Intel Xeon 6747P vs Intel Xeon 678X Comparison

Both processors are 48-core Granite Rapids parts on the same Intel Socket 4710 platform, but they are tuned for very different ends of the server market. The 6747P is the higher-TDP, higher-base-clock workhorse with a larger L3 cache, while the 678X counters with a much higher boost clock, an unlocked multiplier, and significantly more PCIe lanes. The benchmark data shows a clear split: the 6747P dominates multi-threaded and throughput-heavy workloads, while the 678X takes single-threaded performance by a wide margin.

Head-to-Head Benchmarks

The most striking result in the head-to-head data is not a multi-core win but the single-thread gap. In PassMark single-thread, the Xeon 678X scores 3758 against 3236 for the 6747P, a 13.9% advantage. This is a direct consequence of the 678X’s 4.90 GHz boost clock versus the 6747P’s 3.90 GHz. For workloads that depend on per-core responsiveness, this is a decisive margin.

However, the 6747P wins 12 of the 14 head-to-head tests, and the wins are not trivial. In Cinebench R15, R20, and R23 multi-core, the 6747P is consistently 3.2% ahead, scoring 8712, 36301, and 86432 respectively, against 8444, 35185, and 83775 for the 678X. This consistency across all three Cinebench versions suggests a stable architectural advantage in sustained all-core loads, likely tied to its higher 2.70 GHz base clock versus 2.40 GHz.

The largest single delta is in PassMark physics, where the 6747P scores 13398 against 7809 for the 678X, a massive 71.6% lead. This is an outlier compared to other tests, but it indicates that the 6747P’s memory subsystem or scheduling behavior is dramatically better suited to physics simulation. Similarly, in PassMark integer math, the 6747P scores 468518 versus 405075, a 15.7% advantage, and in find prime numbers it leads by 14% (1151 vs 1010). These are compute-heavy integer tasks where the 6747P’s higher base clock and larger 288 MB L3 cache provide a clear edge.

Data throughput tests also favor the 6747P. It leads by 8.4% in data compression (1833378 vs 1690896) and by 8.6% in data encryption (90789 vs 83598). Random string sorting goes to the 6747P by 9.9% (180382 vs 164102). The margin narrows in floating-point math (1.1% lead, 365904 vs 362070) and extended instructions (0.8% lead, 142557 vs 141431), showing that the 678X’s high boost clock can nearly close the gap in vector-heavy workloads. PassMark multithread shows a 3.2% lead for the 6747P (101685 vs 98559), aligning with the Cinebench results.

The 678X’s only other win besides single-thread is the duplicate singlethread test, also by 13.9%. This is a clear profile: the 678X is a single-thread specialist, while the 6747P is the multi-threaded generalist.

Architecture Differences

Both chips are built on Intel’s 5 nm process with a dual-die design measuring 2x 598 mm². They share the same Granite Rapids architecture, the same 112 KB L1 and 2 MB L2 cache per core, and identical eight-channel DDR5 memory support with 409.6 GB/s bandwidth. ECC memory is supported on both.

The key architectural split is in the L3 cache. The 6747P has 288 MB of shared L3, while the 678X has 192 MB. This 96 MB difference is significant for workloads that benefit from large on-die data residency, such as database processing or virtualization. The 6747P’s larger cache likely explains its wins in data compression and encryption, where data reuse is common.

The memory bus is identical, but the PCIe implementation differs substantially. The 6747P provides Gen 5 with 88 lanes (CPU only), while the 678X provides Gen 5 with 128 lanes (CPU only). That is 40 additional lanes on the 678X, which is a major factor for systems requiring high-density NVMe storage or multiple GPU accelerators. Both have no integrated graphics.

Clock behavior is the other major divergence. The 6747P runs a 2.70 GHz base clock and 3.90 GHz boost, while the 678X runs a lower 2.40 GHz base but a much higher 4.90 GHz boost. The 678X also has an unlocked multiplier, allowing overclocking, whereas the 6747P is locked. This makes the 678X a more flexible part for single-thread tuning, despite its lower base clock.

Thermal design power differs as well: the 6747P is rated at 330 W, while the 678X is rated at 300 W. The 6747P consumes more power to sustain its high base clock across all cores, while the 678X’s lower TDP reflects its boost-oriented design. Both are active production parts, with the 6747P released on 2025-02-23 and the 678X on 2026-02-01.

FAQ

Q: Which processor is faster in multi-threaded workloads?

A: The Intel Xeon 6747P wins every multi-threaded benchmark in the head-to-head data. It leads by 3.2% in Cinebench R15, R20, and R23 multi-core, and by 3.2% in PassMark multithread. The largest multi-threaded win is 71.6% in PassMark physics.

Q: Is the Xeon 678X better for single-threaded performance?

A: Yes. The 678X scores 3758 in PassMark single-thread, which is 13.9% higher than the 6747P’s 3236. This is the only benchmark category where the 678X wins, aside from the duplicate singlethread test.

Q: How do the cache sizes compare?

A: The 6747P has 288 MB of shared L3 cache, while the 678X has 192 MB. Both have identical per-core L1 (112 KB) and L2 (2 MB) caches. The 6747P’s larger L3 is likely a factor in its wins in data compression and encryption.

Q: What is the difference in PCIe lane count?

A: The 678X provides Gen 5 with 128 lanes (CPU only), while the 6747P provides Gen 5 with 88 lanes (CPU only). The 678X offers 40 additional lanes, making it better suited for systems with many expansion cards or NVMe devices.

Q: Can either processor be overclocked?

A: Only the 678X has an unlocked multiplier. The 6747P is multiplier-locked. This gives the 678X headroom for manual tuning, which is unusual for a server-class Xeon.

Q: Do both processors support the same memory configuration?

A: Yes. Both support DDR5 with an eight-channel memory bus and 409.6 GB/s bandwidth, and both have ECC memory support. There is no difference in memory capability between the two.

Specification Differences

The core and thread counts are identical (48 cores, 96 threads), as are the L1 and L2 cache sizes. The differences are as follows:

• Base clock: 6747P is 2.70 GHz; 678X is 2.40 GHz.
• Boost clock: 6747P is 3.90 GHz; 678X is 4.90 GHz.
• TDP: 6747P is 330 W; 678X is 300 W.
• L3 cache: 6747P is 288 MB shared; 678X is 192 MB shared.
• PCIe lanes: 6747P is Gen 5, 88 lanes (CPU only); 678X is Gen 5, 128 lanes (CPU only).
• Multiplier: 6747P is locked; 678X is unlocked.
• Release date: 6747P is 2025-02-23; 678X is 2026-02-01.
• Launch MSRP: 6747P is $6497; 678X is $3749.
• Part number: 6747P is SRVEZ; 678X is SA2CX.

The die size, process node, memory bus, memory bandwidth, socket, architecture, and integrated graphics are identical. Both are listed as Server/Workstation market segment parts.

The Verdict

The benchmark data paints a clear picture. The 6747P is the superior multi-threaded processor, winning 12 of 14 head-to-head tests. Its wins range from a narrow 0.8% in extended instructions to a massive 71.6% in physics. The 6747P also has a 50% larger L3 cache (288 MB vs 192 MB) and a higher base clock (2.70 GHz vs 2.40 GHz), which are the likely drivers of its throughput advantage. Its 330 W TDP is higher, but it delivers the performance for that power.

The 678X is the single-thread champion, with a 13.9% lead in PassMark single-thread. It also offers an unlocked multiplier, 128 PCIe lanes versus 88, and a lower 300 W TDP. However, it loses every multi-threaded test, and its 192 MB L3 cache is a significant handicap for data-heavy workloads. The 678X is not a bad processor—it still scores 83775 in Cinebench R23 multi-core—but it is consistently behind the 6747P in all-core scenarios.

For pure compute density, the 6747P is the choice. For single-thread responsiveness or overclocking potential, the 678X is the only option, but you sacrifice a lot of multi-threaded performance to get it.

Where Each One Wins

Intel Xeon 6747P: Choose this for any workload that scales across all cores. The data shows it wins in Cinebench R15/R20/R23 multi-core, PassMark multithread, data compression, data encryption, integer math, find prime numbers, floating-point math, random string sorting, extended instructions, and physics. The physics win (71.6%) is particularly notable for simulation or engineering workloads. Its larger 288 MB L3 cache makes it better for datasets that fit within cache, and its higher base clock ensures consistent all-core performance.

Intel Xeon 678X: Choose this for single-threaded performance or for systems that need more expansion capability. The 13.9% single-thread lead is substantial, and the 128 PCIe lanes allow for more NVMe drives or GPUs than the 6747P’s 88 lanes. The unlocked multiplier is a unique feature for a Xeon, appealing to users who want to manually tune clock speeds. Its lower 300 W TDP also makes it easier to cool in dense chassis, though the benchmark data shows it cannot match the 6747P in multi-threaded throughput. The 678X is also listed with a lower launch MSRP, but that is not a factor in performance analysis.