NVIDIA CMP 90HX vs NVIDIA Tesla V100 PCIe 16 GB Comparison

Head-to-Head Benchmarks

The benchmark data is decisive: the NVIDIA Tesla V100 PCIe 16 GB outperforms the NVIDIA CMP 90HX by a substantial margin in the only direct comparison available. In the Geekbench OpenCL test, the Tesla V100 scores 163,063 points, while the CMP 90HX manages 69,000 points. That represents a 136.3% advantage for the Tesla V100, meaning it more than doubles the CMP 90HX's recorded performance. This is not a close contest by any measure.

The Tesla V100's average benchmark score across all recorded tests stands at 138,063, placing it in the 96th percentile of all GPUs in the database. The CMP 90HX, by contrast, has an average score of 69,000 and sits in the 90th percentile. While both cards rank highly overall, the gap between them is stark. The Tesla V100's nearest rivals include the NVIDIA Tesla V100 SXM2 32 GB (average score 137,731, only 0.2% behind), the AMD Instinct MI100 (139,035, 0.7% ahead), the AMD Radeon PRO V620 (136,472, 1.2% behind), and the AMD Radeon Pro W6800X Duo (135,774, 1.7% behind). The CMP 90HX, on the other hand, trades blows with the Intel Arc A770 (68,809, 0.3% behind), the AMD Radeon Instinct MI25 (68,562, 0.6% behind), the AMD Radeon Pro WX 8200 (69,870, 1.2% ahead), and the NVIDIA Quadro P6000 (69,986, 1.4% ahead).

The data shows that the Tesla V100 competes at a performance tier roughly double that of the CMP 90HX. The CMP 90HX's closest competitors are all within a 1.4% band, indicating that its performance level is well established, but that level is far below what the Tesla V100 delivers. In the head-to-head OpenCL result, the Tesla V100 wins the only recorded comparison, and it wins by a landslide.

Where Each One Wins

The Tesla V100 PCIe 16 GB wins in raw compute throughput, as evidenced by its Geekbench OpenCL score. This card is built for high-performance computing workloads where FP32 and FP16 arithmetic dominate. Its FP32 throughput is 14.13 TFLOPS, and its FP16 output reaches 28.26 TFLOPS (at a 2:1 ratio). The CMP 90HX delivers 21.89 TFLOPS in both FP32 and FP16 (at a 1:1 ratio). Interestingly, the CMP 90HX has a higher raw FP32 number, yet its OpenCL score is far lower. This suggests that the Tesla V100's architecture, memory subsystem, or driver optimizations yield better real-world compute results despite the lower theoretical FP32 peak.

The CMP 90HX does not win any recorded benchmark comparisons against the Tesla V100. It has zero wins in the head-to-head data. However, the CMP 90HX does have its own strengths in the specification sheet. It features 50 RT cores and 200 tensor cores, both of which are absent from the Tesla V100's listed specifications. For workloads that leverage ray tracing or tensor operations, the CMP 90HX has dedicated hardware. The Tesla V100 does have 640 tensor cores, but the CMP 90HX's tensor cores are paired with a newer Ampere architecture, which may offer different efficiency characteristics. The recorded benchmark data, however, does not include any ray tracing or tensor-specific tests, so these advantages remain theoretical in the context of this comparison.

For general compute tasks measured by OpenCL, the Tesla V100 is the clear winner. For tasks that might utilize the CMP 90HX's RT cores or its higher FP32 peak, the data does not provide direct evidence of superiority. The only recorded benchmark favors the Tesla V100.

Architecture Differences

The two cards come from entirely different architectural generations. The Tesla V100 uses the GV100 chip based on the Volta architecture, manufactured on TSMC's 12 nm process. The die measures 815 mm² and contains 21,100 million transistors, yielding a transistor density of 25.9 million per square millimeter. The CMP 90HX uses the GA102 chip based on the Ampere architecture, built on Samsung's 8 nm process. Its die is smaller at 628 mm², but it packs 28,300 million transistors, resulting in a higher density of 45.1 million per square millimeter. The CMP 90HX is a denser, more modern chip.

Memory configurations differ substantially. The Tesla V100 ships with 16 GB of HBM2 memory on a 4096-bit bus, delivering 897.0 GB/s of bandwidth. The CMP 90HX has 10 GB of GDDR6X memory on a 320-bit bus, providing 760.3 GB/s. The Tesla V100 has a wider memory bus and higher bandwidth, which likely contributes to its strong compute performance. Clock speeds also differ: the Tesla V100 runs at 1245 MHz base and 1380 MHz boost, while the CMP 90HX runs at 1500 MHz base and 1710 MHz boost. The CMP 90HX has higher clocks, but the Tesla V100 still wins in the benchmark.

Core configurations diverge as well. The Tesla V100 has 5120 shading units, 320 TMUs, and 128 ROPs, along with 640 tensor cores. The CMP 90HX has 6400 shading units, 200 TMUs, and 80 ROPs, plus 50 RT cores and 200 tensor cores. The CMP 90HX has more shading units but fewer TMUs and ROPs. The Tesla V100's pixel rate is 176.6 GPixel/s, while the CMP 90HX's is 136.8 GPixel/s. The Tesla V100's texture rate is 441.6 GTexel/s, versus 342.0 GTexel/s for the CMP 90HX. Despite having fewer shading units, the Tesla V100 leads in pixel and texture throughput.

The CMP 90HX supports DirectX 12 Ultimate (12_2), while the Tesla V100 supports DirectX 12 (12_1). Both support OpenGL 4.6 and Vulkan 1.4. The CMP 90HX's bus interface is PCIe 1.0 x4, which is unusual and likely reflects its mining-oriented design. The Tesla V100 uses PCIe 3.0 x16. Neither card has display outputs. The CMP 90HX has a physical length of 285 mm (11.2 inches) and height of 112 mm (4.4 inches); the Tesla V100's dimensions are not recorded.

The CMP 90HX is classified under "Mining GPUs" in the database, while the Tesla V100 belongs to the Tesla Volta generation. The Tesla V100 was released on 2017-06-20, and the CMP 90HX on 2021-07-27. Both are end-of-life products.

FAQ

Q: Which card has the higher average benchmark score?

A: The NVIDIA Tesla V100 PCIe 16 GB has an average benchmark score of 138,063, while the NVIDIA CMP 90HX has an average score of 69,000. The Tesla V100 leads by more than double.

Q: How much faster is the Tesla V100 in the OpenCL test?

A: The Tesla V100 scores 163,063 in Geekbench OpenCL, versus 69,000 for the CMP 90HX, a 136.3% difference in favor of the Tesla V100.

Q: Do these cards support the same DirectX version?

A: No. The CMP 90HX supports DirectX 12 Ultimate (12_2), while the Tesla V100 supports DirectX 12 (12_1). Both support OpenGL 4.6 and Vulkan 1.4.

Q: What are the memory capacities and types?

A: The Tesla V100 has 16 GB of HBM2 on a 4096-bit bus with 897.0 GB/s bandwidth. The CMP 90HX has 10 GB of GDDR6X on a 320-bit bus with 760.3 GB/s bandwidth.

Q: Which card has more shading units?

A: The CMP 90HX has 6400 shading units, while the Tesla V100 has 5120. However, the Tesla V100 has more TMUs (320 vs. 200) and ROPs (128 vs. 80).

Q: Do either of these cards have display outputs?

A: No. Both the Tesla V100 and the CMP 90HX have no display outputs.

The Verdict

Based strictly on the recorded benchmark data, the NVIDIA Tesla V100 PCIe 16 GB is the superior compute card. It wins the only head-to-head comparison by 136.3%, and its average benchmark score of 138,063 places it in the 96th percentile of all GPUs, far above the CMP 90HX's 90th percentile. The Tesla V100's nearest rivals, such as the Tesla V100 SXM2 32 GB and the AMD Instinct MI100, are within 1% of its average score, indicating it sits at the top of its performance class.

The CMP 90HX, with an average score of 69,000, is competitive with mid-range professional cards like the Intel Arc A770 and the NVIDIA Quadro P6000, but it is not in the same league as the Tesla V100. Its higher FP32 peak of 21.89 TFLOPS does not translate into better OpenCL results, likely due to architectural and memory subsystem differences. The CMP 90HX's only potential advantages lie in its RT cores and its newer DirectX 12 Ultimate support, but no benchmark data in the database confirms any real-world benefit from these features.

For users prioritizing raw compute performance, the Tesla V100 is the clear choice. For workloads that might leverage ray tracing or the CMP 90HX's specific feature set, the data does not provide evidence of superiority. The verdict is unambiguous: the Tesla V100 PCIe 16 GB outperforms the NVIDIA CMP 90HX in every recorded benchmark.

Specification Differences

| Specification | NVIDIA Tesla V100 PCIe 16 GB | NVIDIA CMP 90HX |

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

| Chip | GV100 | GA102 |

| Architecture | Volta | Ampere |

| Generation | Tesla Volta (Vxx) | Mining GPUs |

| Process Node | 12 nm | 8 nm |

| Foundry | TSMC | Samsung |

| Transistors | 21,100 million | 28,300 million |

| Die Size | 815 mm² | 628 mm² |

| Transistor Density | 25.9M / mm² | 45.1M / mm² |

| Base Clock | 1245 MHz | 1500 MHz |

| Boost Clock | 1380 MHz | 1710 MHz |

| Memory Clock | 876 MHz (1752 Mbps effective) | 1188 MHz (19 Gbps effective) |

| Memory Size | 16 GB | 10 GB |

| Memory Type | HBM2 | GDDR6X |

| Memory Bus Width | 4096 bit | 320 bit |

| Memory Bandwidth | 897.0 GB/s | 760.3 GB/s |

| Shading Units | 5120 | 6400 |

| TMUs | 320 | 200 |

| ROPs | 128 | 80 |

| RT Cores | Not listed | 50 |

| Tensor Cores | 640 | 200 |

| Pixel Rate | 176.6 GPixel/s | 136.8 GPixel/s |

| Texture Rate | 441.6 GTexel/s | 342.0 GTexel/s |

| FP32 | 14.13 TFLOPS | 21.89 TFLOPS |

| FP16 | 28.26 TFLOPS (2:1) | 21.89 TFLOPS (1:1) |

| TDP | 300 W | 320 W |

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

| DirectX | 12 (12_1) | 12 Ultimate (12_2) |

| Dimensions | Not recorded | 285 mm (11.2 inches) length, 112 mm (4.4 inches) height |

| Release Date | 2017-06-20 | 2021-07-27 |

| Predecessor | Tesla Pascal | Not listed |

| Successor | Tesla Turing | Not listed |

| Geekbench OpenCL Score | 163063 | 69000 |

| Average Benchmark Score | 138063 | 69000 |

| Percentile vs All GPUs | 96 | 90 |