AMD Instinct MI350X vs Intel Arc Pro B390 Comparison

Head-to-Head Benchmarks

The recorded database contains no head-to-head benchmark entries for the AMD Instinct MI350X versus the Intel Arc Pro B390. Both products register an average benchmark score of zero, and neither has a list of nearest rivals with comparative scores. The wins counters for both sides stand at zero, meaning no measured performance comparison can be drawn from the available data.

What the data does reveal is a stark contrast in raw compute capabilities. The AMD Instinct MI350X delivers 72.09 TFLOPS of FP32 throughput, while the Intel Arc Pro B390 reaches 7.680 TFLOPS. This places the AMD part approximately 9.4 times higher in single-precision floating-point performance, a figure derived directly from the recorded numbers. In FP16 compute, the MI350X maintains 72.09 TFLOPS with a 1:1 ratio, whereas the Arc Pro B390 reaches 15.36 TFLOPS using a 2:1 ratio. The AMD accelerator leads in half-precision work by a factor of roughly 4.7.

Texture processing follows the same pattern. The MI350X achieves a texture rate of 2,252.8 GTexel/s, compared to 120.0 GTexel/s for the Arc Pro B390. That is an 18.8-fold advantage for the AMD product. Pixel throughput tells a different story, however. The AMD Instinct MI350X records 0 MPixel/s because it has no ROPs and no display outputs, while the Intel Arc Pro B390 produces 60.00 GPixel/s. The Intel part is the only one of the two with any rasterization output capability.

Both products sit at the 50th percentile among all GPUs in the database, a parity that reflects the absence of measured benchmarks rather than equivalent real-world performance. The data simply has no performance scores to rank either part against a wider field.

Architecture Differences

The AMD Instinct MI350X uses the CDNA 4.0 architecture built around the MI350 256CU chip, while the Intel Arc Pro B390 employs the Xe3-LPG architecture with a Panther Lake chip. The AMD product belongs to the Instinct (MIx) generation, and the Intel part belongs to the Arc Graphics-WM (Panther Lake) generation.

Both are fabricated on a 3 nm process, but they use different foundries. TSMC produces the AMD chip, and Intel runs its own fabs for the Arc Pro B390. The AMD die measures 2380 mm² and contains 185,000 million transistors, giving a transistor density of 77.7M per mm². The Intel die size and transistor count are recorded as unknown, so no density figure exists for comparison.

The compute resources differ enormously. The MI350X contains 16,384 shading units and 1,024 TMUs, with zero ROPs. The Arc Pro B390 has 1,536 shading units, 48 TMUs, and 24 ROPs. The Intel part also includes 12 ray tracing cores, while the AMD datasheet lists no RT core count. Neither product lists tensor cores in the recorded data.

Memory architecture is another major divergence. The AMD accelerator carries 288 GB of HBM3e memory on an 8192-bit bus, delivering 8.19 TB/s of bandwidth. The Intel Arc Pro B390 uses system shared memory, with its bus width, bandwidth, and capacity all dependent on the host platform. Memory clock speeds reflect this split: the MI350X runs at 2000 MHz with 8 Gbps effective transfer, while the Arc Pro B390's memory clock is listed as system shared.

The AMD chip has a base clock of 1000 MHz and a boost clock of 2200 MHz. The Intel chip starts at 300 MHz base and boosts to 2500 MHz. Despite the higher boost frequency on the Intel part, the massive difference in shading units and TMUs gives the AMD accelerator the decisive throughput advantage.

Feature support also differs. The MI350X lists no API support for DirectX, OpenGL, or Vulkan, consistent with a compute accelerator that has no display outputs. The Arc Pro B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, and its display outputs are noted as portable device dependent. Power delivery separates the two as well: the MI350X has a TDP of 1000 W with a suggested PSU of 1400 W, while the Arc Pro B390 draws 80 W and has no suggested PSU listed. The AMD part is an OAM Module with no power connectors, and the Intel part is an IGP with no power connectors.

FAQ

Q: Which product has higher FP32 compute throughput?

A: The AMD Instinct MI350X records 72.09 TFLOPS of FP32 performance, which is roughly 9.4 times higher than the 7.680 TFLOPS listed for the Intel Arc Pro B390.

Q: Does the AMD Instinct MI350X support display outputs?

A: No. The MI350X lists no display outputs and has 0 ROPs with a pixel rate of 0 MPixel/s. The Intel Arc Pro B390, by contrast, has 24 ROPs and a pixel rate of 60.00 GPixel/s, with display outputs dependent on the portable device.

Q: What memory configuration does each product use?

A: The AMD Instinct MI350X uses 288 GB of HBM3e memory on an 8192-bit bus with 8.19 TB/s bandwidth. The Intel Arc Pro B390 uses system shared memory, so its capacity, bus width, and bandwidth are system dependent.

Q: Are both products manufactured on the same process node?

A: Yes, both are listed as 3 nm parts. However, the AMD chip is made by TSMC, while the Intel chip is made by Intel's own foundry.

Q: Which product includes ray tracing hardware?

A: The Intel Arc Pro B390 records 12 ray tracing cores. The AMD Instinct MI350X does not list any RT core count in the database.

Q: What is the thermal design power for each?

A: The AMD Instinct MI350X has a TDP of 1000 W and a suggested PSU of 1400 W. The Intel Arc Pro B390 has a TDP of 80 W with no suggested PSU listed.

Specification Differences

| Specification | AMD Instinct MI350X | Intel Arc Pro B390 |

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

| Architecture | CDNA 4.0 | Xe3-LPG |

| Chip | MI350 256CU | Panther Lake |

| Foundry | TSMC | Intel |

| Transistors | 185,000 million | unknown |

| Die Size | 2380 mm² | unknown |

| Transistor Density | 77.7M / mm² | null |

| Base Clock | 1000 MHz | 300 MHz |

| Boost Clock | 2200 MHz | 2500 MHz |

| Memory Clock | 2000 MHz, 8 Gbps effective | System Shared |

| Memory Size | 288 GB | System Shared |

| Memory Type | HBM3e | System Shared |

| Memory Bus Width | 8192 bit | System Shared |

| Memory Bandwidth | 8.19 TB/s | System Dependent |

| Shading Units | 16384 | 1536 |

| TMUs | 1024 | 48 |

| ROPs | 0 | 24 |

| RT Cores | null | 12 |

| Pixel Rate | 0 MPixel/s | 60.00 GPixel/s |

| Texture Rate | 2,252.8 GTexel/s | 120.0 GTexel/s |

| FP32 | 72.09 TFLOPS | 7.680 TFLOPS |

| FP16 | 72.09 TFLOPS (1:1) | 15.36 TFLOPS (2:1) |

| TDP | 1000 W | 80 W |

| Slot Width | OAM Module | IGP |

| Suggested PSU | 1400 W | null |

| Bus Interface | PCIe 5.0 x16 | IGP |

| Display Outputs | No outputs | Portable Device Dependent |

| DirectX | N/A | 12 Ultimate (12_2) |

| OpenGL | N/A | 4.6 |

| Vulkan | N/A | 1.4 |

| Release Date | 2025-06-11 | 2026-01-26 |

| Production Status | null | Active |

Both products share the same 3 nm process node, a PCIe-related interface (though the Intel part is an IGP rather than a discrete card), and the absence of dedicated tensor core listings. Neither has a launch MSRP recorded in the database.

The Verdict

The recorded data indicates two products built for entirely different purposes. The AMD Instinct MI350X is a compute accelerator with massive memory capacity, extremely high FP32 and FP16 throughput, and no graphics output capability. Its 288 GB of HBM3e memory, 8.19 TB/s bandwidth, and 72.09 TFLOPS of FP32 performance place it firmly in the category of high-throughput parallel compute. The absence of ROPs, display outputs, and graphics API support confirms that it is not designed for rendering or display workloads.

The Intel Arc Pro B390 is an integrated graphics processor with 24 ROPs, 12 ray tracing cores, and support for DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. Its 60.00 GPixel/s pixel rate and 7.680 TFLOPS FP32 performance are modest in absolute terms, but the feature set is complete for a graphics-oriented part. The 80 W TDP and system shared memory indicate a low-power integrated solution rather than a discrete accelerator.

Users who need raw compute throughput, massive memory bandwidth, and large capacity for data-parallel workloads should select the AMD Instinct MI350X. Users who need graphics rendering, ray tracing, and display output from a low-power integrated part should select the Intel Arc Pro B390. The data does not support any direct performance comparison because no benchmarks exist, but the architectural specifications make the intended use cases clear.

Where Each One Wins

The AMD Instinct MI350X wins in every compute throughput category recorded. Its FP32 performance of 72.09 TFLOPS is roughly 9.4 times the Intel part's 7.680 TFLOPS. Its FP16 performance of 72.09 TFLOPS is about 4.7 times the Intel part's 15.36 TFLOPS. Texture rate favors the AMD product at 2,252.8 GTexel/s versus 120.0 GTexel/s. Memory capacity and bandwidth are decisively in the AMD column: 288 GB versus system shared, and 8.19 TB/s versus system dependent. The MI350X also has a wider memory bus at 8192 bit, more shading units at 16,384 versus 1,536, and more TMUs at 1,024 versus 48. The AMD part uses a PCIe 5.0 x16 interface, while the Intel part is an IGP.

The Intel Arc Pro B390 wins in graphics-oriented features. It produces 60.00 GPixel/s of pixel rate, while the AMD part produces 0 MPixel/s. It includes 24 ROPs and 12 ray tracing cores, neither of which the AMD datasheet records. It supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4, whereas the AMD part lists N/A for all three APIs. The Intel part has display outputs, though they depend on the portable device, while the AMD part has no outputs at all. The Intel boost clock reaches 2500 MHz, higher than the AMD boost clock of 2200 MHz, though this does not compensate for the far smaller shader count. The Intel TDP of 80 W is substantially lower than the AMD TDP of 1000 W, making it the only viable choice for power-constrained integrated systems.

The release dates also separate the products. The AMD Instinct MI350X launched on 2025-06-11, while the Intel Arc Pro B390 launched on 2026-01-26. The Intel part is marked as active in production status, while the AMD part has no production status recorded. Both sit at the 50th percentile in the database, reflecting the absence of benchmark data rather than any measured equivalence.