Intel Xeon 6745P vs Intel Xeon 674X Comparison

The Intel Xeon 674X and Intel Xeon 6745P are both Granite Rapids server/workstation processors on the Intel Socket 4710 platform, but they are tuned for different priorities. The 674X is the high-frequency specialist, with a 4.90 GHz boost clock and a 14% lead in single-threaded PassMark performance (3933 vs 3450). The 6745P is the core-count and cache champion, offering 32 cores, 64 threads, and a massive 336 MB of L3 cache, which translates into consistent multi-threaded wins across most PassMark workloads. Benchmark results indicate the 6745P is the better all-rounder for heavily threaded server tasks, while the 674X is the choice for workloads that demand raw single-core speed, such as legacy software or lightly threaded simulations. The 674X counters with a 23.5% lead in PassMark physics (7586 vs 6144) and a 1.8% edge in prime number finding (693 vs 681), showing it is not purely a single-core part. Ultimately, the data suggests the 6745P is the safer pick for general server duty, while the 674X is a specialized tool for frequency-sensitive applications.

The Verdict

The Intel Xeon 6745P is the default choice for most buyers. It wins 11 of the 16 head-to-head benchmarks, and its average benchmark score of 154858 is 8.2% higher than the 674X's 143103. The 6745P's advantage is most pronounced in data-heavy tasks: it is 8.6% faster in data compression (1352801 vs 1236272), 8.2% faster in data encryption (66665 vs 61195), and 8.8% faster in floating-point math (267438 vs 243877). These are exactly the sorts of workloads that dominate database, virtualization, and content-rendering servers. With 32 cores and 64 threads, it simply has more parallel resources to throw at multi-threaded jobs.

The Intel Xeon 674X is the pick when single-threaded performance is the bottleneck. Its 4.90 GHz boost clock powers a 14% advantage in PassMark single-thread (3933 vs 3450) and a 23.5% lead in PassMark physics (7586 vs 6144). This makes it the better option for applications that cannot scale across many cores, such as certain EDA tools, legacy database engines, or real-time control systems. The 674X also has a 1.8% edge in prime number finding (693 vs 681), indicating a slight strength in integer-heavy single-threaded loops. However, the 674X's lower core count (28 vs 32) and smaller L3 cache (144 MB vs 336 MB) mean it loses ground in nearly every multi-threaded PassMark test, despite its higher clock speed.

The data is clear: if your workload is parallel, buy the 6745P. If it is serial, buy the 674X. The 6745P is also the safer investment for future-proofing, since its larger cache and more cores will age better as software becomes more parallel. The 674X is a niche part for a specific frequency-critical use case.

FAQ

Q: Which CPU has the higher boost clock?

A: The Intel Xeon 674X boosts to 4.90 GHz, while the 6745P boosts to 4.30 GHz. This explains the 674X's 14% lead in PassMark single-thread performance (3933 vs 3450).

Q: Which CPU has more cores and threads?

A: The Intel Xeon 6745P has 32 cores and 64 threads, while the 674X has 28 cores and 56 threads. The 6745P's 4 extra cores contribute to its wins in multi-threaded benchmarks.

Q: How much larger is the L3 cache on the 6745P?

A: The 6745P has 336 MB of shared L3 cache, which is 192 MB more than the 674X's 144 MB. This large cache difference is a major factor in the 6745P's 8.6% lead in data compression (1352801 vs 1236272).

Q: Is the 674X ever faster than the 6745P?

A: Yes. The 674X wins PassMark single-thread (3933 vs 3450), PassMark physics (7586 vs 6144), and PassMark find prime numbers (693 vs 681). It also edges out the 6745P in Cinebench R15 single-core (1018 vs 1018) by a negligible margin.

Q: Are both CPUs on the same socket and process node?

A: Yes, both use Intel Socket 4710 and are built on Intel's 5 nm process with a 2x 598 mm² die size. They share the same eight-channel DDR5 memory bus and 409.6 GB/s memory bandwidth.

Q: What is the launch MSRP difference?

A: The 674X has a launch MSRP of $2199, while the 6745P has a launch MSRP of $5250. The 6745P is significantly more expensive, but its higher average benchmark score (154858 vs 143103) reflects its superior multi-threaded performance.

Architecture Differences

The two CPUs share the same Granite Rapids architecture, codename, and 5 nm process node from Intel, but they are binned and configured differently. The 674X is part of the Xeon 600 (Granite Rapids-WS) generation, while the 6745P belongs to the Xeon 6 (Granite Rapids-SP) generation. This generation split explains their divergent design goals.

The most significant architectural difference is the L3 cache. The 674X has 144 MB of shared L3, while the 6745P has 336 MB — a 192 MB difference. This is not a minor tweak; it is a fundamental difference in how much data each chip can hold on-die for rapid access. The 6745P's larger cache is the primary reason it excels in data compression and encryption, where repeatedly accessing the same datasets benefits from a bigger cache footprint.

Core counts also differ. The 674X has 28 cores and 56 threads, while the 6745P has 32 cores and 64 threads. Both use the same 112 KB L1 and 2 MB L2 per core, so the per-core cache structure is identical. The 4-core deficit on the 674X is partially offset by its higher boost clock (4.90 GHz vs 4.30 GHz), which helps in latency-sensitive single-threaded tasks.

PCIe lane allocation differs as well. The 674X provides Gen 5 with 128 lanes (CPU only), while the 6745P provides Gen 5 with 88 lanes. This makes the 674X better suited for configurations with many high-speed storage or accelerator cards. The 674X also has an unlocked multiplier, while the 6745P is locked, which is relevant for overclocking, although server platforms rarely leverage this.

Specification Differences

| Specification | Intel Xeon 674X | Intel Xeon 6745P |

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

| Cores | 28 | 32 |

| Threads | 56 | 64 |

| Base Clock | 3.00 GHz | 3.10 GHz |

| Boost Clock | 4.90 GHz | 4.30 GHz |

| TDP | 270 W | 300 W |

| L3 Cache | 144 MB (shared) | 336 MB (shared) |

| PCIe | Gen 5, 128 Lanes (CPU only) | Gen 5, 88 Lanes (CPU only) |

| Generation | Xeon 600 (Granite Rapids-WS) | Xeon 6 (Granite Rapids-SP) |

| Launch MSRP | $2199 | $5250 |

| Multiplier | Unlocked | Locked |

| Part Number | SA2CZ | SRWPAQ7L9 |

Both CPUs share identical specs for L1 cache (112 KB per core), L2 cache (2 MB per core), memory support (DDR5), memory bus (eight-channel), memory bandwidth (409.6 GB/s), ECC memory support, integrated graphics (N/A), market segment (Server/Workstation), production status (Active), process node (5 nm), foundry (Intel), and die size (2x 598 mm²). The 674X was released on 2026-02-01, while the 6745P was released on 2025-02-23.

Head-to-Head Benchmarks

The benchmark data reveals a clear split. The 6745P dominates multi-threaded PassMark workloads, while the 674X wins every single-threaded or latency-sensitive test.

In Cinebench, the two are effectively tied. The 6745P wins R15 multi-core (7214 vs 7213), R20 multi-core (30062 vs 30057), R23 multi-core (71578 vs 71566), and R20 single-core (4244 vs 4243) by a single point each. The 674X wins R15 single-core (1018 vs 1018) by zero points. These are rounding errors, not performance differences.

The real divergence starts with PassMark. The 6745P is 8.6% faster in data compression (1352801 vs 1236272), 8.2% faster in data encryption (66665 vs 61195), 10.1% faster in extended instructions (108326 vs 97373), 8.8% faster in floating-point math (267438 vs 243877), 8.3% faster in integer math (336926 vs 308968), and 4.5% faster in random string sorting (133528 vs 127529). The 6745P also edges out the 674X in PassMark multithread (84210 vs 84196) by a negligible 0% delta.

The 674X's counterattacks are decisive in its own domains. It wins PassMark single-thread with a 14% margin (3933 vs 3450), PassMark physics with a 23.5% margin (7586 vs 6144), and PassMark find prime numbers with a 1.8% margin (693 vs 681). The physics result is particularly striking — the 674X is nearly a quarter faster, despite having 4 fewer cores. This indicates its higher boost clock is far more impactful for physics simulations than the 6745P's extra cores.

Where Each One Wins

The Intel Xeon 6745P is the winner for database workloads, data compression pipelines, encryption tasks, and any multi-threaded integer or floating-point math. Its 8.6% lead in data compression and 8.2% lead in data encryption are directly relevant to storage servers and secure communications. The 10.1% lead in extended instructions (108326 vs 97373) suggests better support for AVX-512 or similar instruction sets, which is valuable for scientific computing and media encoding. With 336 MB of L3 cache, it can hold larger datasets in cache, reducing memory latency in repeated-access patterns. This is the chip for general-purpose virtualization, big-data analytics, and content rendering.

The Intel Xeon 674X is the winner for single-threaded applications, physics simulations, and prime-number-finding workloads. Its 14% lead in PassMark single-thread (3933 vs 3450) is critical for legacy software that cannot use more than one or two cores. The 23.5% lead in PassMark physics (7586 vs 6144) makes it the better choice for real-time physics engines in simulation or gaming servers. The 1.8% lead in find prime numbers (693 vs 681) indicates a slight edge in certain cryptographic or mathematical loops. Additionally, its 128 PCIe Gen 5 lanes (vs 88 on the 6745P) make it the better option for systems with many NVMe drives or multiple high-bandwidth accelerators. The 674X is a niche product for frequency-critical, I/O-heavy deployments where the 6745P's extra cores would go unused.