AMD Instinct MI300 vs Intel Arc Pro B390 Comparison

The AMD Instinct MI300 and the Intel Arc Pro B390 occupy entirely separate corners of the hardware spectrum, a fact made clear by their specifications. The MI300 is a massive accelerator with a 1017 mm² die and 153,000 million transistors, built for high-compute workloads. The Arc Pro B390 is an integrated graphics processor (IGP) on a 3 nm process, drawing a fraction of the power. This analysis compares the two using the recorded database information, focusing on what each component delivers in its intended environment.

FAQ

Q: What are the primary architectures of these two processors?

A: The AMD Instinct MI300 uses the CDNA 3.0 architecture on a 5 nm process from TSMC. The Intel Arc Pro B390 uses the Xe3-LPG architecture on a 3 nm process from Intel.

Q: How do their memory configurations differ?

A: The MI300 has 128 GB of HBM3 memory on an 8192-bit bus, delivering 5.32 TB/s of bandwidth. The Arc Pro B390 uses System Shared memory, meaning its size, type, and bus width are all System Shared, with bandwidth listed as System Dependent.

Q: What is the difference in their FP32 compute performance?

A: The MI300 delivers 47.87 TFLOPS of FP32 performance. The Arc Pro B390 delivers 7.680 TFLOPS, which is significantly lower.

Q: Do both processors support standard graphics APIs?

A: No. The MI300 has no display outputs and lists DirectX, OpenGL, and Vulkan as N/A. The Arc Pro B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Q: What are their power requirements?

A: The MI300 has a TDP of 600 W and requires a 1000 W suggested PSU with 2x 8-pin power connectors. The Arc Pro B390 has a TDP of 80 W and uses no power connectors, as it is an integrated processor.

Q: What is the release timeline for each?

A: The MI300 was released on January 3, 2023. The Arc Pro B390 has a release date of January 26, 2026.

Architecture Differences

The two chips are built on different foundational designs. The AMD Instinct MI300 uses the CDNA 3.0 architecture, a compute-focused design aimed at data center workloads. Its chip is named Aqua Vanjaram, and it is manufactured on a 5 nm process at TSMC. The Intel Arc Pro B390 uses the Xe3-LPG architecture, a low-power graphics architecture, with the chip named Panther Lake, manufactured on a 3 nm process at Intel.

The transistor counts tell a story of scale. The MI300 integrates 153,000 million transistors on a 1017 mm² die, yielding a transistor density of 150.4M per mm². The Arc Pro B390 has unknown transistor and die size figures, but its integration level is that of an IGP, meaning it is built into a larger processor package. The MI300 is a discrete accelerator board at 267 mm in length and 111 mm in height, while the Arc Pro B390 has no dimensions listed because it is not a standalone card.

Memory architecture separates them further. The MI300 uses 128 GB of HBM3 on a massive 8192-bit bus, achieving 5.32 TB/s of bandwidth. The Arc Pro B390 uses System Shared memory, which means it relies on the host system's memory, and its bandwidth is System Dependent. This makes the MI300 a self-contained memory powerhouse, while the Arc Pro B390 is constrained by the platform it is integrated into.

Compute resources also differ sharply. The MI300 has 14,080 shading units and 880 texture mapping units, with no ROPs listed (0). Its pixel rate is 0 MPixel/s, reflecting its lack of a display pipeline. The Arc Pro B390 has 1,536 shading units, 48 TMUs, and 24 ROPs, with a pixel rate of 60.00 GPixel/s. The B390 also has 12 ray tracing cores, while the MI300 lists no ray tracing cores. The MI300's texture rate is 1,496.0 GTexel/s, compared to the B390's 120.0 GTexel/s.

The MI300's FP32 throughput is 47.87 TFLOPS, and its FP16 throughput is also 47.87 TFLOPS on a 1:1 ratio. The Arc Pro B390 achieves 7.680 TFLOPS in FP32 and 15.36 TFLOPS in FP16 on a 2:1 ratio. This indicates the MI300 is designed for consistent heavy compute across precisions, while the B390 doubles its throughput in FP16.

API support is another differentiator. The MI300 lists DirectX, OpenGL, and Vulkan as N/A and 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 Portable Device Dependent. The MI300 is not a graphics card in the traditional sense; it is an accelerator with no video output. The B390 is an integrated GPU meant to drive displays on portable devices.

Head-to-Head Benchmarks

The database contains no direct head-to-head benchmark results between these two processors. Both have an avgBenchmarkScore of 0, and the winsA and winsB fields are both 0. As such, this comparison relies entirely on the recorded specification data to determine where each component holds an advantage.

The most decisive difference is in memory bandwidth. The MI300 offers 5.32 TB/s from its HBM3 stack on an 8192-bit bus. The Arc Pro B390 offers System Dependent bandwidth, which cannot match a dedicated HBM3 implementation in absolute terms. For any workload that is bandwidth-limited, the MI300 has a fundamental advantage.

In raw compute, the MI300's FP32 figure of 47.87 TFLOPS is roughly 6.2 times the Arc Pro B390's 7.680 TFLOPS. The FP16 comparison is closer in ratio: the MI300's 47.87 TFLOPS is about 3.1 times the B390's 15.36 TFLOPS. This gap reflects the MI300's purpose as a data center accelerator, while the B390 is an integrated solution for client devices.

The texture rate also favors the MI300 heavily. At 1,496.0 GTexel/s, it is more than 12 times the B390's 120.0 GTexel/s. The B390 counters in pixel processing with 60.00 GPixel/s, while the MI300 has a pixel rate of 0 MPixel/s. The B390 also has 12 ray tracing cores, a feature the MI300 does not list, making the B390 the only one of the two with any ray tracing capability.

Clock speeds show different design priorities. The MI300 has a base clock of 1000 MHz and a boost of 1700 MHz. The Arc Pro B390 has a base clock of 300 MHz and a boost of 2500 MHz. The B390 boosts much higher, but it has far fewer execution units. The MI300 compensates with a much larger number of shading units and TMUs.

The MI300's memory clock is 1300 MHz with 5.2 Gbps effective speed. The Arc Pro B390's memory clock is System Shared, reinforcing its dependence on host memory. The MI300's 128 GB capacity dwarfs any shared memory allocation typical of an IGP, and its 8192-bit bus width is an order of magnitude larger than what an integrated processor would use.

Power consumption is inversely related to their compute scales. The MI300 is rated at 600 W TDP with a suggested PSU of 1000 W. The Arc Pro B390 is rated at 80 W TDP with no power connectors and no suggested PSU. The B390's efficiency per watt is not directly measurable from the data, but its absolute power draw is far lower.

Specification Differences

The two processors differ in almost every measurable specification. The MI300 uses a 5 nm process; the B390 uses 3 nm. The MI300 is manufactured by TSMC; the B390 by Intel. The MI300 has 153,000 million transistors on a 1017 mm² die; the B390 has unknown transistor and die size figures.

Memory is a major split. The MI300 has 128 GB of HBM3, an 8192-bit bus, and 5.32 TB/s bandwidth. The B390 uses System Shared memory for size, type, and bus width, with System Dependent bandwidth. The MI300 has a memory clock of 1300 MHz (5.2 Gbps effective); the B390 lists System Shared for its memory clock.

The compute units are vastly different in count. The MI300 has 14,080 shading units, 880 TMUs, and 0 ROPs. The B390 has 1,536 shading units, 48 TMUs, and 24 ROPs. The B390 has 12 ray tracing cores; the MI300 has none listed. Pixel rates are 0 MPixel/s for the MI300 and 60.00 GPixel/s for the B390. Texture rates are 1,496.0 GTexel/s versus 120.0 GTexel/s.

FP32 performance is 47.87 TFLOPS for the MI300 and 7.680 TFLOPS for the B390. FP16 performance is 47.87 TFLOPS (1:1) for the MI300 and 15.36 TFLOPS (2:1) for the B390. The MI300's FP16 matches its FP32, while the B390's FP16 is double its FP32.

Power and physical requirements differ completely. The MI300 has a 600 W TDP, 2x 8-pin power connectors, and a 1000 W suggested PSU. The B390 has an 80 W TDP, no power connectors, and no suggested PSU. The MI300 is a discrete card with dimensions of 267 mm by 111 mm; the B390 is an IGP with no dimensions.

The bus interface is PCIe 5.0 x16 for the MI300 and IGP for the B390. Display outputs are absent on the MI300 and Portable Device Dependent on the B390. API support is N/A for the MI300 and includes DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4 for the B390.

The MI300's release date is January 3, 2023, with a predecessor of Radeon Instinct. The B390's release date is January 26, 2026, with a predecessor of HD Graphics-WM. The B390 has a production status of Active, while the MI300's production status is not listed. Neither has a launch MSRP in the database.

Where Each One Wins

The MI300 wins in any workload that demands massive memory capacity and bandwidth. Its 128 GB of HBM3 and 5.32 TB/s bandwidth make it suited for large data sets that cannot fit in the shared memory of an integrated GPU. The 8192-bit bus width provides a memory path that an IGP cannot approach.

The MI300 also wins in raw compute throughput. Its 47.87 TFLOPS FP32 and FP16 figures are multiples of the B390's numbers. For compute-heavy tasks like large-scale matrix operations or scientific simulation, the MI300's 14,080 shading units and 880 TMUs provide the execution resources. The texture rate of 1,496.0 GTexel/s supports heavy texturing workloads, though the MI300 has no display pipeline.

The Arc Pro B390 wins in power efficiency and integration. At 80 W TDP with no external power connectors, it fits into platforms where the MI300's 600 W TDP and 1000 W PSU requirement are impossible. The B390 is an IGP, meaning it requires no separate card, no PCIe slot, and no additional cooling beyond the host system.

The B390 wins in graphics features. It supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, and it has 12 ray tracing cores. Its pixel rate of 60.00 GPixel/s and 24 ROPs give it a functional display and rasterization pipeline. The MI300 has no display outputs and no graphics API support, making it unsuitable for any interactive graphics work.

The B390 also wins on clock speed. Its boost of 2500 MHz is higher than the MI300's 1700 MHz, though it has far fewer cores running at that speed. For single-threaded or lightly threaded graphics tasks, the B390's higher clock can be an advantage.

The Verdict

The data shows two processors with no overlap in purpose. The AMD Instinct MI300 is a discrete accelerator with 128 GB of HBM3, 5.32 TB/s of bandwidth, and 47.87 TFLOPS of FP32 compute. It requires a 600 W power budget and a 1000 W PSU, has no display outputs, and supports no graphics APIs. It is built for compute workloads that need massive memory and throughput.

The Intel Arc Pro B390 is an integrated processor with 7.680 TFLOPS FP32, 15.36 TFLOPS FP16, 12 ray tracing cores, and support for DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. It draws 80 W, uses system memory, and is designed for portable devices, as indicated by its Portable Device Dependent display outputs and IGP bus interface.

For a buyer or system integrator, the choice comes down to the workload. The MI300 is the only option when the task requires 128 GB of dedicated memory, 5.32 TB/s of bandwidth, or FP32 compute beyond 40 TFLOPS. The B390 is the only option when the system needs a graphics processor with ray tracing, API support, and minimal power draw.

The MI300's lack of a display output and graphics API support means it cannot serve as a general-purpose GPU. The B390's System Shared memory and lower compute figures mean it cannot match the MI300 in bandwidth or raw throughput. Neither processor can substitute for the other in its respective role. The recorded data supports a straightforward conclusion: select the MI300 for dedicated compute acceleration, select the B390 for integrated graphics in a power-constrained, portable platform.