NVIDIA CMP 90HX vs NVIDIA TITAN X Pascal Comparison

FAQ

Q: How do the two GPUs compare in overall benchmark scores?

A: The NVIDIA TITAN X Pascal has an average benchmark score of 72,098, while the NVIDIA CMP 90HX averages 69,000. The TITAN X Pascal also holds a 91st percentile rank versus the CMP 90HX’s 90th percentile, placing both firmly in the upper tier of all GPUs.

Q: Which card is faster in the available head-to-head test?

A: In the only direct comparison, Geekbench OpenCL, the CMP 90HX scores 69,000 versus the TITAN X Pascal’s 66,696. That gives the CMP 90HX a 3.3% advantage, and it wins the sole head-to-head benchmark.

Q: What are the nearest performance rivals for each card?

A: The TITAN X Pascal sits near the AMD Radeon Pro Vega 64 (0.4% faster), AMD Radeon RX 6650M (0.5% slower), AMD Radeon RX 6600 LE (1.8% slower), and AMD Radeon Vega Frontier Edition (1.7% faster). The CMP 90HX is bracketed by Intel Arc A770 (0.3% slower), AMD Radeon Instinct MI25 (0.6% slower), AMD Radeon Pro WX 8200 (1.2% faster), and NVIDIA Quadro P6000 (1.4% faster).

Q: Do these cards have display outputs?

A: No, not in the same way. The TITAN X Pascal includes 1x DVI, 1x HDMI 2.0, and 3x DisplayPort 1.4a outputs. The CMP 90HX has no display outputs at all, as it is designed for mining use only.

Q: What are the TDP and power connector requirements?

A: The TITAN X Pascal has a 250 W TDP and requires a 600 W suggested PSU with 1x 6-pin plus 1x 8-pin connectors. The CMP 90HX draws 320 W, needs a 700 W recommended PSU, and uses 2x 8-pin connectors.

Q: Which card is newer and what is their production status?

A: The TITAN X Pascal was released on 2016-08-01, while the CMP 90HX launched on 2021-07-27. Both are end-of-life. The TITAN X Pascal had a launch MSRP of 1,199 USD; the CMP 90HX has no listed launch MSRP.

Architecture Differences

The two cards come from entirely different architectural generations. The TITAN X Pascal uses the GP102 chip on TSMC’s 16 nm process, packing 11,800 million transistors onto a 471 mm² die with a transistor density of 25.1M per mm². The CMP 90HX is built on the GA102 chip using Samsung’s 8 nm node, containing 28,300 million transistors across a 628 mm² die, achieving 45.1M transistors per mm² — nearly double the density.

The CMP 90HX’s Ampere architecture introduces hardware that simply does not exist on the Pascal-based TITAN X. The CMP 90HX carries 50 RT cores and 200 tensor cores, while the TITAN X Pascal has none. This is also reflected in the API support: the CMP 90HX supports DirectX 12 Ultimate (12_2), whereas the TITAN X Pascal is limited to DirectX 12 (12_1). Both cards support OpenGL 4.6 and Vulkan 1.4.

The memory subsystem also differs fundamentally. The TITAN X Pascal uses 12 GB of GDDR5X over a 384-bit bus, yielding 480.4 GB/s bandwidth. The CMP 90HX uses 10 GB of GDDR6X over a narrower 320-bit bus, yet delivers 760.3 GB/s — a 58% increase in bandwidth due to the faster 19 Gbps effective memory speed versus the TITAN X Pascal’s 10 Gbps effective.

Compute capabilities diverge sharply as well. The CMP 90HX has 6,400 shading units, 200 TMUs, and 80 ROPs, against the TITAN X Pascal’s 3,584 shading units, 224 TMUs, and 96 ROPs. The CMP 90HX’s FP32 throughput is 21.89 TFLOPS, exactly double the TITAN X Pascal’s 10.97 TFLOPS. FP16 performance is even more lopsided: the CMP 90HX achieves 21.89 TFLOPS at a 1:1 ratio, while the TITAN X Pascal manages only 171.5 GFLOPS at a 1:64 ratio.

The CMP 90HX also uses a PCIe 1.0 x4 interface, a deliberate restriction for mining use, versus the TITAN X Pascal’s PCIe 3.0 x16. The CMP 90HX has no display outputs, while the TITAN X Pascal offers a full complement of DVI, HDMI 2.0, and DisplayPort 1.4a.

The Verdict

The data points to a clear split in purpose. The TITAN X Pascal is a general-purpose compute and display card; the CMP 90HX is a mining-oriented part with no video outputs and a restricted PCIe interface. For any workload requiring visual output, the CMP 90HX is unusable, and the TITAN X Pascal is the only option.

In raw compute, the CMP 90HX holds the edge. Its FP32 throughput (21.89 TFLOPS) is exactly twice the TITAN X Pascal’s (10.97 TFLOPS), and its memory bandwidth (760.3 GB/s) is 58% higher. The Geekbench OpenCL result reflects this, with the CMP 90HX leading by 3.3%. However, the TITAN X Pascal still wins on average benchmark score (72,098 vs. 69,000) and percentile rank (91 vs. 90), suggesting that real-world OpenCL workloads may not fully utilize the CMP 90HX’s theoretical compute advantage.

The TITAN X Pascal also offers more VRAM (12 GB vs. 10 GB) and a wider 384-bit bus, though the CMP 90HX compensates with faster GDDR6X. For users who need display outputs, DirectX 12_1 support, or a standard PCIe 3.0 x16 interface, the TITAN X Pascal is the only viable choice. For headless compute tasks where raw FP32 and memory bandwidth dominate, the CMP 90HX’s 3.3% benchmark lead and 2x FP32 throughput make it the stronger pick, assuming the PCIe 1.0 x4 bottleneck and lack of outputs are acceptable.

Specification Differences

| Specification | NVIDIA TITAN X Pascal | NVIDIA CMP 90HX |

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

| Chip | GP102 | GA102 |

| Architecture | Pascal | Ampere |

| Generation | GeForce 10 | Mining GPUs |

| Process Node | 16 nm | 8 nm |

| Foundry | TSMC | Samsung |

| Transistors | 11,800 million | 28,300 million |

| Die Size | 471 mm² | 628 mm² |

| Transistor Density | 25.1M / mm² | 45.1M / mm² |

| Base Clock | 1417 MHz | 1500 MHz |

| Boost Clock | 1531 MHz | 1710 MHz |

| Memory Clock | 1251 MHz (10 Gbps effective) | 1188 MHz (19 Gbps effective) |

| Memory Size | 12 GB | 10 GB |

| Memory Type | GDDR5X | GDDR6X |

| Memory Bus | 384 bit | 320 bit |

| Memory Bandwidth | 480.4 GB/s | 760.3 GB/s |

| Shading Units | 3584 | 6400 |

| TMUs | 224 | 200 |

| ROPs | 96 | 80 |

| RT Cores | None | 50 |

| Tensor Cores | None | 200 |

| Pixel Rate | 147.0 GPixel/s | 136.8 GPixel/s |

| Texture Rate | 342.9 GTexel/s | 342.0 GTexel/s |

| FP32 | 10.97 TFLOPS | 21.89 TFLOPS |

| FP16 | 171.5 GFLOPS (1:64) | 21.89 TFLOPS (1:1) |

| TDP | 250 W | 320 W |

| Power Connectors | 1x 6-pin + 1x 8-pin | 2x 8-pin |

| Suggested PSU | 600 W | 700 W |

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

| Display Outputs | 1x DVI, 1x HDMI 2.0, 3x DisplayPort 1.4a | No outputs |

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

| Length | 267 mm (10.5 inches) | 285 mm (11.2 inches) |

| Height | 112 mm (4.4 inches) | 112 mm (4.4 inches) |

| Width | 40 mm (1.6 inches) | Not specified |

| Release Date | 2016-08-01 | 2021-07-27 |

| Launch MSRP | 1,199 USD | Not specified |

Head-to-Head Benchmarks

The only direct benchmark comparison available is Geekbench OpenCL. In this test, the NVIDIA CMP 90HX scores 69,000 against the TITAN X Pascal’s 66,696, producing a 3.3% delta in favor of the CMP 90HX. This is the sole head-to-head result, and it gives the CMP 90HX one win versus zero for the TITAN X Pascal.

The CMP 90HX’s victory in OpenCL is consistent with its architectural advantages. Its FP32 throughput of 21.89 TFLOPS is exactly double the TITAN X Pascal’s 10.97 TFLOPS, and its memory bandwidth of 760.3 GB/s is 58% higher. The 3.3% score delta, however, is far smaller than the 100% FP32 gap, indicating that OpenCL workloads on this test are not purely compute-bound. Memory latency, driver efficiency, or other factors likely dampen the CMP 90HX’s theoretical advantage.

The TITAN X Pascal, despite losing the head-to-head, holds a higher average benchmark score of 72,098 across its two tests (Geekbench OpenCL at 66,696 and Geekbench Vulkan at 77,499). The CMP 90HX has no Vulkan benchmark listed, so its 69,000 average comes solely from the OpenCL result. The TITAN X Pascal’s Vulkan score of 77,499 is its strongest showing and suggests that in Vulkan-based workloads, it may outperform the CMP 90HX, though no direct comparison exists.

Looking at nearest rivals, the TITAN X Pascal’s 72,098 average is within 1.8% of cards like the AMD Radeon RX 6600 LE (70,829) and AMD Radeon RX 6650M (71,768), while trailing the AMD Radeon Pro Vega 64 (72,379) by just 0.4%. The CMP 90HX’s 69,000 average sits within 1.4% of the NVIDIA Quadro P6000 (69,986) and AMD Radeon Pro WX 8200 (69,870), and ahead of the Intel Arc A770 (68,809) by 0.3%. Both cards occupy a similar overall performance tier, with the TITAN X Pascal holding a slight edge in aggregate scores.

The pixel rates tell a different story. The TITAN X Pascal achieves 147.0 GPixel/s versus the CMP 90HX’s 136.8 GPixel/s, a 7.5% advantage for the older card. This is attributable to the TITAN X Pascal’s higher ROP count (96 vs. 80). Texture rates are nearly identical, with the TITAN X Pascal at 342.9 GTexel/s and the CMP 90HX at 342.0 GTexel/s, a negligible 0.3% difference despite the CMP 90HX’s lower TMU count (200 vs. 224).

The CMP 90HX’s clear wins are in FP32 throughput (2x), FP16 throughput (128x), memory bandwidth (58% higher), and raw transistor count (2.4x). The TITAN X Pascal counters with higher pixel rate, more ROPs, more VRAM, and a standard PCIe interface. The benchmark data, however, shows the CMP 90HX ahead in the only available head-to-head test, even if the margin is modest at 3.3%.