NVIDIA CMP 30HX vs NVIDIA P102-100 Comparison

The Verdict

The data splits this matchup cleanly down the middle: the NVIDIA CMP 30HX wins one benchmark decisively, and the NVIDIA P102-100 wins the other by a smaller margin. If your workload leans on OpenCL compute, the CMP 30HX is the clear pick — it beats the P102-100 by 31.4% in Geekbench OpenCL, a massive gap. If Vulkan matters more, the P102-100 takes it, but only by 7.4%. Neither card is a universal winner, so the choice depends entirely on which API your software uses. The CMP 30HX also carries a launch MSRP of 799 USD, while the P102-100 has no listed launch MSRP. Both are end-of-life mining GPUs with no display outputs, so neither is suited for gaming or general desktop use. For raw compute in OpenCL-heavy tasks, pick the CMP 30HX. For Vulkan-centric workloads, the P102-100 is the better option, though its advantage is far less pronounced.

Architecture Differences

The two cards come from different NVIDIA architectures and different process nodes. The CMP 30HX uses the TU116 chip built on Turing architecture at TSMC's 12 nm process. The P102-100 uses the GP102 chip on the older Pascal architecture at TSMC's 16 nm node. This generational gap shows up in several ways. The TU116 packs 6,600 million transistors on a 284 mm² die, giving a transistor density of 23.2M per mm². The GP102 is a much larger chip: 11,800 million transistors on a 471 mm² die, with a density of 25.1M per mm². Despite the older process, the P102-100's bigger die and higher transistor count give it more raw hardware resources.

The CMP 30HX has 1,408 shading units, 88 texture mapping units, and 48 ROPs. The P102-100 nearly doubles that: 3,200 shading units, 200 TMUs, and 80 ROPs. Neither card has RT cores or tensor cores, so ray tracing and AI acceleration are absent from both. In terms of compute throughput, the P102-100's FP32 rating is 10.77 TFLOPS versus 5.027 TFLOPS on the CMP 30HX — more than double. However, the CMP 30HX has a major advantage in FP16: it delivers 10.05 TFLOPS at a 2:1 ratio, while the P102-100 manages only 168.3 GFLOPS at a 1:64 ratio. This is a striking difference; the Turing card supports fast half-precision math, whereas the Pascal card essentially does not. That explains why the CMP 30HX excels in OpenCL, which often leverages FP16 paths. The P102-100's higher FP32 and texture rates (336.6 GTexel/s vs 157.1 GTexel/s) and pixel rate (134.6 GPixel/s vs 85.68 GPixel/s) show its raw fill-rate advantage, but those gains don't translate into a win on the OpenCL test.

Memory also differs fundamentally. The CMP 30HX uses 6 GB of GDDR6 on a 192-bit bus, yielding 336.0 GB/s bandwidth. The P102-100 uses 5 GB of GDDR5X on a wider 320-bit bus, delivering 440.3 GB/s. So the P102-100 has less capacity but more bandwidth. Clock speeds are close: the CMP 30HX boosts to 1785 MHz from a 1530 MHz base, while the P102-100 boosts to 1683 MHz from a 1582 MHz base. Memory clocks differ too: the CMP 30HX runs at 1750 MHz (14 Gbps effective), the P102-100 at 1376 MHz (11 Gbps effective). Both cards use the same PCIe 1.0 x4 bus interface, a peculiar choice for mining-focused hardware, and neither has display outputs.

Head-to-Head Benchmarks

The Geekbench OpenCL result is a blowout. The CMP 30HX scores 65,199 against the P102-100's 49,602, a 31.4% advantage. This is the single biggest gap in the entire comparison. The CMP 30HX's FP16 capability at 2:1 ratio (10.05 TFLOPS) versus the P102-100's negligible FP16 (168.3 GFLOPS at 1:64) likely drives this result. OpenCL workloads often use half-precision arithmetic, and the Turing architecture handles it far more efficiently. The P102-100 cannot compete here despite having over twice the FP32 throughput and 31% more memory bandwidth.

The Geekbench Vulkan result flips the script, but not by as much. The P102-100 scores 67,454 against the CMP 30HX's 62,484, a 7.4% margin. Vulkan tends to favor raw fill rate and shader count, which is where the P102-100's 3,200 shading units and 80 ROPs shine. Its pixel rate of 134.6 GPixel/s and texture rate of 336.6 GTexel/s are substantially higher than the CMP 30HX's 85.68 GPixel/s and 157.1 GTexel/s. Still, a 7.4% win is modest compared to the 31.4% loss in OpenCL. If you average the two scores, the CMP 30HX's combined benchmark average is 63,842, placing it at the 89th percentile of all GPUs. The P102-100's average is 58,528, at the 88th percentile. The CMP 30HX also sits higher among its nearest rivals: its closest competitor is the AMD Radeon RX 9060 XT LP at 63,830 (0% delta), while the P102-100's closest is the AMD Radeon PRO V710 at 58,657 (-0.2% delta). Both cards are within a hair of their nearest rivals, but the CMP 30HX occupies a slightly higher performance tier overall.

Specification Differences

The table below lists only the fields where the two cards differ:

| Field | NVIDIA CMP 30HX | NVIDIA P102-100 |

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

| Chip | TU116 | GP102 |

| Architecture | Turing | Pascal |

| Process Node | 12 nm | 16 nm |

| Transistors | 6,600 million | 11,800 million |

| Die Size | 284 mm² | 471 mm² |

| Transistor Density | 23.2M / mm² | 25.1M / mm² |

| Base Clock | 1530 MHz | 1582 MHz |

| Boost Clock | 1785 MHz | 1683 MHz |

| Memory Clock | 1750 MHz (14 Gbps effective) | 1376 MHz (11 Gbps effective) |

| Memory Size | 6 GB | 5 GB |

| Memory Type | GDDR6 | GDDR5X |

| Memory Bus Width | 192 bit | 320 bit |

| Memory Bandwidth | 336.0 GB/s | 440.3 GB/s |

| Shading Units | 1408 | 3200 |

| TMUs | 88 | 200 |

| ROPs | 48 | 80 |

| Pixel Rate | 85.68 GPixel/s | 134.6 GPixel/s |

| Texture Rate | 157.1 GTexel/s | 336.6 GTexel/s |

| FP32 | 5.027 TFLOPS | 10.77 TFLOPS |

| FP16 | 10.05 TFLOPS (2:1) | 168.3 GFLOPS (1:64) |

| TDP | 125 W | 250 W |

| Power Connectors | 1x 8-pin | 2x 8-pin |

| Suggested PSU | 300 W | 600 W |

| Length | 229 mm (9 inches) | 267 mm (10.5 inches) |

| Height | 111 mm (4.4 inches) | Not specified |

| Width | 35 mm (1.4 inches) | Not specified |

| Release Date | 2021-02-24 | 2018-02-11 |

| Launch MSRP | 799 USD | Not specified |

Fields that are identical include the manufacturer (NVIDIA), generation (Mining GPUs), foundry (TSMC), slot width (Dual-slot), bus interface (PCIe 1.0 x4), display outputs (No outputs), API support (DirectX 12_1, OpenGL 4.6, Vulkan 1.4), and production status (End-of-life). Neither card has RT cores or tensor cores.

FAQ

Q: Which card is faster in OpenCL?

A: The NVIDIA CMP 30HX wins Geekbench OpenCL with a score of 65,199 versus the P102-100's 49,602, a 31.4% advantage.

Q: Which card is faster in Vulkan?

A: The NVIDIA P102-100 wins Geekbench Vulkan with a score of 67,454 versus the CMP 30HX's 62,484, a 7.4% margin.

Q: Why does the CMP 30HX win OpenCL despite having fewer shading units?

A: The CMP 30HX supports FP16 at a 2:1 ratio with 10.05 TFLOPS, while the P102-100 only manages 168.3 GFLOPS at 1:64. OpenCL workloads often use half-precision math, which the Turing architecture handles far more efficiently.

Q: Does the P102-100 have more memory bandwidth?

A: Yes. The P102-100 delivers 440.3 GB/s over a 320-bit GDDR5X bus, versus the CMP 30HX's 336.0 GB/s over a 192-bit GDDR6 bus.

Q: Which card has higher power requirements?

A: The P102-100 has a 250 W TDP and requires a 600 W PSU with 2x 8-pin connectors. The CMP 30HX has a 125 W TDP, needs only a 300 W PSU, and uses a single 8-pin connector.

Q: Do these cards support display output?

A: No. Both cards have no display outputs, as they are designed exclusively for mining workloads.

Where Each One Wins

The CMP 30HX is the pick for OpenCL-based compute tasks. Its 31.4% lead in Geekbench OpenCL is decisive, and its FP16 throughput at 10.05 TFLOPS gives it a massive edge in half-precision workloads. It also draws half the power (125 W vs 250 W), requires a smaller PSU (300 W vs 600 W), and fits in a shorter chassis at 229 mm versus 267 mm. If your mining or compute software uses OpenCL — which is common in many cryptocurrency algorithms and scientific workloads — the CMP 30HX is the clear winner. It also has a higher average benchmark score (63,842 vs 58,528) and a slightly better percentile ranking (89th vs 88th).

The P102-100 wins in Vulkan, though the margin is tighter at 7.4%. Its raw shader count (3,200 vs 1,408), higher FP32 throughput (10.77 TFLOPS vs 5.027 TFLOPS), and greater memory bandwidth (440.3 GB/s vs 336.0 GB/s) make it better suited for Vulkan-based rendering or compute tasks that use full-precision math. It also has superior fill rates: 134.6 GPixel/s versus 85.68 GPixel/s, and 336.6 GTexel/s versus 157.1 GTexel/s. If your workload is Vulkan-centric and you can tolerate the higher power draw and larger physical size, the P102-100 is the better choice.

For a general-purpose mining card, the CMP 30HX looks like the safer bet. It wins the benchmark that matters most (OpenCL) by a huge margin, costs less in power (125 W vs 250 W), and has a smaller footprint. The P102-100's Vulkan win is real but modest, and its strengths in FP32 and fill rate don't translate into a dominant benchmark performance. The data suggests that Turing's architectural improvements in the CMP 30HX — particularly in FP16 — outweigh the P102-100's raw hardware advantage. With both cards at end-of-life, the CMP 30HX's higher average score and better percentile placement make it the more compelling option for most mining scenarios. However, the P102-100 remains relevant if your specific algorithm or application is Vulkan-based and you need maximum FP32 throughput.