NVIDIA CMP 90HX vs NVIDIA Quadro RTX 6000 Comparison

NVIDIA Quadro RTX 6000 and NVIDIA CMP 90HX represent two very different purposes from the same manufacturer. The Quadro RTX 6000 is a workstation-class Turing product aimed at professional visualization, while the CMP 90HX is a mining-oriented Ampere part with no display outputs. The recorded data shows only one overlapping benchmark, Geekbench OpenCL, which provides a narrow but informative comparison.

Head-to-Head Benchmarks

The sole head-to-head benchmark in the database is Geekbench OpenCL. The NVIDIA Quadro RTX 6000 scores 74,179, while the NVIDIA CMP 90HX scores 69,000. This gives the Quadro a 7.5% advantage in compute performance as measured by this test. The delta is modest but consistent with the architectural differences between the two cards.

Looking at the broader context, the Quadro RTX 6000 sits in the 94th percentile of all GPUs in the database, with an average benchmark score of 101,872 across multiple tests. Its nearest rivals include the AMD Radeon Pro Vega II Duo (average score 106,750, which is 4.6% higher) and the AMD Radeon Pro W6600X (average score 107,342, which is 5.1% higher). On the lower side, the AMD Radeon RX 7900M trails by 4.5% (average 97,487) and the AMD Radeon Pro VII trails by 4.9% (average 97,131).

The CMP 90HX, by contrast, sits in the 90th percentile with an average benchmark score of 69,000 from its single OpenCL result. Its nearest rivals are tightly clustered: the Intel Arc A770 is just 0.3% behind (68,809), the AMD Radeon Instinct MI25 is 0.6% behind (68,562), the AMD Radeon Pro WX 8200 is 1.2% ahead (69,870), and the NVIDIA Quadro P6000 is 1.4% ahead (69,986). This suggests the CMP 90HX is competitive with a specific band of professional and consumer GPUs, but it does not reach the higher absolute performance of the Quadro RTX 6000.

The 7.5% lead for the Quadro RTX 6000 in OpenCL is meaningful but not overwhelming. It indicates that despite the CMP 90HX having a higher raw FP32 throughput (21.89 TFLOPS versus 16.31 TFLOPS), the actual measured compute workload favors the Turing architecture. This could stem from driver optimizations, memory subsystem behavior, or the specific characteristics of the OpenCL test. The data does not support a simple clock-for-clock or core-for-core comparison; the measured result is what matters.

FAQ

Q: Which GPU has the higher Geekbench OpenCL score?

A: The NVIDIA Quadro RTX 6000 scores 74,179 versus 69,000 for the NVIDIA CMP 90HX, a 7.5% advantage for the Quadro.

Q: How does the CMP 90HX compare to its nearest rival, the Intel Arc A770?

A: The CMP 90HX averages 69,000, which is 0.3% higher than the Intel Arc A770's average of 68,809.

Q: What percentile does each GPU occupy in the database?

A: The Quadro RTX 6000 is in the 94th percentile of all GPUs, while the CMP 90HX is in the 90th percentile.

Q: Does the CMP 90HX have any display outputs?

A: No, the CMP 90HX has no display outputs, whereas the Quadro RTX 6000 offers 4x DisplayPort 1.4a and 1x USB Type-C.

Q: What is the memory configuration difference?

A: The Quadro RTX 6000 has 24 GB of GDDR6 on a 384-bit bus with 672.0 GB/s bandwidth, while the CMP 90HX has 10 GB of GDDR6X on a 320-bit bus with 760.3 GB/s bandwidth.

Q: Which GPU has a higher FP32 throughput?

A: The CMP 90HX has 21.89 TFLOPS FP32, while the Quadro RTX 6000 has 16.31 TFLOPS, making the CMP 90HX approximately 34% higher on paper, though the measured OpenCL score favors the Quadro.

The Verdict

Based strictly on the recorded benchmark data, the NVIDIA Quadro RTX 6000 is the better choice for raw compute performance in the one test where both are measured. The 7.5% lead in Geekbench OpenCL, combined with its higher 94th percentile standing versus the CMP 90HX's 90th percentile, positions the Quadro as the stronger overall performer in this comparison.

However, the CMP 90HX is not a workstation card. It has no display outputs, which eliminates it from any visual computing use case. The Quadro RTX 6000, with its 4x DisplayPort 1.4a and USB Type-C outputs, is the only viable option for tasks requiring graphics output. The CMP 90HX's purpose is compute-heavy workloads where display output is irrelevant, such as mining, but the benchmark data shows it does not outperform the Quadro in OpenCL despite its higher theoretical FP32 rate.

For users who need a professional GPU with display capability and measured compute performance, the Quadro RTX 6000 is the clear pick. For users who need a mining-specific card and do not require display outputs, the CMP 90HX is the only option in this pair, but the data shows it trails the Quadro in the recorded compute test. The verdict is straightforward: the Quadro RTX 6000 wins the benchmark comparison, while the CMP 90HX wins only in the niche of having no display outputs at all.

Specification Differences

The two cards differ across nearly every major specification. The Quadro RTX 6000 uses a 12 nm TSMC process with 18,600 million transistors on a 754 mm² die, while the CMP 90HX uses an 8 nm Samsung process with 28,300 million transistors on a 628 mm² die. The transistor density reflects this: 24.7M per mm² for the Quadro versus 45.1M per mm² for the CMP.

Clock speeds also differ. The Quadro has a base clock of 1440 MHz and a boost of 1770 MHz, while the CMP has a base of 1500 MHz and a boost of 1710 MHz. Memory clocks are 1750 MHz (14 Gbps effective) for the Quadro and 1188 MHz (19 Gbps effective) for the CMP.

Memory capacity and type diverge sharply: the Quadro offers 24 GB of GDDR6 on a 384-bit bus, while the CMP offers 10 GB of GDDR6X on a 320-bit bus. Bandwidth favors the CMP at 760.3 GB/s versus 672.0 GB/s for the Quadro.

The compute units differ as well. The Quadro has 4,608 shading units, 288 TMUs, 96 ROPs, 72 RT cores, and 576 tensor cores. The CMP has 6,400 shading units, 200 TMUs, 80 ROPs, 50 RT cores, and 200 tensor cores. Pixel rate is higher on the Quadro (169.9 GPixel/s versus 136.8 GPixel/s), while texture rate is higher on the Quadro as well (509.8 GTexel/s versus 342.0 GTexel/s). FP32 throughput is higher on the CMP (21.89 TFLOPS versus 16.31 TFLOPS), and FP16 differs in ratio: the Quadro offers 32.62 TFLOPS with a 2:1 ratio, while the CMP offers 21.89 TFLOPS with a 1:1 ratio.

Power and physical dimensions also vary: the Quadro has a TDP of 260 W with a 1x 6-pin plus 1x 8-pin connector and a suggested PSU of 600 W, while the CMP has a TDP of 320 W with 2x 8-pin connectors and a suggested PSU of 700 W. The Quadro is 267 mm long and 111 mm tall, while the CMP is 285 mm long and 112 mm tall. The bus interface is PCIe 3.0 x16 for the Quadro and PCIe 1.0 x4 for the CMP. The Quadro has display outputs, the CMP has none.

Architecture Differences

The architectures are fundamentally different generations. The Quadro RTX 6000 uses the Turing architecture on the TU102 chip, while the CMP 90HX uses the Ampere architecture on the GA102 chip. The process nodes reflect this generational shift: 12 nm for Turing, 8 nm for Ampere. The foundry changes from TSMC to Samsung.

The transistor count and die size tell a story of scaling. The CMP's GA102 packs 28,300 million transistors into 628 mm², a much denser design than the Quadro's 18,600 million transistors on 754 mm². The density difference is stark: 45.1M per mm² for the CMP versus 24.7M per mm² for the Quadro.

The memory architecture differs in type and width. The Quadro uses GDDR6 on a 384-bit bus, while the CMP uses GDDR6X on a 320-bit bus. The GDDR6X allows higher effective data rates (19 Gbps versus 14 Gbps), which explains why the CMP achieves higher bandwidth despite a narrower bus.

The compute feature sets differ in RT and tensor core counts. The Quadro has 72 RT cores and 576 tensor cores, while the CMP has 50 RT cores and 200 tensor cores. The FP16 implementation differs: the Quadro achieves 32.62 TFLOPS with a 2:1 ratio (meaning FP16 is double the FP32 rate), while the CMP achieves 21.89 TFLOPS with a 1:1 ratio (FP16 equals FP32). This suggests the Quadro's Turing architecture is optimized for mixed-precision workloads, while the CMP's Ampere design treats FP16 and FP32 at the same throughput.

The bus interface is another major architectural difference: the Quadro uses PCIe 3.0 x16, while the CMP uses PCIe 1.0 x4. This severely limits the CMP's host communication bandwidth, which is consistent with its mining-focused design where data transfer is less critical than raw compute.

Where Each One Wins

The Quadro RTX 6000 wins in the measured compute benchmark. Its 7.5% lead in Geekbench OpenCL is the only head-to-head data point, and it also wins on pixel rate (169.9 GPixel/s versus 136.8 GPixel/s), texture rate (509.8 GTexel/s versus 342.0 GTexel/s), and FP16 throughput (32.62 TFLOPS versus 21.89 TFLOPS). It also has more memory capacity (24 GB versus 10 GB) and display outputs, making it the only card of the two suitable for visual workloads.

The CMP 90HX wins on raw FP32 throughput (21.89 TFLOPS versus 16.31 TFLOPS), memory bandwidth (760.3 GB/s versus 672.0 GB/s), and shading unit count (6,400 versus 4,608). It also has a higher base clock (1500 MHz versus 1440 MHz) and a denser transistor design. However, these theoretical advantages do not translate into a higher OpenCL score in the recorded data.

The use-case split is clear. The Quadro RTX 6000 is for professional visualization, rendering, and any task requiring display output or high memory capacity. Its 24 GB frame buffer and 4x DisplayPort outputs make it a workstation card. The CMP 90HX is for compute-heavy, display-less workloads such as mining. Its higher FP32 rate and bandwidth could benefit certain compute kernels, but the measured OpenCL result shows it behind the Quadro. The CMP's PCIe 1.0 x4 interface also makes it unsuitable for general-purpose computing where host transfer speeds matter.

In summary, the Quadro RTX 6000 wins the benchmark comparison and offers broader utility due to display support and memory capacity. The CMP 90HX wins only in niche specifications like FP32 throughput and bandwidth, but the database's recorded performance metric favors the Quadro.