AMD Radeon Instinct MI308X vs Intel Arc Pro B390 Comparison

Where Each One Wins

The AMD Radeon Instinct MI308X and Intel Arc Pro B390 occupy completely different corners of the graphics hardware landscape, and the recorded data makes that split unmistakable. The MI308X is a data center compute accelerator built for massive parallel workloads, while the Arc Pro B390 is an integrated graphics processor designed for mobile and compact systems. There is no benchmark overlap in the database, so the wins are defined by architectural capability rather than direct frame-rate comparisons.

The MI308X wins in raw compute throughput. Its FP32 output is 81.72 TFLOPS, which is over ten times the 7.680 TFLOPS of the Arc Pro B390. In FP16, the gap widens further: the MI308X delivers 653.7 TFLOPS (8:1 ratio) versus 15.36 TFLOPS (2:1 ratio) for the Intel part. This makes the AMD accelerator the clear choice for AI training, scientific simulation, and any workload that can saturate thousands of shader units. The MI308X also offers 192 GB of HBM3 memory on an 8192-bit bus, yielding 10.3 TB/s of bandwidth. That memory subsystem is in a different class entirely from the Arc Pro B390, which uses System Shared memory with bandwidth described as System Dependent.

The Arc Pro B390 wins in integrated efficiency and platform flexibility. Its 80 W TDP is a fraction of the MI308X's 750 W, and it requires no external power connectors. The Intel part is an IGP, meaning it lives on the processor package and uses the system's shared memory. This eliminates the need for a discrete card, a separate power delivery path, or a dedicated cooling solution beyond what the laptop or compact desktop already has. The Arc Pro B390 also brings display outputs, labeled Portable Device Dependent, while the MI308X has no outputs at all. For any system that needs to drive a screen, the Intel part is the only one of the two that can do so.

The Arc Pro B390 also supports a full API stack: DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The MI308X lists no API support in the database, which aligns with its role as a compute-only accelerator with no graphics pipeline. In terms of production status, the Arc Pro B390 is Active, while the MI308X has no recorded status.

Architecture Differences

The two chips are built on fundamentally different architectures with different foundries and process nodes. The MI308X uses CDNA 3.0, AMD's compute-focused architecture, implemented on a 5 nm process at TSMC. The chip, codenamed Aqua Vanjaram, packs 153,000 million transistors on a 1017 mm² die, giving a transistor density of 150.4 million per square millimeter. The Arc Pro B390 uses Intel's Xe3-LPG architecture, built on a 3 nm process at Intel's own foundry. The database lists the transistor count and die size as unknown, but the process node advantage belongs to Intel.

The MI308X has 19,456 shading units, 1,216 texture mapping units, and 0 ROPs. The zero ROP count is telling: this is not a rasterization part. It has no pixel output rate (0 MPixel/s) and no display outputs. The Arc Pro B390, by contrast, has 1,536 shading units, 48 TMUs, and 24 ROPs, producing a pixel rate of 60.00 GPixel/s. It also includes 12 ray tracing cores, a feature entirely absent from the MI308X's specification sheet.

Memory architecture differs completely. The MI308X uses 192 GB of HBM3 on an 8192-bit bus with 10.3 TB/s bandwidth. Memory clock is 2525 MHz, translating to 10.1 Gbps effective. The Arc Pro B390 has no dedicated memory; it relies on System Shared memory with System Dependent bandwidth. This is typical for integrated graphics, where the CPU and GPU share the same memory pool.

Clock behavior also diverges. The MI308X has a base clock of 1000 MHz and a boost clock of 2100 MHz. The Arc Pro B390 idles at 300 MHz base but boosts to 2500 MHz, a higher peak frequency despite the much lower power envelope. Texture rate tells the story of throughput: the MI308X hits 2,553.6 GTexel/s versus 120.0 GTexel/s for the Intel part.

The MI308X uses a PCIe 5.0 x16 bus interface and is an OAM Module in terms of slot width. The Arc Pro B390 is an IGP with no bus interface beyond the processor package. Release dates are far apart: the MI308X launched on December 5, 2023, while the Arc Pro B390 is dated January 26, 2026.

The Verdict

The data supports a clear split. The AMD Radeon Instinct MI308X is for compute-heavy data center workloads where memory bandwidth and FP32/FP16 throughput are the primary constraints. Its 10.3 TB/s bandwidth, 192 GB capacity, and 81.72 TFLOPS FP32 make it suitable for large model training, HPC simulations, and similar tasks that require massive parallel processing. It is not a graphics card in any conventional sense: no display outputs, no ROPs, no API support listed, and a 750 W TDP that demands a serious power delivery system.

The Intel Arc Pro B390 is for systems where integration and low power matter more than absolute compute. Its 80 W TDP, IGP form factor, and portable-device-dependent display outputs make it a fit for laptops and compact devices that need modest graphics capability without a discrete card. The 12 ray tracing cores and full DirectX 12 Ultimate support indicate it can handle modern graphics APIs, while the 300 MHz base clock suggests aggressive power saving at idle.

There is no scenario where these two compete for the same purchase. The MI308X cannot drive a monitor, and the Arc Pro B390 cannot approach the MI308X's compute density. The database shows a 50th percentile ranking for both against all GPUs, but that metric is based on zero recorded benchmark scores for each, so it reflects the absence of comparative data rather than performance equivalence.

FAQ

Q: Which GPU has higher FP32 performance?

A: The AMD Radeon Instinct MI308X delivers 81.72 TFLOPS FP32, compared to 7.680 TFLOPS for the Intel Arc Pro B390.

Q: Can the MI308X output video to a display?

A: No. The MI308X has no display outputs and a pixel rate of 0 MPixel/s. The Arc Pro B390 has display outputs described as Portable Device Dependent.

Q: What memory does each GPU use?

A: The MI308X uses 192 GB of HBM3 on an 8192-bit bus with 10.3 TB/s bandwidth. The Arc Pro B390 uses System Shared memory with System Dependent bandwidth.

Q: How do their power requirements compare?

A: The MI308X has a 750 W TDP and a suggested PSU of 1150 W. The Arc Pro B390 has an 80 W TDP and no suggested PSU listed.

Q: Does the Arc Pro B390 support ray tracing?

A: Yes, it has 12 ray tracing cores. The MI308X lists no ray tracing cores at all.

Q: What are their release dates?

A: The MI308X was released on December 5, 2023. The Arc Pro B390 is dated January 26, 2026.

Head-to-Head Benchmarks

The database contains no direct head-to-head benchmark results between these two parts, and each has zero recorded benchmark scores. The comparison therefore relies on the specification-level data recorded for each.

The largest win for the MI308X is in FP32 throughput: 81.72 TFLOPS versus 7.680 TFLOPS, a factor of 10.6. In FP16, the MI308X's 653.7 TFLOPS (8:1) is 42.6 times the Arc Pro B390's 15.36 TFLOPS (2:1). Texture rate favors the MI308X at 2,553.6 GTexel/s versus 120.0 GTexel/s, a 21.3x advantage. Memory bandwidth is similarly lopsided: 10.3 TB/s versus System Dependent, which in practice will be limited by the host system's memory controller.

The Arc Pro B390 wins in pixel rate, 60.00 GPixel/s versus 0 MPixel/s for the MI308X. It also has a higher boost clock at 2500 MHz versus 2100 MHz, though the MI308X has a higher base clock at 1000 MHz versus 300 MHz. The Intel part has 12 ray tracing cores where the AMD part has none. The Arc Pro B390 also supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4; the MI308X lists no API support.

In terms of physical and electrical characteristics, the Arc Pro B390's 80 W TDP is 670 W lower than the MI308X's 750 W. The MI308X requires a 1150 W suggested PSU, while the Arc Pro B390 needs none beyond the host system. The MI308X is an OAM Module with a PCIe 5.0 x16 interface; the Arc Pro B390 is an IGP with no separate bus interface.

Specification Differences

| Specification | AMD Radeon Instinct MI308X | Intel Arc Pro B390 |

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

| Architecture | CDNA 3.0 | Xe3-LPG |

| Process Node | 5 nm | 3 nm |

| Foundry | TSMC | Intel |

| Transistors | 153,000 million | unknown |

| Die Size | 1017 mm² | unknown |

| Base Clock | 1000 MHz | 300 MHz |

| Boost Clock | 2100 MHz | 2500 MHz |

| Memory Size | 192 GB | System Shared |

| Memory Type | HBM3 | System Shared |

| Memory Bus Width | 8192 bit | System Shared |

| Memory Bandwidth | 10.3 TB/s | System Dependent |

| Shading Units | 19456 | 1536 |

| TMUs | 1216 | 48 |

| ROPs | 0 | 24 |

| Ray Tracing Cores | None listed | 12 |

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

| Texture Rate | 2,553.6 GTexel/s | 120.0 GTexel/s |

| FP32 | 81.72 TFLOPS | 7.680 TFLOPS |

| FP16 | 653.7 TFLOPS (8:1) | 15.36 TFLOPS (2:1) |

| TDP | 750 W | 80 W |

| Slot Width | OAM Module | IGP |

| Power Connectors | None | None |

| Suggested PSU | 1150 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 |

| Production Status | Not listed | Active |

| Release Date | 2023-12-05 | 2026-01-26 |

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