Trending Topics
Trending
Intel broadens Xeon 6 family with launch of Xeon 6700P and 6500P series processors
Following the launch of its first Xeon 6 processor family with P-cores, the Xeon 6900P, Intel today announced the Xeon 6700P and 6500P series processors with P-cores. By doing so, it aims to address the full range of enterprise workloads, including data centers, AI and HPC (high-performance computing).
“We’re trying to solve for the broadest set of workloads, the broadest set of deployments and infrastructure use cases,” said Ronak Singhal, Chief Architect, Xeon Products, during a private pre-briefing event.
He added that Intel wants to allow “our customers to optimise what they’re running from both a power standpoint, an efficiency standpoint and a TCO standpoint, while also looking at emerging workloads like AI”.
For the first time, Intel also publicly stated how it believes its Xeon 6 family compares against AMD’s EPYC chips in the key area of AI. Unsurprisingly, it reckons it does well.
Intel Xeon 6700/6500/6300: completing the Xeon 6 family
As can be seen from the slide above, Intel is clumping the 6700P and 6500P families together. “The difference between 6700 and 6500 is really just different points on the performance spectrum,” said Singhal.
To draw out a couple of key points from that table, the series includes fewer cores than the 6900 (up to 86 rather than 128) but keeps a high number of PCI Express lanes. That, Singhal said, was already proving attractive to companies creating storage solutions.
Unlike the 6900P and Xeon 6 with E-cores, socket support now extends to 4S and 8S rather than just 1S and 2S. Although, again, a single socket solution with up to 136 PCIe lanes is well suited to storage.
Across the family, Intel is attacking network, content delivery and AI solutions. And HPC/supercomputing: this is where the MRDIMM memory becomes a big factor, said Singhal.
“Really the predominant use cases [for MDRDIMM] are obviously people who can benefit from the memory bandwidth,” he said. “So we see it on what I’ll call the classical HPC workloads, as well as some of the emerging AI workloads that really demand that bandwidth.”
Although Intel wanted to focus on the 6700P and 6500P series, Singhal mentioned that the 6300 with P-cores have also been announced. He added that the 6300P series “focused on low core count, more the value side of the spectrum”.
Full details of the entire family are at the foot of this article.
Shape and potential of the AI market
Singhal prefaced his comments by pointing out that we need to define what we mean by AI. While GenAI services have grabbed the headlines, he said that – according to IDC data – the bulk of spending remains in established areas.
“More dollars were still spent [in 2024] on what I’ll call classical AI, so things like machine learning and other analytics,” he said.
“Looking ahead to 2027, “there’s a significant growth in generative AI, you can see here, going from $40 billion to almost 4x higher at $153 billion. But the classical AI space is also growing, growing slower than generative AI, but it’s still a bigger space than generative AI.”
Intel vs AMD in AI (Xeon 6900P vs EPYC 9005)
So, how does Intel believe it fares in this key area? I’ll let this slide do the talking:
“The comparison [above] is with our 6700 series at the 86 cores versus 128 cores of our competition,” said Singhal. “So even though our competition has more cores, you can see that for these AI use cases where the compute’s done on the CPU, we are able to beat them, and that’s because of… the compute capability we have per core, as well as our bandwidth capabilities.”
The above slide shows how the 6980P, part of the Xeon 6900 series with P-cores launched last year, compares to AMD’s EPYC 9965. Deliberately, Singhal is comparing its 128-core part with AMD’s 192-core chip. That’s the highest number of cores in either of the families. z
“This is using AMD just-released recently software optimisations for their CPUs,” said Singhal. “Those software optimizations are actually based on the software optimizations we’ve done, which, you know, that’s part of the open-source ecosystem. So this is using their top core count with their top software optimizations and showing again how we’re able to beat them.”
As always with such presentations, it’s important to note that these are Intel’s chosen benchmarks and that it represents one point in time. We expect AMD to announce the sixth generation of its EPYC processors in October, at which point it will no doubt release similar looking graphs in its favour.
Price of Intel Xeon 6700P and 6500P series
Price of Intel Xeon 6300P series
Price of Intel Xeon 6900P series
Related articles
About The Author
Tim Danton
NEXT UP
