload is on the HPC level like crunching extremely large datasets, training large AI models with the highest efficiency, or running massively compute-intensive scientific simulations, and you want to execute these tasks with the smallest footprint, a Dual CPU configuration might be your answer. Exxact has both Deep Learning and AI Workstations and Servers that house dual CPU configurations. We can help you decide which system is the best system for your needs.

Have any questions?
Contact Exxact to learn more!