AMD Radeon Instinct MI300X vs Intel Arc B390 Comparison

Head-to-Head Benchmarks

The database contains no direct head-to-head benchmark comparison between the AMD Radeon Instinct MI300X and the Intel Arc B390. The two GPUs target fundamentally different market segments, and the recorded data reflects this divergence.

For the Intel Arc B390, the database includes a single benchmark result. In the 3DMark Steel Nomad DX12 test, the Arc B390 scores 1482 points. This places the integrated GPU at the 9th percentile among all GPUs in the database. Its nearest recorded rivals are all older NVIDIA parts. The Arc B390 sits 1.3% ahead of the GeForce GT 520MX (1463 points), 1.5% ahead of the GeForce 800M (1460 points), 2.5% ahead of the GeForce GT 625 OEM (1446 points), and 2.7% ahead of the GeForce GT 710 (1443 points). These margins are narrow, indicating that the Arc B390 performs at the level of entry-level discrete graphics from several generations ago.

The AMD Radeon Instinct MI300X has no benchmark scores recorded in the database. Its average benchmark score is listed as 0, and it has no nearest rivals. The percentile field shows the MI300X at the 50th percentile, but this value is not derived from any actual benchmark run. The absence of data means the MI300X cannot be compared directly to the Arc B390 in any synthetic test within the database.

The compute capabilities of the two parts, however, are documented in the specification fields. The MI300X delivers 81.72 TFLOPS of FP32 compute, while the Arc B390 delivers 7.680 TFLOPS. In FP16, the MI300X reaches 653.7 TFLOPS using an 8:1 ratio, while the Arc B390 manages 15.36 TFLOPS using a 2:1 ratio. The MI300X therefore offers roughly 10.6 times the FP32 throughput and 42.6 times the FP16 throughput of the Arc B390. These are theoretical peak rates, not measured benchmark scores, but they indicate the scale of the performance gap.

Memory bandwidth shows a similar disparity. The MI300X has 192 GB of HBM3 memory on a 8192-bit bus, delivering 10.3 TB/s of bandwidth. The Arc B390 uses system shared memory, with bandwidth listed as system dependent. The MI300X also has a vastly larger memory pool, 192 GB versus shared system memory for the Arc B390, though the shared configuration means the Intel part can draw on whatever system RAM is available.

The pixel and texture rates also differ sharply. The MI300X lists a pixel rate of 0 MPixel/s, as it has no ROPs and no display outputs, while the Arc B390 has 24 ROPs and a pixel rate of 60.00 GPixel/s. The texture rate for the MI300X is 2,553.6 GTexel/s, compared to 120.0 GTexel/s for the Arc B390. The MI300X has 19,456 shading units and 1,216 TMUs, while the Arc B390 has 1,536 shading units, 48 TMUs, and 12 ray tracing cores.

The Verdict

The data shows two products with no overlapping use cases. The AMD Radeon Instinct MI300X is a compute accelerator for data center workloads. It has no display outputs, no RT cores, no recorded benchmark scores, and a 750 W TDP. It uses an OAM module slot and requires a 1150 W suggested power supply. Its memory configuration, 192 GB of HBM3 with 10.3 TB/s bandwidth, targets large model inference and training tasks.

The Intel Arc B390 is an integrated GPU within the Panther Lake processor. It has a 80 W TDP, is classified as IGP, and supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. It has display outputs described as portable device dependent. Its single benchmark result places it in the lowest performance tier of the database, at the 9th percentile, competing with GPUs from over a decade ago.

Who should pick which depends entirely on the workload. The MI300X suits anyone running compute-heavy tasks that require massive memory bandwidth and FP32 or FP16 throughput, such as large-scale AI inference or scientific simulation. The Arc B390 suits portable devices needing basic graphics output and light 3D acceleration, with support for modern API features like ray tracing and DirectX 12 Ultimate.

The MI300X has zero wins in the head-to-head field, and the Arc B390 also has zero wins. This is not a competitive matchup. The MI300X does not even have display outputs, making it unusable for standard desktop graphics tasks. The Arc B390, conversely, lacks the memory bandwidth and compute density for any serious data center work.

FAQ

Q: Which GPU has higher FP32 compute performance?

A: The AMD Radeon Instinct MI300X delivers 81.72 TFLOPS of FP32 compute, compared to 7.680 TFLOPS for the Intel Arc B390.

Q: What memory configuration does each GPU use?

A: The MI300X has 192 GB of HBM3 memory on a 8192-bit bus with 10.3 TB/s bandwidth. The Arc B390 uses system shared memory, with system dependent bandwidth.

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

A: Yes, the Arc B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The MI300X has no listed API support for DirectX, OpenGL, or Vulkan.

Q: What is the power consumption of each GPU?

A: The MI300X has a TDP of 750 W and a suggested power supply of 1150 W. The Arc B390 has a TDP of 80 W and no suggested power supply listed.

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

A: The Arc B390 scores 1482 in 3DMark Steel Nomad DX12, which is 1.3% ahead of the GeForce GT 520MX, 1.5% ahead of the GeForce 800M, 2.5% ahead of the GeForce GT 625 OEM, and 2.7% ahead of the GeForce GT 710.

Q: Does the MI300X have any display outputs?

A: No, the MI300X has no display outputs. The Arc B390 has display outputs described as portable device dependent.

Specification Differences

The two GPUs differ in nearly every specification field. The AMD Radeon Instinct MI300X uses a 5 nm process node from TSMC, with 153,000 million transistors on a 1017 mm² die. The Intel Arc B390 uses a 3 nm process node from Intel, with transistor count and die size listed as unknown.

Clock speeds differ significantly. The MI300X has a base clock of 1000 MHz and a boost clock of 2100 MHz. The Arc B390 has a base clock of 300 MHz and a boost clock of 2500 MHz. The memory clock for the MI300X is 2525 MHz, which equates to 10.1 Gbps effective. The Arc B390's memory clock is listed as system shared.

Memory specifications are entirely different. The MI300X has 192 GB of HBM3 with a 8192-bit bus and 10.3 TB/s bandwidth. The Arc B390 has system shared memory, system shared type, system shared bus width, and system dependent bandwidth.

The compute unit counts diverge sharply. The MI300X has 19,456 shading units, 1,216 TMUs, and 0 ROPs. The Arc B390 has 1,536 shading units, 48 TMUs, 24 ROPs, and 12 ray tracing cores. The MI300X has no RT cores listed.

Rates and outputs differ as well. The MI300X has a pixel rate of 0 MPixel/s and a texture rate of 2,553.6 GTexel/s. The Arc B390 has a pixel rate of 60.00 GPixel/s and a texture rate of 120.0 GTexel/s. The MI300X has no display outputs, while the Arc B390 has outputs described as portable device dependent.

Power and form factor differ completely. The MI300X has a 750 W TDP, uses an OAM Module slot, has no power connectors, and requires a 1150 W suggested power supply. The Arc B390 has an 80 W TDP, uses an IGP slot, has no power connectors, and has no suggested power supply. The MI300X uses a PCIe 5.0 x16 bus interface, while the Arc B390 uses IGP.

Release dates differ by over two years. The MI300X was released on December 5, 2023. The Arc B390 has a release date of January 26, 2026. The MI300X lists its predecessor as FirePro Data Center, with no successor. The Arc B390 has no predecessor and is marked as active production status. The MI300X has no production status listed.

Architecture Differences

The AMD Radeon Instinct MI300X is built on CDNA 3.0 architecture, using the Aqua Vanjaram chip. This is a compute-focused architecture designed for data center acceleration. The Intel Arc B390 uses Xe3-LPG architecture, based on the Panther Lake chip. Xe3-LPG is a low-power graphics architecture intended for integrated use in mobile processors.

The manufacturing processes reflect this divergence. The MI300X uses a 5 nm process at TSMC, resulting in 153,000 million transistors on a 1017 mm² die. The Arc B390 uses a 3 nm process at Intel, with transistor count and die size unknown. The transistor density for the MI300X is 150.4 million transistors per square millimeter, while no density figure exists for the Arc B390.

The core configurations are built for different purposes. The MI300X has 19,456 shading units and 1,216 TMUs, but zero ROPs. This is characteristic of a compute accelerator that does not render pixels. The Arc B390 has 1,536 shading units, 48 TMUs, and 24 ROPs, plus 12 ray tracing cores, making it a complete graphics solution for rendering tasks.

Memory architecture also differs fundamentally. The MI300X uses 192 GB of HBM3 on a 8192-bit bus, achieving 10.3 TB/s bandwidth. This is a dedicated, high-bandwidth memory pool. The Arc B390 has no dedicated memory and instead uses system shared memory, with bandwidth dependent on the host system.

The feature sets are tailored to their roles. The MI300X has no listed API support for DirectX, OpenGL, or Vulkan, and no display outputs. It is not designed for graphics rendering. The Arc B390 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, and has display outputs for portable devices.

The MI300X's FP16 performance of 653.7 TFLOPS uses an 8:1 ratio relative to FP32, indicating a strong bias toward reduced-precision compute workloads. The Arc B390's FP16 performance of 15.36 TFLOPS uses a 2:1 ratio, which is typical for graphics-oriented architectures that need balanced precision for shading and rendering.

Where Each One Wins

The AMD Radeon Instinct MI300X wins decisively in raw compute throughput. Its FP32 output of 81.72 TFLOPS and FP16 output of 653.7 TFLOPS dwarf the Arc B390's 7.680 TFLOPS and 15.36 TFLOPS. For any workload that depends on massive parallel floating-point calculations, such as large language model training or scientific simulation, the MI300X is the clear choice.

The MI300X also wins on memory capacity and bandwidth. With 192 GB of HBM3 and 10.3 TB/s bandwidth, it can hold and process datasets that would be impossible for the Arc B390's system shared memory. The Arc B390's memory bandwidth is system dependent, meaning it is constrained by the host platform's memory subsystem.

The Intel Arc B390 wins on graphics features. It has 12 ray tracing cores, 24 ROPs, and a pixel rate of 60.00 GPixel/s. The MI300X has zero ROPs and a pixel rate of 0 MPixel/s. The Arc B390 also supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, while the MI300X lists no graphics API support.

The Arc B390 wins on power efficiency and form factor. Its 80 W TDP is a fraction of the MI300X's 750 W TDP. The Arc B390 uses an IGP slot, making it suitable for integration into portable devices. The MI300X requires an OAM Module slot and a 1150 W suggested power supply, restricting it to data center racks.

The Arc B390 also wins on measured benchmark performance, as it has a recorded 3DMark Steel Nomad DX12 score of 1482. The MI300X has no benchmark scores in the database. However, this is a function of testing methodology rather than capability, as the MI300X targets a different workload class.

For display output, the Arc B390 wins by having outputs, described as portable device dependent. The MI300X has no display outputs at all. Anyone needing to connect a monitor must choose the Arc B390.

The MI300X wins on transistor count and die size, with 153,000 million transistors on a 1017 mm² die. The Arc B390's figures are unknown. The MI300X also has a larger memory bus at 8192 bits, compared to system shared for the Arc B390.

In summary, the MI300X wins for compute density, memory capacity, and bandwidth. The Arc B390 wins for graphics rendering, API support, power efficiency, and portability. The data does not support using either GPU for the other's intended purpose.