Intel Xeon 6730P vs Intel Xeon 674X Comparison

Where Each One Wins

The benchmark data splits these two Granite Rapids Xeons into very different use cases. The Intel Xeon 674X wins 15 of 16 head-to-head tests, while the Intel Xeon 6730P takes only a single victory. That one win, however, is revealing. The 6730P leads in PassMark physics by 11.9% (8606 versus 7586), which suggests its higher core count (32 versus 28) gives it an edge in workloads that scale purely with core count and memory-level parallelism rather than clock speed or per-core efficiency.

The 674X, by contrast, dominates everything else. Its single-thread advantage is massive: 31.3% ahead in PassMark single-thread (3933 versus 2995) and 13.6% ahead in every Cinebench single-core test. This is a processor built for latency-sensitive, lightly threaded tasks where the 4.90 GHz boost clock versus 3.80 GHz makes a decisive difference. The 674X also leads in data compression (8.6%), encryption (9.3%), floating-point math (7.5%), integer math (6.3%), and random string sorting (11.9%). Those are the workloads that benefit from higher clock speeds and the larger per-core L2 cache (2 MB per core on both, but the 674X has fewer cores to share the L3 pool, which we will examine later).

For server administrators, the split is practical: if the workload is physics simulation or any parallel compute that stresses the scheduler and memory controller evenly across many cores, the 6730P's 32 cores and 64 threads provide a measurable advantage. If the workload is database transactions, single-threaded application logic, or mixed integer/floating-point HPC, the 674X's clock advantage wins consistently. The 674X also holds the average benchmark score edge: 143103 versus 124756, a difference of roughly 14.7% in its favor. Its percentile placement is 98 versus 97 for the 6730P, both elite CPUs globally, but the 674X sits one tier higher.

FAQ

Q: Which CPU has the higher single-thread performance?

A: The Intel Xeon 674X. It scores 3933 in PassMark single-thread versus 2995 for the 6730P, a 31.3% lead. In Cinebench R23 single-core, the 674X scores 8893 (the 6730P's R23 single-core score is also 8893, but the head-to-head table lists the 674X at 71566 for multicore; the single-core R23 result for the 6730P is not in the head-to-head, so we rely on the PassMark delta). The R15 and R20 single-core tests show 13.6% advantages for the 674X (1018 versus 896, and 4243 versus 3735, respectively).

Q: Does the 6730P ever beat the 674X?

A: Yes, in exactly one recorded test: PassMark physics. The 6730P scores 8606 versus 7586 for the 674X, an 11.9% margin. This is the only head-to-head win for the 6730P out of 16 recorded benchmarks.

Q: How do their average benchmark scores compare?

A: The 674X averages 143103 across all recorded benchmarks, while the 6730P averages 124756. That is a 14.7% gap in favor of the 674X. The nearest rival to the 674X is the Intel Xeon 6732P at 143444 (0.2% higher), while the 6730P's closest rival is the AMD Ryzen Threadripper PRO 5975WX at 124171 (0.5% lower).

Q: What memory bandwidth do both support?

A: Both support DDR5 with an eight-channel memory bus and a recorded memory bandwidth of 409.6 GB/s. There is no difference in the memory subsystem between the two.

Q: Which has more L3 cache?

A: The 6730P has 288 MB of shared L3 cache, while the 674X has 144 MB. Both have 112 KB L1 per core and 2 MB L2 per core. The 6730P's larger L3 pool is notable given its higher core count.

Q: Are both CPUs currently in production?

A: Yes, both are listed as Active production status. The 674X released in February 2026, and the 6730P released in February 2025.

Head-to-Head Benchmarks

The most striking result is the consistent 13.6% margin across all five Cinebench tests. The 674X leads in R15 multicore (7213 versus 6349), R15 single-core (1018 versus 896), R20 multicore (30057 versus 26458), R20 single-core (4243 versus 3735), and R23 multicore (71566 versus 62996). That uniformity suggests a clock-for-clock efficiency advantage, not just a boost clock difference. The 674X runs a 3.00 GHz base and 4.90 GHz boost, while the 6730P runs 2.50 GHz base and 3.80 GHz boost. The delta in boost clocks is 1.10 GHz, which translates into the 13.6% Cinebench edge.

PassMark results show a wider spread. The single-thread test shows the largest gap: 31.3% (3933 versus 2995). This is the headline number for anyone running legacy or poorly threaded applications. The random string sorting test shows an 11.9% lead (127529 versus 113919), which is a classic memory-latency-sensitive workload; the 674X's smaller L3 (144 MB) but higher clock appears to be the deciding factor. Data encryption shows a 9.3% lead (61195 versus 55964), data compression 8.6% (1236272 versus 1138470), floating-point math 7.5% (243877 versus 226838), and integer math 6.3% (308968 versus 290740).

The extended instructions test is close: 1.2% in favor of the 674X (97373 versus 96204). Find prime numbers is even closer: 1.0% (693 versus 686). These small margins indicate that for AVX-512 or prime-sieving workloads, the two CPUs are nearly interchangeable, and the 6730P's extra cores almost offset the 674X's clock advantage.

The physics test is the outlier. The 6730P wins 8606 versus 7586, an 11.9% margin. Physics workloads in PassMark often respond to core count and memory bandwidth equally, and the 6730P's 32 cores versus 28 cores, combined with its larger 288 MB L3, explains the result. The multithread test, however, goes the other way: 13.6% to the 674X (84196 versus 74113). That is surprising given the 6730P has 4 more cores and 8 more threads. The data suggests the 674X's higher per-core throughput outweighs the core-count deficit in generic multithreaded workloads.

Specification Differences

The most obvious difference is core count: 28 cores and 56 threads for the 674X versus 32 cores and 64 threads for the 6730P. Clock speeds differ substantially: the 674X runs 3.00 GHz base and 4.90 GHz boost, while the 6730P runs 2.50 GHz base and 3.80 GHz boost. TDP is 270 watts for the 674X and 250 watts for the 6730P.

L3 cache is the other major split: 144 MB shared for the 674X versus 288 MB shared for the 6730P. L1 and L2 remain identical at 112 KB per core and 2 MB per core, respectively. PCIe lane counts differ: the 674X supports Gen 5 with 128 lanes (CPU only), while the 6730P supports Gen 5 with 88 lanes (CPU only). Both use the same Intel Socket 4710.

The launch MSRP is $2199 for the 674X and $3726 for the 6730P. The 674X has an unlocked multiplier, while the 6730P is locked. The part numbers are SA2CZ for the 674X and SRV5R for the 6730P. The 6730P belongs to the Xeon 6 (Granite Rapids-SP) generation, while the 674X belongs to the Xeon 600 (Granite Rapids-WS) generation. Both are server/workstation parts with ECC memory support and no integrated graphics.

Architecture Differences

Both CPUs share the Granite Rapids architecture, the 5 nm process node, and the same 2x 598 mm² die size. The foundry is Intel for both. The difference lies in the product segmentation: the 674X is a workstation-class part (Granite Rapids-WS, Xeon 600 generation), while the 6730P is a scalable processor (Granite Rapids-SP, Xeon 6 generation). This explains the unlocked multiplier on the 674X and the larger PCIe allocation (128 lanes versus 88), both workstation-oriented features.

The cache hierarchy differs in L3 only. The 6730P's 288 MB L3 is double the 674X's 144 MB. With 32 cores, the 6730P has 9 MB of L3 per core, while the 674X has approximately 5.1 MB per core. For datasets that fit within the larger L3, the 6730P should reduce DRAM traffic, but the benchmark data shows the 674X still wins most memory-sensitive tests, likely due to its clock advantage. The 674X also has a higher TDP (270 versus 250), which aligns with its higher boost clock.

Both support DDR5 with eight-channel memory and identical 409.6 GB/s bandwidth. PCIe generation is Gen 5 for both, but lane count differs as noted. The release dates are roughly a year apart: February 2025 for the 6730P, February 2026 for the 674X. The 674X is the newer design, and its benchmark profile reflects a focus on single-thread and mixed workload performance rather than raw core scaling.

The Verdict

For single-thread-bound server applications, the Intel Xeon 674X is the clear choice. Its 31.3% PassMark single-thread lead and 13.6% Cinebench single-core lead over the 6730P are decisive. The 674X also wins every Cinebench multicore test and every PassMark multithread test, despite having 4 fewer cores. Its unlocked multiplier and 128 PCIe Gen 5 lanes make it the more flexible workstation platform.

The Intel Xeon 6730P should be selected only for workloads that match its physics benchmark win: compute patterns that favor core count and large L3 cache over clock speed. Its 32 cores and 288 MB L3 provide a real advantage in that specific PassMark physics test, and the 1.2% and 1.0% margins in extended instructions and prime numbers show it can hold its own in SIMD-heavy and number-theoretic tasks. However, the 6730P loses by 13.6% or more in 11 of the 16 recorded benchmarks, and its average score is 14.7% lower.

The data does not support a general-purpose recommendation for the 6730P. The 674X wins 15 of 16 head-to-head tests, and its single-thread dominance alone justifies its selection for most mixed server workloads. The 6730P's only niche is physics simulation or similar core-scaled parallel compute, where its 11.9% physics win and larger L3 can be exploited. For everything else, the 674X is the benchmark leader.