Intel Core Ultra 9 290K Plus vs Intel Xeon 654 Comparison

The Verdict

The benchmark data presents a clear split: the Intel Xeon 654 wins 12 of 17 head-to-head tests, while the Intel Core Ultra 9 290K Plus wins only 5. Yet the average benchmark scores tell a nuanced story. The Xeon 654 averages 90,717 points across its benchmark suite, placing it in the 96th percentile of all CPUs. The Core Ultra 9 290K Plus averages 84,003 points, also in the 96th percentile. The Xeon holds an edge, but the gap is not enormous in aggregate.

The Xeon 654 belongs to the Granite Rapids generation, built for Server/Workstation duty. It wins decisively in integer math (25% ahead), extended instructions (21.4% ahead), and data compression (17.3% ahead). Its physics score is a standout: 5,596 versus 3,315, a 68.8% advantage. This is a processor for computational workloads, database operations, and any task that stresses multi-threaded throughput.

The Core Ultra 9 290K Plus, built on Arrow Lake Refresh, is the desktop challenger. It wins the single-thread battle decisively. PassMark single-thread shows 4,823 versus 3,778, a 21.7% advantage. It also leads in floating-point math by 24.1%, in prime number finding by 22.5%, and in encryption by 22.6%. The data suggests this is the chip for latency-sensitive desktop tasks, scientific floating-point work, and security operations.

The verdict is straightforward. The Xeon 654 is for server racks and workstation farms where multi-core throughput dominates. The Core Ultra 9 290K Plus is for a desktop system where single-core response and floating-point performance matter more. Neither chip is clearly "better" in every scenario. The choice depends on the workload profile. The Xeon's wins are more numerous, but the Core Ultra's wins are concentrated in areas that matter for interactive use. Buyers who run compression, compression, integer math, or physics simulations should choose the Xeon. Those who run floating-point math, encryption, or single-threaded applications should choose the Core Ultra.

FAQ

Q: Which processor has more cores?

The Intel Core Ultra 9 290K Plus has 24 cores and 24 threads, while the Intel Xeon 654 has 18 cores and 36 threads. The Xeon relies on hyper-threading to reach 36 threads from fewer physical cores.

Q: Which chip has the higher boost clock?

The Core Ultra 9 290K Plus boosts to 5.80 GHz, while the Xeon 654 boosts to 4.80 GHz. The Core also has a higher base clock at 3.70 GHz versus 3.10 GHz.

Q: How do they compare in Cinebench multi-core tests?

The Xeon 654 wins every Cinebench multicore test, but by a narrow 0.8% margin. In Cinebench R23, the Xeon scores 52,150 while the Core Ultra scores 51,731.

Q: Which chip is better for single-threaded performance?

The Core Ultra 9 290K Plus is the clear winner. PassMark single-thread shows 4,823 for the Core versus 3,778 for the Xeon, a 21.7% advantage.

Q: What kind of memory bandwidth does each support?

The Xeon 654 uses an eight-channel DDR5 memory bus with 409.6 GB/s bandwidth. The Core Ultra 9 290K Plus uses a dual-channel bus with 115.2 GB/s. This is a major architectural difference, likely explaining the Xeon's compression advantage.

Q: Does the Core Ultra have integrated graphics?

Yes, the Core Ultra 9 290K Plus includes Arc Xe-LPG Graphics 64EU. The Xeon 654 has no integrated graphics, listed as N/A.

Architecture Differences

The two processors come from different design lineages. The Xeon 654 is built on Granite Rapids, the 600 series, and uses a 5nm process from Intel. The Core Ultra 9 290K Plus comes from Arrow Lake Refresh, part of the Core Ultra Series 2, and uses a 3nm process from TSMC. The fabrication difference is significant: 5nm versus 3nm, indicating a more advanced transistor node for the desktop chip.

The Xeon has a massive die size of 2x 598 mm², while the Core Ultra has a single 243 mm² die. The Xeon's dual-die design supports its server role. The Core Ultra packs 17,800 million transistors into that smaller die. The Xeon's transistor count is not listed in the database.

Cache architecture differs substantially. The Xeon provides 112 KB of L1 per core, 2 MB of L2 per core, and 72 MB of shared L3. The Core Ultra has 192 KB of L1 per core, 3 MB of L2 per core, and 36 MB of shared L3. The Xeon has double the shared L3 cache. The Core Ultra has larger per-core caches, which likely explains its single-thread advantage.

Memory support is a major architectural split. Both support DDR5, but the Xeon uses an eight-channel memory bus with 409.6 GB/s bandwidth. The Core Ultra uses a dual-channel bus with 115.2 GB/s. That is a 3.6x difference in theoretical memory bandwidth, a direct result of the server versus desktop design philosophy.

PCIe lanes also differ. The Xeon provides Gen 5 with 128 lanes (CPU only), while the Core Ultra provides Gen 5 with 20 lanes (CPU only). The Xeon is meant to feed many expansion cards, storage controllers, and network interfaces. The Core Ultra is a desktop part with limited expansion.

Both chips support ECC memory, a surprising feature for a desktop chip. Both have unlocked multipliers, allowing overclocking. The Xeon's market segment is Server/Workstation, the Core Ultra's is Desktop. The Xeon was released on 2026-02-01, while the Core Ultra has no release date in the database.

Specification Differences

The table below highlights where the two processors differ:

| Specification | Intel Xeon 654 | Intel Core Ultra 9 290K Plus |

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

| Cores | 18 | 24 |

| Threads | 36 | 24 |

| Base Clock | 3.10 GHz | 3.70 GHz |

| Boost Clock | 4.80 GHz | 5.80 GHz |

| TDP | 200 W | 125 W |

| Socket | Intel Socket 4710 | Intel Socket 1851 |

| Process Node | 5 nm | 3 nm |

| Foundry | Intel | TSMC |

| Die Size | 2x 598 mm² | 243 mm² |

| L1 Cache | 112 KB (per core) | 192 KB (per core) |

| L2 Cache | 2 MB (per core) | 3 MB (per core) |

| L3 Cache | 72 MB (shared) | 36 MB (shared) |

| Memory Bus | Eight-channel | Dual-channel |

| Memory Bandwidth | 409.6 GB/s | 115.2 GB/s |

| PCIe Lanes (CPU only) | Gen 5, 128 Lanes | Gen 5, 20 Lanes |

| Integrated Graphics | N/A | Arc Xe-LPG Graphics 64EU |

| Market Segment | Server/Workstation | Desktop |

The Xeon has fewer cores (18 versus 24) but more threads (36 versus 24), because it enables hyperthreading. The Core Ultra uses no hyperthreading, so threads equal cores. The Core Ultra has higher clocks, both base and boost, and a lower TDP of 125 W versus the Xeon's 200 W. The Xeon uses Intel Socket 4710, the Core Ultra uses Intel Socket 1851, so they are not interchangeable.

The Xeon has more L3 cache (72 MB shared versus 36 MB shared), but the Core Ultra has larger L1 and L2 caches per core. The Xeon's memory bandwidth is 409.6 GB/s versus 115.2 GB/s, a significant difference. The Xeon has no integrated graphics, while the Core Ultra includes Arc Xe-LPG Graphics 64EU.

The Xeon has a launch MSRP of $1199. The Core Ultra has no listed launch MSRP in the database.

Head-to-Head Benchmarks

The Cinebench suite shows an extremely tight race. In R15, R20, and R23, both multicore and singlecore, the Xeon 654 wins every test by exactly 0.8%. For example, Cinebench R23 multicore: Xeon 52,150 versus Core Ultra 51,731. Cinebench R23 singlecore: Xeon 7,362 versus Core Ultra 7,303. This is a consistent but small edge.

The Xeon's biggest win is in PassMark physics, where it scores 5,596 versus 3,315, a 68.8% advantage. This indicates a large lead in physics simulation workloads. The second biggest win is in integer math: Xeon 207,745 versus Core Ultra 166,194, a 25% advantage. The Xeon also wins extended instructions at 63,539 versus 52,338, a 21.4% gain. Data compression shows the Xeon with 818,902 versus 698,346, which is 17.3% ahead.

The Xeon also wins random string sorting by 3.9% (82,828 versus 79,744) and the PassMark multithread test by 0.8% (61,353 versus 60,860). In every Cinebench test, the Xeon wins by 0.8%. The Xeon wins a total of 12 tests; the Core Ultra wins 5.

The Core Ultra's biggest win is in PassMark floating point math: 214,760 versus 163,093, a 24.1% advantage. It wins prime number finding by 22.5% (503 versus 390). It wins data encryption by 22.6% (52,563 versus 40,675). Finally, it wins single-threaded PassMark by 21.7% (4,823 versus 3,778). Those are significant wins, but they are fewer in number.

Notably, the Core Ultra's single-thread score of 4,823 is not just higher, it is 21.7% higher. This is the largest performance gap in the entire head-to-head table. The Xeon's physics win at 68.8% is the largest proportional difference in the other direction.

Both chips are at the 96th percentile of all CPUs, so they are both high-end parts. The Xeon's average benchmark score of 90,717 is higher than the Core Ultra's 83,003, but the Xeon's nearest rivals include the AMD Ryzen AI Max+ 392 at 90,541 (0.2% behind) and the AMD Ryzen 9 9955HX at 91,199 (0.5% ahead). The Core Ultra's nearest rivals include the Intel Core Ultra 9 285K at 83,807 (0.2% behind) and the AMD EPYC 4584PX at 83,090 (1.1% behind).

Where Each One Wins

The Xeon 654 is the winner in all Cinebench tests, all multithreaded workloads, compression, integer math, extended instructions, and physics. This is the chip for server-side data processing, scientific simulation, database compression, and any workload that scales with thread count. The physics win of 68.8% is a standout, making the Xeon the obvious choice for physics engine calculations, finite element analysis, or any compute that relies on rigid body dynamics.

The Xeon also wins data compression by 17.3%, which is important for file servers, database storage, and any application that compresses data on the fly. Its massive memory bandwidth (409.6 GB/s) and 72 MB of L3 cache likely drive these wins. The Xeon's eight-channel memory bus is a clear advantage for memory-hungry server workloads.

The Core Ultra 9 290K Plus is the winner in floating point math, encryption, prime number calculation, and single-thread performance. The floating point win of 24.1% makes it the better choice for scientific computing, 3D graphics, and any task that uses FPU-heavy code. The encryption win of 22.6% is significant for security protocols, VPN, and any workload that encrypts data continuously.

The single-thread win of 21.7% is crucial for desktop responsiveness. Legacy programs, office applications, and many games rely on single-thread performance. The 5.80 GHz boost clock and larger per-core caches (192 KB L1, 3 MB L2) support this win. The Core Ultra's higher clocks (3.70 GHz base, 5.80 GHz boost) give it a clear edge in latency-sensitive, single-threaded tasks.

The Core Ultra also wins in prime finding (22.5% over) and floating point (24.1% over), suggesting that its memory architecture and large per-core caches help in certain math-heavy workloads. The Core's dual-channel memory bus is narrower, but for these particular tasks, it does not hurt.

In summary: the Xeon wins 12 of 17 tests, including all Cinebench, and the Core wins 5 tests, including the largest single-thread gap. The Xeon is for parallel throughput, the Core is for single-thread speed and floating-point efficiency. The data does not show one chip dominating the other across the board; it shows two different optimization philosophies, one for server density and one for desktop agility.