Intel Xeon 6747P vs Intel Xeon 6980P Comparison

The Intel Xeon 6980P and Intel Xeon 6747P are both Granite Rapids server processors, but they target different ends of the scaling spectrum. The 6980P is the flagship with 128 cores, while the 6747P is a 48-core part. Benchmark data reveals a fascinating split: the 6747P dominates in Cinebench and several Passmark tests, while the 6980P takes the lead in data-heavy and math workloads. This analysis breaks down exactly where each chip wins and what that means for potential deployment.

Head-to-Head Benchmarks

The most striking result is the Cinebench sweep. The 6747P wins all three Cinebench multicore tests by an identical margin of 26.9%. In Cinebench R23 multicore, the 6747P scores 86432 against the 6980P's 63175. The same 26.9% delta appears in both Cinebench R20 (36301 vs 26533) and Cinebench R15 (8712 vs 6367). This consistency suggests a fundamental scaling advantage for the smaller chip in this render workload.

Passmark multithreaded performance follows the same pattern. The 6747P scores 101685, beating the 6980P's 74324 by 26.9% — again the exact same delta. Passmark physics is even more lopsided: 13398 for the 6747P versus 3350 for the 6980P, a 75% deficit for the flagship. Prime number finding also favors the 6747P, which scores 1151 compared to the 6980P's 555, a 51.8% advantage.

Single-thread performance is another clear win for the 6747P. In Passmark single-thread, it scores 3236 versus 1681 for the 6980P, a 48.1% lead. This is notable given both chips share the same 3.90 GHz boost clock. The 6747P's higher 2.70 GHz base clock versus 2.00 GHz on the 6980P likely explains much of this gap, but the magnitude is substantial.

However, the 6980P strikes back decisively in data-centric workloads. Passmark data compression shows the 6980P scoring 2364519 against 1833378, a 29% win. Data encryption is even better: 125246 versus 90789, a 38% advantage. Extended instructions deliver the largest margin for the 6980P, with a score of 214794 against 142557 — a 50.7% lead.

Math workloads also favor the 128-core chip. Floating point math scores 501720 for the 6980P versus 365904 for the 6747P, a 37.1% delta. Integer math shows a similar 36.1% advantage, with scores of 637476 and 468518 respectively. Random string sorting rounds out the 6980P's wins, with 240792 against 180382, a 33.5% margin.

The final tally is 8 wins for the 6747P and 6 for the 6980P across the 14 head-to-head benchmarks. But the nature of those wins matters. The 6747P's victories are concentrated in rendering, physics simulation, and single-thread latency — workloads that often scale poorly with core count. The 6980P's wins are in throughput-oriented tasks like encryption, compression, and math, where its massive core count can be fully utilized.

Architecture Differences

Both processors share the Granite Rapids architecture, manufactured on Intel's 5 nm process. The core design is identical, with each core featuring 112 KB of L1 cache and 2 MB of L2 cache. The fundamental difference lies in the chip design: the 6980P uses three 598 mm² dies, while the 6747P uses two. This die count difference explains the core count disparity — 128 cores and 256 threads versus 48 cores and 96 threads.

The L3 cache scales accordingly. The 6980P has 504 MB of shared L3 cache, while the 6747P has 288 MB. This 216 MB difference is substantial and likely contributes to the 6980P's strong showing in data compression and encryption, where large working sets benefit from cache capacity.

Memory architecture is another major divergence. The 6980P supports twelve-channel DDR5 memory with a theoretical bandwidth of 614.4 GB/s. The 6747P drops to eight-channel memory, providing 409.6 GB/s. That's a 204.8 GB/s bandwidth advantage for the 6980P, which directly impacts memory-throughput-sensitive workloads. Both support ECC memory, which is standard for server platforms.

PCIe lanes also differ. The 6980P provides 96 Gen 5 lanes, while the 6747P offers 88. This is a modest difference of 8 lanes, but it can matter for systems with many GPUs or NVMe drives. Neither chip has integrated graphics, and both are locked multipliers.

The socket situation is critical for platform planning. The 6980P uses Intel Socket 7529, while the 6747P fits Intel Socket 4710. These are not interchangeable, meaning the choice of processor dictates the motherboard and overall system design.

Power envelopes diverge significantly. The 6980P has a 500 W TDP, while the 6747P draws 330 W. This 170 W difference has major implications for cooling, power delivery, and operating costs. The 6747P's lower TDP also correlates with its higher base clock — the smaller chip can run all cores at 2.70 GHz, while the 6980P has to drop to 2.00 GHz to stay within its power budget.

FAQ

Q: Why does the 48-core Xeon 6747P beat the 128-core Xeon 6980P in Cinebench R23 multicore?

A: The 6747P scores 86432 versus 63175 for the 6980P, a 26.9% advantage. This is likely due to the 6747P's higher 2.70 GHz base clock and lower 330 W TDP, which allows sustained all-core boost behavior. The 6980P's 2.00 GHz base clock at 500 W may cause thermal or power throttling in this workload.

Q: Which processor has better single-thread performance?

A: The 6747P wins decisively. In Passmark single-thread, it scores 3236 against 1681 for the 6980P, a 48.1% margin. Both chips boost to 3.90 GHz, but the 6747P's higher base clock and lower core count give it a latency advantage.

Q: How do the two chips compare in memory bandwidth?

A: The 6980P has a clear advantage with twelve-channel memory and 614.4 GB/s bandwidth, versus eight-channel and 409.6 GB/s for the 6747P. This explains why the 6980P wins data compression (29% faster) and encryption (38% faster) workloads.

Q: Are these processors socket-compatible?

A: No. The 6980P uses Intel Socket 7529, while the 6747P uses Intel Socket 4710. They require completely different motherboards and platform designs.

Q: What is the cache difference between the two?

A: Both share the same per-core L1 (112 KB) and L2 (2 MB) caches. The L3 cache differs significantly: the 6980P has 504 MB shared, while the 6747P has 288 MB shared.

Q: Which chip has more PCIe lanes?

A: The 6980P provides 96 Gen 5 lanes, while the 6747P offers 88. This gives the 6980P room for 8 additional lanes of Gen 5 connectivity.

The Verdict

The data presents a nuanced picture. The Xeon 6747P is the better choice for workloads that prioritize single-thread speed, physics simulation, and rendering. Its Cinebench dominance (26.9% ahead in all three versions) and 75% lead in Passmark physics make it compelling for engineering workstations and interactive simulation. The 48.1% single-thread advantage also matters for latency-sensitive database queries and legacy software that scales poorly.

The Xeon 6980P is the pick for throughput-critical, memory-hungry applications. Its 50.7% lead in extended instructions, 38% advantage in encryption, and 37.1% edge in floating-point math show where its 128 cores and 504 MB L3 cache shine. The 614.4 GB/s memory bandwidth versus 409.6 GB/s gives it a fundamental advantage in large dataset processing.

The 6747P wins more head-to-head benchmarks (8 versus 6), but the 6980P wins the workloads that typically matter most in high-end server environments. The 6980P also sits in the 99th percentile of all CPUs, as does the 6747P. The 6980P's average benchmark score of 251516 places it 5.6% above the 6747P's 238263, and it runs 3% ahead of the AMD EPYC 9634. The 6747P trails the EPYC 9634 by 2.5% but beats the AMD EPYC 9455P by 9.4% and the Intel Xeon w9-3595X by 13.5%.

Specification Differences

| Specification | Intel Xeon 6980P | Intel Xeon 6747P |

|---|---|---|

| Cores | 128 | 48 |

| Threads | 256 | 96 |

| Base Clock | 2.00 GHz | 2.70 GHz |

| TDP | 500 W | 330 W |

| Socket | Intel Socket 7529 | Intel Socket 4710 |

| Generation | Xeon 6 (Granite Rapids-AP) | Xeon 6 (Granite Rapids-SP) |

| Die Size | 3x 598 mm² | 2x 598 mm² |

| L3 Cache | 504 MB (shared) | 288 MB (shared) |

| Memory Bus | Twelve-channel | Eight-channel |

| Memory Bandwidth | 614.4 GB/s | 409.6 GB/s |

| PCIe | Gen 5, 96 Lanes (CPU only) | Gen 5, 88 Lanes (CPU only) |

| Launch MSRP | $12460 | $6497 |