Intel Xeon 6736P in Real Servers: Performance of the 36-Core Granite Rapids CPU

The Intel Xeon 6736P is a 36-core Granite Rapids server processor with 72 threads, a maximum clock speed of 4.1 GHz, and a 205 W TDP. It is used in both single- and dual-socket systems designed for virtualization, databases, application servers, and other enterprise workloads.

In terms of core count, the Xeon 6736P sits in the middle of the Xeon 6 lineup. It cannot match the highest-core-count models in peak multithreaded performance, but it offers higher clock speeds and requires fewer software licenses in applications priced per processor core.

Intel Xeon 6736P Benchmarks in Production Servers

The table below shows SPEC CPU 2017 results from commercially available servers. Every configuration uses two Xeon 6736P processors, providing a total of 72 cores and 144 threads.

SPECint_rate2017 measures system throughput in integer workloads, while SPECfp_rate2017 covers floating-point calculations. These are results for the complete dual-socket server, not for a single processor.

The difference between the fastest and slowest server is approximately 6%. Since the processors are identical, the variation comes from memory configuration, BIOS settings, power profiles, cooling, and the software environment.

The CPU model alone does not determine the exact performance of a complete server. A system with all memory channels active and a properly configured performance profile can noticeably outperform another server using the same processors but a less effective configuration.

Clock Speeds Under Load

The Xeon 6736P has a base clock of 2.0 GHz and a maximum Turbo Boost frequency of 4.1 GHz. Intel specifies an all-core frequency of up to 3.4 GHz, although the actual operating speed depends on temperature, power consumption, and the settings of the individual server platform.

The processor supports Intel Speed Select Technology. Server manufacturers and administrators can choose profiles that prioritize higher performance or lower power consumption. As a result, two servers equipped with the Xeon 6736P may sustain different clock speeds under the same workload.

This model does not compete with higher-end Xeon 6 processors on core count. Its main advantage is a set of 36 performance cores with boost speeds reaching 4.1 GHz. That matters in mixed workloads where some operations scale across many threads while others remain sensitive to clock speed and memory latency.

Dual-Socket Configuration and NUMA

A system with two Xeon 6736P processors provides 72 physical cores, 144 threads, and a combined processor TDP of 410 W. This configuration suits virtual machine consolidation, databases, rendering, analytics, and other workloads capable of using a large number of threads.

Adding a second processor does not guarantee twice the performance. Memory in a dual-socket server is divided between two NUMA nodes. When an application frequently accesses memory attached to the other processor, latency increases and scaling efficiency declines.

The hypervisor, operating system, and applications must distribute threads and data correctly between the sockets. Without proper NUMA configuration, part of the second processor’s potential is lost. For software with poor NUMA support, one more powerful CPU can sometimes be more efficient than two lower-performance processors.

Memory and Expansion Devices

Each Xeon 6736P supports eight channels of DDR5 ECC memory. A dual-socket server therefore offers sixteen memory channels, but maximum bandwidth is available only when the modules are distributed evenly between the processors and channels.

Total memory capacity alone says little about memory subsystem performance. A server with a large amount of RAM but too few memory modules can restrict the processors in analytics, simulation, rendering, and large-database workloads.

PCI Express 5.0 support allows servers to accommodate numerous NVMe drives, high-speed network adapters, and compute accelerators. The Xeon 6736P is therefore used not only in general-purpose servers but also in storage platforms and multi-GPU compute nodes.

Where the Intel Xeon 6736P Makes Sense

The processor is suitable for virtualization, enterprise databases, application servers, software-defined storage, and medium-density compute clusters.

One specific argument in favor of the Xeon 6736P is software licensed by processor core count. Moving to a 64- or 86-core CPU can sharply increase licensing costs, even when the additional cores are not fully utilized. In such environments, the 36-core model may provide a better balance between performance and total system cost.

Higher-core-count Xeon 6 processors are better suited to maximizing virtual machine density. Workloads that depend heavily on single-thread performance may benefit from models with fewer cores and higher base frequencies.

The Xeon 6736P makes the most sense when software licensing costs and per-core performance matter more than maximum compute density. A dual-socket configuration provides 72 cores, but it also requires a properly configured memory subsystem and NUMA topology. Without that preparation, part of the platform’s available performance will remain unused.