Intel Arc A580 vs NVIDIA CMP 90HX Comparison

Head-to-Head Benchmarks

The only directly comparable benchmark recorded for both GPUs is Geekbench OpenCL, and the result is decisive. The Intel Arc A580 scores 91,657, while the NVIDIA CMP 90HX scores 69,000. That gives Intel a 24.7% lead in this specific test. The delta is substantial, and it flips the expected narrative if one were to assume the higher-tier NVIDIA part would dominate outright.

The NVIDIA CMP 90HX sits at the 90th percentile among all GPUs in the database, with an average benchmark score of 69,000. Its nearest rivals are tightly clustered: the Intel Arc A770 trails by just 0.3%, the AMD Radeon Instinct MI25 trails by 0.6%, while the AMD Radeon Pro WX 8200 and NVIDIA Quadro P6000 lead by 1.2% and 1.4%, respectively. This tells us the CMP 90HX is firmly in a competitive mid-to-high tier, but it is not a standout performer even among its immediate peers.

The Intel Arc A580, by contrast, records an average benchmark score of 57,756, placing it at the 87th percentile. Its nearest rivals include the AMD Radeon RX 5600 OEM (0.6% ahead), the AMD Radeon RX 9070 GRE (0.7% behind), the Intel Arc A570M (0.8% ahead), and the AMD Radeon RX 6950 XT (1.1% ahead). The A580's average score is dragged down by its lower-scoring 3DMark Steel Nomad DX12 result of 2,229, but its Geekbench OpenCL score of 91,657 is clearly its strongest showing.

The head-to-head data contains only one benchmark, so the overall win count is 1 for Intel and 0 for NVIDIA. The 24.7% delta in OpenCL is not a small margin; it is a whole-tier gap in compute performance. However, the CMP 90HX's 21.89 TFLOPS FP32 rating versus the A580's 12.29 TFLOPS FP32 suggests the NVIDIA part should be faster in raw shader work, yet the OpenCL result contradicts that expectation. The A580's higher boost clock of 2000 MHz versus 1710 MHz, combined with its 96 ROPs versus 80, may explain part of the discrepancy. The A580 also has a pixel rate of 192.0 GPixel/s and a texture rate of 384.0 GTexel/s, both exceeding the CMP 90HX's 136.8 GPixel/s and 342.0 GTexel/s.

The Verdict

Based strictly on the recorded data, the Intel Arc A580 wins the only direct comparison available. In Geekbench OpenCL, it outperforms the NVIDIA CMP 90HX by 24.7%, a margin that cannot be dismissed as noise. The A580 also carries a lower power draw (175 W versus 320 W), a smaller suggested power supply (450 W versus 700 W), and a more modern process node (6 nm versus 8 nm). It is an active product with display outputs, while the CMP 90HX is end-of-life and has no display outputs at all.

The NVIDIA CMP 90HX, however, is not without its own strengths in the database. Its average benchmark score of 69,000 is higher than the A580's 57,756, and it sits at the 90th percentile versus the A580's 87th. The CMP 90HX also offers 10 GB of GDDR6X memory with 760.3 GB/s of bandwidth, compared to the A580's 8 GB of GDDR6 at 512.0 GB/s. For workloads that depend heavily on memory bandwidth, the CMP 90HX has a clear theoretical edge. Its FP32 throughput of 21.89 TFLOPS is also 78% higher than the A580's 12.29 TFLOPS, which suggests that in pure compute tasks that do not hit the OpenCL path, the NVIDIA part may be faster.

Who should pick which? The data points to a split decision. If the task is general OpenCL compute, the Arc A580 is the better choice, with a 24.7% measured advantage and lower system power requirements. If the task involves large memory transfers, high-bandwidth data movement, or raw FP32 shader work, the CMP 90HX's 760.3 GB/s bandwidth and 21.89 TFLOPS make it the more capable part on paper. The CMP 90HX also has a higher average benchmark score across all recorded tests, which reflects its overall standing in the database. The A580's single high OpenCL score is impressive, but it is one data point. The CMP 90HX's average is built on a broader set of measurements, even if only one is listed here, and it places higher in the percentile ranking.

FAQ

Q: Which GPU scores higher in Geekbench OpenCL?

A: The Intel Arc A580 scores 91,657, which is 24.7% higher than the NVIDIA CMP 90HX's 69,000.

Q: What is the average benchmark score for each GPU?

A: The NVIDIA CMP 90HX has an average benchmark score of 69,000, while the Intel Arc A580 has an average of 57,756.

Q: How do the two GPUs compare in terms of power consumption?

A: The NVIDIA CMP 90HX has a TDP of 320 W and suggests a 700 W power supply. The Intel Arc A580 has a TDP of 175 W and suggests a 450 W power supply.

Q: Does the NVIDIA CMP 90HX support display outputs?

A: No, the CMP 90HX has no display outputs. The Intel Arc A580 has 1x HDMI 2.1 and 3x DisplayPort 2.0 outputs.

Q: What are the memory configurations?

A: The CMP 90HX has 10 GB of GDDR6X on a 320-bit bus with 760.3 GB/s bandwidth. The A580 has 8 GB of GDDR6 on a 256-bit bus with 512.0 GB/s bandwidth.

Q: Which GPU has a higher FP32 throughput?

A: The NVIDIA CMP 90HX has 21.89 TFLOPS FP32, while the Intel Arc A580 has 12.29 TFLOPS FP32.

Specification Differences

The two cards diverge on nearly every major specification. The NVIDIA CMP 90HX uses 10 GB of GDDR6X memory with a 320-bit bus and 760.3 GB/s bandwidth. The Intel Arc A580 uses 8 GB of GDDR6 with a 256-bit bus and 512.0 GB/s bandwidth. The CMP 90HX has 6,400 shading units, 200 TMUs, and 80 ROPs, while the A580 has 3,072 shading units, 192 TMUs, and 96 ROPs. Ray tracing cores differ as well: the CMP 90HX has 50 RT cores and 200 tensor cores, while the A580 has 24 RT cores and no tensor cores listed.

Clock speeds favor Intel. The A580 has a base clock of 1700 MHz and a boost clock of 2000 MHz, compared to the CMP 90HX's 1500 MHz base and 1710 MHz boost. Memory clocks also differ: the CMP 90HX runs at 1188 MHz with 19 Gbps effective, while the A580 runs at 2000 MHz with 16 Gbps effective. Pixel rate and texture rate both favor the A580 (192.0 GPixel/s and 384.0 GTexel/s versus 136.8 GPixel/s and 342.0 GTexel/s). FP16 performance flips the ratio: the CMP 90HX does 21.89 TFLOPS at 1:1, while the A580 does 24.58 TFLOPS at 2:1.

Power and physical requirements differ significantly. The CMP 90HX draws 320 W and requires a 700 W power supply, while the A580 draws 175 W and requires a 450 W power supply. Both are dual-slot cards with 2x 8-pin connectors. The bus interface differs: the CMP 90HX uses PCIe 1.0 x4, while the A580 uses PCIe 4.0 x16. Display outputs are present only on the A580. The CMP 90HX is 285 mm long and 112 mm tall; the A580 has no recorded dimensions.

Architecture Differences

The NVIDIA CMP 90HX is built on the GA102 chip using the Ampere architecture, fabricated on Samsung's 8 nm process with 28,300 million transistors on a 628 mm² die, giving a transistor density of 45.1 million per mm². The Intel Arc A580 uses the DG2-512 chip with the Xe-HPG architecture, fabricated on TSMC's 6 nm process with 21,700 million transistors on a 406 mm² die, giving a higher density of 53.4 million per mm². The density difference shows Intel squeezing more transistors into less space, but the CMP 90HX still has more total transistors by 6,600 million.

The CMP 90HX belongs to NVIDIA's Mining GPUs generation, released on July 27, 2021, and is now end-of-life. The A580 belongs to the Alchemist (Arc 5) generation, released on October 9, 2023, and is still active. The A580 has a predecessor (Xe Graphics) and a successor (Battlemage), while the CMP 90HX has neither. Both support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, so API compatibility is identical.

The CMP 90HX includes tensor cores (200 of them), which the A580 does not list. The A580's FP16 throughput of 24.58 TFLOPS at 2:1 ratio indicates shader-based FP16 execution, whereas the CMP 90HX's 1:1 ratio means it processes FP16 at the same rate as FP32. The A580's pixel rate advantage (192.0 GPixel/s versus 136.8 GPixel/s) suggests a more efficient rasterization pipeline, despite having fewer shading units. The CMP 90HX's memory subsystem, with 760.3 GB/s bandwidth, is 48.5% faster than the A580's 512.0 GB/s, which is the single largest architectural advantage for the NVIDIA part.

Where Each One Wins

The Intel Arc A580 wins in any scenario where the Geekbench OpenCL score is indicative of real-world compute performance. Its 24.7% lead in that test, combined with a lower power draw and a 450 W PSU requirement, makes it the more practical choice for systems with modest power delivery. The A580's display outputs (1x HDMI 2.1, 3x DisplayPort 2.0) mean it can drive monitors, while the CMP 90HX cannot. Its higher pixel rate and texture rate also suggest better performance in fill-rate-bound workloads, such as certain post-processing effects or high-resolution rendering passes. The A580's 6 nm process node and higher transistor density indicate a more modern design with potentially better efficiency per clock.

The NVIDIA CMP 90HX wins in scenarios that leverage its memory bandwidth and FP32 throughput. The 760.3 GB/s bandwidth is 48.5% higher than the A580's 512.0 GB/s, which matters for large data sets, high-resolution textures, or compute workloads that stream data heavily. The 21.89 TFLOPS FP32 rating is 78% higher than the A580's 12.29 TFLOPS, giving the CMP 90HX a clear edge in raw shader compute if the software path is optimized for NVIDIA's architecture. Its 10 GB GDDR6X memory also provides 2 GB more capacity than the A580's 8 GB, which can help with larger working sets. The CMP 90HX also has a higher percentile ranking (90th versus 87th) and a higher average benchmark score (69,000 versus 57,756), indicating stronger overall standing in the database.

The production status difference is also meaningful. The CMP 90HX is end-of-life, so it is a legacy part with no ongoing driver focus for gaming or display workloads. The A580 is active, with a successor planned (Battlemage), which suggests continued software support. For a system that needs a GPU today with modern connectivity and driver updates, the A580 is the data-backed choice. For a compute node that prioritizes bandwidth and FP32 throughput above all else, the CMP 90HX retains a theoretical advantage, but the measured OpenCL result shows the A580 can outperform it in at least one major API. The recorded data gives Intel the only head-to-head win, and that is the fact that stands out.