AMD Ryzen AI Embedded P164

The AMD Ryzen AI Embedded P164 is an 8-core, 16-thread mobile processor built on TSMC’s 4 nm process, belonging to the Gorgon Point generation and the Ryzen AI Embedded (Zen 5 / Zen 5c) family. With a 28 W TDP and a boost clock of 5.00 GHz, it posts an average benchmark score of 52901, placing it in the 91st percentile among all CPUs. This data positions the P164 as a high-efficiency part that delivers desktop-class throughput in a mobile envelope, though its exact workload fit depends on how its individual scores break down across single-thread, multi-thread, and specialized instruction tests.

Who Should Consider It

The benchmark data indicates the P164 is best suited for users whose workloads blend high-frequency single-thread responsiveness with solid multi-core throughput, all within a strict power envelope. The PassMark single-thread score of 4029 is exceptionally strong — it places the chip among the top tier for lightly-threaded tasks, making it an excellent choice for office productivity, web browsing, and application launch scenarios where latency per core matters more than raw core count. For software developers, the integer math score of 87940 and extended instructions score of 24193 suggest robust compiler and scripting performance, while the floating-point math score of 55799 indicates capable scientific and engineering computation for a 28 W part.

For creative professionals, the data shows a more mixed picture. The multithread score of 25889 is respectable but not class-leading, meaning video rendering, 3D scene compositing, and large batch photo editing — tasks that scale across all 16 threads — will complete efficiently but without the headroom of higher-TDP desktop rivals. The random string sorting score of 34801 and data compression score of 327891 indicate strong database and archival workloads, making the P164 a viable option for embedded servers or network-attached storage devices that require fast data manipulation. Gamers, meanwhile, benefit from the high single-thread score and the integrated Radeon 880M graphics, which enables 1080p esports titles and older AAA games without a discrete GPU, though the lack of a dedicated graphics score in the data prevents precise frame-rate estimates.

The P164 is not the right choice for users who prioritize maximum multi-threaded throughput above all else — the data shows it trails the AMD Ryzen 9 7900X in average score — nor for those who need massive memory bandwidth beyond the 89.6 GB/s dual-channel DDR5/LPDDR5X configuration. Instead, it fits systems where power efficiency, compact footprint, and single-thread agility are paramount, such as industrial automation controllers, portable workstations, and edge AI appliances that rely on the integrated Radeon 880M for display output.

Power and Thermals

The P164 carries a 28 W TDP, a figure that classifies it firmly in the ultra-low-power mobile segment. This TDP level implies that a capable air cooler — such as a slim laptop heatsink or a small low-profile cooler in a mini-PC — is entirely sufficient to manage thermals under sustained load. The data does not include thermal throttling metrics, but the combination of a 4 nm process and a modest core count (8 cores) suggests that the chip can maintain its 5.00 GHz boost clock in short bursts, with sustained all-core loads likely settling to lower frequencies within the 28 W envelope.

The 28 W TDP also means the P164 is well-suited for fanless or semi-passive designs in embedded chassis, where acoustic noise and dust ingress are concerns. The integrated Radeon 880M graphics share the same thermal budget, so heavy GPU workloads will reduce the CPU's available power headroom; benchmark results do not isolate CPU-only vs. iGPU-only power draws, but the overall platform is designed for thermal continuity rather than peak performance. Compared to the rival AMD Ryzen 5 9500F, which achieves a similar average score (52873) but is a desktop part with a presumably higher TDP, the P164 delivers comparable throughput at a fraction of the power — a trade-off that favors mobile and embedded deployments.

How It Compares

AMD Ryzen 5 9500F: The P164 edges out this desktop rival by a razor-thin 0.1% in average benchmark score (52901 vs. 52873). This near-parity is remarkable given the P164's 28 W TDP versus the 9500F's desktop-class power envelope. The data indicates that for multi-threaded workloads, the two chips are effectively interchangeable, but the P164's superior single-thread score (4029) gives it an edge in latency-sensitive tasks, while the 9500F likely relies on higher sustained power delivery for longer all-core workloads.

Intel Xeon 634: The P164 trails this Intel server processor by 0.1% in average score (52901 vs. 52974). The delta is within noise, but the architectural differences are significant: the Xeon 634 targets server reliability with ECC memory support and higher core counts, whereas the P164 offers the same ECC capability (eccMemory: true) in a lower-power mobile form factor. The P164's single-thread advantage over the Xeon is likely substantial, given its 5.00 GHz boost clock, making it more responsive for interactive workloads.

AMD EPYC 7313P: The P164 is 0.6% behind this server-grade EPYC part (52901 vs. 53206). The EPYC 7313P is designed for datacenter density, with massive multi-threaded throughput and high memory bandwidth, so the fact that a 28 W mobile chip comes within 0.6% reflects the P164's architectural efficiency. However, the EPYC likely crushes the P164 in memory-bound workloads due to its multi-channel memory controllers, whereas the P164's dual-channel 89.6 GB/s bandwidth is a clear limitation for server-class applications.

AMD Ryzen 9 7900X: The P164 is 0.7% behind this high-end desktop Ryzen (52901 vs. 53288). The 7900X is a 12-core part with a much higher TDP, so the performance gap is far smaller than the power gap would suggest. This indicates that the P164's Zen 5 / Zen 5c core architecture delivers excellent instructions-per-clock, allowing it to nearly match a desktop flagship in average score. The 7900X will pull ahead in heavily multi-threaded benchmarks with more than 16 threads, but the P164 holds its own in single-thread and lightly-threaded tests.

FAQ

Q: Does the AMD Ryzen AI Embedded P164 support ECC memory?

A: Yes, the P164 supports ECC memory, which is confirmed by the eccMemory: true field in the specifications. This makes it suitable for embedded and server applications where data integrity is critical.

Q: What is the memory bandwidth of the P164?

A: The P164 supports dual-channel DDR5 and LPDDR5X memory with a maximum bandwidth of 89.6 GB/s. This is a fixed specification from the fact pack, and the actual achievable bandwidth will depend on the specific memory modules used.

Q: How does the P164's single-thread performance compare to its multi-thread performance?

A: The P164 scores 4029 in PassMark single-thread tests and 25889 in PassMark multithread tests. The single-thread score is exceptionally high for a 28 W part, while the multithread score is modest, indicating that the chip is better suited for latency-sensitive tasks than for heavily parallelized workloads.

Q: What integrated graphics does the P164 include?

A: The P164 includes the Radeon 880M integrated graphics. The fact pack does not provide specific GPU benchmark scores, so gaming or compute performance of the iGPU cannot be quantified from this data.

Q: Is the P164's multiplier unlocked for overclocking?

A: No, the multiplierUnlocked field is false, meaning the P164 does not support user overclocking. Users must rely on the stock 2.00 GHz base clock and 5.00 GHz boost clock.

Q: What is the process node and codename for the P164?

A: The P164 is built on TSMC's 4 nm process and uses the Gorgon Point codename. It belongs to the Ryzen AI Embedded (Zen 5 / Zen 5c) generation, which combines Zen 5 and Zen 5c cores in an 8-core, 16-thread layout.

Benchmark Performance

The P164's average benchmark score of 52901 places it in the 91st percentile of all CPUs, a strong showing for a 28 W mobile part. Digging into the individual PassMark subtests, the chip demonstrates a lopsided performance profile. The single-thread score of 4029 is the standout figure, indicating excellent branch prediction and high IPC from the Zen 5 / Zen 5c cores. In contrast, the multithread score of 25889 — while respectable — shows that the 8-core/16-thread configuration cannot match the parallel throughput of higher-core-count rivals, which is expected given the TDP constraint.

The integer math score of 87940 is notably high, suggesting that the P164 excels at general-purpose arithmetic, database operations, and encryption workloads. The extended instructions score of 24193 indicates robust support for AVX-512 or similar vector extensions, which benefits scientific simulations and media encoding. Floating-point math scores 55799, which is moderate — likely reflecting the thermal limits on sustained FPU operation. The data compression score of 327891 is a massive outlier, indicating that the chip's memory subsystem and cache hierarchy are exceptionally well-tuned for compression algorithms. Conversely, the find prime numbers score of 71 is the weakest subtest, suggesting that the chip's integer division and modulo operations are not optimized for this specific workload.

Against rivals, the P164 shows remarkable efficiency. It is 0.1% ahead of the AMD Ryzen 5 9500F (52901 vs. 52873), a desktop part with a much higher TDP. It trails the Intel Xeon 634 by 0.1% (52901 vs. 52974), the AMD EPYC 7313P by 0.6% (52901 vs. 53206), and the AMD Ryzen 9 7900X by 0.7% (52901 vs. 53288). These deltas are all within the margin of error for benchmark noise, meaning that the P164 effectively matches four very different desktop and server processors in average score. The key differentiator is the power envelope: the P164 achieves this performance at 28 W, whereas the rivals are desktop or server parts with significantly higher power budgets.

Single-Thread vs Multi-Thread Behavior

The P164 presents an unusual split between single-thread and multi-thread performance. The single-thread score of 4029 is elite, ranking among the fastest mobile processors for lightly-threaded tasks. This is driven by the 5.00 GHz boost clock and the high IPC of the Zen 5 / Zen 5c cores. In contrast, the multithread score of 25889 is only about 6.4 times the single-thread score (25889 / 4029 ≈ 6.4), which is far below the theoretical 16x scaling from 16 threads. This indicates that the 28 W TDP severely limits all-core boost frequencies, causing the chip to throttle significantly when all 16 threads are active.

For real workloads, this means the P164 will feel extremely snappy for web browsing, document editing, and code compilation of small projects — tasks that rely on one or two cores. It will also handle moderately threaded workloads like video conferencing, light photo editing, and simultaneous background tasks without strain. However, for heavily threaded workloads such as 4K video rendering, large-scale data analysis, or multi-threaded game physics, the P164 will underperform relative to its single-thread capabilities. The physics score of 1210 is notably low, suggesting that gaming physics engines — which are typically multi-threaded — will be a bottleneck. Users should pair this chip with GPU-accelerated rendering or compute offload to compensate for the limited multi-thread CPU throughput.

Platform and Compatibility

The P164 uses the AMD Socket FP8, a mobile socket designed primarily for thin-and-light laptops and embedded systems. It supports DDR5 and LPDDR5X memory in a dual-channel configuration, with a maximum memory bandwidth of 89.6 GB/s. ECC memory is supported, which is a critical feature for embedded applications that require error-free data processing. The memory bus is dual-channel, meaning that users should populate two memory modules to achieve full bandwidth; a single module will halve the available bandwidth.

For expansion, the P164 provides PCIe Gen 4 with 16 lanes from the CPU. This is sufficient for a single discrete GPU or multiple NVMe SSDs, though the lane count is lower than desktop platforms that offer 20 or 24 lanes. The integrated Radeon 880M graphics handle display output, so no discrete GPU is required for basic operation. The production status is active, and the release date is March 8, 2026, indicating that this is a current-generation product. The socket FP8 is not upgradeable to future generations in the same way that desktop sockets are, so the P164 is best viewed as a fixed-spec solution rather than a platform with a long upgrade path. The lack of an unlocked multiplier (multiplierUnlocked: false) further reinforces that this is a locked, embedded-focused part.

Architecture and Design

The P164 is built on TSMC's 4 nm process node, a leading-edge manufacturing technology that enables the 28 W TDP while still achieving a 5.00 GHz boost clock. The die size is 233 mm², which is relatively large for a mobile part, suggesting that the chip includes substantial integrated graphics and cache resources. The codename Gorgon Point indicates this is a specific embedded variant within the Ryzen AI Embedded lineup, and the generation is listed as "Ryzen AI Embedded (Zen 5 / Zen 5c)", meaning the 8 cores are a mix of full Zen 5 cores and compact Zen 5c cores. This heterogeneous core layout allows the chip to balance high-performance tasks (on Zen 5 cores) with power-efficient background tasks (on Zen 5c cores), though the fact pack does not specify the exact core distribution.

The cache hierarchy consists of 80 KB of L1 cache per core, 1 MB of L2 cache per core, and 8 MB of shared L3 cache. This yields a total of 8 MB L3, which is modest compared to desktop Ryzen parts that often feature 32 MB or more. The smaller L3 cache likely contributes to the lower multi-thread scaling, as threads competing for shared L3 bandwidth will experience higher latency. The L1 and L2 cache sizes are generous per core, which supports the strong single-thread performance. The Radeon 880M integrated graphics are included on-die, and while the fact pack does not provide GPU-specific specifications, the 233 mm² die size suggests that the GPU takes up a significant portion of the silicon. The process node and core design together enable the P164 to achieve near-desktop performance in a 28 W envelope, making it a compelling option for embedded systems that require both computational density and power efficiency.