Company
Upgrade, Refresh or Wait? A Look Into the New Intel and AMD CPUs
May 18, 2021 — Bill Conrades

Practical considerations around the new Intel and AMD CPUs
Amid the noise surrounding the battle between AMD and Intel® for chip supremacy is the plain fact that both companies are working hard to improve their products.
MBX engineers and designers have researched both product roadmaps and completed some comparisons of the latest architectures. There are areas where we think it’s worth upgrading or switching brand alliances—but not without careful consideration around how such a change would affect the rest of your hardware ecosystem.
First, let’s take a high-level view of the new architectures from Intel and AMD, then look at areas where we think a hardware update or refresh for your platform might make sense.
Chips, cores, and more
AMD announced its third-generation Epyc™ CPUs—codenamed Milan—in February 2021. At the time, the company said these chips were twice as fast as Intel’s second-gen Xeon® Scalable processors. Compared with its own second-gen chip (Rome), AMD claims Milan offers:
- 2.2x performance increase vs. Rome
- Up to 64 physical CPU cores
- Up to 3.7GHz base clock frequency & up to 4.1GHz boost frequency
- Up to 32Mb level 3cache that is now shared between the cores of each CPU, enhancing on-CPU cache efficiency
- Higher quantity of PCIe lanes than Intel Ice Lake (described below), offering increased connection options and bus performance:
- PCIe Generation 4.0 used in Milan (and Rome) is 2x the bus speed vs. the generation 3.0 used in Intel’s second generation scalable processors
- Intel is now supporting PCIe Generation 4.0 with its new Ice Lake offering
- Higher quantity of PCIe lanes per processor socket (128 lanes vs 64 of Intel Ice Lake) enables solutions with densely packed storage and I/O solutions
- PCIe Generation 4.0 used in Milan (and Rome) is 2x the bus speed vs. the generation 3.0 used in Intel’s second generation scalable processors
Use cases for the Milan CPU include storage solutions, applications that require high networking bandwidth such as content delivery, and servers hosting virtual machines. For instance, a single socket AMD Epyc Milan-based server can host an increased quantity of the latest generation of NVMe drives and high speed network connections because of the PCIe lanes available per socket. In addition, AMD Milan CPU’s are attractive for virtual machine instances due to the higher physical core count.
Milan also provides enhanced security with built-in silicon Root of Trust. Overall, these are follow-on changes and enhancements from Rome. AMD seems to be on a stable trajectory for improvements.
Then in April 2021, Intel launched its third-gen Xeon Scalable CPUs, codenamed Ice Lake. Intel claims it has higher-speed processing and provides better performance for intense workloads, such as artificial intelligence (AI), HPC and cryptography processing. For example, compared with the Intel Cascade Lake generation, Ice Lake provides:
- Up to 40% performance improvement
- Up to 40 physical CPU cores
- Up to 3.9GHz base clock frequency & up to 4.4GHz boost frequency
- Deep-learning boost with built-in AI and crypto acceleration
- 1.6x memory bandwidth increase
- 2.66x memory capacity increase
Despite a significant difference in the number of cores—Milan has up to 64, Ice Lake has up to 40—Intel has embedded accelerators that are software-optimized specifically for AI and cryptography. For some AI workloads, the processors could serve as an alternative to GPUs. Intel also features a higher clock speed, which is an advantage for single threaded legacy applications or applications that thrive on clock speed.
In both cases, these improvements set the stage for further releases. While we see AMD’s Milan chips as the next in its series, we think Intel has taken a bigger step forward with Ice Lake: Intel has leveled the playing field in most areas by adding PCIe gen 4 and moving to 8 memory lanes, and has given customers more options.
Upgrade? Refresh? AMD or Intel?
So, where does that leave companies that either need to refresh their hardware due to end-of-life components or who are considering an upgrade?
First, because of the serious logjam across today’s hardware supply chain, we suggest making a component determination based on critical needs. For example, we’re seeing orders for GPUs that take 20+ weeks, and we are urgently advising our customers to forecast now and get in the order line. We can help with forecasting and offer alternates where inventory is scarce, even making alternate hardware platform design suggestions if needed.
On the flipside, because the Intel and AMD third-gen processors have been in the works and in their fabs, they may be a little easier to obtain than their predecessors over the coming months.
You’ll want to make your decision based on what’s best for your product. For example, the new Ice Lake CPUs require a complete architecture upgrade for your hardware platform, while the Milan CPU is a drop-in replacement with a BIOS upgrade. Also, you need to think about optimization—if you are switching from Intel to AMD, or vice versa, it’s recommended that you recompile your application code.
Another consideration: Our take is that the new generation of chips may necessitate creative cooling solutions, like liquid cooling in certain designs. As performance increases, so does the thermal design power (TDP), which could affect the price and design of your hardware platform.
Finally, if you’re unsure whether AMD or Intel provides the right CPU for your needs, ask us to provide an assessment. There may be instances where it makes sense to wait, and other cases that necessitate an urgent upgrade.
Contact MBX today for a customized review of the new Intel and AMD CPUs
Additional Resources
7 Reasons to Keep Applications & Data On-premise

Telecom
ARM-Based Processors Are The Future: Delivering High Performance, Low Cost, Sustainable Computing Solutions

Industrial
MBX Expands Edge AI Hardware Portfolio with NVIDIA IGX Orin-Based Platforms

Medical
AHEAD Engineered Solutions

Telecom
Why Pre-Trained Models Matter for Machine Learning

Medical