Components & Peripherals News
Oracle Bets On AMD, Ampere CPUs; Ellison Says ‘Intel x86 Architecture’ Is ‘Reaching Its Limit’
Oracle Chairman Larry Ellison reportedly ommitted Intel’s name when discussing the firm’s processor spending plans Wednesday: ‘This year, Oracle will buy GPUs and CPUs from three companies. We will buy GPUs from Nvidia, and we’re buying billions of dollars of those. We will spend three times that on CPUs from Ampere and AMD.’
Oracle reportedly plans to spend significant amounts of money this year on CPUs from AMD and chip startup Ampere Computing for new infrastructure as the company’s founder and chairman, Larry Ellison, said that the “old Intel x86 architecture” is “reaching its limit.”
The Austin, Texas-based infrastructure giant announced new initiatives with AMD and Ampere Computing in separate announcements over the past week. The moves underline the increased competition faced not just by Intel but also AMD, which makes x86 chips too, as Arm-based CPUs designed by companies like Ampere Computing proliferate deeper into the cloud and data center markets.
Oracle, which made $50 billion in revenue for its 2023 fiscal year, is also following rivals in the cloud and data center markets by investing deeper into GPUs from Nvidia, whose chips are the primary engines behind popular generative AI applications like ChatGPT.
At a Wednesday event, Ellison said the company is spending “billions” of dollars on Nvidia GPUs and investing significantly more into CPUs from AMD and Ampere Computing, Reuters reported.
“This year, Oracle will buy GPUs and CPUs from three companies,” Ellison reportedly said. “We will buy GPUs from Nvidia, and we’re buying billions of dollars of those. We will spend three times that on CPUs from Ampere and AMD. We still spend more money on conventional compute.”
Oracle Dumps Long-Time Ally Intel For AMD With Exadata X10M
Last week, Oracle revealed that it is fully eschewing Intel CPUs in favor of AMD CPUs for the company’s purpose-built database infrastructure, Exadata, across on-premises and cloud offerings for the first time in the platform’s 14-year history.
The company said that fourth-generation AMD EPYC processors, which launched last fall, will power the 12th generation of its Exadata platform, the X10M, across its two on-prem offerings: Oracle Exadata Machine and the fully managed Oracle Exadata Cloud@Customer solution.
An Oracle spokesperson told CRN that the X10M version of its third Exadata offering, Oracle Exadata Cloud Infrastructure, will also rely on AMD processors when the service is announced at a later date.
X10M represents a significant change in CPU vendor for Oracle’s Exadata platform, which is used by nearly 90 percent of the Fortune Global 100, according to previous statements by the company.
Holger Mueller, principal analyst at Constellation Research, told CRN that Exadata serves as a vehicle for the company’s database services and, as a result, helps drive Platform-as-a-Service revenue.
“Exadata has a long tail of revenue,” he said in an email.
Franky Faust, a lead Oracle database consultant at Ottawa, Ontario-based Pythian, said Exadata is popular with Oracle customers because it’s easy to use and fast for database workloads.
“Once you start using it, it’s hard to get away from it. It’s easy to use and easy to work with, and it gives you a lot of performance for your workload that you wouldn’t find anywhere else,” he said.
Oracle Tapped AMD For Higher Core Counts
For most of Exadata’s history, Oracle relied on Intel Xeon processors to power the on-premises versions of the platform since its 2008 debut. But the company’s preferences began to change with the X9M platform, which debuted in 2021.
While Oracle chose Intel Xeon for the X9M version Oracle Exadata Machine and Oracle Exadata Cloud@Customer, the company opted to mix Intel CPUs and AMD CPUs for that generation’s iteration of Oracle Exadata Cloud Infrastructure in early 2022. Third-gen EPYC CPUs powered the cloud offering’s database servers while third-gen Intel Xeon CPUs covered the processing for storage servers.
Now with X10M, Oracle is letting go of Intel CPUs entirely and betting on AMD CPUs for the latest generation of the Exadata platform, which the company said delivers “extreme scale and dramatically improved price performance.”
Compared to the X9M platform, X10M provides three times more cores in database servers and two times more cores in storage servers, which help the platform provide up to three times higher transaction throughput and as much as 3.6 times faster analytics queries, according to Oracle.
This large uptick in cores is due to the high core counts of fourth-gen AMD EPYC processors, which provide up to 96 cores on a single CPU. This means a dual-socket server can sport up to 192. Fourth-gen Intel Xeon, on the other hand, only goes up to 60 cores per CPU, requiring more CPUs and, therefore, more servers to achieve similar levels of core density.
“Tripling to 96 cores was long unthinkable outside of the world of specialized eight-socket machines. But with Exadata X10M, Oracle has shifted gears to work with AMD which has broken the scaling barrier on standard two-socket units,” said Tony Baer, principal at research firm dbInsight, in a statement.
An Oracle spokesperson told CRN that for every generation of Exadata the company chooses component suppliers “that we believe provide the best performance, reliability, and price for Exadata workloads.”
“In this release, the CPU vendor we chose was AMD,” the spokesperson said while leaving the door open for Oracle to choose competitors for future Exadata platforms. “We will continue to work closely with other CPU and component vendors on future Exadata products,” the representative added.
Faust, the database consultant at Pythian, told CRN that he doesn’t think Exadata customers “will care that much” about Oracle shifting the platform from Intel to AMD because, “in the end, they will get better performance for the same money.”
He did note, however, that because X10M has three times more cores than the previous generation, licensing all the CPU cores will be “more expensive.” But, he added, Exadata “offers a capacity-on-demand model” that will let customers pay as they grow.
On the other hand, Faust said, AMD could see increased acceptance with enterprises due to Oracle’s adoption of the chip designer’s CPUs for the latest Exadata platform.
“I think it helps more AMD than Oracle. Once AMD references that their chips are being used on Oracle’s Exadata Database Machine, it is more likely that AMD will get better acceptance across different enterprises,” he said.
Intel Used Exadata As Showcase For Now-Dead Optane Memory
Intel previously leaned on Oracle’s Exadata platform as a showcase for its Optane persistent memory, a now-discontinued technology for servers and PCs that combined the persistent qualities of storage with performance that nearly rivaled DRAM.
This began in 2018, when Oracle announced that the X8M Exadata platform would use Optane persistent memory combined with second-gen Intel Xeon Scalable processors to provide what it called the “fastest database machine in the world.”
“Our collaboration with Intel sets a new industry standard for supporting databases with the highest performance and availability,” Juan Loaiza, executive vice president of mission-critical database technologies at Oracle, said at the time.
Oracle continued to support Intel Optane with its X9M platform in 2021. When the platform was announced, Intel data center leader Sandra Rivera said the chipmaker’s “partnership with Oracle has never been more important.” She added that Intel “uses Exadata in our most critical factory operations within our manufacturing group.”
Despite Oracle’s support and reported customer wins elsewhere, Intel was losing hundreds of millions of dollars on Optane while generating revenues that were miniscule in comparison to the billions generated by the company’s CPUs. According to a report by storage publication Blocks and Files, Intel’s Optane business experience a $576 million operating loss in 2020 and then another $271 million loss for the first nine months of 2021 while revenue had only grown to $392 million for that whole year.
Then, less than a year after the X9M launch, Intel announced that it would exit the Optane business to focus on its core chip businesses. In place of Optane, the company is shifting memory-related work to Compute Express Link, an emerging standard for new servers that can enable features such as memory expansion, memory pooling and memory sharing, Intel CEO Pat Gelsinger said at the time.
Mueller, the analyst at Constellation Research, said Intel’s exit from Optane wasn’t a key factor in Oracle’s decision to choose AMD for the X10M platform but added that it “certainly did not help.”
Intel Touts ‘Long-Standing Relationship With Oracle’ As Ellison Praises Arm
An Intel spokesperson told CRN last Friday that the semiconductor giant continues to have a fruitful relationship with Oracle that includes providing new chips for the infrastucture giant’s cloud services.
“We value our long-standing relationship with Oracle, and we look forward to continuing our collaboration to deploy future industry-leading technologies across Oracle’s cloud infrastructure,” the spokesperson said in a statement.
The chipmaker added that “global adoption and momentum for Intel’s 4th Gen Xeon processors remains strong with the top 10 cloud service providers deploying cloud instances now and throughout 2023.”
However, Ellison’s comments about AMD, Ampere Computing and Nvidia as well as the omission of Intel’s name in his company’s processor spending plans indicate that the Oracle chairman is more keen on building out partnerships with Intel’s rivals.
The tech titan made his comments at an event where Oracle announced that its flagship database software, Oracle Database 19c Enterprise Edition, has been modified to run on CPUs based on Arm’s instruction set architecture, starting with chips from Ampere Computing, which has raised hundreds of millions of dollars in capital from Oracle.
According to Reuters, Ellison said chips made by Ampere Computing are much more power efficient than those from AMD and Nvidia, which will help Oracle’s data centers where electricity consumption is limited by regulations.
A blog post by Oracle also highlighted how the single thread per core architecture of Ampere Computing’s processors enable organizations to “run workloads with consistent and predictable performance while achieving excellent performance scaling.”
“It’s a major commitment to move to a new supplier. We’ve moved to a new architecture and we’ve moved to new supplier,” Ellison reportedly said. “We think that this is the future. The old Intel x86 architecture, after many decades in the market, is reaching its limit.”
In a new statement to CRN on Thursday, an Intel spokesperson said the chipmaker has been shipping its fourth-gen Xeon processors to Oracle “for many months and plans to continue that trend with current and next-gen Xeon well in the future.”
“We look forward to fostering that collaboration for years to come, working together on both hardware and software solutions for customers,” the representative added.
Oracle did not respond to a request for comment about its spending plans with Intel this year.
Faust, the database consultant at Pythian, called Oracle’s decision to enable Arm-based CPUs on its database software a “long-awaited move” and said the instruction set architecture—which powers most smartphones and, more recently, all of Apple’s latest Mac computers—“plays a vital role due to its energy efficiency without compromising performance.”
But despite Ellison’s dim view on the future of x86, which powers CPUs made by Intel and AMD, Faust doesn’t think Oracle’s latest bet on Arm will hinder its relationship with AMD anytime soon. There’s also the fact that Intel and AMD plan to challenge the capabilities of Arm-based processors with cloud-optimized CPUs tuned for high core density and efficiency.
“It will take some time for companies to adopt Arm-based servers or cloud instances. While we are witnessing a growing trend towards Arm, developers still need to invest time in coding and refactoring their applications to be compatible with Arm architecture, and this process cannot be rushed,” the database consultant said. “Apple’s transition to Arm-based Macs over the past three years has shown that it takes time for software optimization.”