AMD Ryzen AI Embedded P164 vs Intel Core Ultra 5 235H Comparison

FAQ

Q: Which processor has more cores and threads?

A: The Intel Core Ultra 5 235H has 14 cores and 14 threads, while the AMD Ryzen AI Embedded P164 has 8 cores and 16 threads. The AMD part uses simultaneous multithreading to reach 16 threads from 8 cores.

Q: What are the boost clock speeds of these two mobile processors?

A: Both processors share the same 5.00 GHz boost clock. Their base clocks differ, with the AMD part at 2.00 GHz and the Intel part at 2.40 GHz.

Q: Which chip has the larger L3 cache?

A: The Intel Core Ultra 5 235H has 18 MB of shared L3 cache, while the AMD Ryzen AI Embedded P164 has 8 MB of L3 cache. The Intel part also has larger per-core L1 and L2 caches.

Q: Do both processors support ECC memory?

A: No. The AMD Ryzen AI Embedded P164 supports ECC memory, while the Intel Core Ultra 5 235H does not.

Q: Which CPU has the higher average benchmark score?

A: The AMD Ryzen AI Embedded P164 has an average benchmark score of 52901, which places it in the 91st percentile of all CPUs. The Intel Core Ultra 5 235H has an average score of 37522, placing it in the 85th percentile.

Q: How many benchmark wins does each processor have in the head-to-head comparison?

A: The Intel Core Ultra 5 235H wins 8 of the 11 head-to-head tests. The AMD Ryzen AI Embedded P164 wins 3 tests.

Architecture Differences

The two processors come from different design philosophies. The AMD Ryzen AI Embedded P164 uses the Gorgon Point codename and belongs to the Ryzen AI Embedded generation built on Zen 5 and Zen 5c cores. It is manufactured on a 4 nm process at TSMC with a die size of 233 mm². The Intel Core Ultra 5 235H uses the Arrow Lake architecture with the Arrow Lake-H codename, belongs to the Core Ultra Series 2 generation, and is built on a 3 nm process, also at TSMC. The Intel part does not have a recorded die size in the database.

Core organization differs significantly. The AMD chip has 8 cores and 16 threads, relying on simultaneous multithreading to double thread count. The Intel chip has 14 cores and 14 threads, meaning it does not use multithreading and instead relies on a higher physical core count. Cache layouts reflect this: the AMD part has 80 KB of L1 per core, 1 MB of L2 per core, and 8 MB of L3. The Intel part has 192 KB of L1 per core, 3 MB of L2 per core, and 18 MB of shared L3.

The memory controllers also differ. Both support DDR5 and LPDDR5X over a dual-channel bus, but the AMD chip has a recorded memory bandwidth of 89.6 GB/s and supports ECC memory. The Intel chip has a higher recorded memory bandwidth of 102.4 GB/s but no ECC support. PCIe connectivity favors the AMD chip in lane count with Gen 4 and 16 lanes (CPU only), while the Intel chip uses Gen 5 with 8 lanes (CPU only). Integrated graphics differ as well: the AMD part uses the Radeon 880M, and the Intel part uses Arc Graphics 140T. The AMD processor is locked, as is the Intel part, so neither offers an unlocked multiplier.

Release timing puts the Intel chip earlier in the database, with a release date in January 2025, while the AMD chip has a release date in March 2026. Both are listed as Active in production status and target the Mobile market segment.

The Verdict

The benchmark data points to clear roles for each processor. The Intel Core Ultra 5 235H is the stronger all-around performer in raw throughput tests, winning 8 of 11 head-to-head comparisons. Its wins include multithread performance, physics, floating-point math, and single-thread performance. The AMD Ryzen AI Embedded P164 counters with a much higher average benchmark score of 52901 versus 37522, driven largely by its standout results in data compression and integer math. That average score also places the AMD part in the 91st percentile of all CPUs, while the Intel chip sits in the 85th percentile.

For workloads that stress integer operations, encryption, and compression, the AMD part is the better choice. For general computing, floating-point math, prime number searches, and physics simulations, the Intel part delivers higher scores. The AMD chip also brings ECC memory support, which matters for reliability-sensitive applications. The Intel chip brings a larger L3 cache and higher memory bandwidth. The choice depends on which benchmark categories match the intended workload.

Specification Differences

| Specification | AMD Ryzen AI Embedded P164 | Intel Core Ultra 5 235H |

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

| Cores | 8 | 14 |

| Threads | 16 | 14 |

| Base clock | 2.00 GHz | 2.40 GHz |

| Boost clock | 5.00 GHz | 5.00 GHz |

| TDP | 28 W | 28 W |

| Socket | AMD Socket FP8 | Intel BGA 2049 |

| Codename | Gorgon Point | Arrow Lake-H |

| Generation | Ryzen AI Embedded (Zen 5 / Zen 5c) | Ultra 5 (Arrow Lake-H) |

| Process node | 4 nm | 3 nm |

| Foundry | TSMC | TSMC |

| Die size | 233 mm² | Not recorded |

| L1 cache | 80 KB per core | 192 KB per core |

| L2 cache | 1 MB per core | 3 MB per core |

| L3 cache | 8 MB | 18 MB shared |

| Memory bandwidth | 89.6 GB/s | 102.4 GB/s |

| ECC memory | Yes | No |

| PCIe | Gen 4, 16 lanes (CPU only) | Gen 5, 8 lanes (CPU only) |

| Integrated graphics | Radeon 880M | Arc Graphics 140T |

| Part number | Unknown | SRQAP |

Head-to-Head Benchmarks

The Intel Core Ultra 5 235H takes the majority of the head-to-head tests, but the AMD Ryzen AI Embedded P164 wins the two largest individual gaps. In data compression, the AMD chip scores 327891 against 301979 for Intel, a delta of 8.6%. In integer math, the AMD chip scores 87940 against 74247, a delta of 18.4%. The AMD chip also wins extended instructions with 24193 against 23354, a 3.6% margin. These three wins give the AMD part its edge in the average benchmark score.

The Intel chip wins the other eight tests, with several by large margins. The biggest gap is in find prime numbers, where Intel scores 239 against 71, a delta of 70.3%. Floating-point math also favors Intel heavily: 93509 against 55799, a 40.3% delta. Physics shows Intel at 1985 against 1210, a 39% delta. Data encryption goes to Intel at 23121 against 16055, a 30.6% delta. Multithread performance favors Intel at 30091 against 25889, a 14% delta. Single-thread performance favors Intel at 4359 against 4029, a 7.6% delta. Random string sorting goes to Intel at 36208 against 34801, a 3.9% delta.

The single-thread result is notable because both processors share the same 5.00 GHz boost clock. The Intel chip still manages a 7.6% higher single-thread score, indicating better per-clock efficiency in that workload. The multithread result is also interesting because the AMD chip has 16 threads against Intel's 14, yet Intel wins by 14%, suggesting the higher physical core count and larger cache help overcome the thread deficit.

The average benchmark scores in the database tell a different story than the head-to-head win count. The AMD part averages 52901, which is 41% higher than Intel's 37522 average. This happens because the AMD wins in data compression and integer math are large absolute scores that pull up its average, while several Intel wins are on tests with smaller absolute point values. The percentile rankings reflect this: AMD sits at 91, Intel at 85.

Where Each One Wins

The AMD Ryzen AI Embedded P164 wins in workloads that rely on integer throughput and data compression. The 18.4% lead in integer math and the 8.6% lead in data compression make it the stronger choice for database operations, file archiving, and general integer-heavy processing. Its 3.6% edge in extended instructions adds another point in favor of applications that use specialized instruction sets. The ECC memory support also gives it an advantage for tasks where data integrity is critical, such as financial calculations or long-running server-style workloads. The higher average benchmark score of 52901 and the 91st percentile ranking reinforce its position as the higher-scoring chip overall in the database.

The Intel Core Ultra 5 235H wins in most other categories. The 70.3% lead in find prime numbers points to a strong advantage in workloads that stress branch prediction and simple integer loops. The 40.3% lead in floating-point math makes it the better choice for scientific computing, 3D rendering, and any workload with heavy FPU usage. The 39% lead in physics suggests an edge in simulation and game physics calculations. The 30.6% lead in data encryption covers security-related workloads. The 14% lead in multithread performance and the 7.6% lead in single-thread performance give it the overall throughput advantage in general productivity tasks. The larger 18 MB L3 cache and higher 102.4 GB/s memory bandwidth support these wins by reducing memory latency and feeding the cores faster.

The Intel chip also wins in random string sorting by 3.9%, a smaller margin that still indicates an edge in text processing and sorting algorithms. Its 14 cores provide more physical hardware threads for parallel workloads, even without multithreading. The 3 nm process node gives it a manufacturing advantage over the AMD part's 4 nm node, which may contribute to its higher base clock of 2.40 GHz against AMD's 2.00 GHz.

For a mixed workload that includes floating-point math, encryption, and physics, the Intel Core Ultra 5 235H delivers more wins. For integer-heavy tasks, compression work, and applications requiring ECC memory, the AMD Ryzen AI Embedded P164 is the stronger option. The data does not support a single universal winner; it supports two distinct profiles for different task types.