Intel Xeon 6747P

The Intel Xeon 6747P is a 48-core, 96-thread server processor from the Granite Rapids architecture, part of the Xeon 6 (Granite Rapids-SP) generation. It carries a 330W TDP, an eight-channel DDR5 memory interface rated at 409.6 GB/s, and 88 PCIe Gen 5 lanes, making it a compute- and bandwidth-focused part for the server/workstation segment. The database records no benchmark scores for this processor and places it at the 50th percentile among all CPUs, so its evaluation rests on architectural specifications rather than measured results. Its launch MSRP is $6497.

Who Should Consider It

The 6747P is built for workloads that scale across many cores and demand large memory bandwidth. Its 48 cores and 96 threads make it a strong fit for heavy multi-threaded compute: database workloads, large-scale virtualization, data analytics, and HPC-style simulations all benefit from the combination of core count and the 288 MB of shared L3 cache. The eight-channel DDR5 interface with 409.6 GB/s of bandwidth means memory-bandwidth-bound applications, such as in-memory databases and scientific computing, are a natural fit, because many cores can be kept fed without stalling on memory access.

Rendering and media creation workloads that parallelize across threads will also find the 48-core design compelling. The large shared cache reduces the cost of repeated data access across threads, which helps in ray tracing, video encoding, and simulation workloads where many threads touch overlapping data sets. For office productivity and light desktop use, however, the 6747P is dramatically overprovisioned; its server/workstation segment designation and 330W TDP make it unsuitable for typical desktop or small-office deployments, and the platform cost and cooling requirements would be wasted on such tasks.

The processor is not aimed at gaming. Its architecture, core count, and platform requirements are oriented toward sustained multi-threaded throughput, not low-latency single-thread responsiveness. Buyers who need ECC memory reliability, massive core counts, and high memory bandwidth, and who operate in a server or workstation chassis, are the intended audience. The ECC memory support is a meaningful feature for workloads where data integrity is non-negotiable, such as financial processing or long-running scientific computations.

Power and Thermals

The 6747P carries a 330W TDP, placing it in the high-power tier of server processors. This is a part that requires a substantial server-grade cooling solution; the data does not specify a cooler type, but the power envelope implies a chassis designed for high airflow or liquid cooling. The processor's die is large, two dies at 598 mm² each, which spreads heat across a wide area but still demands robust thermal management to maintain the 3.90 GHz boost behavior.

The multiplier is locked, so there is no overclocking headroom; the 330W TDP is the operating envelope users must plan around. The 5nm process node, fabricated by Intel, helps contain power relative to the core count, but 48 cores at up to 3.90 GHz still produce significant heat. System builders should plan for power delivery and cooling infrastructure commensurate with a 330W-class part, including appropriate chassis fans, heatsinks, and airflow paths. The active production status indicates this is a current part, so thermal solutions are likely available from server vendors, but the data does not enumerate them. A dense deployment of multiple such processors in one chassis would require careful attention to overall system thermal capacity.

Benchmark Performance

The database lists no benchmark scores for the 6747P, the average benchmark score is recorded as 0, and its percentile among all CPUs is 50. This means the processor sits at the midpoint of the database's CPU population, though the absence of measured scores means this percentile likely reflects its recorded standing rather than a performance measurement. Without benchmark results, performance analysis must rely on the architectural specification, which is substantial.

The 48 cores at a 3.90 GHz boost clock, paired with 288 MB of shared L3 cache, indicate a processor designed for high-throughput multi-threaded execution. The eight-channel DDR5 interface at 409.6 GB/s of memory bandwidth removes a common bottleneck for large datasets; many server workloads are memory-bound rather than compute-bound, and the bandwidth figure is among the highest in the server class. In workloads that scale across cores, database queries, virtualization hosts, batch processing, the combination of core count, cache, and bandwidth is the primary performance driver.

There are no rival scores or delta percentages in the data, so direct percentage comparisons are not possible. The 50th percentile ranking suggests that, across the entire database of CPUs, the 6747P sits in the middle, but the database population includes consumer and desktop parts, so a server-focused 48-core processor being at the midpoint reflects the breadth of that population, not a weakness in its intended segment. The absence of recorded benchmarks means the 0 average score should not be read as a performance result; it is a data gap. Benchmark results, when published, would provide concrete deltas; the data currently does not include them.

How It Compares

The FACT PACK records no nearest rivals for the 6747P, so a direct rival-by-rival comparison is not possible from the available data. Its position must be inferred from its own specifications and segment. As a Granite Rapids-SP part in the Xeon 6 generation, it sits within Intel's server lineup, distinguished by its 48-core count, 330W TDP, and eight-channel DDR5 support.

Within the server/workstation segment, the 6747P's 48 cores and 96 threads place it in the high-core-count tier, while its 288 MB of shared L3 cache is a substantial on-die resource that reduces reliance on main memory for many workloads. The 88 PCIe Gen 5 lanes provide extensive I/O connectivity, which is typical of server processors designed for expansion-heavy deployments with multiple accelerators or high-speed network interfaces. Its 50th percentile standing across all CPUs in the database indicates a mid-pack position in the broader population, but the database's mix of consumer and server parts means this is not a segment-specific comparison.

The processor's active production status and 2025-02-23 release date indicate it is a current-generation product. Without rival data, the analysis is limited to its architectural positioning: a high-core, high-bandwidth server part aimed at workloads that can exploit parallel execution and large memory throughput. In the absence of named competitors, the 6747P's value proposition rests on its own specification sheet, 48 cores, 288 MB of L3, and 409.6 GB/s of memory bandwidth, rather than on head-to-head deltas.

FAQ

Q: What socket does the Intel Xeon 6747P use?

A: It uses Intel Socket 4710.

Q: How much cache does the 6747P have?

A: It has 112 KB of L1 cache per core, 2 MB of L2 cache per core, and 288 MB of shared L3 cache.

Q: What memory does it support?

A: It supports DDR5 memory over an eight-channel interface with 409.6 GB/s of bandwidth, and it supports ECC memory.

Q: How many PCIe lanes does it provide?

A: It provides 88 PCIe Gen 5 lanes from the CPU only.

Q: What is the TDP of the 6747P?

A: The TDP is 330W.

Q: When was it released, and is it still in production?

A: The release date is 2025-02-23, and production status is active.

Q: Is the multiplier unlocked?

A: No, the multiplier is locked.

Platform and Compatibility

The 6747P is built for Intel Socket 4710, which places it in the Granite Rapids-SP platform within the Xeon 6 generation. The architecture is Granite Rapids, fabricated on a 5nm process at Intel, with a die composed of two 598 mm² pieces. The platform supports DDR5 memory across eight channels, with ECC support and a total memory bandwidth of 409.6 GB/s. This makes the platform suitable for memory-reliability-critical workloads such as databases and financial applications, where a single bit error can be costly.

For I/O, the processor provides 88 PCIe Gen 5 lanes from the CPU, enabling substantial expansion: high-speed networking, storage controllers, and accelerators can all be attached directly without a separate chipset bottleneck. The server/workstation market segment designation means the platform assumes a chassis and chipset designed for enterprise use, with the cooling and power delivery to match the 330W TDP. The part number is SRVEZ, and the production status is active, indicating ongoing availability. The locked multiplier means no overclocking; platform upgrades would occur within the Granite Rapids-SP generation, and the data does not specify cross-generation compatibility beyond the Socket 4710 interface.

Single-Thread vs Multi-Thread Behavior

The 6747P's base clock is 2.70 GHz, with a boost clock of 3.90 GHz. The boost clock represents the maximum single-thread ceiling, while the 48 cores and 96 threads define the multi-threaded capacity. The gap between base and boost, from 2.70 to 3.90 GHz, shows that the processor can raise frequency when fewer cores are active, but the design is clearly oriented toward sustained multi-core operation rather than peak single-core speed.

The cache hierarchy reinforces this split: 112 KB of L1 and 2 MB of L2 per core provide fast per-core access, while the 288 MB of shared L3 cache enables data sharing across all 48 cores. For single-thread-bound workloads, the 3.90 GHz boost clock is the relevant figure, but the processor's core count and shared cache are the defining traits. Workloads that cannot use many threads will leave most of the processor idle, and the high TDP will be spent without corresponding throughput. Conversely, workloads that scale, rendering, virtualization, large-scale analytics, will exploit the full 48-core design and the large shared cache to reduce memory latency and inter-thread communication overhead. The eight-channel DDR5 interface further supports multi-threaded scenarios where many cores simultaneously demand memory access. In short, the 6747P is a multi-thread specialist; its single-thread capability is adequate for server tasks, but its value lies in parallel throughput, and buyers should match workloads to that strength.