Intel Arc A570M vs Intel Arc Pro B390 Comparison

FAQ

Q: What is the Intel Arc A570M based on?

A: The Intel Arc A570M is built on the DG2-256 chip using the Xe-HPG architecture, belongs to the Alchemist (Arc 5 Mobile) generation, and is manufactured on a 6 nm process at TSMC.

Q: What is the Intel Arc Pro B390 based on?

A: The Intel Arc Pro B390 uses the Panther Lake chip with the Xe3-LPG architecture, is part of the Arc Graphics-WM (Panther Lake) generation, and is manufactured on a 3 nm process at Intel.

Q: How do the memory configurations differ?

A: The Arc A570M has 8 GB of GDDR6 memory on a 128-bit bus with 224.0 GB/s bandwidth. The Arc Pro B390 uses System Shared memory, with the bus width and bandwidth listed as System Shared and System Dependent, respectively.

Q: What is the difference in shading units and compute power?

A: The Arc A570M has 2048 shading units and delivers 5.325 TFLOPS FP32. The Arc Pro B390 has 1536 shading units but delivers a higher 7.680 TFLOPS FP32, along with 15.36 TFLOPS FP16 compared to 10.65 TFLOPS FP16 for the A570M.

Q: Which GPU has a higher boost clock?

A: The Arc Pro B390 boosts up to 2500 MHz, while the Arc A570M boosts to 1300 MHz. The A570M has a higher base clock at 900 MHz versus 300 MHz for the B390.

Q: What benchmark data is available for each GPU?

A: The Arc A570M has a recorded Geekbench OpenCL score of 58239, placing it in the 88th percentile among all GPUs. The Arc Pro B390 has no recorded benchmark scores in the database and sits at the 50th percentile with an average score of 0.

Where Each One Wins

The recorded data shows a clear split between the two GPUs in terms of benchmark readiness and architectural positioning. The Intel Arc A570M is the only one of the two with an actual measured benchmark score. Its Geekbench OpenCL result of 58239 places it in the 88th percentile of all GPUs in the database. That score sits within a tight cluster of rivals: it is 0.3% behind the AMD Radeon RX 6950 XT (average score 58392), 0.3% ahead of the AMD Radeon RX 5600 OEM (average score 58085), 0.5% behind the NVIDIA P102-100 (average score 58528), and 0.7% behind the AMD Radeon PRO V710 (average score 58657). The A570M effectively lands in a performance band where the nearest competitors are within a single percentage point, meaning its compute result is representative of that tier rather than an outlier.

The Intel Arc Pro B390, by contrast, has no benchmark entries in the database. Its average benchmark score is recorded as 0, and its percentile rank is 50, which reflects the absence of measured data rather than a performance verdict. The B390 cannot be said to win any benchmark comparison because no scores exist for it. However, its specification sheet indicates where it would likely compete: the FP32 compute rating of 7.680 TFLOPS is 44% higher than the A570M's 5.325 TFLOPS, and its FP16 rating of 15.36 TFLOPS is 44% higher as well. The B390 also carries a higher boost clock (2500 MHz versus 1300 MHz) and a lower base clock (300 MHz versus 900 MHz), which suggests a design tuned for bursty, thermally constrained workloads rather than sustained baseline throughput.

In terms of raw pixel and texture throughput, the A570M wins on both metrics. It produces 83.20 GPixel/s against the B390's 60.00 GPixel/s, a 38.7% advantage. Its texture rate of 166.4 GTexel/s is 38.7% higher than the B390's 120.0 GTexel/s. These differences stem from the A570M having more TMUs (128 versus 48) and more ROPs (64 versus 24), even though the B390 has a higher clock ceiling. For workloads dominated by fill rate and texture sampling, the A570M holds a decisive edge.

The A570M also wins on memory bandwidth in absolute terms. Its dedicated 8 GB GDDR6 pool delivers 224.0 GB/s, whereas the B390 relies on system shared memory with bandwidth described as System Dependent. In a mobile or integrated context, the B390's memory performance will vary with the host platform, while the A570M provides a fixed, predictable bandwidth figure.

The B390 wins on process technology and compute density. Its 3 nm Intel process is a generation ahead of the A570M's 6 nm TSMC node. The B390 achieves higher FP32 throughput with 24% fewer shading units (1536 versus 2048), indicating a more efficient compute design per unit. It also supports the same DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4 APIs, so feature-level parity is maintained.

Architecture Differences

The two GPUs come from different Intel architecture families. The Arc A570M uses the Xe-HPG architecture on the DG2-256 chip, which is part of the Alchemist generation for mobile Arc 5 products. The Arc Pro B390 uses the Xe3-LPG architecture on the Panther Lake chip, belonging to the Arc Graphics-WM generation. This represents a generational jump from the first discrete Arc architecture to a newer, integrated-class design.

The fabrication process differs substantially. The A570M is built on a 6 nm TSMC process with 11,500 million transistors on a 269 mm² die, giving a transistor density of 42.8M per mm². The B390 is built on Intel's 3 nm process, but its transistor count, die size, and density are recorded as unknown or null in the database. The 3 nm node signifies a newer manufacturing technology, which helps explain the B390's higher clock ceiling and improved compute efficiency.

The compute resource layout differs as well. The A570M has 2048 shading units, 128 TMUs, 64 ROPs, and 16 ray tracing cores. The B390 has 1536 shading units, 48 TMUs, 24 ROPs, and 12 ray tracing cores. The A570M has 33% more shading units, 167% more TMUs, 167% more ROPs, and 33% more RT cores. Despite this, the B390 achieves higher FP32 throughput due to its higher boost clock and architectural efficiency.

The memory architecture is fundamentally different. The A570M is a discrete GPU with dedicated 8 GB GDDR6 memory on a 128-bit interface. The B390 is an integrated GPU (IGP) that shares system memory, with memory type, bus width, and bandwidth all dependent on the host platform. This is a structural difference, not just a specification gap, as it affects latency, capacity flexibility, and power behavior.

The power envelope is similar but not identical. The A570M has a TDP of 75 W, while the B390 is rated at 80 W. Both use an IGP slot width, and neither has a separate power connector listed (the B390 is explicitly listed with "None" for power connectors). The A570M uses a PCIe 4.0 x8 bus interface, while the B390 uses an IGP bus interface, reflecting its integrated nature.

Clock behavior differs significantly. The A570M runs at a base of 900 MHz and boosts to 1300 MHz, a modest boost range. The B390 runs at a base of 300 MHz and boosts to 2500 MHz, a much wider range that allows it to idle or run low-load tasks at very low clocks while ramping aggressively under load. This suggests different power management strategies: the A570M is designed for sustained mobile operation, while the B390 is built for burst responsiveness in integrated settings.

Specification Differences

The table below highlights the fields where the two GPUs differ.

| Specification | Intel Arc A570M | Intel Arc Pro B390 |

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

| Chip | DG2-256 | Panther Lake |

| Architecture | Xe-HPG | Xe3-LPG |

| Generation | Alchemist (Arc 5 Mobile) | Arc Graphics-WM (Panther Lake) |

| Process Node | 6 nm | 3 nm |

| Foundry | TSMC | Intel |

| Transistors | 11,500 million | unknown |

| Die Size | 269 mm² | unknown |

| Transistor Density | 42.8M / mm² | null |

| Base Clock | 900 MHz | 300 MHz |

| Boost Clock | 1300 MHz | 2500 MHz |

| Memory Size | 8 GB | System Shared |

| Memory Type | GDDR6 | System Shared |

| Memory Bus Width | 128 bit | System Shared |

| Memory Bandwidth | 224.0 GB/s | System Dependent |

| Shading Units | 2048 | 1536 |

| TMUs | 128 | 48 |

| ROPs | 64 | 24 |

| RT Cores | 16 | 12 |

| Pixel Rate | 83.20 GPixel/s | 60.00 GPixel/s |

| Texture Rate | 166.4 GTexel/s | 120.0 GTexel/s |

| FP32 | 5.325 TFLOPS | 7.680 TFLOPS |

| FP16 | 10.65 TFLOPS | 15.36 TFLOPS |

| TDP | 75 W | 80 W |

| Power Connectors | null | None |

| Bus Interface | PCIe 4.0 x8 | IGP |

| Release Date | 2023-07-31 | 2026-01-26 |

| Predecessor | null | HD Graphics-WM |

| Percentile vs All GPUs | 88 | 50 |

| Avg Benchmark Score | 58239 | 0 |

The release dates differ by over two years, with the A570M launching on 2023-07-31 and the B390 on 2026-01-26. The B390 has a recorded predecessor (HD Graphics-WM), while the A570M has none.

Head-to-Head Benchmarks

No direct head-to-head benchmark entries exist in the database for these two GPUs. The A570M has a single Geekbench OpenCL score of 58239, while the B390 has no benchmark scores at all. The wins and losses must therefore be inferred from the recorded specification data and the A570M's measured performance relative to its nearest rivals.

The A570M's measured OpenCL score of 58239 places it in the 88th percentile of all GPUs. Its closest rival, the AMD Radeon RX 6950 XT, averages 58392, which is only 0.3% higher. The NVIDIA P102-100 averages 58528, 0.5% higher, and the AMD Radeon PRO V710 averages 58657, 0.7% higher. The AMD Radeon RX 5600 OEM averages 58085, which is 0.3% lower than the A570M. These deltas are all within one percentage point, indicating that the A570M sits squarely in a competitive band where no single rival dominates by a meaningful margin.

For the B390, the absence of a benchmark score means its percentile rank of 50 is a placeholder based on no data. No comparison can be made between the two GPUs using measured benchmark results, because the database contains no overlapping test results.

Looking at computed throughput, the B390 has the advantage in FP32 and FP16. Its FP32 rating of 7.680 TFLOPS is 44.2% higher than the A570M's 5.325 TFLOPS. Its FP16 rating of 15.36 TFLOPS is 44.2% higher than the A570M's 10.65 TFLOPS. These are substantial margins that would likely translate into wins for the B390 in compute-bound workloads such as machine learning inference, shader-heavy rendering, or general-purpose GPU compute, provided the shared memory bandwidth is sufficient.

The A570M wins decisively on pixel fill rate and texture fill rate. Its 83.20 GPixel/s is 38.7% higher than the B390's 60.00 GPixel/s. Its 166.4 GTexel/s is 38.7% higher than the B390's 120.0 GTexel/s. These margins are driven by the A570M's larger ROP and TMU counts, and they matter for rasterization-heavy tasks, high-resolution framebuffer writes, and texture-heavy scenes.

The A570M also wins on memory bandwidth certainty. Its 224.0 GB/s is fixed and dedicated. The B390's bandwidth is System Dependent, meaning its actual throughput depends on the host platform's memory configuration, which is not specified in the database. In a best-case scenario with fast system memory, the B390 could approach or exceed the A570M's bandwidth, but no recorded data supports that claim.

Clock behavior favors the B390 for burst workloads. A boost clock of 2500 MHz versus 1300 MHz gives the B390 a 92.3% higher peak clock. However, its base clock of 300 MHz is 66.7% lower than the A570M's 900 MHz, indicating that sustained low-load performance would be weaker on the B390.

The A570M's 2048 shading units and 128 TMUs give it a structural advantage in parallel texture and pixel work. The B390's 1536 shading units and 48 TMUs are fewer, but the higher boost clock and newer 3 nm process allow it to reach higher FP32 throughput. The data shows two different design philosophies: the A570M is a discrete mobile GPU with fixed resources and predictable bandwidth, while the B390 is an integrated GPU with fewer units, a higher clock ceiling, and shared memory.

The A570M's measured score of 58239 is the only benchmark data point in this comparison. It confirms that the A570M performs within 1% of several high-end desktop GPUs from AMD and NVIDIA, based on OpenCL compute. The B390 has no recorded data, so its performance cannot be validated against any rival or against the A570M. The specification differences suggest the B390 would win in raw FP32 compute and boost-clock-sensitive tasks, while the A570M would win in fill-rate-bound and memory-bandwidth-bound scenarios, but those conclusions rely on the recorded specs rather than direct measurements.