AMD Instinct MI300 vs Intel Arc B390 Comparison

FAQ

Q: What are the core specifications of the AMD Instinct MI300 and Intel Arc B390?

A: The AMD Instinct MI300 uses the CDNA 3.0 architecture, built on a 5 nm TSMC process with 153,000 million transistors on a 1017 mm² die. It has 14,080 shading units, 880 TMUs, and 0 ROPs. The Intel Arc B390 uses the Xe3-LPG architecture, built on a 3 nm Intel process, with 1,536 shading units, 48 TMUs, 24 ROPs, and 12 ray tracing cores.

Q: How do their memory configurations compare?

A: The AMD Instinct MI300 features 128 GB of HBM3 memory on an 8192-bit bus, delivering 5.32 TB/s bandwidth. The Intel Arc B390 uses system shared memory, with bandwidth listed as system dependent, and no dedicated VRAM.

Q: What are the clock speeds and power draws?

A: The AMD Instinct MI300 runs at a 1000 MHz base and 1700 MHz boost, with a 600 W TDP and 2x 8-pin power connectors, requiring a 1000 W suggested PSU. The Intel Arc B390 runs at a 300 MHz base and 2500 MHz boost, with an 80 W TDP, no power connectors, and no suggested PSU, as it is an integrated graphics processor.

Q: What is the performance difference in their recorded benchmarks?

A: The Intel Arc B390 has a recorded 3DMark Steel Nomad DX12 score of 1482, placing it in the 9th percentile of all GPUs. The AMD Instinct MI300 has no recorded benchmarks and sits in the 50th percentile with an average benchmark score of 0.

Q: How does the Intel Arc B390 compare to its nearest rivals?

A: The Intel Arc B390 scores 1.3% higher than the NVIDIA GeForce GT 520MX (1463), 1.5% higher than the NVIDIA GeForce 800M (1460), 2.5% higher than the NVIDIA GeForce GT 625 OEM (1446), and 2.7% higher than the NVIDIA GeForce GT 710 (1443).

Q: What are the API support levels for each?

A: The AMD Instinct MI300 lists N/A for DirectX, OpenGL, and Vulkan, indicating no consumer graphics API support. The Intel Arc B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Architecture Differences

The AMD Instinct MI300 and Intel Arc B390 are fundamentally different processors with distinct design goals. The MI300 is a data center accelerator built on CDNA 3.0, optimized for compute workloads rather than graphics output. Its 5 nm TSMC process packs 153,000 million transistors into a 1017 mm² die, achieving a transistor density of 150.4 million per mm². This massive die houses 14,080 shading units and 880 texture mapping units, but zero ROPs, confirming its non-rendering purpose. The architecture uses HBM3 memory with a 128 GB capacity and an 8192-bit bus, delivering 5.32 TB/s of bandwidth. The MI300 has no display outputs and no consumer graphics API support, as DirectX, OpenGL, and Vulkan are all listed as N/A.

The Intel Arc B390 is an integrated graphics processor based on the Xe3-LPG architecture, built on Intel's 3 nm process. It uses a "Panther Lake" chip and is part of the Arc Graphics-M generation. The B390 has 1,536 shading units, 48 TMUs, 24 ROPs, and 12 dedicated ray tracing cores, making it a fully capable consumer graphics solution. It supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The B390 relies on system shared memory, meaning its performance is heavily dependent on the host system's RAM and memory bandwidth. Its transistor count and die size are unknown in the database.

Clock behavior also differs significantly. The MI300 has a 1000 MHz base and 1700 MHz boost, while the B390 has a much lower 300 MHz base but a higher 2500 MHz boost. The MI300's power draw is 600 W, requiring a 1000 W suggested PSU and 2x 8-pin connectors, whereas the B390 draws only 80 W and requires no external power connectors. The MI300 connects via PCIe 5.0 x16, while the B390 uses an IGP bus interface. The MI300 has physical dimensions of 267 mm length and 111 mm height, while the B390 has no recorded dimensions as it is integrated into a portable device.

The Verdict

The data shows two products with no overlap in intended use. The AMD Instinct MI300 is a server-grade accelerator with no graphics APIs, no display outputs, and zero ROPs. Its 128 GB HBM3 memory and 5.32 TB/s bandwidth target compute tasks like large-scale data processing and AI workloads. The Intel Arc B390 is an integrated GPU for portable devices, with a full consumer API stack, 12 ray tracing cores, and a 60.00 GPixel/s pixel rate.

For users needing a data center compute accelerator, the MI300 is the only choice from this pair. Its 47.87 TFLOPS FP32 performance and 1:1 FP16 ratio indicate strong compute throughput. For users needing a consumer graphics solution with rendering and API support, the B390 is the only option. Its 7.680 TFLOPS FP32 and 15.36 TFLOPS FP16 (2:1 ratio) show a focus on graphics efficiency. The MI300's 50th percentile standing versus the B390's 9th percentile reflects their different positioning; the MI300 has no recorded benchmarks, while the B390's single 3DMark Steel Nomad score of 1482 places it far below most GPUs.

Specification Differences

The recorded specifications where the two differ are substantial:

• Architecture: CDNA 3.0 (MI300) versus Xe3-LPG (B390)
• Process node: 5 nm TSMC (MI300) versus 3 nm Intel (B390)
• Transistors: 153,000 million (MI300) versus unknown (B390)
• Die size: 1017 mm² (MI300) versus unknown (B390)
• Base clock: 1000 MHz (MI300) versus 300 MHz (B390)
• Boost clock: 1700 MHz (MI300) versus 2500 MHz (B390)
• Memory size: 128 GB HBM3 (MI300) versus System Shared (B390)
• Memory bus width: 8192 bit (MI300) versus System Shared (B390)
• Memory bandwidth: 5.32 TB/s (MI300) versus System Dependent (B390)
• Shading units: 14,080 (MI300) versus 1,536 (B390)
• TMUs: 880 (MI300) versus 48 (B390)
• ROPs: 0 (MI300) versus 24 (B390)
• Ray tracing cores: null (MI300) versus 12 (B390)
• Pixel rate: 0 MPixel/s (MI300) versus 60.00 GPixel/s (B390)
• Texture rate: 1,496.0 GTexel/s (MI300) versus 120.0 GTexel/s (B390)
• FP32: 47.87 TFLOPS (MI300) versus 7.680 TFLOPS (B390)
• FP16: 47.87 TFLOPS 1:1 (MI300) versus 15.36 TFLOPS 2:1 (B390)
• TDP: 600 W (MI300) versus 80 W (B390)
• Power connectors: 2x 8-pin (MI300) versus None (B390)
• Suggested PSU: 1000 W (MI300) versus null (B390)
• Bus interface: PCIe 5.0 x16 (MI300) versus IGP (B390)
• Display outputs: No outputs (MI300) versus Portable Device Dependent (B390)
• DirectX: N/A (MI300) versus 12 Ultimate 12_2 (B390)
• OpenGL: N/A (MI300) versus 4.6 (B390)
• Vulkan: N/A (MI300) versus 1.4 (B390)
• Dimensions: 267 mm x 111 mm (MI300) versus null (B390)
• Production status: null (MI300) versus Active (B390)
• Release date: 2023-01-03 (MI300) versus 2026-01-26 (B390)

Head-to-Head Benchmarks

The database contains no head-to-head benchmark entries between the AMD Instinct MI300 and Intel Arc B390, and the MI300 has no standalone benchmarks recorded. The only recorded performance data is the B390's 3DMark Steel Nomad DX12 score of 1482.

This score places the B390 in the 9th percentile of all GPUs, indicating it sits near the bottom of the performance distribution. Its nearest rivals in the database are all NVIDIA products with comparable low-end scores. The B390 outperforms the GT 520MX by 1.3%, the GeForce 800M by 1.5%, the GT 625 OEM by 2.5%, and the GT 710 by 2.7%. These deltas are narrow, showing that the B390's performance is closely clustered with these legacy low-end parts.

The MI300's lack of benchmark data means no direct numerical comparison is possible. Its 50th percentile standing is derived from its position in the database, but with an average benchmark score of 0, it has not demonstrated measured graphics performance. The MI300's compute-oriented design, with zero ROPs and no graphics APIs, explains the absence of graphics benchmarks. Its FP32 throughput of 47.87 TFLOPS is over six times the B390's 7.680 TFLOPS, but this metric reflects raw compute rather than rendering capability.

The wins tally shows 0 for each product, as no head-to-head tests exist. The B390's single benchmark confirms it functions as a graphics device, while the MI300's specifications confirm it does not.

Where Each One Wins

The AMD Instinct MI300 wins in raw compute throughput. Its 47.87 TFLOPS FP32 and identical 47.87 TFLOPS FP16 (1:1 ratio) indicate balanced compute performance for workloads that do not require graphics rendering. The 128 GB HBM3 memory with 5.32 TB/s bandwidth provides massive data movement capacity, suited for large models and datasets. The 1,496.0 GTexel/s texture rate and 8192-bit memory bus further support high-throughput compute operations. The MI300 also wins on scale, with 14,080 shading units and 880 TMUs compared to the B390's 1,536 and 48 respectively.

The Intel Arc B390 wins in every graphics-related category. It has 24 ROPs versus the MI300's 0, allowing pixel output at 60.00 GPixel/s. Its 12 ray tracing cores enable hardware-accelerated ray tracing, which the MI300 cannot perform. The B390 supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4, while the MI300 lists N/A for all three. The B390 also wins on power efficiency, drawing 80 W versus 600 W, and on physical integration, as it requires no power connectors and fits into portable devices.

The B390 wins on clock speed, with a 2500 MHz boost versus the MI300's 1700 MHz. It also wins on production status, listed as Active, while the MI300 has no recorded status. The B390's release date of 2026-01-26 is newer than the MI300's 2023-01-03.

The MI300 wins on process node density, with 150.4 million transistors per mm² on a 1017 mm² die. The B390's transistor density is unknown. The MI300 also wins on memory bandwidth, delivering 5.32 TB/s versus the B390's system-dependent bandwidth, and on FP16 throughput when scaled for compute, though the B390's 2:1 FP16 ratio indicates a graphics-oriented design.

For users with compute workloads, the MI300's data shows clear advantages in memory capacity, bandwidth, and FP32 throughput. For users with graphics needs, the B390's data shows it is the only functional option, with API support, ray tracing, and pixel output capabilities that the MI300 lacks entirely.