AMD Instinct MI350X vs Intel Arc Pro B370 Comparison

# AMD Instinct MI350X vs Intel Arc Pro B370

The AMD Instinct MI350X and Intel Arc Pro B370 occupy entirely different segments of the GPU market, and the recorded data reflects this clearly. The MI350X is a 1000 W OAM module designed for compute workloads, built on AMD's CDNA 4.0 architecture with 16,384 shading units. The Arc Pro B370 is a 25 W integrated graphics processor within Intel's Panther Lake chip, using Xe3-LPG architecture with 1,280 shading units. Both products sit at the 50th percentile in the database's GPU rankings, but with zero recorded head-to-head benchmarks and no average benchmark scores, the comparison relies on specifications and architectural characteristics rather than direct performance measurements.

Where Each One Wins

The AMD Instinct MI350X wins decisively in raw compute throughput. Its FP32 output of 72.09 TFLOPS dwarfs the Arc Pro B370's 6.144 TFLOPS, a difference of roughly 11.7 times. The MI350X also delivers 2,252.8 GTexel/s texture fill rate against the B370's 96.00 GTexel/s. Memory capacity and bandwidth follow the same pattern: 288 GB of HBM3e with 8.19 TB/s bandwidth versus system-shared memory with system-dependent bandwidth. Any workload that scales with raw floating-point operations, texture throughput, or memory bandwidth favors the MI350X.

The Intel Arc Pro B370 wins in integration and power efficiency. Its 25 W TDP is 40 times lower than the MI350X's 1000 W. The B370 is an IGP with no external power connectors, while the MI350X requires a 1400 W suggested power supply. The B370 also provides display outputs, described as portable device dependent, whereas the MI350X has no display outputs at all. The B370 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4; the MI350X lists no API support in the database. For graphics rendering and portable integration, the B370 is the functional choice.

FAQ

Q: Which GPU has higher FP32 compute performance?

A: The AMD Instinct MI350X delivers 72.09 TFLOPS FP32, while the Intel Arc Pro B370 delivers 6.144 TFLOPS. The MI350X is approximately 11.7 times faster in this metric.

Q: How do their memory configurations differ?

A: The MI350X has 288 GB of HBM3e memory on an 8192-bit bus with 8.19 TB/s of bandwidth. The Arc Pro B370 uses system-shared memory with a system-shared bus and system-dependent bandwidth.

Q: What is the power requirement for each?

A: The MI350X has a 1000 W TDP and a suggested PSU of 1400 W. The Arc Pro B370 has a 25 W TDP and no suggested PSU listed, consistent with its IGP form factor.

Q: Can either GPU output to a display?

A: The Arc Pro B370 has display outputs described as portable device dependent. The MI350X has no display outputs.

Q: What process nodes and foundries are used?

A: Both use 3 nm process technology. The MI350X is fabricated by TSMC; the Arc Pro B370 is fabricated by Intel.

Q: Which GPU supports modern graphics APIs?

A: The Arc Pro B370 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The MI350X lists no API support in the database.

Head-to-Head Benchmarks

Direct benchmark comparisons between the two are absent from the database, so the analysis falls to specification-level differences. The largest single gap is in FP32 throughput. The MI350X's 72.09 TFLOPS is exactly 11.73 times the B370's 6.144 TFLOPS. Texture rate tells a similar story: 2,252.8 GTexel/s divided by 96.00 GTexel/s yields a 23.47 times advantage for the MI350X. Pixel rate inverts the comparison: the B370 produces 48.00 GPixel/s, while the MI350X records 0 MPixel/s, reflecting its lack of a raster output stage.

Memory bandwidth represents perhaps the most extreme difference. The MI350X's 8.19 TB/s is a fixed hardware specification. The B370's bandwidth is system dependent, meaning it inherits whatever memory bandwidth the host platform provides. In no plausible system configuration would shared system memory approach 8.19 TB/s. The B370's memory clock is likewise system shared, while the MI350X runs its HBM3e at 2000 MHz with 8 Gbps effective. Clock speeds for the compute cores are closer: the MI350X boosts to 2200 MHz, the B370 to 2400 MHz. The B370 actually holds a 200 MHz boost advantage, though with far fewer shaders.

The TDP difference is the most operationally significant number for deployment decisions. The MI350X consumes 1000 W, the B370 only 25 W. This 40 times power gap means the MI350X requires a 1400 W suggested PSU, while the B370's suggested PSU field is null. A system housing the MI350X also needs an OAM module slot, whereas the B370 is an IGP integrated directly into the processor package.

Specification Differences

The MI350X and B370 differ in nearly every measurable specification. Shading units: 16,384 versus 1,280. Texture mapping units: 1,024 versus 40. Raster output units: 0 versus 20. Ray tracing cores: none listed for the MI350X, 10 for the B370. FP16 throughput: the MI350X achieves 72.09 TFLOPS at a 1:1 ratio with FP32; the B370 achieves 12.29 TFLOPS at a 2:1 ratio, meaning its FP16 rate is double its FP32 rate.

Physical dimensions differ as well. The MI350X measures 102 mm in length and 165 mm in width. The B370 has no listed dimensions, consistent with an IGP that occupies no separate board space. The MI350X uses a PCIe 5.0 x16 bus interface; the B370's interface is IGP. Neither product has a launch MSRP in the database. The MI350X's production status is not listed, while the B370's status is active. Release dates place the MI350X on June 11, 2025, and the B370 on January 26, 2026.

Architecture Differences

The MI350X uses AMD's CDNA 4.0 architecture, specifically the MI350 256CU chip. It contains 185,000 million transistors on a 2380 mm² die, yielding a transistor density of 77.7 million per square millimeter. This is a massive compute-oriented die fabricated by TSMC. The B370 uses Intel's Xe3-LPG architecture on the Panther Lake chip, with transistor count and die size listed as unknown. Its TDP of 25 W indicates a power-efficient integrated design.

The MI350X belongs to the Instinct (MIx) generation and succeeds the Radeon Instinct product line. The B370 belongs to the Arc Graphics-WM (Panther Lake) generation and succeeds HD Graphics-WM. The MI350X's API support fields are all N/A, confirming its role as a compute accelerator rather than a graphics device. The B370, by contrast, supports the full modern graphics stack including DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4.

Memory architecture separates the two fundamentally. The MI350X uses dedicated HBM3e with an 8192-bit bus. The B370 uses system-shared memory, meaning it has no dedicated VRAM and depends on the host's main memory for all data storage. This makes the B370's memory performance entirely dependent on the surrounding platform, while the MI350X's 8.19 TB/s is fixed and self-contained. The MI350X has no display outputs, reinforcing its compute-only design; the B370 is explicitly designed to drive displays in portable devices.

The Verdict

The data separates these products into distinct roles with no meaningful overlap. The MI350X is a compute accelerator with 72.09 TFLOPS FP32, 288 GB of HBM3e, and 8.19 TB/s bandwidth, packaged as a 1000 W OAM module with no display support. The B370 is a 25 W integrated GPU with 6.144 TFLOPS FP32, 10 ray tracing cores, full modern graphics API support, and portable device display outputs. The MI350X's FP32 output is 11.73 times higher, its texture rate is 23.47 times higher, and its memory bandwidth is fixed at 8.19 TB/s versus the B370's system-dependent figure.

The B370's boost clock of 2400 MHz exceeds the MI350X's 2200 MHz, and its pixel rate of 48.00 GPixel/s contrasts with the MI350X's 0 MPixel/s. These numbers confirm that the B370 is optimized for graphics output and integration, while the MI350X is optimized for raw compute throughput. The MI350X records no API support, no display outputs, and no raster operations; the B370 records all three. The production status of the MI350X is unlisted, while the B370 is active.

Organizations with compute-heavy workloads requiring massive memory capacity and bandwidth would align with the MI350X. Systems requiring graphics output, modern API support, and minimal power draw would align with the B370. The 40 times power difference and the 11.73 times FP32 difference define the boundaries of each product's intended use case. No benchmark scores exist in the database to refine this comparison further, so the specification data stands as the definitive record.