NVIDIA CMP 30HX vs NVIDIA Tesla P100 PCIe 16 GB Comparison

The Verdict

The data presents a clear hierarchy between these two NVIDIA cards, but the correct choice depends entirely on workload. The NVIDIA Tesla P100 PCIe 16 GB is the decisive performance winner, taking the sole head-to-head benchmark with a 22.1% lead over the CMP 30HX in Geekbench OpenCL. Its average benchmark score of 79,605 places it in the 92nd percentile of all GPUs, while the CMP 30HX sits at an 89th percentile with an average score of 63,842.

For compute-heavy tasks like scientific simulation, machine learning training, or any workload that leverages OpenCL, the Tesla P100 is the obvious pick. It offers roughly 1.9 times the FP32 throughput of the CMP 30HX (9.526 TFLOPS versus 5.027 TFLOPS), and its 16 GB of HBM2 memory with 732.2 GB/s bandwidth dwarfs the CMP 30HX's 6 GB GDDR6 at 336.0 GB/s. The P100 also carries a launch MSRP of 5,699 USD.

The CMP 30HX, however, is not without its niche. Its 125 W TDP is exactly half the P100's 250 W, and it requires only a 300 W suggested PSU compared to the P100's 600 W. For mining operations or compute workloads where power efficiency is the primary constraint, the CMP 30HX (launch MSRP of 799 USD) could be the more pragmatic choice. It also supports Vulkan 1.4, a newer API version than the P100's 1.3. But in raw performance, the data is unambiguous: the Tesla P100 dominates.

Architecture Differences

The two cards come from different architectural generations and are built for entirely different purposes. The Tesla P100 uses the GP100 chip based on the Pascal architecture, fabricated on a 16 nm process at TSMC. The die is massive at 610 mm², housing 15,300 million transistors, which yields a transistor density of 25.1 million per mm². The CMP 30HX, in contrast, uses the TU116 chip from the Turing architecture, built on a more advanced 12 nm process, also at TSMC. Its die is considerably smaller at 284 mm², with 6,600 million transistors and a density of 23.2 million per mm².

Memory architecture is a fundamental differentiator. The P100 utilizes 16 GB of HBM2 across a 4096-bit bus, delivering 732.2 GB/s of bandwidth. The CMP 30HX uses 6 GB of GDDR6 on a 192-bit bus, achieving 336.0 GB/s. The P100's memory clock runs at 715 MHz (1430 Mbps effective), while the CMP 30HX's memory runs at 1750 MHz (14 Gbps effective), showing that the CMP 30HX compensates for its narrower bus with faster memory speed.

Compute resources are heavily skewed toward the P100. It has 3,584 shading units, 224 texture mapping units, and 96 ROPs. The CMP 30HX has 1,408 shading units, 88 TMUs, and 48 ROPs. Both cards lack RT cores and tensor cores. Clock speeds tell a different story: the CMP 30HX runs at a 1530 MHz base and 1785 MHz boost, significantly higher than the P100's 1190 MHz base and 1329 MHz boost. Yet the P100's raw throughput advantage prevails: 9.526 TFLOPS FP32 versus 5.027 TFLOPS, and 19.05 TFLOPS FP16 versus 10.05 TFLOPS (both at 2:1 ratio).

The P100's pixel rate is 127.6 GPixel/s versus 85.68 GPixel/s for the CMP 30HX, and its texture rate is 297.7 GTexel/s versus 157.1 GTexel/s. The P100 measures 267 mm (10.5 inches) in length, while the CMP 30HX is shorter at 229 mm (9 inches) but has documented height and width of 111 mm (4.4 inches) and 35 mm (1.4 inches). The P100's height and width are not specified. Both are dual-slot cards with a single 8-pin power connector, and neither has display outputs, confirming their compute/mining orientation.

FAQ

Q: Which card is faster in OpenCL, and by how much?

A: The NVIDIA Tesla P100 PCIe 16 GB wins the only head-to-head OpenCL benchmark with a score of 79,605 versus 65,199 for the CMP 30HX, a 22.1% advantage.

Q: How do the two cards compare in overall performance percentiles?

A: The Tesla P100 sits in the 92nd percentile of all GPUs, while the CMP 30HX is in the 89th percentile. The P100's average benchmark score is 79,605, and the CMP 30HX's is 63,842.

Q: What are the power requirements for each card?

A: The Tesla P100 has a 250 W TDP and suggests a 600 W PSU. The CMP 30HX has a 125 W TDP and suggests a 300 W PSU. Both use a single 8-pin power connector.

Q: Which card has more memory bandwidth?

A: The Tesla P100 has 732.2 GB/s from 16 GB of HBM2 on a 4096-bit bus. The CMP 30HX has 336.0 GB/s from 6 GB of GDDR6 on a 192-bit bus.

Q: Are there any API differences between the two?

A: Both support DirectX 12 (12_1) and OpenGL 4.6. The CMP 30HX supports Vulkan 1.4, while the Tesla P100 supports Vulkan 1.3.

Q: Which card was released more recently?

A: The CMP 30HX was released on 2021-02-24, while the Tesla P100 was released on 2016-06-19. Both are now end-of-life products.

Specification Differences

| Specification | Tesla P100 PCIe 16 GB | CMP 30HX |

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

| Architecture | Pascal | Turing |

| Process Node | 16 nm | 12 nm |

| Transistors | 15,300 million | 6,600 million |

| Die Size | 610 mm² | 284 mm² |

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

| Base Clock | 1190 MHz | 1530 MHz |

| Boost Clock | 1329 MHz | 1785 MHz |

| Memory Clock | 715 MHz (1430 Mbps effective) | 1750 MHz (14 Gbps effective) |

| Memory Size | 16 GB | 6 GB |

| Memory Type | HBM2 | GDDR6 |

| Memory Bus Width | 4096 bit | 192 bit |

| Memory Bandwidth | 732.2 GB/s | 336.0 GB/s |

| Shading Units | 3584 | 1408 |

| TMUs | 224 | 88 |

| ROPs | 96 | 48 |

| Pixel Rate | 127.6 GPixel/s | 85.68 GPixel/s |

| Texture Rate | 297.7 GTexel/s | 157.1 GTexel/s |

| FP32 | 9.526 TFLOPS | 5.027 TFLOPS |

| FP16 | 19.05 TFLOPS (2:1) | 10.05 TFLOPS (2:1) |

| TDP | 250 W | 125 W |

| Suggested PSU | 600 W | 300 W |

| Bus Interface | PCIe 3.0 x16 | PCIe 1.0 x4 |

| Vulkan API | 1.3 | 1.4 |

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

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

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

| Release Date | 2016-06-19 | 2021-02-24 |

| Launch MSRP | 5,699 USD | 799 USD |

Head-to-Head Benchmarks

The single head-to-head benchmark available is Geekbench OpenCL, and it is not close. The Tesla P100 scores 79,605, while the CMP 30HX manages 65,199. That is a 22.1% margin in favor of the Pascal card. This result aligns with the broader benchmark data: the P100's average score across all tests is 79,605, while the CMP 30HX averages 63,842 across its two recorded benchmarks (65,199 in OpenCL and 62,484 in Vulkan).

The P100's nearest rivals underline its strong positioning. It sits just 0.3% behind the GeForce RTX 5090 (79,842) and only 0.3% ahead of the Tesla P100 PCIe 12 GB (79,396). It beats the AMD Radeon RX 6850M XT by 0.8% (78,940) and trails the AMD Radeon Pro Vega 64X by 1.7% (80,959). These are tight margins, indicating the P100 is competitive with much newer hardware.

The CMP 30HX's rival comparisons are similarly tight but at a lower performance tier. It is exactly level with the AMD Radeon RX 9060 XT LP (63,830), 0.1% ahead of the AMD Radeon RX 7600M (63,775), 0.2% ahead of the AMD Radeon Pro Vega 56 (63,693), and 0.6% behind the AMD Radeon Pro WX 9100 (64,212).

The performance gap between the two cards is not merely academic. The P100 delivers 89.4% more FP32 throughput, 89.4% more FP16 throughput, 48.9% higher pixel rate, and 89.5% higher texture rate than the CMP 30HX. Its memory bandwidth is 117.9% higher. The CMP 30HX counters with a 28.6% higher base clock and a 34.3% higher boost clock, but these advantages are insufficient to overcome the P100's massive resource advantage in shading units (2.5 times more), TMUs (2.5 times more), and ROPs (2 times more).

The CMP 30HX does win on form factor and power efficiency. It is shorter by 38 mm, has a defined height and width for chassis compatibility, consumes exactly half the power, and requires a 300 W PSU instead of 600 W. For a multi-GPU mining rig or a power-constrained compute node, those differences matter. But for anyone prioritizing compute performance, the Tesla P100 PCIe 16 GB is the clear victor, with a 22.1% lead in the only direct comparison available.