AMD Radeon Instinct MI300 vs Intel Arc Pro B390 Comparison

Where Each One Wins

The recorded data shows no overlapping use cases for these two parts. The AMD Radeon Instinct MI300 is a data center accelerator with no display outputs, built for compute workloads that require massive memory capacity and bandwidth. It carries 128 GB of HBM3 memory across an 8192-bit bus, delivering 6.55 TB/s of bandwidth. That configuration targets large model inference, scientific computing, and other memory-bound tasks where the GPU must hold substantial datasets locally.

The Intel Arc Pro B390 is an integrated graphics processor (IGP) designed for portable devices. It uses system shared memory, meaning its memory capacity and bandwidth are entirely dependent on the host platform. It has display outputs, though they are portable device dependent. This part targets everyday graphics, media playback, and light content creation on laptops and compact systems. The two do not compete for the same socket, chassis, or workload profile.

Benchmark wins are zero for both parts in the head-to-head comparison data, which reflects the absence of any shared test suite rather than parity in performance. The AMD part has no benchmark entries at all, and the Intel part has no benchmark entries either. Both sit at the 50th percentile versus all GPUs in the database, a placeholder ranking that does not reflect measured performance.

Architecture Differences

The MI300 uses the CDNA 3.0 architecture on TSMC's 5 nm process. The chip, codenamed Aqua Vanjaram, integrates 153,000 million transistors on a 1017 mm² die. Transistor density calculates to 150.4 million per square millimeter. The Intel Arc Pro B390 uses the Xe3-LPG architecture on Intel's 3 nm process, built around the Panther Lake chip. Transistor count and die size are listed as unknown, so no direct density comparison is possible.

The compute arrays differ sharply. The MI300 has 14,080 shading units, 880 texture mapping units, and no raster operation units, which gives it a pixel rate of 0 MPixel/s. It is not designed to rasterize frames. Its texture rate is 1,496.0 GTexel/s. The Intel part has 1,536 shading units, 48 TMUs, and 24 ROPs. It delivers 60.00 GPixel/s and 120.0 GTexel/s. The MI300 has no listed ray tracing cores, while the Intel part includes 12 RT cores.

Memory architecture separates them completely. The MI300 uses dedicated HBM3 with 128 GB capacity and a 6.55 TB/s bandwidth figure. The Intel part uses system shared memory for capacity, type, bus width, and bandwidth, all marked system dependent. Clock behavior also differs. The MI300 runs at a 1000 MHz base and 1700 MHz boost, with memory at 1600 MHz (6.4 Gbps effective). The Intel part has a 300 MHz base and a 2500 MHz boost, with no dedicated memory clock.

Power and integration also diverge. The MI300 is a 600 W part requiring two 8-pin power connectors and a 1000 W suggested power supply. It uses a PCIe 5.0 x16 interface. The Intel part is an IGP with an 80 W TDP, no power connectors, and an IGP bus interface. The MI300 has no API support listed for DirectX, OpenGL, or Vulkan, which matches its compute-only role. The Intel part lists DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Head-to-Head Benchmarks

No head-to-head benchmark entries exist in the database for these two products. The MI300 has no measured scores, and the Arc Pro B390 has no measured scores. Consequently, there are no delta percentages, no win counts, and no percentile deltas to report. The nearest rivals lists are empty for both parts.

What the data does allow is a specification-level comparison of theoretical throughput. The MI300 reaches 47.87 TFLOPS FP32 and 383.0 TFLOPS FP16 (8:1). The Intel part reaches 7.680 TFLOPS FP32 and 15.36 TFLOPS FP16 (2:1). The FP32 gap is roughly 6.2 times in favor of the MI300 based on these figures. The FP16 gap is larger, about 24.9 times, though the ratio conventions differ (8:1 versus 2:1), so those numbers are not directly comparable in efficiency terms.

Memory bandwidth tells a similar story. The MI300's 6.55 TB/s dwarfs the Intel part's system dependent bandwidth, which has no fixed value in the database. Texture throughput favors the MI300 at 1,496.0 GTexel/s versus 120.0 GTexel/s, a 12.5 times difference. Rasterization is the one area where the Intel part has any output capability. The MI300 produces no pixels, while the Intel part produces 60.00 GPixel/s. That distinction reflects their design intents: one moves data, the other moves frames.

The Intel part has a significantly higher boost clock at 2500 MHz versus 1700 MHz for the MI300, but that advantage is irrelevant across different architectures and power envelopes. The MI300's 600 W TDP enables its higher absolute throughput, while the Intel part's 80 W TDP fits its integrated role.

FAQ

Q: Which GPU has more memory?

A: The AMD Radeon Instinct MI300 has 128 GB of HBM3. The Intel Arc Pro B390 uses system shared memory, so its capacity is system dependent and not fixed.

Q: Can the MI300 output video to a display?

A: No. The MI300 lists "No outputs" for display outputs and has 0 ROPs with a 0 MPixel/s pixel rate. The Intel Arc Pro B390 has display outputs, listed as portable device dependent.

Q: What is the power consumption difference?

A: The MI300 has a 600 W TDP and requires two 8-pin power connectors with a 1000 W suggested power supply. The Intel Arc Pro B390 has an 80 W TDP, no power connectors, and no suggested PSU listed.

Q: Do these two GPUs share any API support?

A: The MI300 lists no DirectX, OpenGL, or Vulkan support. The Intel Arc Pro B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Q: Which architecture is newer?

A: The MI300 uses CDNA 3.0 on a 5 nm TSMC process with a release date of 2023-01-03. The Intel Arc Pro B390 uses Xe3-LPG on a 3 nm Intel process with a release date of 2026-01-26.

Q: How do their FP32 compute figures compare?

A: The MI300 delivers 47.87 TFLOPS FP32. The Intel Arc Pro B390 delivers 7.680 TFLOPS FP32, which is roughly 6.2 times lower based on the recorded figures.

The Verdict

The database paints a clear picture: these are not competing products. The AMD Radeon Instinct MI300 belongs in a server room. Its 128 GB HBM3 pool, 6.55 TB/s bandwidth, and 47.87 TFLOPS FP32 target memory-heavy compute workloads. It has no display outputs, no raster hardware, and no graphics API support. Anyone selecting this part needs a compute card, not a display adapter.

The Intel Arc Pro B390 belongs in a portable device. Its 80 W TDP, IGP bus interface, and system shared memory make it suitable for integrated graphics in laptops. It includes 12 RT cores, DirectX 12 Ultimate support, and display outputs. It exists to render frames on a screen, not to process large datasets.

The MI300 has no rasterization capability, so it cannot replace the Intel part in any graphics role. The Intel part has no dedicated memory and a fraction of the compute throughput, so it cannot approach the MI300's bandwidth or FP32 throughput. The 600 W power requirement alone places the MI300 outside any portable design. The data confirms that each part wins only in its own domain, and those domains do not intersect.

Specification Differences

| Specification | AMD Radeon Instinct MI300 | Intel Arc Pro B390 |

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

| Architecture | CDNA 3.0 | Xe3-LPG |

| Process Node | 5 nm (TSMC) | 3 nm (Intel) |

| Transistors | 153,000 million | unknown |

| Die Size | 1017 mm² | unknown |

| Base Clock | 1000 MHz | 300 MHz |

| Boost Clock | 1700 MHz | 2500 MHz |

| Memory Size | 128 GB HBM3 | System Shared |

| Memory Bus Width | 8192 bit | System Shared |

| Memory Bandwidth | 6.55 TB/s | System Dependent |

| Shading Units | 14080 | 1536 |

| TMUs | 880 | 48 |

| ROPs | 0 | 24 |

| RT Cores | not listed | 12 |

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

| Texture Rate | 1,496.0 GTexel/s | 120.0 GTexel/s |

| FP32 | 47.87 TFLOPS | 7.680 TFLOPS |

| FP16 | 383.0 TFLOPS (8:1) | 15.36 TFLOPS (2:1) |

| TDP | 600 W | 80 W |

| Power Connectors | 2x 8-pin | None |

| Suggested PSU | 1000 W | not listed |

| Bus Interface | PCIe 5.0 x16 | IGP |

| Display Outputs | No outputs | Portable Device Dependent |

| DirectX | not listed | 12 Ultimate (12_2) |

| OpenGL | not listed | 4.6 |

| Vulkan | not listed | 1.4 |

| Release Date | 2023-01-03 | 2026-01-26 |

| Production Status | not listed | Active |

| Predecessor | FirePro Data Center | HD Graphics-WM |