AMD Radeon Instinct MI25 vs Intel Arc A580 Comparison

Head-to-Head Benchmarks

The only directly comparable benchmark between the AMD Radeon Instinct MI25 and the Intel Arc A580 is Geekbench OpenCL, and the result is decisive. The Intel Arc A580 scores 91,657, while the AMD Radeon Instinct MI25 scores 68,562. That puts Intel ahead by 25.2%, a substantial margin in raw compute throughput for this particular workload. The MI25 trails the Arc A580 by that same 25.2% delta, meaning the gap is not marginal but a clear tier separation in OpenCL performance.

Looking at the MI25’s position among its nearest rivals, the data shows it sits just below a cluster of similarly performing cards. Its score of 68,562 is only 0.4% behind the Intel Arc A770 (68,809), 0.6% behind the NVIDIA CMP 90HX (69,000), 1.9% behind the AMD Radeon Pro WX 8200 (69,870), and 2.0% behind the NVIDIA Quadro P6000 (69,986). These are tight margins, all within roughly 2% of each other, indicating that the MI25 is competitive with that specific group despite its older architecture. The percentile ranking of 90 for the MI25 means it outperforms 90% of all GPUs in the database, a strong showing for a 2017-era accelerator.

The Arc A580, by contrast, sits in the 87th percentile across all GPUs, which is slightly lower than the MI25’s 90th percentile. However, this percentile is based on its average benchmark score of 57,756, which is dragged down by its other benchmark results. The Arc A580 also has a 3DMark Steel Nomad DX12 score of 2,229 and a Geekbench Vulkan score of 79,381, both of which are not available for the MI25, so a direct comparison there is not possible from the recorded data. The average benchmark score of 57,756 for the Arc A580 is actually lower than the MI25’s average of 68,562, but this is misleading because the A580’s average includes multiple tests, while the MI25 only has the single OpenCL result. The head-to-head OpenCL comparison remains the cleanest apples-to-apples metric, and there Intel wins outright.

The win count in the head-to-head table shows Intel taking 1 win and AMD taking 0, which aligns with the OpenCL delta of 25.2%. There is no benchmark in the database where the MI25 comes out ahead of the Arc A580. That said, the MI25’s nearest rival data shows it is no slouch; it is within 2% of cards like the Quadro P6000, which is a professional workstation GPU. The Arc A580’s nearest rivals include the AMD Radeon RX 5600 OEM (58,085 average, 0.6% behind), the AMD Radeon RX 9070 GRE (57,367 average, 0.7% ahead), the Intel Arc A570M (58,239 average, 0.8% behind), and the AMD Radeon RX 6950 XT (58,392 average, 1.1% behind). These deltas are all small, meaning the A580’s average score places it in a tightly packed mid-range group, but its OpenCL score alone would place it far above that group.

FAQ

Q: Which GPU wins the Geekbench OpenCL benchmark?

A: The Intel Arc A580 wins with a score of 91,657 versus the AMD Radeon Instinct MI25’s 68,562, a delta of 25.2% in Intel’s favor.

Q: How does the MI25 compare to its closest rivals?

A: The MI25’s score of 68,562 is 0.4% behind the Intel Arc A770 (68,809), 0.6% behind the NVIDIA CMP 90HX (69,000), 1.9% behind the AMD Radeon Pro WX 8200 (69,870), and 2.0% behind the NVIDIA Quadro P6000 (69,986).

Q: What is the Arc A580’s standing among its nearest rivals?

A: The Arc A580’s average benchmark score of 57,756 is 0.6% behind the AMD Radeon RX 5600 OEM (58,085), 0.7% ahead of the AMD Radeon RX 9070 GRE (57,367), 0.8% behind the Intel Arc A570M (58,239), and 1.1% behind the AMD Radeon RX 6950 XT (58,392).

Q: Does the MI25 have any benchmark scores other than OpenCL?

A: No, the database contains only one recorded benchmark for the MI25, which is the Geekbench OpenCL score of 68,562. The Arc A580 has three recorded benchmarks: OpenCL, Vulkan, and 3DMark Steel Nomad DX12.

Q: Which GPU has the higher percentile ranking?

A: The MI25 ranks in the 90th percentile of all GPUs, while the Arc A580 ranks in the 87th percentile. This is based on their respective average benchmark scores.

Q: What is the Arc A580’s Vulkan score compared to its OpenCL score?

A: The Arc A580’s Geekbench Vulkan score is 79,381, which is notably lower than its OpenCL score of 91,657. The MI25 has no Vulkan score recorded.

Architecture Differences

The two GPUs come from entirely different design philosophies and manufacturing processes. The AMD Radeon Instinct MI25 uses the Vega 10 chip built on GCN 5.0 architecture, fabricated on a 14 nm process by GlobalFoundries. It packs 12,500 million transistors on a die size of 495 mm², resulting in a transistor density of 25.3 million per mm². The Intel Arc A580, by contrast, uses the DG2-512 chip based on Xe-HPG architecture, built on a 6 nm process by TSMC. It contains 21,700 million transistors on a smaller die of 406 mm², giving it a much higher transistor density of 53.4 million per mm². This density difference reflects the newer manufacturing node and the architectural efficiency gains Intel achieved with Xe-HPG.

The MI25 is configured with 4,096 shading units, 256 texture mapping units, and 64 render output units. It has no dedicated ray tracing cores and no tensor cores listed. The Arc A580 has fewer shading units at 3,072, fewer TMUs at 192, but more ROPs at 96. It also includes 24 ray tracing cores, a feature entirely absent from the MI25. The absence of ray tracing hardware on the MI25 is a significant architectural gap, as the Arc A580 can handle hardware-accelerated ray tracing workloads while the MI25 cannot.

The memory subsystems also differ fundamentally. The MI25 uses 16 GB of HBM2 memory on a 2048-bit bus, delivering a bandwidth of 436.2 GB/s. The Arc A580 uses 8 GB of GDDR6 memory on a 256-bit bus, but achieves higher bandwidth at 512.0 GB/s. The MI25’s HBM2 offers twice the capacity and a much wider bus, but the Arc A580’s faster GDDR6 clocks compensate, resulting in higher overall bandwidth. The MI25’s memory clock is listed at 852 MHz with 1704 Mbps effective, while the Arc A580’s memory runs at 2000 MHz with 16 Gbps effective. The Arc A580’s memory speed advantage is a key factor in its higher bandwidth figure.

The MI25’s compute rates are identical to the Arc A580 in raw terms: both deliver 12.29 TFLOPS FP32 and 24.58 TFLOPS FP16 (2:1). The pixel rate differs, however, with the MI25 at 96.00 GPixel/s and the Arc A580 at 192.0 GPixel/s, a doubling that reflects the A580’s higher ROP count and clock speeds. The texture rate is the same for both at 384.0 GTexel/s, despite the MI25 having more TMUs, because the Arc A580’s higher boost clock compensates. The MI25’s base clock is 1400 MHz with a boost of 1500 MHz, while the Arc A580 runs at 1700 MHz base and 2000 MHz boost, giving Intel a substantial clock speed advantage.

Specification Differences

The process node differs: the MI25 uses 14 nm from GlobalFoundries, while the Arc A580 uses 6 nm from TSMC. Transistor counts are 12,500 million for AMD and 21,700 million for Intel. Die size is 495 mm² for AMD versus 406 mm² for Intel. Transistor density is 25.3M per mm² versus 53.4M per mm². The MI25’s base clock is 1400 MHz versus 1700 MHz for the Arc A580, and boost clocks are 1500 MHz versus 2000 MHz. Memory clocks differ substantially: the MI25 runs at 852 MHz (1704 Mbps effective), while the Arc A580 runs at 2000 MHz (16 Gbps effective).

Memory capacity is 16 GB for the MI25 versus 8 GB for the Arc A580. Memory type is HBM2 for AMD and GDDR6 for Intel. Bus width is 2048-bit for AMD and 256-bit for Intel. Bandwidth is 436.2 GB/s for AMD and 512.0 GB/s for Intel. Shading units are 4,096 versus 3,072. TMUs are 256 versus 192. ROPs are 64 versus 96. The Arc A580 has 24 ray tracing cores; the MI25 has none. Pixel rate is 96.00 GPixel/s for AMD and 192.0 GPixel/s for Intel.

TDP is 300 W for the MI25 versus 175 W for the Arc A580. The suggested PSU is 700 W for AMD and 450 W for Intel. The bus interface is PCIe 3.0 x16 for AMD and PCIe 4.0 x16 for Intel. The MI25 has no display outputs, while the Arc A580 has 1x HDMI 2.1 and 3x DisplayPort 2.0. The MI25 supports DirectX 12 (12_1), while the Arc A580 supports DirectX 12 Ultimate (12_2). Vulkan support is 1.3 for AMD and 1.4 for Intel. OpenGL is 4.6 for both. The MI25 is end-of-life with a release date of June 26, 2017, while the Arc A580 is active with a release date of October 9, 2023. The MI25’s predecessor is FirePro Data Center, and the Arc A580’s predecessor is Xe Graphics and its successor is Battlemage.

The Verdict

The data points to a clear winner for general compute workloads: the Intel Arc A580 outperforms the AMD Radeon Instinct MI25 by 25.2% in Geekbench OpenCL. For anyone prioritizing raw OpenCL throughput, the Arc A580 is the better choice. Its higher clocks, newer architecture, and faster memory contribute to this result. Additionally, the Arc A580 offers ray tracing cores, a feature the MI25 lacks entirely, and it does so with a lower TDP of 175 W versus 300 W, requiring a less powerful PSU (450 W versus 700 W). The Arc A580 also supports newer standards, including PCIe 4.0, DirectX 12 Ultimate, and Vulkan 1.4.

The MI25, however, has advantages that matter in specific contexts. It offers 16 GB of HBM2 memory, which is double the Arc A580’s 8 GB GDDR6. For workloads with large memory footprints, such as certain data center or professional compute tasks, that capacity could be decisive despite the lower bandwidth. The MI25 also has a higher percentile ranking (90th versus 87th), and it sits within 2% of professional cards like the NVIDIA Quadro P6000 and AMD Radeon Pro WX 8200. Its 4,096 shading units give it a wider compute width, even if its older architecture and lower clocks hold it back in the OpenCL test.

The verdict from the recorded data is that the Arc A580 is the stronger all-around GPU for modern workloads, especially those that benefit from ray tracing, higher bandwidth, and newer API support. The MI25 remains relevant only for scenarios where 16 GB of memory capacity and its specific compute profile are more valuable than the Arc A580’s substantial performance lead. The Arc A580 wins the head-to-head, and the only benchmark shared between the two confirms that margin. For most users, the Arc A580 is the sensible pick; for those needing double the memory in a compute accelerator with no display outputs, the MI25 still has a niche, but it is an end-of-life product with no active support path.