NVIDIA CMP 40HX

The NVIDIA CMP 40HX is a Turing-architecture mining accelerator built on the 12 nm TU106 chip, featuring 2,304 shading units, 8 GB of GDDR6 memory, and a 185 W TDP. With an average benchmark score of 85,637, it sits in the 94th percentile of all GPUs, placing it in a competitive bracket against both professional and consumer cards. Its benchmark results, derived from Geekbench OpenCL and Vulkan tests, reveal a device that is surprisingly versatile despite its mining-focused design and lack of display outputs.

How It Compares

The closest rival, the AMD Radeon PRO W7600, posts an average score of 85,851, which is just 0.2% higher than the CMP 40HX. This margin is negligible in real-world terms, meaning the two cards are effectively performance equals in compute workloads. The data suggests that for raw number-crunching tasks, the CMP 40HX matches a modern professional workstation card, despite being built on older 12 nm technology.

Against the NVIDIA GeForce RTX 5090, the CMP 40HX actually leads by 1.6%, with the rival scoring 84,306. This is a counterintuitive result, as the RTX 5090 is a flagship consumer GPU, yet the benchmark data shows the mining card holding a slight edge. The implication is that the CMP 40HX's compute-oriented design, without display output overhead, may contribute to its higher raw throughput in these specific tests.

The NVIDIA GeForce RTX 5090 D trails by a similar margin, scoring 84,241, which is 1.7% lower than the CMP 40HX. This variant, likely with some feature restrictions, still cannot surpass the mining card in the averaged scores. The data indicates that the CMP 40HX's performance profile is not easily predicted by its market segment, as it outpaces even newer flagship hardware in these benchmarks.

The final rival, the NVIDIA GeForce RTX 5050 Mobile, scores 84,171, also 1.7% behind the CMP 40HX. This comparison is particularly telling, as the mobile chip likely operates at lower clocks due to thermal constraints, while the CMP 40HX benefits from a dual-slot desktop design with dedicated cooling. The 1.7% lead reinforces that the mining card delivers consistent, sustained performance without the power limitations of laptop components.

Memory Subsystem

The CMP 40HX is equipped with 8 GB of GDDR6 memory on a 256-bit bus, yielding a bandwidth of 448.0 GB/s. This memory configuration is well-suited for high-resolution workloads, as the substantial bus width allows for efficient data transfer between the GPU and memory. The 14 Gbps effective memory speed ensures that large datasets, common in compute and rendering tasks, can be fed to the processing cores without becoming a bottleneck.

For high-resolution scenarios, the 8 GB capacity is a limiting factor in some modern applications, but the bandwidth helps mitigate this by moving data quickly. The 448.0 GB/s figure is competitive with many workstation cards of the same era, and the 256-bit interface provides a balanced ratio of capacity to throughput. Benchmark results indicate that the memory subsystem performs admirably, contributing to the card's strong overall scores in OpenCL and Vulkan tests.

The choice of GDDR6 over newer memory types means the card relies on mature technology, but the 14 Gbps effective speed is sufficient to support the 7.603 TFLOPS of FP32 compute power. In memory-intensive tasks, the data shows no obvious weakness, as the card's average score remains within 0.2% of the Radeon PRO W7600, which likely uses more modern memory. The practical impact is that users can expect consistent performance across memory-heavy workloads up to the 8 GB limit.

Ray Tracing and Feature Set

The CMP 40HX includes 36 RT cores and 288 tensor cores, bringing Turing's dedicated ray tracing and AI acceleration hardware to a mining-focused product. This is unusual for a card with no display outputs, as these features are typically associated with gaming or creative workloads. The presence of these cores suggests the card was designed to handle a variety of compute tasks beyond simple cryptographic hashing.

API support is robust, with DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4 all listed. This means the card can run modern graphics workloads, even though it cannot output video. The DirectX 12 Ultimate support indicates hardware-level ray tracing capabilities, which the 36 RT cores can accelerate, while the 288 tensor cores provide AI inference and DLSS-style processing power, though the latter is irrelevant without display outputs.

The inclusion of these features in a mining card raises questions about NVIDIA's intent. The data shows that the card's compute performance is strong, and the RT and tensor cores add versatility for non-graphics tasks like machine learning inference or scientific simulation. The Vulkan 1.4 support further enhances its utility as a general-purpose compute device, allowing it to work with a wide range of APIs and frameworks.

FAQ

Q: What is the average benchmark score of the NVIDIA CMP 40HX?

A: The average benchmark score is 85,637, based on Geekbench OpenCL and Vulkan tests, placing it in the 94th percentile of all GPUs.

Q: How does the CMP 40HX compare to the AMD Radeon PRO W7600?

A: The CMP 40HX scores 85,637, which is 0.2% lower than the Radeon PRO W7600's 85,851, making them effectively equal in performance.

Q: Does the CMP 40HX support ray tracing?

A: Yes, it includes 36 RT cores and supports DirectX 12 Ultimate (12_2), which enables hardware-accelerated ray tracing in compatible applications.

Q: What is the memory bandwidth of the CMP 40HX?

A: The card has a 256-bit memory bus with GDDR6 memory, delivering 448.0 GB/s of bandwidth.

Q: What power supply is recommended for the CMP 40HX?

A: The suggested PSU is 450 W, and the card requires a single 8-pin power connector.

Q: Can the CMP 40HX output video to a display?

A: No, the card has no display outputs, as it was designed specifically for mining compute tasks.

Benchmark Performance

The Geekbench OpenCL score of 93,395 and Vulkan score of 77,879 combine to produce an average of 85,637. The OpenCL result is notably higher, suggesting the card excels in general compute workloads, while the Vulkan score is lower but still respectable. This discrepancy indicates that the CMP 40HX's architecture is optimized for certain types of workloads, possibly those that leverage the tensor cores or memory bandwidth extensively.

Comparing to rivals, the CMP 40HX leads the RTX 5090 by 1.6% and the RTX 5090 D by 1.7%, which is surprising given the generational gap. The RTX 5090, with an average score of 84,306, is a modern flagship, yet the mining card edges it out. This suggests that the CMP 40HX's 7.603 TFLOPS of FP32 performance and 448.0 GB/s bandwidth are well-tuned for the specific operations in Geekbench, or that the mining card's lack of display output overhead provides a slight efficiency advantage.

The RTX 5050 Mobile trails by 1.7% with a score of 84,171, which is expected given its mobile form factor and likely lower clocks. The CMP 40HX's 1,650 MHz boost clock, combined with its dual-slot cooling, allows it to maintain performance under sustained load, something that mobile parts often struggle with. The data implies that the CMP 40HX is a reliable performer, even if its mining-focused design might lead to assumptions of lower compute capability.

The 0.2% gap to the Radeon PRO W7600 is within statistical noise, meaning the two cards are interchangeable in terms of raw compute. This is a strong result for the CMP 40HX, as the Radeon PRO W7600 is a professional workstation card with presumably more modern architecture. The benchmark data collectively shows that the CMP 40HX, despite its niche purpose, delivers performance that rivals or exceeds dedicated compute and gaming hardware.

Power and Cooling

The CMP 40HX has a TDP of 185 W, which is modest for a card with 2,304 shading units and 36 RT cores. This power draw is manageable with a suggested PSU of 450 W, and the card requires only a single 8-pin power connector. The dual-slot cooling solution, with dimensions of 229 mm in length, 111 mm in height, and 35 mm in width, suggests a standard heatsink and fan design typical of mid-range GPUs.

The 185 W TDP is lower than many comparable cards, which may be a result of the mining-focused design prioritizing efficiency over raw clocks. The base clock of 1,470 MHz and boost clock of 1,650 MHz are moderate, and the 12 nm process, while older, is well-understood for thermal management. The data indicates that the card should run cool and quiet under typical loads, given the dual-slot cooler and reasonable power envelope.

The power connector requirement of 1x 8-pin is standard, and the 450 W PSU recommendation leaves adequate headroom for a typical system. The card's end-of-life production status means it is no longer manufactured, but its power characteristics remain relevant for those using it in compute rigs. The 35 mm width confirms it is a true dual-slot card, requiring two expansion slots for installation, which is typical for this class of GPU.

Who Should Consider It

The CMP 40HX is positioned for users who need raw compute performance without the need for display output. Its average score of 85,637, which is 1.6% ahead of the RTX 5090, makes it a compelling option for headless compute servers or mining rigs where graphics output is unnecessary. The 8 GB of VRAM and 448.0 GB/s bandwidth are sufficient for many scientific and AI workloads, though the capacity may limit the largest datasets.

For high-resolution compute tasks, the card's memory bandwidth is a strong asset, but the 8 GB capacity means users should carefully assess their VRAM needs. The 94th percentile ranking indicates it outperforms the vast majority of GPUs, making it suitable for tasks like rendering, simulation, or machine learning inference. The lack of display outputs is a clear constraint, but for users who access results remotely, this is not an issue.

The card's 7.603 TFLOPS of FP32 performance and 288 tensor cores make it a viable option for AI inference, where the tensor cores can accelerate matrix operations. The 36 RT cores, while designed for ray tracing, also provide additional compute capability. Users with existing infrastructure that can handle a card without video outputs will find the CMP 40HX to be a high-performance addition, based on the benchmark data showing it rivals and often exceeds modern flagship GPUs.