Intel Arc A550M vs NVIDIA CMP 30HX Comparison

The database pits two unusual silicon stories against each other: a desktop mining card stripped of display outputs and a mobile graphics chip with a modern feature set. On paper the NVIDIA CMP 30HX looks like a narrow specialist, while the Intel Arc A550M carries newer architecture and ray tracing hardware. Yet the recorded benchmark data tells a surprisingly one-sided story, and the reasons why are worth digging into.

Head-to-Head Benchmarks

Both measured tests went to the CMP 30HX, and by wide margins.

In Geekbench OpenCL, the CMP 30HX scored 65,199 against 49,894 for the Arc A550M, a 30.7% advantage. The Vulkan test narrows the gap slightly but does not change the outcome: 62,484 versus 49,580, a 26% win for the NVIDIA card. Two tests, two wins for item A, zero for item B.

What makes this interesting is the context each score sits in. The CMP 30HX's average benchmark score of 63,842 places it in the 89th percentile of all GPUs in the database, and its nearest rivals cluster almost perfectly around it: the AMD Radeon RX 9060 XT LP averages 63,830 (a 0% delta), the Radeon RX 7600M sits at 63,775 (0.1% behind), the Radeon Pro Vega 56 at 63,693 (0.2% behind), and the Radeon Pro WX 9100 actually edges ahead at 64,212 (-0.6%). That is a remarkably tight grouping, suggesting the CMP 30HX's compute performance is solidly established at that tier.

The Arc A550M's average of 49,737 puts it in the 86th percentile, and its rival neighborhood is noisier. The GeForce RTX 5070 Ti averages 49,957 (-0.4%), the Radeon RX Vega 64 averages 50,001 (-0.5%), the RX 6900 XT pulls ahead at 50,951 (-2.4%), and the RX 6800 XT trails at 48,477 (2.6%). The spread of plus or minus a few percent around the A550M hints at a chip whose results swing depending on the workload, which fits a mobile part with a modest 60 W TDP.

So the headline gap is real: roughly 26 to 31% in favor of the mining card across both measured APIs, a spread larger than the entire rival band either GPU occupies.

Where Each One Wins

Strictly from the recorded data, the CMP 30HX wins every measured benchmark. The A550M's wins are structural rather than score-based, visible in the spec sheet rather than the results table.

For raw throughput, the CMP 30HX leads in memory bandwidth, 336.0 GB/s against 224.0 GB/s, thanks to its 192-bit bus compared to the 128-bit bus on the Arc, even though both run GDDR6 at 14 Gbps effective. That bandwidth edge plausibly underpins the compute results above.

The A550M counters with newer API support: DirectX 12 Ultimate (12_2) versus 12 (12_1) on the NVIDIA card, which matters for workloads relying on the newest graphics features. It also carries 16 RT cores where the CMP 30HX has none listed, doubles the memory capacity at 8 GB versus 6 GB, and offers a vastly more modern host interface, PCIe 4.0 x16 against an unusual PCIe 1.0 x4 link on the CMP card. And the A550M can actually drive a display, with outputs described as portable device dependent, while the CMP 30HX has no display outputs at all, a consequence of its mining-focused design.

In short: measured compute goes entirely to the CMP 30HX; feature set, capacity, and connectivity favor the Arc A550M.

FAQ

Q: Which GPU is faster in the recorded benchmarks?

A: The NVIDIA CMP 30HX, winning both tests. It leads by 30.7% in Geekbench OpenCL (65,199 vs 49,894) and 26% in Geekbench Vulkan (62,484 vs 49,580).

Q: How do the two compare against the wider database?

A: The CMP 30HX sits in the 89th percentile with an average score of 63,842; the Arc A550M sits in the 86th percentile with an average of 49,737.

Q: Does either GPU support ray tracing?

A: The Arc A550M has 16 RT cores. The CMP 30HX lists no RT cores and also has no tensor cores.

Q: Which has more memory?

A: The Arc A550M, with 8 GB of GDDR6. The CMP 30HX has 6 GB. Both use GDDR6 at 14 Gbps effective, but the CMP 30HX's wider 192-bit bus gives it 336.0 GB/s of bandwidth versus 224.0 GB/s.

Q: Can these drive a monitor?

A: The Arc A550M's display outputs are portable device dependent. The CMP 30HX has no display outputs at all, reflecting its origin as a dedicated mining card.

Q: How do their power envelopes differ?

A: The CMP 30HX has a 125 W TDP with a single 8-pin connector and a suggested 300 W PSU. The Arc A550M runs at 60 W as an integrated mobile part with no power connectors.

Specification Differences

The two diverge nearly everywhere.

The CMP 30HX boosts to 1785 MHz from a 1530 MHz base on a 125 W TDP, in a dual-slot card measuring 229 mm long, 111 mm high, and 35 mm wide, powered by one 8-pin connector. The Arc A550M runs a lower 900 MHz base but a higher 2050 MHz boost inside a 60 W envelope as an IGP with no dimensions recorded and no power connectors.

Memory differs in capacity and bus: 6 GB on a 192-bit bus (336.0 GB/s) versus 8 GB on a 128-bit bus (224.0 GB/s), both GDDR6 at 14 Gbps effective.

The interface gap is stark: PCIe 1.0 x4 versus PCIe 4.0 x16. Outputs split just as sharply, none versus portable device dependent. API support differs at the DirectX level, 12 (12_1) versus 12 Ultimate (12_2), while both list OpenGL 4.6 and Vulkan 1.4. The CMP 30HX carries a launch MSRP of 799 USD; no launch MSRP is recorded for the Arc A550M. Both are marked end-of-life, and only the CMP 30HX has a recorded release date, February 24, 2021.

Architecture Differences

These chips come from opposite design philosophies. The CMP 30HX uses NVIDIA's Turing architecture on the TU116 die, a 12 nm TSMC part with 6,600 million transistors on a 284 mm² die, yielding a density of 23.2M per mm². It fields 1408 shading units, 88 TMUs, and 48 ROPs, with no RT or tensor cores listed, which fits a die repurposed for mining rather than gaming.

The Arc A550M is Intel's Xe-HPG architecture on the DG2-512 die, built on a much newer 6 nm TSMC process. It packs 21,700 million transistors into 406 mm² for a density of 53.4M per mm², more than double the NVIDIA card's. It also fields more of everything execution-related: 2048 shading units, 128 TMUs, 64 ROPs, and 16 RT cores.

The theoretical numbers follow: the A550M claims 8.397 TFLOPS FP32 and 131.2 GPixel/s pixel rate against the CMP 30HX's 5.027 TFLOPS and 85.68 GPixel/s, plus 262.4 GTexel/s against 157.1 GTexel/s. FP16 lands at 16.79 TFLOPS versus 10.05 TFLOPS, both at a 2:1 ratio.

Here is the puzzle the data raises: the Arc A550M dominates on every theoretical metric, yet loses both measured benchmarks by 26% or more. The plausible explanations sit elsewhere in the table: the CMP 30HX's 50% bandwidth advantage, its desktop-class 125 W power budget versus a 60 W mobile envelope, and possibly its PCIe 1.0 x4 link being less of a handicap for compute than the A550M's mobile constraints are for sustained throughput. The lesson is a familiar one, theoretical horsepower does not guarantee measured results.

The Verdict

For compute workloads as captured by the database's OpenCL and Vulkan tests, the CMP 30HX is the clear pick: 30.7% and 26% wins, an 89th percentile standing, and a rival neighborhood it matches almost exactly. Buyers chasing measured throughput at that tier should note it performs on par with cards like the Radeon RX 9060 XT LP and Radeon Pro WX 9100.

The Arc A550M is the pick for everything the benchmarks do not capture: 8 GB of memory, DirectX 12 Ultimate support, 16 RT cores, a PCIe 4.0 x16 interface, actual display capability, and a 60 W mobile envelope. The CMP 30HX's lack of display outputs alone disqualifies it from any visual workload.

The data frames it simply. The mining card wins the numbers; the mobile Arc wins the feature set. Neither is a general-purpose purchase today, both being end-of-life, but the head-to-head is a neat case study in why a spec sheet advantage and a benchmark win are not the same thing.