The expectation for Intel’s next-generation Xeon ‘Diamond Rapids’ CPUs to lack multithreading came into view after its CEO said last year that he had ‘taken steps to correct past mistakes regarding multithreading capabilities on our P-cores’ by reintroducing the feature.
With AMD’s next-generation EPYC “Venice” processors set to launch this year, the company is optimistic that the server chips will extend its performance lead over Intel’s Xeon CPUs and allow it to continue taking market share from the rival.
This is according to Madhu Rangarajan, corporate vice president of compute and enterprise AI products at AMD, who told CRN in an interview last month that the chip designer is especially bullish about its server CPUs prospects this year because Intel’s upcoming Xeon “Diamond Rapids” chips are expected to lack multithreading, a common feature in x86 CPUs.
[Related: Intel Exec Confirms CPU Price Increases For OEMs Amid Supply Crunch]
AMD’s current generation of EPYC CPUs, code-named “Turin,” has a “pretty big gap” over Intel’s latest Xeon 6 “Granite Rapids” processors, Rangarajan (pictured above) said. The gap in question refers to AMD’s goal to deliver leadership in performance, performance-per-watt and performance-per-dollar, according to the executive.
“And then Venice is coming soon, and you can expect the gap to be opened up even further,” he said. “And we understand that our competitor has made some interesting choices on multithreading and so on.”
“That’s going to help us gain even more market share in a broader enterprise market where that has implications to your licensing and many other costs associated with it,” Rangarajan added.
In the fourth quarter of last year, AMD’s share of x86-based server CPU shipments grew 3.1 points year over year and 1.1 points sequentially to 28.8 percent against Intel’s 71.2 percent, marketing a record high for the company, according to Mercury Research. This, in turn, resulted in AMD’s revenue share hitting 41.3 percent in servers, the company previously said.
In response to Rangarajan’s comments, Intel fellow Srini Krishna said in a statement to CRN that the semiconductor giant continues to “see positive momentum with our data center customers—from cloud services providers to telcos to broad enterprise companies with the performance, security, and efficiency of [the] Intel Xeon 6 family.”
“Customers have choice and we have to earn their business, but we are confident in the direction of our road map and we remain focused on delivering improvements that matter in real deployments,” added Krishna, who works in Intel’s Data Center Group.
The Intel fellow said the company will share details about its road map “at a later date.”
“Ultimately, winning in the data center is about the aggregate value we deliver across security, efficiency, consistency, quality, and overall platform experience,” Krishna said.
Lip-Bu Tan Pushes Intel To Reintroduce Multithreading
The lack of simultaneous multithreading (SMT) in Diamond Rapids, which is set to launch this year, came into view after Intel CEO Lip-Bu Tan said last summer that he had “taken steps to correct past mistakes regarding multithreading capabilities on our P-cores” by reintroducing the feature for future Xeon processors.
“Moving away from SMT put us at a competitive disadvantage. Bringing it back will help us close performance gaps,” he wrote in a public letter to employees last July.
Tan, who made the comments after joining Intel in March 2025, mentioned this in the context of its data center business, which has bifurcated its road map between CPUs using its performance cores (P-cores) such as Granite Rapids and those using its efficient cores (E-cores) such as “Sierra Forest.”
A mainstay in most Intel processors for decades, the company’s implementation of SMT is called Hyper-Threading Technology, and it allows each CPU core to act as two logical cores that can handle multiple tasks at once, improving the processor’s overall throughput. AMD has been using SMT in its processors since the debut of its Zen architecture in 2017.
While Intel has not explicitly stated that Diamond Rapids would lack SMT, a Wall Street analyst called out the detail in the company’s January earnings call when asking about its competitiveness. In response, Tan said Intel is “laser-focused” on Diamonds Rapids processors with 16 memory channels as it pushes to “accelerate the introduction” of the subsequent Xeon generation, code-named “Coral Rapids,” which will reintroduce SMT.
“Overall, we are very positive. The team is in place. Now, the roadmap is very clear, and we are very decisive [about] doing that,” he said.
‘We’re Going to Take Advantage Of It’
Rangarajan, who worked at Intel for more than 10 years earlier in his career, said it’s also his understanding that Diamond Rapids won’t come with SMT. Such architectural decisions, he noted, are made years in advance, which is why it’s difficult for Intel or any CPU design company to reverse the removal of such a feature within a single generation.
While Intel plans to bring back SMT in the generations after Diamond Rapids, Rangarajan and fellow AMD executive Steve Berg expect their company to “capitalize” on Intel skipping over the feature for the next CPU product cycle.
“We’re going to take advantage of it,” said Berg, a corporate vice president who runs AMD’s server CPU business with hyperscale customers, in the same interview with CRN.
Rangarajan—whose previous employer, Ampere Computing, eschewed multithreading for its Arm-based CPUs—said the performance uplift that such capabilities enable for cloud and enterprise applications is worth the minor increase in silicon space required.
“If you look across a lot of cloud workloads that have shared data sets, they benefit a lot from multithreading, because when one of the threads is stalled, the other thread can make progress,” he said. “And on some key workloads that maybe some of us are using every day, we often see up to 50 percent performance upside just by having that second thread available for a very little minimal tax on the overall CPU.”
“It’s a huge bang for the buck in terms of what it gives you versus how much it takes in terms of silicon area and so on,” Rangarajan added.
Multithreading Helps With Software License Costs
Multithreading also helps enterprise customers get the most out of their software licenses, particularly when it comes to those licensed on a per-core basis, the executive said.
“Especially when you have workloads that are relatively idle, if you have two threads, the second thread can make progress while the first thread is idle and vice versa, so your software license goes a lot further,” Rangarajan said.
“So if you now look at it from a [total cost of ownership] perspective that includes software, there’s a significant advantage to just having that second thread available,” he added. “So that’s why we think there is a really strong value prop to SMT, and we will continue to have that and continue to improve that and continue to sell that.”
AMD Faces Growing Arm Competition In The Cloud And Elsewhere
While AMD sees opportunity to take more server CPU share away from Intel, both companies must contend with growing competition from those who are developing Arm-based processors, which includes hyperscale customers like Amazon Web Services; their biggest AI chip rival, Nvidia; and, starting later this year, Arm itself.
Rangarajan and Berg spoke to CRN before Nvidia revealed its ambitions to build a multibillion-dollar CPU business starting with its custom Vera processor and before Arm announced the AGI CPU, the rival’s first silicon product. Both were unveiled earlier this month with the aim of targeting agentic AI workloads in data centers.
In addressing hyperscalers, Rangarajan said that AMD’s goal is to “make sure we are continuing to deliver so much performance and performance-per-watt and supply and getting the products to them on time that it becomes a really hard decision for them.”
“Should I take on the pain of creating of my own CPU and managing my supply chain, especially in the environment right now, when AMD is offering leadership in performance and performance-per-watt that you can see born out on multiple cases,” he said.
Berg said AMD is also offering to make custom silicon products for hyperscalers, though he noted that no such projects are currently underway.
The company’s goal, according to the executive, is to be the “single most collaborative silicon provider on the planet” for hyperscalers.
“The relationship is very high touch, both from a technical and a business perspective. We have engineers that go in and work directly with their architects and their hardware engineers to identify both the solutions that they need, the speeds and feeds that they might need and the feature sets [they require],” he said.
This relationship also extends to co-selling motions, where AMD sends in sales teams with hyperscalers’ service providers “to identify the opportunities that exist within their fleet to migrate or modernize their customers,” according to the executive.
“They either move net new customers or help their customers save dollars,” he said.
