AMD Ryzen AI Embedded P164 vs Intel Xeon Phi 7290 Comparison

Head-to-Head Benchmarks

The benchmark data paints a clear picture of two dramatically different processors. The Intel Xeon Phi 7290 and the AMD Ryzen AI Embedded P164 trade blows across the Passmark suite, with the Intel part winning 6 tests and the AMD part winning 5. However, the margins of victory tell a more interesting story than the win count alone.

The Xeon Phi 7290's victories are frequently lopsided. In random string sorting, it scores 68,593 against the P164's 34,801, a 97.1% advantage. Physics performance is similarly dominant: 2,257 versus 1,210, an 86.5% lead. Data compression shows a 72.2% gap (564,535 vs 327,891), and extended instructions land at 71.6% ahead (41,517 vs 24,193). Integer math favors the Intel chip by 44.3% (126,922 vs 87,940), while prime number finding is 60.6% better (114 vs 71). These are not subtle differences; they represent fundamental throughput advantages in heavily parallel workloads.

The AMD Ryzen AI Embedded P164 counters with equally decisive wins in other areas. Single-thread performance is the most dramatic: 4,029 versus 485, an 88% advantage for the AMD part. This is a staggering gap, reflecting the architectural chasm between a 2016 many-core accelerator and a modern Zen 5 mobile processor. The P164 also wins multithreaded Passmark scoring 25,889 versus 17,839, a 31.1% lead, despite having only 8 cores and 16 threads against the Xeon Phi's 72 cores and 288 threads. Floating-point math goes to AMD at 55,799 versus 47,417, a 15% edge. Data encryption is also AMD's, at 16,055 versus 12,505, a 22.1% advantage.

What is notable is that the AMD part wins the aggregate Passmark multithread test despite losing several individual compute tests. This suggests that the multithread score is not a simple sum of the other sub-scores, and that the P164's efficiency per thread allows it to outperform in the overall threaded workload. The Xeon Phi's 288 threads produce raw throughput in specific operations, but the AMD chip's higher per-core performance and superior memory subsystem (89.6 GB/s bandwidth, dual-channel DDR5/LPDDR5X) keep it competitive or ahead in mixed workloads.

The Verdict

The data supports a straightforward conclusion: these are tools for entirely different jobs. The Intel Xeon Phi 7290 is a server/workstation part that excels at massively parallel, compute-heavy tasks with high thread counts. Its wins in physics, compression, integer math, and string sorting make it suitable for workloads that can utilize its 72 cores and 288 threads. If your application scales across many threads and relies on integer or extended instruction throughput, the Xeon Phi is the clear performer, with leads ranging from 44.3% to 97.1% over the P164.

The AMD Ryzen AI Embedded P164 is a different beast entirely. It is a mobile-class processor with a 28W TDP, active production status, and a 2026 release date. Its single-thread score of 4,029 dwarfs the Xeon Phi's 485, and it wins multithreaded Passmark overall. This chip is for workloads that prioritize responsiveness, encryption, floating-point math, and power efficiency. It also includes integrated Radeon 880M graphics, which the Xeon Phi lacks entirely.

Looking at the rival comparisons, both CPUs sit at the 91st percentile versus all CPUs, with nearly identical average benchmark scores. The Xeon Phi 7290 averages 53,469, while the P164 averages 52,901. The Xeon Phi's nearest rival, the Intel Core i7-14700F, scores 53,620, just 0.3% higher. The P164's nearest rival, the AMD Ryzen 5 9500F, scores 52,873, a 0.1% difference. Both are positioned in the same performance tier overall, but they reach it through opposite strategies: raw core count versus high per-core efficiency.

Choose the Xeon Phi 7290 for heavy parallel compute where thread count is king. Choose the Ryzen AI Embedded P164 for general-purpose computing, single-threaded tasks, encryption, and floating-point workloads, especially where power consumption and modern features matter.

Architecture Differences

The architectural divide between these two processors is vast. The Intel Xeon Phi 7290 is built on Knights Landing, a 14nm design from Intel's own foundry. It packs 72 cores and 288 threads, with a base clock of 1500 MHz and a boost clock of 1700 MHz. The transistor count is 8,000 million. It uses Intel Socket 3647 and supports DDR4 memory with ECC. There is no integrated graphics. This is a many-core accelerator designed for high-throughput computing, released in June 2016.

The AMD Ryzen AI Embedded P164 is codenamed Gorgon Point, built on a 4nm process by TSMC. It has 8 cores and 16 threads, with a base clock of 2000 MHz and a boost clock of 5000 MHz. The die size is 233 mm². It uses AMD Socket FP8 and supports DDR5 and LPDDR5X memory with dual-channel configuration and 89.6 GB/s bandwidth. ECC memory is supported. It includes Radeon 880M integrated graphics and PCIe Gen 4 with 16 lanes (CPU only). The generation is listed as Ryzen AI Embedded (Zen 5 / Zen 5c), and it was released in March 2026.

Cache configurations differ significantly. The Xeon Phi has 32 KB L1 and 512 KB L2 per core, with no L3 cache listed. The P164 has 80 KB L1 and 1 MB L2 per core, plus 8 MB of shared L3 cache. The larger per-core caches on the AMD part contribute to its massive single-thread advantage. The Xeon Phi's lack of L3 cache is typical of Knights Landing, which relies on high-bandwidth memory and many cores rather than a large shared cache hierarchy.

The manufacturing process gap is enormous: 14nm versus 4nm. The foundry also differs: Intel for the Xeon Phi, TSMC for the AMD part. The Xeon Phi's market segment is Server/Workstation, while the P164 is classified as Mobile. The P164 is marked as Active in production status, while the Xeon Phi has no production status listed. The Xeon Phi's part number is SR2WY; the P164's is unknown.

Specification Differences

The two CPUs differ in nearly every specification field. Core count: 72 versus 8. Threads: 288 versus 16. Base clock: 1500 MHz versus 2000 MHz. Boost clock: 1700 MHz versus 5000 MHz. TDP: 245W versus 28W. Socket: Intel Socket 3647 versus AMD Socket FP8. Process node: 14nm versus 4nm. Foundry: Intel versus TSMC. Transistors: 8,000 million versus not listed. Die size: not listed versus 233 mm².

Cache: L1 is 32 KB per core on the Xeon Phi versus 80 KB per core on the P164. L2 is 512 KB per core versus 1 MB per core. L3 is absent on the Xeon Phi, while the P164 has 8 MB shared. Memory support: DDR4 versus DDR5/LPDDR5X. Memory bus: not listed versus dual-channel. Memory bandwidth: not listed versus 89.6 GB/s. ECC: true for both. PCIe: not listed for the Xeon Phi versus Gen 4, 16 lanes (CPU only) for the P164. Integrated graphics: none versus Radeon 880M.

Release date: June 2016 versus March 2026. Market segment: Server/Workstation versus Mobile. Multiplier unlocked: false for both. Launch MSRP: not listed for either. Part number: SR2WY versus unknown. The Xeon Phi has no codename listed beyond Knights Landing architecture, while the P164 has the Gorgon Point codename.

FAQ

Q: Which CPU has better single-thread performance?

A: The AMD Ryzen AI Embedded P164 dominates, scoring 4,029 in Passmark single-thread tests versus the Intel Xeon Phi 7290's 485. This is an 88% advantage for the AMD part.

Q: How do they compare in multithreaded workloads?

A: The AMD Ryzen AI Embedded P164 wins Passmark multithread with 25,889 versus 17,839 for the Intel Xeon Phi 7290, a 31.1% lead. However, the Xeon Phi wins several specific multithreaded sub-tests, including physics (86.5% lead), data compression (72.2% lead), and integer math (44.3% lead).

Q: What is the core and thread count difference?

A: The Intel Xeon Phi 7290 has 72 cores and 288 threads. The AMD Ryzen AI Embedded P164 has 8 cores and 16 threads. Despite having 9 times fewer cores, the AMD part wins the overall multithreaded Passmark score.

Q: Which processor supports ECC memory?

A: Both the Intel Xeon Phi 7290 and the AMD Ryzen AI Embedded P164 support ECC memory.

Q: Does either CPU have integrated graphics?

A: Only the AMD Ryzen AI Embedded P164 has integrated graphics, featuring Radeon 880M. The Intel Xeon Phi 7290 has no integrated graphics listed.

Q: What are the power consumption figures?

A: The Intel Xeon Phi 7290 has a TDP of 245W. The AMD Ryzen AI Embedded P164 has a TDP of 28W, making it dramatically more power-efficient.

Where Each One Wins

The Intel Xeon Phi 7290 is the clear winner in workloads that demand massive parallel integer throughput. Its physics score of 2,257 versus 1,210 (86.5% ahead) makes it suitable for simulation and physics-heavy applications. Data compression at 564,535 versus 327,891 (72.2% ahead) suits archival and database workloads. Integer math at 126,922 versus 87,940 (44.3% ahead) benefits general computation. Extended instructions at 41,517 versus 24,193 (71.6% ahead) and prime number finding at 114 versus 71 (60.6% ahead) further cement its role in scientific and mathematical computing. Random string sorting at 68,593 versus 34,801 (97.1% ahead) is its largest margin, ideal for sorting and data manipulation tasks.

The AMD Ryzen AI Embedded P164 wins where per-core performance and modern architecture matter. Its single-thread score of 4,029 versus 485 (88% ahead) makes it the obvious choice for applications that are not highly parallel. Data encryption at 16,055 versus 12,505 (22.1% ahead) suits security-related workloads. Floating-point math at 55,799 versus 47,417 (15% ahead) benefits scientific computing that relies on FP operations. The overall multithread score of 25,889 versus 17,839 (31.1% ahead) shows that for many real-world threaded applications, the AMD chip's efficiency wins out over raw core count.

For a system builder, the choice hinges on whether the workload scales across hundreds of threads or relies on a few fast cores with modern features like integrated graphics, high memory bandwidth, and low power draw. The Xeon Phi 7290 is a specialized compute accelerator. The Ryzen AI Embedded P164 is a versatile, efficient processor for general and embedded use. The data does not suggest one is universally better; it suggests they are optimized for opposite ends of the computing spectrum.