AMD Ryzen AI Z2 Extreme GPU vs Intel Arc Pro B390 Comparison

The Verdict

The recorded data positions the AMD Ryzen AI Z2 Extreme GPU and the Intel Arc Pro B390 as two distinct integrated graphics solutions, each with its own clear strengths. The AMD part is a low-power, high-efficiency design built on a mature 4 nm node, while the Intel part is a higher-power, higher-throughput design on a newer 3 nm node. Neither part has a recorded benchmark score, but the architectural specifications alone provide a clear picture of where each silicon excels.

The AMD Ryzen AI Z2 Extreme GPU is the choice for sustained, thermally constrained workloads. Its 28 W TDP, combined with a 256-bit memory bus and 16 GB of dedicated LPDDR5X, makes it suitable for compact systems where power draw and heat dissipation are the primary constraints. The Intel Arc Pro B390, with its 80 W TDP and 1536 shading units, is the better option for raw compute throughput when power limits are not the main concern. Its 7.680 TFLOPS of FP32 performance and 15.36 TFLOPS of FP16 performance indicate a clear advantage in parallel compute tasks.

The data suggests no universal winner. The AMD part wins on efficiency and memory bandwidth relative to its power envelope. The Intel part wins on sheer shading power and FP16 throughput. The choice depends entirely on the system's power budget and the workload's compute requirements.

Architecture Differences

The two GPUs come from different manufacturers, different foundries, and different architectural generations. The AMD Ryzen AI Z2 Extreme GPU uses the Strix Point chip, built on RDNA 3.5 architecture, manufactured by TSMC on a 4 nm process. It contains 34,000 million transistors on a 233 mm² die, resulting in a transistor density of 145.9 million per square millimeter. The Intel Arc Pro B390 uses the Panther Lake chip, built on Xe3-LPG architecture, manufactured by Intel on a 3 nm process. Intel does not disclose transistor counts or die size for this part, so a density comparison is not possible from the recorded data.

The AMD GPU has 1024 shading units, 64 texture mapping units, and 48 raster output units. It carries 16 ray tracing cores. The Intel GPU has 1536 shading units, 48 TMUs, and 24 ROPs, with 12 ray tracing cores. The Intel part has 50% more shading units than the AMD part, but the AMD part has 33% more TMUs and double the ROPs. This divergence points to different rendering philosophies: AMD allocates more hardware to texture and pixel processing, while Intel throws more raw shader horsepower at the problem.

Memory architecture differs fundamentally. The AMD part uses 16 GB of LPDDR5X on a 256-bit bus, delivering a fixed 256.0 GB/s of bandwidth. The Intel part uses system shared memory, with bus width and capacity dependent on the host system. Memory bandwidth for the Intel GPU is listed as system dependent, so no fixed comparative figure exists. The AMD part's dedicated memory pool is an advantage in scenarios where the system's main memory is also used for CPU tasks.

Clock behavior also diverges significantly. The AMD GPU has a base clock of 800 MHz and a boost clock of 2700 MHz. The Intel GPU has a base clock of 300 MHz and a boost clock of 2500 MHz. The AMD part's higher base clock suggests better sustained performance under load, while the Intel part's lower base clock indicates a more aggressive power management profile that relies on boosting when thermal headroom allows.

Both GPUs support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The AMD part has a single USB Type-C display output, while the Intel part's display outputs are portable device dependent. The Intel part is an IGP, meaning it is integrated into the host processor package, while the AMD part is a standalone console GPU with no bus interface listed. The AMD part has no power connectors, consistent with its 28 W TDP, and the Intel part also has no power connectors, despite its 80 W TDP, because it is an integrated solution.

Where Each One Wins

The AMD Ryzen AI Z2 Extreme GPU wins in scenarios that favor memory bandwidth, texture throughput, and pixel fill. Its 256.0 GB/s of dedicated memory bandwidth is a fixed resource that does not compete with CPU memory access. Its 172.8 GTexel/s texture rate and 129.6 GPixel/s pixel rate are substantially higher than the Intel part's 120.0 GTexel/s and 60.00 GPixel/s. For games or applications that rely heavily on texturing and fill-rate-limited rendering, the AMD part has a measurable advantage.

The AMD part also wins on power efficiency. Its 28 W TDP is 65% lower than the Intel part's 80 W TDP. This makes it suitable for fanless designs, thin-and-light chassis, or any system where the thermal solution is limited. The AMD part's base clock of 800 MHz, more than 2.5 times the Intel part's 300 MHz base, suggests it can maintain higher performance without relying on boost states.

The Intel Arc Pro B390 wins on raw compute throughput. Its 7.680 TFLOPS of FP32 performance is 39% higher than the AMD part's 5.530 TFLOPS. Its FP16 performance of 15.36 TFLOPS is nearly three times the AMD part's 5.530 TFLOPS (which runs FP16 at a 1:1 ratio with FP32). This makes the Intel part dramatically better suited for workloads that use reduced-precision arithmetic, such as machine learning inference, image processing, or scientific simulations that tolerate FP16 inputs.

The Intel part also wins on shading unit count. With 1536 shading units versus the AMD part's 1024, the Intel architecture can handle more concurrent threads in compute-heavy workloads. Its 12 ray tracing cores, while fewer than the AMD part's 16, still provide dedicated hardware for ray-traced effects. The Intel part's 3 nm process node is more advanced than the AMD part's 4 nm node, though the lack of Intel transistor data prevents a full comparison.

The Intel part's system shared memory is a disadvantage in fixed-function scenarios but an advantage in unified memory architectures where the CPU and GPU share a single pool. In systems with fast main memory, the Intel part can access the same memory as the CPU without copying data across a bus. The AMD part's dedicated 16 GB pool avoids system memory contention but introduces a fixed memory size that cannot be expanded.

FAQ

Q: Which GPU has higher FP32 compute performance?

A: The Intel Arc Pro B390 delivers 7.680 TFLOPS of FP32 performance, which is 39% higher than the AMD Ryzen AI Z2 Extreme GPU's 5.530 TFLOPS.

Q: Which GPU has higher memory bandwidth?

A: The AMD Ryzen AI Z2 Extreme GPU has a fixed memory bandwidth of 256.0 GB/s, while the Intel Arc Pro B390's bandwidth is system dependent and cannot be compared as a fixed figure.

Q: What are the power requirements of each GPU?

A: The AMD Ryzen AI Z2 Extreme GPU has a TDP of 28 W, while the Intel Arc Pro B390 has a TDP of 80 W. Neither part requires external power connectors.

Q: Which GPU has more shading units?

A: The Intel Arc Pro B390 has 1536 shading units, compared to 1024 shading units on the AMD Ryzen AI Z2 Extreme GPU.

Q: Do both GPUs support the same graphics APIs?

A: Yes, both GPUs support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Q: Which GPU has a higher pixel fill rate?

A: The AMD Ryzen AI Z2 Extreme GPU has a pixel rate of 129.6 GPixel/s, more than double the Intel Arc Pro B390's 60.00 GPixel/s.

Head-to-Head Benchmarks

No direct head-to-head benchmark results are recorded in the database, and neither GPU has an average benchmark score. The percentile field for both GPUs is 50, indicating they fall in the middle of the database's distribution, but this is based on their specification profiles rather than measured performance. The absence of benchmark data means the comparison must rely entirely on architectural specifications and theoretical throughput figures.

The largest win for the Intel Arc Pro B390 is in FP16 compute. The Intel part delivers 15.36 TFLOPS, which is 2.78 times the AMD part's 5.530 TFLOPS. This is the single biggest performance gap in either direction. The Intel part's FP16 runs at a 2:1 ratio relative to FP32, while the AMD part runs at a 1:1 ratio. This means the Intel architecture can double its throughput when precision is halved, a feature that the AMD part does not offer.

The Intel part also wins decisively in FP32 compute. Its 7.680 TFLOPS is 2.150 TFLOPS higher than the AMD part's 5.530 TFLOPS, a 38.9% advantage. This raw throughput advantage comes from the Intel part's 1536 shading units, which outnumber the AMD part's 1024 by 50%. The Intel part's lower clock speed does not fully offset its shading unit advantage.

The AMD Ryzen AI Z2 Extreme GPU wins decisively in pixel fill rate. Its 129.6 GPixel/s is more than double the Intel part's 60.00 GPixel/s. This advantage comes from the AMD part's 48 ROPs, exactly double the Intel part's 24 ROPs, combined with the AMD part's higher 2700 MHz boost clock. The AMD part also wins in texture fill rate, delivering 172.8 GTexel/s versus the Intel part's 120.0 GTexel/s, a 44% advantage driven by its 64 TMUs versus the Intel part's 48.

Memory bandwidth is another clear AMD win. The AMD part's 256.0 GB/s is a fixed, dedicated resource, while the Intel part's bandwidth is system dependent. In a system with shared memory, the Intel part's bandwidth could be higher or lower depending on the host memory configuration, but the AMD part's dedicated pool guarantees a consistent bandwidth level.

Clock speed behavior favors the AMD part's sustained performance. The AMD part's 800 MHz base clock is 2.67 times the Intel part's 300 MHz base clock. While the boost clocks are closer (2700 MHz versus 2500 MHz), the AMD part's higher base clock means it can maintain higher performance without requiring boost states to be active. This is particularly relevant in thermally constrained environments where boost clocks may be throttled.

The AMD part's 16 GB of dedicated LPDDR5X memory is a fixed capacity that cannot be reduced by other system processes. The Intel part's system shared memory is flexible but subject to contention from CPU workloads. For graphics workloads that require predictable memory availability, the AMD part's dedicated pool is a structural advantage.

Ray tracing hardware differs in count but not in architecture. The AMD part has 16 ray tracing cores, while the Intel part has 12. The AMD part's higher count, combined with its higher clock speeds, suggests better ray tracing throughput relative to the Intel part, though no benchmark data exists to confirm this. The Intel part's lower ray tracing core count is partially offset by its higher shading unit count, which can handle some ray tracing-related work in software.

The process node comparison favors Intel on manufacturing technology. The Intel part uses a 3 nm process, while the AMD part uses 4 nm. The Intel part's smaller node allows higher transistor density, though Intel does not disclose its transistor count. The AMD part's 34,000 million transistors on 233 mm² gives a density of 145.9M per square millimeter, a figure that the Intel part cannot match in the recorded data because its die size is unknown.

The power envelope is the clearest differentiator for system design. The AMD part's 28 W TDP allows passive cooling in many chassis, while the Intel part's 80 W TDP requires active cooling in most cases. The AMD part's lower power draw also reduces system-level thermal load, which can improve sustained performance in multi-core CPU workloads running concurrently with GPU tasks. The Intel part's higher power draw enables its higher compute throughput, but at the cost of greater thermal management requirements.