Intel Xeon 6747P vs Intel Xeon 6960P Comparison

Intel Xeon 6960P and Intel Xeon 6747P are both Granite Rapids server processors, but they target different segments of the market. The 6960P is the higher-core-count part, while the 6747P is a lower-core-count offering in the same family. The benchmark data shows a clear performance hierarchy between the two.

Head-to-Head Benchmarks

The Intel Xeon 6960P wins every recorded head-to-head benchmark against the Intel Xeon 6747P. The largest victory comes in PassMark's random string sorting test, where the 6960P scores 371795 versus 180382, a 106.1% advantage. This is the only test where the 6960P more than doubles the 6747P's output.

The second-largest gap is in PassMark physics, where the 6960P scores 24937 against 13398, a delta of 86.1%. This workload scales heavily with core count, and the 6960P's additional cores show clearly.

PassMark data encryption shows a 78.5% lead for the 6960P, with scores of 162013 versus 90789. Integer math follows at 55.3% ahead, scoring 727750 versus 468518. Data compression shows a 52.6% gap, with the 6960P at 2797724 and the 6747P at 1833378.

The floating point math test shows a 44.2% advantage, with scores of 527473 and 365904 respectively. Extended instructions come in at a 35.7% lead, scoring 193404 versus 142557. Find prime numbers shows a 28.9% gap, with scores of 1484 and 1151.

The Cinebench suite shows consistent 28.5% deltas across all three versions. Cinebench R15 multicore scores 11194 for the 6960P and 8712 for the 6747P. Cinebench R20 multicore scores 46645 versus 36301. Cinebench R23 multicore scores 111060 versus 86432. PassMark multithread also shows a 28.5% delta, with scores of 130659 and 101685.

The narrowest margin is in single-thread performance. The 6960P scores 3287 in PassMark single thread, while the 6747P scores 3236, a 1.6% lead. Both processors share the same 2.70 GHz base clock and 3.90 GHz boost clock, which explains why single-threaded performance is nearly identical.

Where Each One Wins

The Xeon 6960P wins in every measured category, so the choice depends on workload intensity and core-count scaling. For highly parallel workloads like data compression, encryption, physics simulations, and random string sorting, the 6960P offers dramatic improvements. Random string sorting more than doubles, physics nearly doubles, and encryption and integer math show gains above 50%.

For general multi-threaded rendering and compute tasks, the Cinebench suite shows a steady 28.5% advantage. This is a substantial gain for workloads that scale linearly with cores, such as video rendering, 3D animation, and scientific computing.

The 6747P's closest performance to the 6960P is in single-threaded tasks. With only a 1.6% difference in PassMark single thread, the 6747P is nearly equivalent for lightly threaded workloads, database queries that rely on single-core speed, or applications that cannot use many cores.

For memory-bandwidth-sensitive workloads, the 6960P benefits from twelve-channel memory with 614.4 GB/s bandwidth, versus the 6747P's eight-channel memory at 409.6 GB/s. The 6747P also has fewer PCIe lanes, 88 versus 96, and a smaller L3 cache at 288 MB versus 432 MB. These differences affect workloads that are memory-bound or require extensive I/O.

The 6747P is the more efficient choice for rack density and power-constrained environments. Its 330 W TDP is lower than the 6960P's 500 W TDP. For workloads that do not need 72 cores, the 6747P delivers most of the single-thread performance at lower power draw.

FAQ

Q: How much faster is the Xeon 6960P in Cinebench R23 multicore?

A: The 6960P scores 111060 versus the 6747P's 86432, a 28.5% advantage.

Q: Is there any benchmark where the Xeon 6747P wins?

A: No. The 6960P wins all 14 recorded head-to-head benchmarks. The closest result is PassMark single thread, where the 6960P leads by only 1.6%.

Q: What is the largest performance gap between the two processors?

A: PassMark random string sorting shows a 106.1% difference, with the 6960P scoring 371795 versus 180382 for the 6747P.

Q: Do both processors have the same clock speeds?

A: Yes, both have a 2.70 GHz base clock and a 3.90 GHz boost clock. This explains the near-identical single-thread scores.

Q: How do the core counts compare?

A: The 6960P has 72 cores and 144 threads, while the 6747P has 48 cores and 96 threads. This 50% core advantage drives the multi-threaded performance gaps.

Q: Which processor has more memory bandwidth?

A: The 6960P has twelve-channel memory with 614.4 GB/s bandwidth. The 6747P has eight-channel memory with 409.6 GB/s bandwidth.

Specification Differences

The core and thread counts differ significantly. The 6960P has 72 cores and 144 threads, while the 6747P has 48 cores and 96 threads. Both share the same 2.70 GHz base clock and 3.90 GHz boost clock.

TDP differs substantially. The 6960P is rated at 500 W, while the 6747P is rated at 330 W. The sockets differ as well: the 6960P uses Intel Socket 7529, while the 6747P uses Intel Socket 4710.

L3 cache differs. The 6960P has 432 MB shared, while the 6747P has 288 MB shared. Both have 112 KB L1 per core and 2 MB L2 per core.

Memory channels and bandwidth differ. The 6960P supports twelve-channel DDR5 with 614.4 GB/s bandwidth. The 6747P supports eight-channel DDR5 with 409.6 GB/s bandwidth.

PCIe lane counts differ. The 6960P has Gen 5 with 96 lanes (CPU only), while the 6747P has Gen 5 with 88 lanes (CPU only).

The die configuration differs. The 6960P uses a 3x 598 mm² layout, while the 6747P uses a 2x 598 mm² layout. Both use Intel's 5 nm process node.

The part numbers and launch dates differ. The 6960P is part number SRPKX, released on 2024-09-23. The 6747P is part number SRVEZ, released on 2025-02-23.

The launch MSRP for the 6960P is $9625. The launch MSRP for the 6747P is $6497.

Architecture Differences

Both processors share the Granite Rapids architecture and Intel 5 nm process node. The 6960P is part of the Granite Rapids-AP generation, while the 6747P is part of the Granite Rapids-SP generation. This AP versus SP distinction reflects the different physical configurations.

The 6960P uses three dies of 598 mm² each, while the 6747P uses two dies of 598 mm² each. The additional die in the 6960P provides the extra cores and L3 cache. The 6960P has 432 MB of shared L3 cache versus 288 MB in the 6747P.

Both support DDR5 memory and ECC memory. Both support PCIe Gen 5. Neither has integrated graphics. Both are active production parts with locked multipliers.

The memory subsystem differs significantly. The 6960P's twelve-channel memory controller delivers 614.4 GB/s, which is 50% more bandwidth than the 6747P's eight-channel controller at 409.6 GB/s. This affects memory-bound workloads beyond what core counts alone would predict.

The I/O capabilities differ. The 6960P provides 96 PCIe Gen 5 lanes, while the 6747P provides 88 lanes. This gives the 6960P more headroom for high-speed storage and networking expansion.

The Verdict

The data supports a clear split based on workload requirements. The Xeon 6960P is the choice for maximum multi-threaded throughput. It wins every benchmark, with particularly large leads in random string sorting, physics, encryption, and integer math. Its 72 cores, 432 MB L3 cache, and twelve-channel memory make it suited for large-scale virtualization, in-memory databases, scientific computing, and any workload that can scale across many cores.

The Xeon 6747P is the choice for single-threaded and moderately threaded workloads where power efficiency matters. Its single-thread score trails the 6960P by only 1.6%, making it nearly identical for lightly threaded applications. The 330 W TDP versus 500 W means it can fit in denser configurations with less cooling overhead. Its eight-channel memory and 88 PCIe lanes still provide robust I/O for most server applications.

The 6747P also positions well against its own rivals. The database shows it 9.4% ahead of the AMD EPYC 9455P and 13.5% ahead of the Intel Xeon w9-3595X in average benchmark score. It trails the AMD EPYC 9634 by 2.5% and the Intel Xeon 6980P by 5.3%.

The 6960P sits at the top of its competitive set. Its average benchmark score of 365194 puts it 0.2% ahead of the AMD EPYC 9754, 4.1% ahead of the AMD EPYC 9475F, and within 2.2% of the AMD EPYC 9655. It ranks in the 100th percentile among all CPUs, while the 6747P ranks in the 99th percentile.

For builders assembling a new server platform, the socket difference is critical. The 6960P requires Intel Socket 7529, while the 6747P requires Intel Socket 4710. These are not interchangeable platforms. The choice between them is not just a processor decision but a platform decision.

In summary, the 6960P offers roughly 28.5% more multi-threaded performance across the Cinebench suite and much larger gains in specific workloads, at the cost of higher power draw and a larger platform footprint. The 6747P offers nearly identical single-thread performance, lower power consumption, and a more modest platform requirement. Choose the 6960P when core count and memory bandwidth are the primary constraints. Choose the 6747P when power density and single-thread performance are the priorities.