Battle of the Decade: AMD EPYC vs. Intel XeonDecember 27, 2019
Multi-processor servers have been around for decades, but with today’s technological advancements, a single-socket server can now do the work of a dual processor system. AMD EPYC 7000 series processors were announced to the tech world in 2017. The launch of these high-performance processors was expected to make a significant impact in the data center. Two and a half years later, they’ve created a whole new set of single-socket systems across several manufacturers challenging Intel’s dominance in this space.
For the past few years, AMD EPYC has established an enviable reputation as an industry leader providing unique compute solutions and security capabilities to accelerate workloads. But even so, Intel remains in the game outperforming EPYC particularly in database applications with their higher clock speeds. Marking a new era in the datacenter, AMD EPYC processors are seeing much wider utilization because of their performance specifications paired with a small price tag and lower power requirements— definitely giving Intel Xeon Scalable processors a real run for their money.
AMD EPYC vs. Intel Xeon Prices
The company kicked off the EPYC line with nine 7000-series processors, code name Naples, ranging from $400 to $4,000. Relatively speaking, these are about half the cost of Intel’s offerings, in general. Known for “2-socket features and performance, with a 1-socket budget,” AMD EPYC processors are at the top of their game providing up to 32 cores, 8 memory channels, and 128 PCIe 3.0 lanes per CPU, unlocking capabilities and performance previously available only in 2-socket platforms. Compared to Intel’s top of the line 8280 processors, that’s 4 more cores, 2 more memory channels, and 80 more PCIe 3.0 lanes.
Furthermore, power consumption is also much less than Scalable processors. You may have noticed that we have not addressed the 2nd generation Intel Xeon 9200 series with 56-cores, and that’s because it’s not available in standard x86 offerings. We also haven’t even discussed the second-generation AMD “ROME” processors, yet…more on that in a few paragraphs.
Unfortunately, Intel Xeon Scalable processors have fallen behind in performance and affordability. Their top-of-the-line platinum processors, also released in 2017, offered a mere 28 cores, 6 memory channels, and a maximum of 48 PCIe 3.0 lanes per CPU.
Intel Xeon processors can set you back from $8,700 to $10,000 depending on your needs. According to AMD, at every targeted price point for two-socket processors, EPYC outperforms the competition with up to 70% more performance in the $800 price range, and up to 47% greater performance at the high-end of the market of $4,000 or more.
So, who makes the best processors? It’s been an ongoing discussion, even Rick the scientist from Rick and Morty, an American adult animated science fiction show on Cartoon Network, has joined the AMD versus Intel debate. In the episode “The Old Man and the Seat,” Rick is preparing to perform a procedure when we see his computer screen.
As he boots up his system, we get a peek at his configuration and guess which CPU is listed? You guessed it, AMD—a 128-bit AMD CPU QX3700+ (@7.99hz) to be exact. I should mention that for us in the real world the QX3700+ does not exist and we still only have 64-bit systems and are not capable of supporting 3,584,825,480GB of RAM. But wouldn’t that be cool! He also uses Linux Debian—so not a fan of Windows—but that’s another story.
Decades of investment in software optimizations have continued to put Intel in the dominant position despite its new competition. Intel admits that AMD EPYC performs much better in memory intensive applications and also supports more RAM slots, but the performance is variable because of the way AMD created its cores. When AMD designed its new Zen architecture, it made a bold decision to move forward with a multi-chip design that relies on several small chips to form one large processor.
High single-core performance and a few cores go a long way for Intel because of the “bus structure.” In other words, having to wait for real-time calculations to complete before another calculation can start makes a difference when computing. It doesn’t improve performance with additional cores because most would have to wait until the next calculation is ready. If you have a higher IPC (instructions per cycle) clock rate like Intel, the wait time is shortened.
In addition, Intel also points out the issues with bandwidth and memory channels. AMD cores have direct local memory access to only 2 double data rate (DDR) channels, while Intel cores have the same access to all 6 DDR memory channels. AMD offers local memory access only to memory connected to a single die and data accessed from all other memory channels comes with a remote latency penalty. On the other hand, Intel offers uniform and consistent memory access for all cores and there are no compromises for memory bandwidth and capacity.
According to Intel, the Xeon Scalable processors have broken over 100 new world record performance benchmarks among popular systems like Dell, HPE, and Lenovo. Dell’s high-performance conjugate gradient (HPCG) performance study with Intel Skylake processors revealed about 65% better performance for a single node and about 67% better performance for both two and four nodes. Intel Skylake processors also give the Dell PowerEdge R940 a performance boost in comparison to its previous generation, the PowerEdge R930, where we can see more than 45% performance improvement across all the applications.
Because many IT organizations across the world face budget and space limitations, AMD jumped on the idea of creating a high-performance, low-cost solution without compromising performance. In the last few years, the no-compromise, single-socket capabilities of AMD EPYC have improved Dell’s EMC PowerEdge platforms, delivering up to 20% lower total cost of ownership in a single-socket, four node VSAN-ready configuration. The HPE ProLiant DL325 Gen 10 single-socket server also received a record-setting title as the most energy efficient AMD server and showed a 27% lower cost per virtual machine when compared to the top of the line Intel Xeon Platinum processor.
Typically, 2-socket servers help to overcome an imbalance of resources given the limitation of the CPU, but with a 1-socket AMD EPYC server you can satisfy workload needs and reduce capital, power, and cooling expenses—without additional capital investment. Choosing to go with a single-socket server can help save thousands of dollars in hardware, but also software. For example, you can cut licensing costs up to 50% with “per socket software” such as VMware vSphere or vSAN.
With its high core count, superior bandwidth, plus unparalleled support for high-speed channels in a single chip, EPYC aims to revolutionize the dual-socket server market while reshaping the expectations for single-socket servers.
“EPYC processors offer uncompromising performance for single-socket systems while scaling dual-socket server performance to new heights, outperforming the competition at every price point,” said Lisa Su, president and CEO of AMD. “We are proud to bring choice and innovation back to the datacenter with the strong support of our global ecosystem partners.”
AMD: The Unsung Hero?
Some of the world’s largest server manufacturers have reached performance milestones using AMD’s new EPYC processors; and include Dell, Supermicro, HPE, Gigabyte, and Lenovo, to name a few. Because of brand loyalty, AMD has been perceived as a cheap alternative to Intel. Being the best budget option isn’t seen as an advantage, but rather more like buying a store brand generic product. For many years, we have come to recognize Intel as a household name. The products we associate with being the best, like Mac products and Amazon web services, run on Intel. Since it carries that “brand-name” feeling, Intel has to be the clear winner, right?
What The Future Holds
In August of this year, AMD EPYC introduced their 1P second-generation “Rome” processors with 64 cores and about half the power consumption of the previous generation, with a price point ranging from $4,425 to $6,950. They are also compatible with the new PCIe 4.0 standard across all 128 PCIe lanes. The 7nm server processor uses the same SP3 socket and infrastructure as the 14nm version of EPYC Naples, which should make it easy to migrate to the new version. Although “Rome” wasn’t built in a day, the company is certainly laying the bricks to introduce “Milan” in 2020, and you guessed it, they will keep pushing the boundaries on core counts.
A few months before AMD’s Rome launch, Intel released their second-generation Xeon processors with a 56-core version, the 9200 series, which have reached price points of about $13,000. That said, the 9200 series is only available for custom implementations and are not available on the standard x86 systems we’ve come to know and love. Intel plans to introduce another set of processors during the first half of 2020, code-named Cooper Lake, with up to 56 cores. Cooper Lake is said to be the first x86 processor to deliver built-in AI training acceleration.
All in all, the two companies excel in creating game-changing products, but AMD has definitely kept Intel on its toes. It’s important to keep in mind that if you’re looking to build a system you will need to have compatible parts. You can’t put a Ferrari engine into a Honda and assume it will work the same, so if your software is expensive, everything changes. You’ll definitely want to forget about the price tags of the Xeons in order for your network to support your business without any problems. Intel Xeon Scalable processors deliver the highest single threaded performance and high throughput without much hardware tuning. You’ll be able to have Intel’s reassurance of their proven and reliable track record.
By comparison, AMD processors tend to be more affordable than Intel, but they both offer a number of strengths. As the “brain” of your computer, it’s important to find a processor that fits your specific needs−you don’t want to pay for features that you don’t need. Overall, Intel has focused on higher clock speeds and efficiency, whereas AMD has created high core counts and energy efficiency. AMD and Intel might be able to coexist in the long run since they cater to different audiences. Looking at the most notable releases this year, it seems like AMD will continue to leave Intel in its dust for the next couple of years, but as Intel has showed us before, the race to greatness is a marathon, not a sprint.