AMD Instinct MI350X vs Intel Arc A380E x2 Comparison

FAQ

Q: What are the core architectural differences between the AMD Instinct MI350X and the Intel Arc A380E x2?

A: The MI350X uses AMD’s CDNA 4.0 architecture on a 3 nm TSMC process with an MI350 256CU chip, while the Arc A380E x2 uses Intel’s Xe-HPG architecture (Alchemist, Arc 3 generation) on a 6 nm TSMC process with a DG2-128 chip. The MI350X is designed for compute acceleration with no display outputs, whereas the Arc A380E x2 includes 8x mini-DisplayPort 2.0 outputs.

Q: How do the memory subsystems compare?

A: The MI350X has 288 GB of HBM3e memory on a 8192-bit bus, delivering 8.19 TB/s bandwidth. The Arc A380E x2 has 6 GB of GDDR6 memory on a 96-bit bus, providing 186.0 GB/s bandwidth. The MI350X’s memory bandwidth is approximately 44 times higher.

Q: What are the raw compute figures for each GPU?

A: The MI350X delivers 72.09 TFLOPS FP32 and 72.09 TFLOPS FP16 (1:1 ratio). The Arc A380E x2 delivers 4.096 TFLOPS FP32 and 8.192 TFLOPS FP16 (2:1 ratio). Thus, the MI350X is about 17.6 times faster in FP32 and about 8.8 times faster in FP16.

Q: How do the physical specifications differ?

A: The MI350X is an OAM module measuring 102 mm by 165 mm, with a 1000 W TDP and no power connectors (relying on the module slot). The Arc A380E x2 is a single-slot card measuring 265 mm by 127 mm by 20 mm, with a 130 W TDP and one 6-pin power connector. The MI350X uses PCIe 5.0 x16, while the Arc A380E x2 uses PCIe 4.0 x8.

Q: Which GPU has higher pixel and texture throughput?

A: The Arc A380E x2 has a pixel rate of 64.00 GPixel/s and a texture rate of 128.0 GTexel/s. The MI350X has a pixel rate of 0 MPixel/s (no ROPs) and a texture rate of 2,252.8 GTexel/s. The MI350X is roughly 17.6 times faster in texture rate, but the Arc A380E x2 is the only one with pixel output capability.

Q: What are the API support differences?

A: The MI350X lists N/A for DirectX, OpenGL, and Vulkan. The Arc A380E x2 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. This makes the Arc A380E x2 suitable for graphics workloads, while the MI350X is compute-focused.

Architecture Differences

The AMD Instinct MI350X and Intel Arc A380E x2 represent fundamentally different design philosophies. The MI350X uses AMD’s CDNA 4.0 architecture, a compute-optimized design with no graphics pipeline. This is evident from its 0 ROPs, 0 MPixel/s pixel rate, and the absence of any display outputs. The chip is built on a 3 nm TSMC process with 185,000 million transistors on a 2380 mm² die, yielding a transistor density of 77.7M per mm². The MI350X features 16,384 shading units, 1,024 TMUs, and a 8192-bit memory bus for HBM3e.

The Intel Arc A380E x2 uses the Xe-HPG architecture within the Alchemist generation (Arc 3). It is built on a 6 nm TSMC process with 7,200 million transistors on a 157 mm² die, for a density of 45.9M per mm². The chip has 1,024 shading units, 64 TMUs, 32 ROPs, and 8 ray tracing cores. Unlike the MI350X, it includes a full graphics feature set with DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4 support, plus 8x mini-DisplayPort 2.0 outputs.

The transistor count difference is stark: the MI350X packs 185,000 million transistors versus 7,200 million for the Arc A380E x2. That is over 25 times more transistors. The die size difference is also large, with the MI350X at 2380 mm² compared to 157 mm². The MI350X’s higher transistor density (77.7M vs 45.9M per mm²) reflects the more advanced 3 nm process.

Clock behavior differs significantly. The MI350X has a base clock of 1000 MHz and a boost clock of 2200 MHz. The Arc A380E x2 runs at a constant 2000 MHz for both base and boost. Memory clocks also differ: the MI350X uses 2000 MHz (8 Gbps effective), while the Arc A380E x2 uses 1937 MHz (15.5 Gbps effective). Despite the higher effective memory clock on the Arc, the MI350X’s vastly wider bus and HBM3e technology deliver far greater bandwidth.

The MI350X has no RT cores listed, while the Arc A380E x2 has 8. The MI350X also lacks any tensor core specification. The power delivery approach is completely different: the MI350X is an OAM module with no power connectors, drawing up to 1000 W through the module interface, whereas the Arc A380E x2 is a single-slot card with a 130 W TDP and one 6-pin connector.

Head-to-Head Benchmarks

The database contains no recorded head-to-head benchmark results for these two GPUs. Both have an avgBenchmarkScore of 0 and an identical percentileVsAllGpus of 50. The winsA and winsB fields are both 0, indicating no direct comparative measurements exist in the database. This absence of data itself is informative: the two products target entirely different segments, and no shared workload has been recorded for both.

However, the specification data allows for meaningful comparisons of theoretical peak performance. In FP32 compute, the MI350X delivers 72.09 TFLOPS versus 4.096 TFLOPS for the Arc A380E x2. That represents a 17.6x advantage for the MI350X. In FP16, the MI350X again delivers 72.09 TFLOPS (1:1 ratio), while the Arc A380E x2 delivers 8.192 TFLOPS (2:1 ratio). Here the MI350X leads by 8.8x.

Texture rate follows a similar pattern. The MI350X achieves 2,252.8 GTexel/s compared to 128.0 GTexel/s for the Arc A380E x2, a 17.6x difference. The pixel rate tells the opposite story: the MI350X has 0 MPixel/s, while the Arc A380E x2 has 64.00 GPixel/s. This reflects the MI350X’s lack of ROPs, making it incapable of traditional rasterization output.

Memory bandwidth shows the largest relative gap. The MI350X provides 8.19 TB/s, while the Arc A380E x2 provides 186.0 GB/s. That is a 44x difference in favor of the MI350X. Memory capacity also favors the MI350X: 288 GB versus 6 GB, a 48x difference.

The Arc A380E x2 holds advantages in certain areas. Its pixel rate of 64.00 GPixel/s is entirely absent on the MI350X. The Arc also has 8 RT cores, which the MI350X lacks entirely. The Arc supports a full API stack (DirectX 12 Ultimate, OpenGL 4.6, Vulkan 1.4), whereas the MI350X lists N/A for all three. The Arc’s display outputs (8x mini-DisplayPort 2.0) are nonexistent on the MI350X.

Power efficiency, when comparing peak compute per watt, favors the Arc A380E x2. The Arc delivers 4.096 TFLOPS FP32 at 130 W, or approximately 0.0315 TFLOPS per watt. The MI350X delivers 72.09 TFLOPS at 1000 W, or approximately 0.0721 TFLOPS per watt. Thus, the MI350X is about 2.3x more efficient in FP32 per watt, despite its much higher absolute power draw.

Specification Differences

| Specification | AMD Instinct MI350X | Intel Arc A380E x2 |

|---|---|---|

| Chip | MI350 256CU | DG2-128 |

| Architecture | CDNA 4.0 | Xe-HPG |

| Generation | Instinct (MIx) | Alchemist (Arc 3) |

| Process Node | 3 nm | 6 nm |

| Transistors | 185,000 million | 7,200 million |

| Die Size | 2380 mm² | 157 mm² |

| Transistor Density | 77.7M / mm² | 45.9M / mm² |

| Base Clock | 1000 MHz | 2000 MHz |

| Boost Clock | 2200 MHz | 2000 MHz |

| Memory Clock | 2000 MHz, 8 Gbps effective | 1937 MHz, 15.5 Gbps effective |

| Memory Size | 288 GB | 6 GB |

| Memory Type | HBM3e | GDDR6 |

| Memory Bus Width | 8192 bit | 96 bit |

| Memory Bandwidth | 8.19 TB/s | 186.0 GB/s |

| Shading Units | 16,384 | 1,024 |

| TMUs | 1,024 | 64 |

| ROPs | 0 | 32 |

| RT Cores | None | 8 |

| Pixel Rate | 0 MPixel/s | 64.00 GPixel/s |

| Texture Rate | 2,252.8 GTexel/s | 128.0 GTexel/s |

| FP32 | 72.09 TFLOPS | 4.096 TFLOPS |

| FP16 | 72.09 TFLOPS (1:1) | 8.192 TFLOPS (2:1) |

| TDP | 1000 W | 130 W |

| Slot Width | OAM Module | Single-slot |

| Power Connectors | None | 1x 6-pin |

| Suggested PSU | 1400 W | 300 W |

| Bus Interface | PCIe 5.0 x16 | PCIe 4.0 x8 |

| Display Outputs | No outputs | 8x mini-DisplayPort 2.0 |

| DirectX | N/A | 12 Ultimate (12_2) |

| OpenGL | N/A | 4.6 |

| Vulkan | N/A | 1.4 |

| Length | 102 mm (4 inches) | 265 mm (10.4 inches) |

| Height | Not listed | 127 mm (5 inches) |

| Width | 165 mm (6.5 inches) | 20 mm (0.8 inches) |

| Production Status | Not listed | End-of-life |

| Release Date | 2025-06-11 | 2024-03-31 |

| Predecessor | Radeon Instinct | Xe Graphics |

| Successor | None listed | Battlemage |

Where Each One Wins

The AMD Instinct MI350X wins decisively in raw compute throughput. Its 72.09 TFLOPS FP32 and FP16 performance dwarfs the Arc A380E x2’s 4.096 and 8.192 TFLOPS respectively. For any workload that stresses dense matrix math, deep learning inference, or scientific simulation, the MI350X’s massive shading unit count (16,384 vs 1,024) and texture rate (2,252.8 GTexel/s vs 128.0 GTexel/s) make it the overwhelmingly stronger candidate. The 288 GB HBM3e memory pool with 8.19 TB/s bandwidth is designed for holding large datasets in memory, such as large language model weights or high-resolution scientific arrays. The MI350X also uses a newer PCIe 5.0 x16 interface versus PCIe 4.0 x8, reducing data transfer bottlenecks.

The Intel Arc A380E x2 wins in graphics-specific capabilities. It has 32 ROPs and a 64.00 GPixel/s pixel rate, which the MI350X lacks entirely (0 MPixel/s). The Arc’s 8 RT cores provide hardware ray tracing support, absent on the MI350X. Its API support (DirectX 12 Ultimate, OpenGL 4.6, Vulkan 1.4) enables standard graphics software stacks, while the MI350X lists N/A for all three. The Arc’s 8x mini-DisplayPort 2.0 outputs allow direct display connectivity, making it suitable for multi-monitor setups or embedded visualization systems. The Arc also has a far lower TDP (130 W vs 1000 W) and a smaller physical footprint (265 mm length, 20 mm width) compared to the MI350X’s OAM module form factor.

The Arc A380E x2’s production status is end-of-life, while the MI350X’s status is not listed. The MI350X was released on 2025-06-11, over a year after the Arc’s 2024-03-31 release. The MI350X’s predecessor is Radeon Instinct, and the Arc’s successor is Battlemage.

The Verdict

The data indicates two products with almost no overlap in intended use. The AMD Instinct MI350X is a compute accelerator with no graphics pipeline, no display outputs, and no consumer API support. Its 72.09 TFLOPS FP32, 288 GB HBM3e, and 8.19 TB/s bandwidth position it for high-performance computing, AI training, and inference workloads where memory capacity and bandwidth are critical. The 1000 W TDP and OAM module form factor require specialized server infrastructure, as indicated by the 1400 W suggested PSU.

The Intel Arc A380E x2 is a compact graphics card with a full feature set: 64.00 GPixel/s pixel rate, 8 RT cores, DirectX 12 Ultimate, OpenGL 4.6, Vulkan 1.4, and 8 display outputs. Its 130 W TDP and single-slot design make it suitable for embedded systems or multi-GPU configurations where space and power are constrained. The 6 GB GDDR6 memory and 186.0 GB/s bandwidth are modest but adequate for graphics rendering tasks.

A user requiring pure compute throughput should select the MI350X, as its FP32 performance is 17.6x higher and its memory bandwidth is 44x higher than the Arc A380E x2. A user requiring graphics output, ray tracing, or standard API compatibility should select the Arc A380E x2, as the MI350X offers none of these features. The MI350X’s 3 nm process and 185,000 million transistors show a design aimed at maximum throughput, while the Arc A380E x2’s 6 nm process and 7,200 million transistors show a design aimed at balanced functionality within a low power envelope. There is no benchmark data to suggest either GPU outperforms the other in a shared workload, because no shared workload exists in the database. The choice comes down to compute versus graphics, and the specifications make that distinction unambiguous.