Intel puts more than 46,000 Xeon 6+ cores in one 133.45 kW rack
Intel says select Xeon 6+ rack configurations scale beyond 46,000 cores at 133.45 kW total rack power. AMD's rival Helion CPU-only rack tops out at 32,768 cores, and AMD did not publish a power rating for it.
Written for builders.
Intel says select Xeon 6+ rack configurations scale beyond 46,000 cores. AMD's CPU-only Helion rack supports up to 32,768 CPU cores. Intel's rack figure works out to no more than about 2.9 W per core.
| Figure | Value | Note | Claim |
|---|---|---|---|
| Intel Xeon 6+ cores per rack, beyond | 46,000 cores | 1 | |
| AMD CPU-only Helion rack, up to | 32,768 CPU cores | 6th Gen AMD EPYC processors | 6 |
| Power per Intel core, about | 2.9 W per core | no more than, from Intel's rack figure | 12 |
What happened
- Intel also detailed its GPU D-series: up to 480 GB of LPDDR5X, 1.5 TB/s of memory bandwidth and 1.3 petaflops of FP8 compute in a 350-watt envelope.
- AMD's second Helion design is an all-in-one rack that puts EPYC CPU nodes, Instinct MI350P GPU nodes and general-purpose compute on one platform.
- AMD says Helion configurations are still under development with ecosystem partners and are planned to be available in November 2026.
Why it matters
- Who has to decide decision Operators now choose a CPU-to-GPU split per rack. AMD offers a CPU-only design and an integrated one, so that choice comes before any shared benchmark exists.
- Who's exposed exposure A buyer mixing vendors has no stated common software layer. Intel's portability claim covers Intel GPU platforms, and AMD says only that its blueprint supports open software.
- What it sets up precedent Intel's racks lean on ORv3 and AMD plans to contribute its Helion spec to OCP. We'd expect rack-level specs to be where the two vendors first become comparable on paper.
Divide the rack's power by its cores and Intel's figure works out to no more than 2.9 W per core [12]. It is a ceiling on watts because Intel wrote "beyond" 46,000 cores, and the figure covers only select Xeon 6+ configurations [1]. We would use total rack power as the denominator because that is what the facility has to feed.
Against AMD's CPU-only Helion rack, Intel's core count is at least 1.4 times higher [13][6]. For that ratio to mean 1.4 times the work, a Xeon 6+ core and a 6th Gen EPYC core would have to finish the same agent task in the same time at the same power. Intel itself says the yardstick is moving: performance is increasingly measured by the number of concurrent agents a rack can sustain at target latency [2].
The vendors agree on the shape of the problem. Dan McNamara, AMD's senior vice president and general manager of Compute and Enterprise AI, said: "Agentic AI is not a single workload." [10] AMD's CPU-only rack hands model inference to separate infrastructure [6]. Intel describes Xeon as the part that orchestrates workloads and manages data movement so accelerators stay busy [11].
Neither release reports a concurrent-agent count or any other software measurement.
What to watch
- Whether partners shipping Helion racks in November publish a power rating for the 32,768-core CPU-only configuration.
- Whether Intel or AMD reports concurrent agents at target latency, the metric Intel says now defines performance.
- What software interface, if any, the Helion specification names when AMD submits it to OCP.
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
At the Open Compute Project Global Summit, Intel said select Xeon 6+ configurations scale beyond 46,000 cores per rack at 133.45 kW total rack power.
- [2]
Intel's release says AI infrastructure performance is increasingly measured by the number of concurrent agents a rack can sustain at target latency.
- [3]
Intel GPU D-series (code-named Crescent Island), built on the Xe3P architecture, combines up to 480 GB of LPDDR5X memory, 1.5 TB/s of memory bandwidth and 1.3 petaflops of FP8 compute within a 350-watt power envelope.
- [4]
Intel GPU D-series has a standard PCIe form factor, an air-cooled design and a common Xe software foundation that supports application portability across Intel GPU platforms.
- [5]
Many of Intel's new Xeon-based rack designs leverage the Open Compute Project's Open Rack v3 (ORv3) standard.
- [6]
AMD Helion includes a CPU-only rack powered by 6th Gen AMD EPYC processors that supports up to 32,768 CPU cores, designed for hyperscale customers with dedicated capacity for agent execution and connections to separate model inference infrastructure.
- [7]
AMD Helion also includes an all-in-one rack that combines AMD EPYC CPU nodes, AMD Instinct MI350P GPU nodes and general-purpose compute on one platform, aimed at enterprise agentic AI.
- [8]
AMD Helion agentic rack configurations are under development with ecosystem partners and are planned to be available in November 2026.
- [9]
AMD plans to contribute the AMD Helion specification to OCP, and says the rack blueprint supports open software.
- [10]
Dan McNamara, senior vice president and general manager, Compute and Enterprise AI, AMD, said: "Agentic AI is not a single workload."
- [11]
Intel says that, working alongside GPUs and specialized accelerators, Xeon orchestrates workloads, executes tasks and manages data movement to keep the entire system productive.
- [12]
Intel's rack figure works out to no more than about 2.9 W per core.
- [13]
Intel's rack core count is at least 1.4 times AMD's CPU-only Helion maximum.
Sources & coverage · 2 publishers
The reporting this story was synthesized from, earliest first. Every link goes to the original.