Ampere’s 128-Core Processor Challenges Intel and AMD in a Cloud-Based Processor ShowdownJune 26, 2020 by Antonio Anzaldua Jr.
Intel and AMD have historically been the leaders of data-intensive processors. But now, Ampere's Altra Max 128-core processor may beat out the two in managing data center workloads.
Ampere recently announced that its 128-core cloud-native processor will be leading the way for addressing the demanding workloads in data centers.
Ampere’s Altra Max will be sampling in the fourth quarter of 2020. Image used courtesy of Ampere
Because workloads in data centers are rough on servers, processor speed and adaptability are critical to prevent bottlenecks. The workloads simultaneously consist of analytics, high capacity management, and application testing verification. Ampere’s Altra processors drive efficiency in the data center’s infrastructure workloads with a specialized design methodology and precise EDA tools.
Earlier this year, Ampere was able to design the world’s first cloud-native processor, a system built for cloud computing with a modern 64-bit Arm server-based architecture. Ampere’s Altra processor family is planning to continue to give users the freedom to accelerate the delivery of all cloud-computing applications. Ampere recently shared the news of its Altra Max 7 nm, 128-core processor that will launch at the end of this year—an effort to alleviate data centers.
Ampere initially launched an 80-core processor called Altra earlier this year. This processor is said to address infrastructure workloads found in data centers. Ampere’s Altra Max is the expansion of the newly-released Altra family.
The Altra Max processor will be useful for applications that take advantage of scale-out and elastic cloud architectures. The highly anticipated 128-core processor will also be compatible with its predecessor's robust rack servers.
There is a big lingering question to answer: how will Ampere’s new processors stand against competition such as Intel or AMD?
Intel’s Gold Processors Against Ampere
Intel’s Xeon Gold 6238R processor provides 28 cores and 56 threads. When these cores and threads are combined, the device provides performance similar to an 84-core processor. A processor’s amount of threads is a hardware support line; if a workload is running on each core and it stalls due to a memory access issue, a thread can start executing on the free core with minor setbacks.
Ampere Altra avoids hyper-threading, which is the method used by Intel to compensate for a lower amount of cores by breaking up physical cores into virtual ones to increase performance. However, in cases where the user needs to improve the cloud workloads and requirements found in data centers, there is little to no room for error or setbacks that can be brought on by relying on threads more than physical cores.
Intel’s Xeon Gold 6238R processor features 28 cores and 56 threads on 14 nm lithography. Image used courtesy of Intel
Intel has its 3rd generation of processors available that are built specifically to run complex artificial intelligence (AI) workloads on the same hardware as existing workloads, stepping up embedded hardware performance. Ampere’s family of Altra products claims to improve cloud workloads by utilizing all 128 high-performing cores and high-memory bandwidth while power management works toward low consumption.
A Close Race Between Ampere and AMD
AMD’s EPYC 7662 is a second-gen 64-core processor coupled with 128 threads—the real competition in addressing data center workloads.
AMD’s EPYC processors offer a consistent set of features across the product line, allowing you to optimize the number of cores required for the workload without sacrificing features like memory channels, memory capacity, or I/O lanes.
AMD’s 2nd-gen processors are ideal for virtualized and cloud computing environments. Image used courtesy of AMD
Ampere’s Chief Executive Jeff Wittich states, “If you can scale out to a ton of cores with 128, Ampere Altra Max is going to give you the highest socket performance and the highest overall performance for those applications.” Regardless of the number of physical cores per socket, AMD and Ampere surpass Intel in per-core performance.
AMD holds a slight advantage in terms of cache memory; the way AMD’s EPYC handles copies of data closer to the processor’s core is at memory size of 4 MB at its level 1 cache while Ampere’s Altra is at 64KB. There are three levels of cache memory. Each level gets slower as it increases in memory size but improves the performance of previous levels. AMD’s EPYC has more cache memory space to store frequently opened programs than Ampere’s Altra.
It All Depends on the Application
Ampere’s Altra addresses many data center workloads including data analytics, artificial intelligence (AI), database storage, edge computing, and web hosting. It is an apt choice for data centers since it avoids relying on threads by offering a higher number of physical cores.
However, for AI-based workloads, Intel’s Xeon Platinum and 3rd-gen scalable processors provide an accelerated inference performance for these deep learning workloads.
The rollout of Ampere's Altra family. Image used courtesy of Ampere
Ampere, AMD, and Intel all have processors that are pushing the boundaries to provide clients with dependable, reactant, high-performing processors. Each manufacturer has processors designed for high-performance computing workloads with a supporting eight-channel DDR4-3200 MHz memory. But for addressing the demanding workloads of data centers, evaluating memory storage, speed, and per-core performance, Ampere and AMD may be the most fruitful options.