AMD Radeon Instinct MI308X vs Intel Arc B390 Comparison

AMD Radeon Instinct MI308X and Intel Arc B390 occupy completely different segments of the graphics hardware spectrum. The MI308X is a data center compute module built on CDNA 3.0, while the Arc B390 is an integrated graphics processor for mobile platforms. The recorded data shows the Arc B390 has a single benchmark entry, scoring 1482 in 3DMark Steel Nomad DX12, placing it at the 9th percentile of all GPUs. The MI308X has no benchmark entries in the database and sits at the 50th percentile by default. This comparison is defined by compute capacity, memory architecture, and intended deployment rather than direct performance parity.

Where Each One Wins

The AMD Radeon Instinct MI308X wins in every measure of raw compute throughput. Its FP32 output reaches 81.72 TFLOPS, which is more than ten times the Arc B390's 7.680 TFLOPS. In FP16 workloads, the gap widens dramatically: the MI308X delivers 653.7 TFLOPS using an 8:1 ratio, while the Arc B390 produces 15.36 TFLOPS at 2:1. The MI308X also dominates texture processing with 2,553.6 GTexel/s versus 120.0 GTexel/s, a difference of over 20 times. Memory bandwidth tells a similar story, with the MI308X offering 10.3 TB/s of dedicated HBM3 bandwidth, whereas the Arc B390 relies on system-shared memory with bandwidth described as system dependent.

The Intel Arc B390 wins in areas that matter for integrated graphics deployment. It has a 2500 MHz boost clock compared to 2100 MHz on the MI308X, and its pixel rate of 60.00 GPixel/s exceeds the MI308X's 0 MPixel/s, which reflects the latter's absence of display outputs. The Arc B390 also supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, while the MI308X lists no API support in the database. Power consumption heavily favors Intel: the Arc B390 uses 80 W, while the MI308X requires 750 W and a suggested 1150 W power supply. The Arc B390 is an IGP with no power connectors, while the MI308X is an OAM module, also with no power connectors but requiring external power delivery through its module interface.

FAQ

Q: Which GPU has higher FP32 compute performance?

A: The AMD Radeon Instinct MI308X delivers 81.72 TFLOPS, compared to 7.680 TFLOPS for the Intel Arc B390. This makes the MI308X roughly 10.6 times faster in single-precision floating-point workloads.

Q: How does memory capacity compare between the two?

A: The MI308X has 192 GB of HBM3 memory on a 8192-bit bus, providing 10.3 TB/s of bandwidth. The Arc B390 uses system-shared memory with no dedicated capacity, bus width, or bandwidth figures listed; its memory performance is system dependent.

Q: What is the power requirement for each GPU?

A: The MI308X has a 750 W TDP and a suggested power supply of 1150 W. The Arc B390 has an 80 W TDP and no suggested power supply listed, consistent with its integrated design.

Q: Does the Arc B390 support modern graphics APIs?

A: Yes, the database lists DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4 support for the Arc B390. The MI308X has no API support entries recorded.

Q: What benchmark scores are available for these GPUs?

A: The Arc B390 scored 1482 in 3DMark Steel Nomad DX12, placing it at the 9th percentile of all GPUs. The MI308X has no benchmark scores recorded in the database.

Q: Which GPU has display outputs?

A: The Arc B390 provides outputs that are portable device dependent, while the MI308X has no display outputs at all. The MI308X is designed for compute workloads, not rendering to a screen.

Head-to-Head Benchmarks

No direct head-to-head benchmark entries exist in the database for these two products. The only recorded performance measurement is the Arc B390's 3DMark Steel Nomad DX12 score of 1482. That score places it 1.3% ahead of the NVIDIA GeForce GT 520MX, 1.5% ahead of the NVIDIA GeForce 800M, 2.5% ahead of the NVIDIA GeForce GT 625 OEM, and 2.7% ahead of the NVIDIA GeForce GT 710. These nearest rivals all sit within a narrow band of scores from 1443 to 1463, indicating the Arc B390's performance class is entry-level. The MI308X has no comparable benchmark data, so its percentile of 50 reflects an unmeasured default rather than demonstrated performance.

The compute figures provide the only meaningful cross-product comparison. The MI308X's FP32 throughput of 81.72 TFLOPS versus 7.680 TFLOPS for the Arc B390 shows a difference of approximately 10.6 times. In FP16, the MI308X's 653.7 TFLOPS is about 42.5 times the Arc B390's 15.36 TFLOPS, though the ratios differ (8:1 versus 2:1), so this comparison is not directly proportional. Texture rate differs by roughly 21.3 times in favor of AMD. The Arc B390's pixel rate of 60.00 GPixel/s stands against the MI308X's 0 MPixel/s, confirming that the Intel part has rasterization capability while the AMD part does not.

The power efficiency comparison is stark. The Arc B390 produces 7.680 TFLOPS at 80 W, while the MI308X produces 81.72 TFLOPS at 750 W. Per watt, the Arc B390 delivers 0.096 TFLOPS/W, and the MI308X delivers 0.109 TFLOPS/W, a modest advantage for the AMD module. However, the MI308X requires a 1150 W suggested power supply, and its OAM form factor means it cannot be installed in a standard desktop slot. The Arc B390's IGP design consumes no additional board space and uses system memory, making it suitable for compact mobile devices.

Specification Differences

The two products differ in nearly every specification category. The MI308X uses a 5 nm TSMC process with 153,000 million transistors on a 1017 mm² die, achieving a transistor density of 150.4 million per mm². The Arc B390 uses a 3 nm Intel process with unknown transistor count and die size. Clock speeds differ: the MI308X runs at 1000 MHz base and 2100 MHz boost, while the Arc B390 runs at 300 MHz base and 2500 MHz boost. Memory is the largest divergence: the MI308X has 192 GB of HBM3 on a 8192-bit bus with 10.3 TB/s bandwidth at 2525 MHz (10.1 Gbps effective), while the Arc B390 uses system-shared memory with no dedicated specifications.

The compute unit counts are also far apart. The MI308X has 19,456 shading units, 1,216 TMUs, and 0 ROPs, with no RT cores or tensor cores listed. The Arc B390 has 1,536 shading units, 48 TMUs, 24 ROPs, and 12 RT cores, with no tensor cores listed. The MI308X's pixel rate is 0 MPixel/s, while the Arc B390 manages 60.00 GPixel/s. The MI308X supports PCIe 5.0 x16, while the Arc B390 uses an IGP bus interface. The MI308X is an OAM module with no display outputs, while the Arc B390 is an IGP with portable device dependent outputs. The MI308X was released on 2023-12-05, and the Arc B390 on 2026-01-26. The MI308X's predecessor is listed as FirePro Data Center, while the Arc B390 has no predecessor. The Arc B390's production status is active, while the MI308X has no production status recorded. Neither product has a launch MSRP in the database.

Architecture Differences

The MI308X is built on CDNA 3.0 architecture with the Aqua Vanjaram chip, part of the Radeon Instinct (MIx) generation. CDNA is AMD's compute-focused architecture line, designed for data center acceleration. The Arc B390 uses Xe3-LPG architecture with the Panther Lake chip, part of the Arc Graphics-M (Panther Lake) generation. Xe3-LPG is Intel's low-power graphics architecture for integrated use. The MI308X relies on TSMC's 5 nm process, while the Arc B390 uses Intel's 3 nm process. The MI308X has no RT cores listed, but the Arc B390 includes 12 RT cores for hardware ray tracing. The MI308X's FP16 compute is rated at 653.7 TFLOPS with an 8:1 ratio, meaning it trades FP32 throughput for FP16 at a high multiplier. The Arc B390's FP16 is 15.36 TFLOPS at a 2:1 ratio, indicating a more balanced relationship between FP32 and FP16. The MI308X's memory architecture uses HBM3 with a massive 8192-bit bus, while the Arc B390 has no dedicated memory interface. The MI308X has no display outputs, consistent with a compute accelerator, while the Arc B390 integrates display capabilities for portable devices. The MI308X is a 750 W OAM module, while the Arc B390 is an 80 W IGP. The MI308X has no API support listed, while the Arc B390 supports DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4.

The Verdict

The data defines two separate products with no overlap in intended use. The AMD Radeon Instinct MI308X is a data center compute module with 192 GB of HBM3, 81.72 TFLOPS FP32, and 653.7 TFLOPS FP16, designed for workloads that demand massive memory bandwidth and compute throughput. Its 750 W TDP and OAM form factor place it in server racks, not desktops. The Intel Arc B390 is an integrated GPU for mobile devices, with 7.680 TFLOPS FP32, 15.36 TFLOPS FP16, 12 RT cores, and a 60.00 GPixel/s pixel rate. Its 80 W TDP and IGP form factor target portable systems where power efficiency and small footprint matter.

Anyone selecting a GPU for high-performance compute should choose the MI308X based on its overwhelming compute and memory advantages. Anyone building a compact mobile device with integrated graphics should choose the Arc B390, which offers modern API support, hardware ray tracing, and display outputs. The Arc B390's benchmark score of 1482 places it near the NVIDIA GeForce GT 520MX and GeForce 800M, confirming entry-level performance. The MI308X has no benchmark scores, so its percentile ranking cannot be validated through recorded measurements. The specification sheet, however, leaves no ambiguity about which product delivers more raw compute capability.