AMD Instinct MI325X vs Intel Arc Pro B390 Comparison

AMD Instinct MI325X and Intel Arc Pro B390 occupy opposite extremes of the GPU spectrum, and the recorded data confirms a fundamental split in purpose, architecture, and capability. The MI325X is a 1000 W compute accelerator built on CDNA 3.0, while the Arc Pro B390 is an 80 W integrated graphics processor with full API support. The database shows no head-to-head benchmark results, no wins for either part, and identical 50th percentile rankings among all GPUs, which means the comparison must rely entirely on the physical and architectural specifications recorded.

Head-to-Head Benchmarks

The database contains no direct benchmark scores for these two products. The avgBenchmarkScore for both is 0, and the headToHeadBenchmarks array is empty. This absence of measured data is itself informative: the two parts are designed for such different workloads that no common benchmark suite has been run on both. The MI325X has no display outputs and no graphics API support, so it cannot run standard graphics tests. The Arc Pro B390 has no standalone compute benchmarks recorded, and its memory bandwidth is listed as "System Dependent," which makes any fixed comparison impossible.

The one measurable distinction in the data is the percentile ranking. Both cards sit at the 50th percentile against all GPUs in the database. That identical rank does not indicate equal performance; it reflects that neither card has been scored in the database's benchmark suite. The MI325X's FP32 throughput of 81.72 TFLOPS versus the Arc Pro B390's 7.680 TFLOPS indicates a 10.6x difference in raw single-precision compute. In FP16, the MI325X delivers 81.72 TFLOPS at a 1:1 ratio, while the Arc Pro B390 delivers 15.36 TFLOPS at a 2:1 ratio, a 5.3x gap. These figures come from the specification sheets, not from executed benchmarks, but they represent the only quantitative comparison available.

Texture rate further separates the two. The MI325X reaches 2,553.6 GTexel/s, while the Arc Pro B390 records 120.0 GTexel/s, a 21.3x difference. Pixel rate tells a different story: the MI325X lists 0 MPixel/s because it has no ROPs and no display pipeline, while the Arc Pro B390 produces 60.00 GPixel/s. This inversion reveals the core design split. The MI325X is a pure compute engine, and the Arc Pro B390 is a graphics-oriented processor that can also do compute.

FAQ

Q: Which GPU has higher FP32 performance?

A: The AMD Instinct MI325X records 81.72 TFLOPS FP32, which is 10.6x the 7.680 TFLOPS of the Intel Arc Pro B390.

Q: Does the Arc Pro B390 support modern graphics APIs?

A: Yes. The Arc Pro B390 lists DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4 support. The MI325X lists N/A for all three APIs.

Q: What memory configurations do the two GPUs use?

A: The MI325X uses 256 GB of HBM3e on an 8192 bit bus with 6.14 TB/s bandwidth. The Arc Pro B390 uses system shared memory with system dependent bandwidth.

Q: Are these GPUs similar in power requirements?

A: No. The MI325X has a 1000 W TDP and suggests a 1400 W PSU, while the Arc Pro B390 has an 80 W TDP and no suggested PSU listed.

Q: Which GPU has more shading units?

A: The MI325X has 19,456 shading units, compared to 1,536 on the Arc Pro B390, a 12.7x difference.

Q: Can either GPU output to a display?

A: The MI325X has no display outputs. The Arc Pro B390 has display outputs listed as "Portable Device Dependent."

Architecture Differences

The MI325X uses the Aqua Vanjaram chip built on CDNA 3.0 architecture. It is fabricated by TSMC on a 5 nm process and contains 153,000 million transistors on a 1017 mm² die. The transistor density works out to 150.4M per mm². The Arc Pro B390 uses the Panther Lake chip built on Xe3-LPG architecture. It is fabricated by Intel on a 3 nm process, with transistor count and die size listed as unknown.

The CDNA 3.0 architecture in the MI325X is designed for data center compute. It has 19,456 shading units, 1,216 TMUs, and zero ROPs. It also has no ray tracing cores and no tensor cores listed. The Xe3-LPG architecture in the Arc Pro B390 is a graphics-oriented design with 1,536 shading units, 48 TMUs, 24 ROPs, and 12 ray tracing cores. This is the only ray tracing hardware present in either product.

The memory architecture differs completely. The MI325X uses 256 GB of HBM3e with an 8192 bit bus, delivering 6.14 TB/s of bandwidth. The Arc Pro B390 uses system shared memory with no dedicated bus width and bandwidth described only as system dependent. The MI325X is an OAM module with no power connectors and no display outputs. The Arc Pro B390 is an integrated graphics processor with no power connectors and display outputs described as portable device dependent.

The API support also marks a major architectural fork. The MI325X lists N/A for DirectX, OpenGL, and Vulkan. The Arc Pro B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The MI325X has a 1000 MHz base clock and a 2100 MHz boost clock, while the Arc Pro B390 has a 300 MHz base clock and a 2500 MHz boost clock. The memory clock on the MI325X is 1500 MHz with 6 Gbps effective speed, while the Arc Pro B390 lists system shared memory.

Specification Differences

The process node differs: the MI325X uses 5 nm from TSMC, the Arc Pro B390 uses 3 nm from Intel. The MI325X has 153,000 million transistors on a 1017 mm² die; the Arc Pro B390 has unknown transistor count and die size. Transistor density is 150.4M per mm² for the MI325X and null for the Arc Pro B390.

The shading unit count is 19,456 versus 1,536. TMUs are 1,216 versus 48. ROPs are 0 versus 24. The MI325X has no ray tracing cores, while the Arc Pro B390 has 12. FP32 compute is 81.72 TFLOPS versus 7.680 TFLOPS. FP16 compute is 81.72 TFLOPS at 1:1 versus 15.36 TFLOPS at 2:1. Texture rate is 2,553.6 GTexel/s versus 120.0 GTexel/s. Pixel rate is 0 MPixel/s versus 60.00 GPixel/s.

Memory size is 256 GB versus system shared. Memory type is HBM3e versus system shared. Bus width is 8192 bit versus system shared. Bandwidth is 6.14 TB/s versus system dependent. TDP is 1000 W versus 80 W. The MI325X suggests a 1400 W PSU, the Arc Pro B390 lists none. Slot width is OAM Module versus IGP. Bus interface is PCIe 5.0 x16 versus IGP. Display outputs are none versus portable device dependent. The MI325X has no API support, while the Arc Pro B390 supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4. The MI325X released on 2024-10-09, the Arc Pro B390 on 2026-01-26. The MI325X lists Radeon Instinct as a predecessor, the Arc Pro B390 lists HD Graphics-WM.

The Verdict

The data identifies two distinct products with no overlapping use cases. The MI325X is a data center compute accelerator with 256 GB of HBM3e, 6.14 TB/s of memory bandwidth, and 81.72 TFLOPS of FP32 compute. It has no display outputs, no graphics API support, and no ROPs. The Arc Pro B390 is an integrated graphics processor with full DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4 support, 24 ROPs, and 12 ray tracing cores. It uses system shared memory and consumes 80 W.

The 1000 W TDP of the MI325X versus the 80 W TDP of the Arc Pro B390 shows a 12.5x power difference. The FP32 compute gap is 10.6x, and the texture rate gap is 21.3x. These are not competing products. The MI325X targets compute workloads that do not require graphics output, and the Arc Pro B390 targets portable devices that need full graphics API support with minimal power draw. The release dates reinforce this: the MI325X launched in October 2024, the Arc Pro B390 in January 2026.

Where Each One Wins

The MI325X wins in every compute-related metric. It has more shading units, more TMUs, higher FP32 and FP16 throughput, and massively higher memory bandwidth. Its 256 GB HBM3e pool with an 8192 bit bus is a data center class memory subsystem. The 2,553.6 GTexel/s texture rate indicates extreme fill-rate potential for compute workloads that use texture sampling. The 1000 W TDP and 1400 W suggested PSU reflect a design that assumes dedicated infrastructure. Its release in October 2024 places it in the current data center generation.

The Arc Pro B390 wins in every graphics and integration metric. It has 24 ROPs where the MI325X has none, and it delivers 60.00 GPixel/s of pixel throughput. The 12 ray tracing cores provide hardware acceleration that the MI325X entirely lacks. Full DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4 support mean it can run modern graphics workloads. The IGP slot width and system shared memory mean it requires no separate memory purchase. The 80 W TDP makes it suitable for portable devices, and its display outputs are described as portable device dependent. The 3 nm process from Intel indicates a modern, power-efficient fabrication node.

The data does not support a single winner. The MI325X is the choice for compute tasks that demand 81.72 TFLOPS of FP32 and 6.14 TB/s of bandwidth. The Arc Pro B390 is the choice for graphics workloads that need ray tracing, pixel output, and API compatibility at 80 W. The 50th percentile ranking for both in the database reflects the absence of shared benchmark data, not equivalence in capability. The recorded specifications make the intended deployment environments explicit: one is an OAM module for a server chassis, the other is an integrated processor for a portable device.