AMD Ryzen AI Max 390 vs Intel Core Ultra 9 386H Comparison

Benchmark results in the database split sharply between the AMD Ryzen AI Max 390 and the Intel Core Ultra 9 386H. The AMD part wins 11 head-to-head tests, the Intel part wins 6, and the average benchmark score is 56273 for AMD versus 43210 for Intel. AMD also holds the 91st percentile among all CPUs, while Intel holds the 88th. These are two mobile processors with different core counts, thread counts, memory designs, and process nodes.

Where Each One Wins

The AMD Ryzen AI Max 390 owns the entire Cinebench set. It wins Cinebench R15 multi-core, R15 single-core, R20 multi-core, R20 single-core, R23 multi-core, and R23 single-core. It also wins PassMark multithread, integer math, data compression, extended instructions, and random string sorting. That is an 11 to 6 win split in favor of AMD.

The Intel Core Ultra 9 386H takes PassMark single-thread, floating-point math, physics, data encryption, and find prime numbers. In use-case terms, AMD dominates multi-threaded throughput, integer-heavy workloads, compression, and sorting. Intel leads in floating-point math, encryption, prime-number search, physics, and the PassMark single-thread metric. The two workloads disagree on single-core performance: AMD wins every Cinebench single-core test, while Intel wins PassMark single-thread.

Architecture Differences

AMD uses Zen 5 (Strix Halo) on a 4 nm TSMC process. Intel uses Panther Lake on a 3 nm Intel process. AMD has 12 cores and 24 threads; Intel has 16 cores and 16 threads. AMD's base clock is 3.20 GHz and boost is 5.00 GHz. Intel's base is 2.10 GHz and boost is 4.90 GHz. The AMD part is rated at a TDP of 55, the Intel part at 25.

AMD's cache layout uses 80 KB L1 and 1 MB L2 per core plus 64 MB shared L3. Intel uses 192 KB L1 and 2.5 MB L2 per core plus 18 MB shared L3. Memory support also differs: AMD uses LPDDR5X on a quad-channel bus with 256.0 GB/s bandwidth; Intel supports DDR5 and LPDDR5X on a dual-channel bus with 115.2 GB/s bandwidth. AMD enables ECC memory, Intel does not. AMD offers PCIe Gen 4 with 16 CPU lanes, Intel offers PCIe Gen 5 with 12 CPU lanes.

Integrated graphics are Radeon 8050S for AMD and Intel Xe3 Graphics for Intel. The AMD socket is FP11; Intel uses BGA 2540. Release dates are 2025-01-05 for AMD and 2026-01-04 for Intel. AMD has no series entry in the database, while Intel is listed in Core Ultra Series 3. AMD's die size is recorded as 2x 70.6 mm²; Intel has no die size entry.

Head-to-Head Benchmarks

| Test | AMD Ryzen AI Max 390 | Intel Core Ultra 9 386H | Winner | Delta |

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

| Cinebench R15 multi-core | 3635 | 3223 | AMD | 12.8% |

| Cinebench R15 single-core | 513 | 303.5 | AMD | 69% |

| Cinebench R20 multi-core | 15146 | 12820 | AMD | 18.1% |

| Cinebench R20 single-core | 2138 | 1809 | AMD | 18.2% |

| Cinebench R23 multi-core | 36064 | 20547 | AMD | 75.5% |

| Cinebench R23 single-core | 5091 | 2071.5 | AMD | 145.8% |

| PassMark data compression | 487145 | 352365 | AMD | 38.3% |

| PassMark data encryption | 25097 | 27150 | Intel | -7.6% |

| PassMark extended instructions | 38716 | 29138 | AMD | 32.9% |

| PassMark find prime numbers | 316 | 341 | Intel | -7.3% |

| PassMark floating-point math | 90594 | 108527 | Intel | -16.5% |

| PassMark integer math | 146519 | 87284 | AMD | 67.9% |

| PassMark multithread | 41737 | 35399 | AMD | 17.9% |

| PassMark physics | 2761 | 3028 | Intel | -8.8% |

| PassMark random string sorting | 53113 | 42135 | AMD | 26.1% |

| PassMark single-thread | 4028 | 4218 | Intel | -4.5% |

| PassMark singlethread | 4028 | 4218 | Intel | -4.5% |

The largest AMD margin is Cinebench R23 single-core: 5091 against 2071.5, a 145.8% lead. The largest AMD multi-threaded margin is Cinebench R23 multi-core: 36064 against 20547, a 75.5% lead. PassMark integer math follows at 67.9% ahead, and Cinebench R15 single-core is 69% ahead. Intel's largest win is PassMark floating-point math: 108527 against 90594, a 16.5% lead. The other Intel wins are narrower: physics by 8.8%, data encryption by 7.6%, find prime numbers by 7.3%, and PassMark single-thread by 4.5%. PassMark single-thread and PassMark singlethread record identical scores in the database.

FAQ

Q: Which processor has the higher average benchmark score?

A: The AMD Ryzen AI Max 390 records an average benchmark score of 56273, while the Intel Core Ultra 9 386H records 43210. The AMD also sits at the 91st percentile of all CPUs versus 88th for Intel.

Q: Does the Intel part have more cores?

A: Yes. The Intel Core Ultra 9 386H has 16 cores and 16 threads, while the AMD Ryzen AI Max 390 has 12 cores and 24 threads. Despite fewer cores, AMD has more threads.

Q: Which CPU wins Cinebench R23 single-core?

A: AMD wins by a large margin. It scores 5091 versus 2071.5, a 145.8% advantage. In PassMark single-thread, however, Intel wins 4218 to 4028, a 4.5% advantage.

Q: What memory bandwidth does each support?

A: AMD uses LPDDR5X on a quad-channel bus with 256.0 GB/s bandwidth. Intel supports DDR5 and LPDDR5X on a dual-channel bus with 115.2 GB/s bandwidth.

Q: Is ECC memory supported?

A: Yes for AMD, no for Intel. The AMD Ryzen AI Max 390 lists ECC memory support; the Intel Core Ultra 9 386H does not.

Q: How do the two compare to their nearest rivals?

A: AMD's average score of 56273 is 0.1% below the Ryzen AI 9 HX PRO 470 (56306) and 0.5% above the Core i9-14900HX (56004). Intel's average of 43210 is 0.3% below the Ryzen AI Max PRO 385 (43326) and 0.9% above the Core i9-12900KF (42830).

Specification Differences

| Field | AMD Ryzen AI Max 390 | Intel Core Ultra 9 386H |

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

| Series | No entry | Core Ultra Series 3 |

| Cores | 12 | 16 |

| Threads | 24 | 16 |

| Base clock | 3.20 GHz | 2.10 GHz |

| Boost clock | 5.00 GHz | 4.90 GHz |

| TDP | 55 | 25 |

| Socket | AMD Socket FP11 | Intel BGA 2540 |

| Architecture | Zen 5 | Panther Lake |

| Codename | Strix Halo | Panther Lake |

| Generation | Ryzen AI Max (Zen 5 (Strix Halo)) | Ultra 9 (Panther Lake-H) |

| Process node | 4 nm | 3 nm |

| Foundry | TSMC | Intel |

| Die size | 2x 70.6 mm² | No entry |

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

| L2 cache | 1 MB (per core) | 2.5 MB (per core) |

| L3 cache | 64 MB (shared) | 18 MB (shared) |

| Memory support | LPDDR5X | DDR5, LPDDR5X |

| Memory bus | Quad-channel | Dual-channel |

| Memory bandwidth | 256.0 GB/s | 115.2 GB/s |

| ECC memory | Yes | No |

| PCIe | Gen 4, 16 Lanes (CPU only) | Gen 5, 12 Lanes (CPU only) |

| Integrated graphics | Radeon 8050S | Intel Xe3 Graphics |

| Release date | 2025-01-05 | 2026-01-04 |

| Part number | 100-000001423 | SA4R5Q9EH |

The Verdict

The data gives a clear workload split. AMD wins 11 of the head-to-head tests and carries the higher average benchmark score, 56273 versus 43210, plus the higher percentile, 91 versus 88. Its advantages are largest in Cinebench R23 single-core, Cinebench R23 multi-core, and integer math. Intel's 6 wins are concentrated in floating-point math, physics, encryption, prime-number search, and PassMark single-thread, and it does so at a TDP of 25 versus AMD's 55.

The data points to AMD for workloads that resemble Cinebench, compression, sorting, and integer throughput. The data points to Intel for workloads that resemble floating-point math, encryption, physics, and lower TDP operation.