AMD Instinct MI300X vs Intel Arc Pro B390 Comparison

FAQ

Q: What is the AMD Instinct MI300X designed for?

A: The AMD Instinct MI300X is an OAM Module with no display outputs, a 750 W TDP, and a suggested PSU of 1150 W. It uses the CDNA 3.0 architecture and is built for compute workloads, as indicated by its 192 GB of HBM3 memory and 5.32 TB/s bandwidth.

Q: What is the Intel Arc Pro B390 designed for?

A: The Intel Arc Pro B390 is an IGP (integrated graphics processor) with a 80 W TDP and portable device dependent display outputs. It uses the Xe3-LPG architecture and supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, making it suitable for integrated graphics duties in mobile or compact systems.

Q: How do their memory configurations compare?

A: The MI300X has a fixed 192 GB of HBM3 memory with an 8192-bit bus and 5.32 TB/s bandwidth. The Arc Pro B390 uses system shared memory, with its bandwidth being system dependent and no dedicated VRAM.

Q: Which GPU has a higher boost clock?

A: The Intel Arc Pro B390 has a higher boost clock at 2500 MHz, while the AMD Instinct MI300X boosts to 2100 MHz. However, the MI300X has a much higher base clock at 1000 MHz versus 300 MHz for the Arc Pro B390.

Q: What are the API support differences?

A: The Intel Arc Pro B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The AMD Instinct MI300X has no API support listed (N/A for DirectX, OpenGL, and Vulkan), reflecting its compute-focused nature.

Q: What is the process node difference?

A: The AMD Instinct MI300X is built on a 5 nm process at TSMC, with 153,000 million transistors on a 1017 mm² die. The Intel Arc Pro B390 uses a 3 nm process at Intel, with transistor count and die size listed as unknown.

Architecture Differences

The AMD Instinct MI300X and Intel Arc Pro B390 represent fundamentally different design philosophies. The MI300X uses the CDNA 3.0 architecture, which is optimized for data center compute acceleration. This is evident in its massive 153,000 million transistor count spread across a 1017 mm² die, produced on TSMC's 5 nm process. The chip, codenamed Aqua Vanjaram, delivers 19,456 shading units and 1,216 texture mapping units.

In contrast, the Intel Arc Pro B390 uses the Xe3-LPG architecture, which is designed for power-efficient integrated graphics. Built on Intel's 3 nm process, it is part of the Arc Graphics-WM (Panther Lake) generation. The B390 has 1,536 shading units, 48 TMUs, and 24 ROPs, along with 12 ray tracing cores. Its transistor count and die size are not recorded in the database.

The MI300X has no ROPs (0) and a pixel rate of 0 MPixel/s, indicating it is not designed for rasterization output. Its texture rate is 2,553.6 GTexel/s, reflecting its compute throughput. The B390, by contrast, has a pixel rate of 60.00 GPixel/s and a texture rate of 120.0 GTexel/s, showing it can handle traditional graphics rendering.

Memory architecture also differs sharply. The MI300X uses dedicated HBM3 with a 8192-bit bus and 5.32 TB/s bandwidth. The B390 relies entirely on system shared memory, with no dedicated VRAM. The MI300X has a base clock of 1000 MHz and boost of 2100 MHz; the B390 has a base of 300 MHz and boost of 2500 MHz.

The MI300X has no display outputs, no power connectors (it uses an OAM Module slot), and no API support for DirectX, OpenGL, or Vulkan. The B390 has portable device dependent outputs, uses no power connectors as an IGP, and supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. The MI300X has a TDP of 750 W, while the B390 draws only 80 W.

The Verdict

The recorded data shows two products aimed at completely different segments. The AMD Instinct MI300X is a data center accelerator with a benchmark score of 317,994 in Geekbench OpenCL, placing it in the 100th percentile of all GPUs. It sits 5% behind the NVIDIA H200 NVL (334,891), 8% behind the NVIDIA B200 (345,482), but 7.5% ahead of the NVIDIA L40S (295,763) and 10.7% ahead of the NVIDIA RTX 6000 Ada Generation (287,237).

The Intel Arc Pro B390 has no recorded benchmark scores, an average benchmark score of 0, and sits in the 50th percentile. It has no nearest rivals listed in the database, making direct performance comparisons impossible.

For compute-heavy workloads requiring massive memory capacity and bandwidth, the MI300X is the clear choice from the data. Its 192 GB HBM3 and 5.32 TB/s bandwidth, combined with 81.72 TFLOPS FP32 performance, position it for large-scale AI and scientific computing. The B390, with its 7.680 TFLOPS FP32 and integrated design, serves a different purpose: providing graphics capability in compact systems where a discrete GPU is not feasible.

The MI300X cannot render graphics (no ROPs, no display outputs), while the B390 supports modern graphics APIs. Users needing raw compute should select the MI300X; those needing integrated graphics with low power consumption should select the B390.

Specification Differences

The database records the following differences between the two products:

• Process Node: MI300X uses 5 nm (TSMC); B390 uses 3 nm (Intel).
• Transistors: MI300X has 153,000 million; B390 is unknown.
• Die Size: MI300X is 1017 mm²; B390 is unknown.
• Transistor Density: MI300X is 150.4M / mm²; B390 is null.
• Base Clock: MI300X is 1000 MHz; B390 is 300 MHz.
• Boost Clock: MI300X is 2100 MHz; B390 is 2500 MHz.
• Memory: MI300X has 192 GB HBM3 with 8192-bit bus and 5.32 TB/s bandwidth; B390 uses System Shared memory with system dependent bandwidth.
• Shading Units: MI300X has 19,456; B390 has 1,536.
• TMUs: MI300X has 1,216; B390 has 48.
• ROPs: MI300X has 0; B390 has 24.
• RT Cores: MI300X has none (null); B390 has 12.
• Pixel Rate: MI300X is 0 MPixel/s; B390 is 60.00 GPixel/s.
• Texture Rate: MI300X is 2,553.6 GTexel/s; B390 is 120.0 GTexel/s.
• FP32: MI300X is 81.72 TFLOPS; B390 is 7.680 TFLOPS.
• FP16: MI300X is 81.72 TFLOPS (1:1); B390 is 15.36 TFLOPS (2:1).
• TDP: MI300X is 750 W; B390 is 80 W.
• Slot Width: MI300X is OAM Module; B390 is IGP.
• Suggested PSU: MI300X is 1150 W; B390 has none.
• Bus Interface: MI300X is PCIe 5.0 x16; B390 is IGP.
• Display Outputs: MI300X has none; B390 is Portable Device Dependent.
• APIs: MI300X has N/A for DirectX, OpenGL, Vulkan; B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, Vulkan 1.4.
• Release Date: MI300X is 2023-12-05; B390 is 2026-01-26.
• Production Status: MI300X is null; B390 is Active.
• Predecessor: MI300X is Radeon Instinct; B390 is HD Graphics-WM.

Head-to-Head Benchmarks

The database contains no direct head-to-head benchmark results between these two products. The MI300X has a single recorded Geekbench OpenCL score of 317,994, while the B390 has no benchmark entries.

However, the MI300X's score can be contextualized against its nearest rivals. It trails the NVIDIA H200 NVL by 5% (334,891 versus 317,994) and the NVIDIA B200 by 8% (345,482 versus 317,994). It leads the NVIDIA L40S by 7.5% (317,994 versus 295,763) and the NVIDIA RTX 6000 Ada Generation by 10.7% (317,994 versus 287,237).

The B390's 50th percentile placement with zero average benchmark score indicates it has not been tested in the same manner. Its FP32 throughput of 7.680 TFLOPS is roughly 9.4% of the MI300X's 81.72 TFLOPS, based on the recorded figures. The FP16 comparison shows the B390 at 15.36 TFLOPS versus 81.72 TFLOPS for the MI300X, though the MI300X runs FP16 at a 1:1 ratio while the B390 uses a 2:1 ratio.

The MI300X's memory bandwidth of 5.32 TB/s is orders of magnitude higher than the B390's system dependent bandwidth, which cannot be quantified from the data.

Where Each One Wins

AMD Instinct MI300X wins on raw compute throughput. Its 81.72 TFLOPS FP32 and FP16 (1:1) performance dwarfs the B390's 7.680 TFLOPS FP32 and 15.36 TFLOPS FP16 (2:1). The texture rate of 2,553.6 GTexel/s versus 120.0 GTexel/s confirms the MI300X is built for massive parallel workloads.

MI300X wins on memory capacity and bandwidth. With 192 GB of HBM3 and 5.32 TB/s bandwidth, it can hold and feed large datasets. The B390's system shared memory is flexible but has no dedicated bandwidth figure.

MI300X wins on benchmark percentile. It sits in the 100th percentile of all GPUs with a Geekbench OpenCL score of 317,994, while the B390 is in the 50th percentile with no recorded score.

Intel Arc Pro B390 wins on power efficiency. Its 80 W TDP is dramatically lower than the MI300X's 750 W. The B390 also has a higher boost clock at 2500 MHz versus 2100 MHz.

B390 wins on graphics capability. It has 24 ROPs and 12 ray tracing cores, supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4, and provides display outputs. The MI300X has zero ROPs and no display support.

B390 wins on integration. As an IGP with no power connectors, it fits into compact systems without a discrete GPU slot. The MI300X requires an OAM Module slot and a 1150 W suggested PSU.

B390 wins on process node. It uses a newer 3 nm process at Intel, versus 5 nm at TSMC for the MI300X, though the MI300X packs far more transistors.

MI300X wins on transistor scale. Its 153,000 million transistors on a 1017 mm² die with a density of 150.4M / mm² shows a level of complexity the B390 does not approach.

The release dates also differ: the MI300X launched on 2023-12-05, while the B390 is dated 2026-01-26 and is marked as Active in production.