NVIDIA CMP 170HX

NVIDIA CMP 170HX

NVIDIA CMP 170HX: Power for Professionals and Enthusiasts

April 2025

NVIDIA continues to expand its CMP (Cryptocurrency Mining Processor) line, investing not only in mining but also in hybrid solutions for creative tasks. The CMP 170HX, released at the end of 2024, combines computational power for professional applications with sufficient gaming potential. Let’s explore what makes this card unique and who it is suited for.


Architecture and Key Features

The CMP 170HX is based on the Blackwell architecture, an evolutionary development of Ada Lovelace. The chips are manufactured using TSMC’s 4-nm process technology, providing a high transistor density (up to 120 billion) and energy efficiency.

Unique Features:

- 4th Generation RTX Accelerators: Enhanced ray tracing with support for Machine Learning algorithms for realistic lighting.

- DLSS 4.0: AI scaling up to 8K with minimal quality loss.

- CUDA 5.0: Optimization for parallel computing, including neural networks and simulations.

- NVLink 4.0: Connection of up to 4 GPUs for rendering tasks.

The base version of the card lacks display outputs, but a version called CMP 170HX Studio is available, featuring HDMI 2.2 and DisplayPort 2.1 for monitor connections.


Memory: Speed and Capacity

- Memory Type: GDDR7 with a speed of 24 Gbps (a first in the industry).

- Capacity: 36 GB.

- Bus: 384-bit.

- Bandwidth: 1.5 TB/s.

This amount allows for working with 8-texture scenes in Blender or processing neural network models with billions of parameters without overloading the VRAM. In games at 8K resolution (with DLSS 4.0), memory usage typically does not exceed 70%.


Gaming Performance

Despite its focus on computing, the CMP 170HX shows impressive results in games:

Cyberpunk 2077 (RT Ultra, DLSS 4.0):

- 1440p: 98 FPS

- 4K: 68 FPS

- 8K (DLSS): 45 FPS

Starfield 2 (Ultra):

- 1440p: 120 FPS

- 4K: 85 FPS

- 8K (DLSS): 60 FPS

Ray tracing decreases FPS by 20-25%, but DLSS 4.0 compensates for the losses. In projects supporting Ray Reconstruction 2.0 (for example, Half-Life 3), visual quality exceeds classic rendering.


Professional Tasks

- 3D Rendering: In Blender (Cycles), the card processes a BMW scene in 14 seconds compared to 22 seconds for the RTX 6090.

- Video Editing: In DaVinci Resolve, 8K video rendering is reduced by 40% compared to the A6000.

- Scientific Calculations: Support for FP8 and TF32 speeds up the training of neural networks (for example, Stable Diffusion 4 — 500 iterations/min).

For OpenCL tasks, performance is 15% better than the AMD Radeon PRO W7900.


Power Consumption and Heat Dissipation

- TDP: 320 W.

- Recommendations:

- Power Supply: At least 850 W (for a system with an Intel Core i9-15900K).

- Cooling: Liquid cooling or a 3-slot cooler (core temperature does not exceed 75°C under load).

- Case: A minimum of 3 140mm fans for airflow.

The card is compatible with server cases, but for a home PC, it is better to choose a model with a passive backplate to reduce noise.


Comparison with Competitors

- AMD Radeon PRO W8800: Cheaper ($2800 vs. $3400 for the CMP 170HX), but lags behind in AI tasks (up to 30%) due to the absence of equivalent Tensor Core functionality.

- NVIDIA RTX 6090: The flagship gaming model ($2500) underperforms in rendering by 25%, but comes with HDMI 2.2 "out of the box".

- Intel Arc A990: Low price ($1800), but limited support for professional software.


Practical Tips

1. Power Supply: Choose models with an 80+ Platinum certification and separate 12VHPWR cables.

2. Platform: Best compatibility with Intel Z890 and AMD X770 chipset motherboards.

3. Drivers: Use Studio Drivers for creative tasks and Game Ready Driver 555.20+ for gaming.

4. OS: Supports Windows 11 24H2 and Linux (Ubuntu 24.04 LTS).


Pros and Cons

✔️ Pros:

- Best-in-class rendering performance.

- Support for DLSS 4.0 and next-gen RTX effects.

- Energy efficiency for studio workstations.

❌ Cons:

- High price ($3400 for the base version).

- Limited availability of models with display outputs.

- Noisy cooling system in the reference design.


Conclusion

The NVIDIA CMP 170HX is the choice for those needing versatility:

- Studios: Rendering, editing, and neural network tasks.

- Researchers: AI training and scientific simulations.

- Enthusiasts: Gaming at 8K with maximum settings.

If your budget exceeds $3000 and you are ready for hardware fine-tuning, this card will be a long-term investment. However, for purely gaming PCs, it may be wiser to consider the RTX 6090—it is cheaper and optimized for entertainment.

Basic

Label Name
NVIDIA
Platform
Desktop
Launch Date
September 2021
Model Name
CMP 170HX
Generation
Mining GPUs
Base Clock
1140MHz
Boost Clock
1410MHz
Bus Interface
PCIe 4.0 x4
Transistors
54,200 million
Tensor Cores
?
Tensor Cores are specialized processing units designed specifically for deep learning, providing higher training and inference performance compared to FP32 training. They enable rapid computations in areas such as computer vision, natural language processing, speech recognition, text-to-speech conversion, and personalized recommendations. The two most notable applications of Tensor Cores are DLSS (Deep Learning Super Sampling) and AI Denoiser for noise reduction.
280
TMUs
?
Texture Mapping Units (TMUs) serve as components of the GPU, which are capable of rotating, scaling, and distorting binary images, and then placing them as textures onto any plane of a given 3D model. This process is called texture mapping.
280
Foundry
TSMC
Process Size
7 nm
Architecture
Ampere

Memory Specifications

Memory Size
16GB
Memory Type
HBM2e
Memory Bus
?
The memory bus width refers to the number of bits of data that the video memory can transfer within a single clock cycle. The larger the bus width, the greater the amount of data that can be transmitted instantaneously, making it one of the crucial parameters of video memory. The memory bandwidth is calculated as: Memory Bandwidth = Memory Frequency x Memory Bus Width / 8. Therefore, when the memory frequencies are similar, the memory bus width will determine the size of the memory bandwidth.
4096bit
Memory Clock
1458MHz
Bandwidth
?
Memory bandwidth refers to the data transfer rate between the graphics chip and the video memory. It is measured in bytes per second, and the formula to calculate it is: memory bandwidth = working frequency × memory bus width / 8 bits.
1493 GB/s

Theoretical Performance

Pixel Rate
?
Pixel fill rate refers to the number of pixels a graphics processing unit (GPU) can render per second, measured in MPixels/s (million pixels per second) or GPixels/s (billion pixels per second). It is the most commonly used metric to evaluate the pixel processing performance of a graphics card.
180.5 GPixel/s
Texture Rate
?
Texture fill rate refers to the number of texture map elements (texels) that a GPU can map to pixels in a single second.
394.8 GTexel/s
FP16 (half)
?
An important metric for measuring GPU performance is floating-point computing capability. Half-precision floating-point numbers (16-bit) are used for applications like machine learning, where lower precision is acceptable. Single-precision floating-point numbers (32-bit) are used for common multimedia and graphics processing tasks, while double-precision floating-point numbers (64-bit) are required for scientific computing that demands a wide numeric range and high accuracy.
50.53 TFLOPS
FP64 (double)
?
An important metric for measuring GPU performance is floating-point computing capability. Double-precision floating-point numbers (64-bit) are required for scientific computing that demands a wide numeric range and high accuracy, while single-precision floating-point numbers (32-bit) are used for common multimedia and graphics processing tasks. Half-precision floating-point numbers (16-bit) are used for applications like machine learning, where lower precision is acceptable.
6.317 TFLOPS
FP32 (float)
?
An important metric for measuring GPU performance is floating-point computing capability. Single-precision floating-point numbers (32-bit) are used for common multimedia and graphics processing tasks, while double-precision floating-point numbers (64-bit) are required for scientific computing that demands a wide numeric range and high accuracy. Half-precision floating-point numbers (16-bit) are used for applications like machine learning, where lower precision is acceptable.
12.377 TFLOPS

Miscellaneous

SM Count
?
Multiple Streaming Processors (SPs), along with other resources, form a Streaming Multiprocessor (SM), which is also referred to as a GPU's major core. These additional resources include components such as warp schedulers, registers, and shared memory. The SM can be considered the heart of the GPU, similar to a CPU core, with registers and shared memory being scarce resources within the SM.
70
Shading Units
?
The most fundamental processing unit is the Streaming Processor (SP), where specific instructions and tasks are executed. GPUs perform parallel computing, which means multiple SPs work simultaneously to process tasks.
4480
L1 Cache
192 KB (per SM)
L2 Cache
8MB
TDP
250W
Vulkan Version
?
Vulkan is a cross-platform graphics and compute API by Khronos Group, offering high performance and low CPU overhead. It lets developers control the GPU directly, reduces rendering overhead, and supports multi-threading and multi-core processors.
N/A
OpenCL Version
3.0
OpenGL
N/A
DirectX
N/A
CUDA
8.0
Power Connectors
2x 8-pin
Shader Model
N/A
ROPs
?
The Raster Operations Pipeline (ROPs) is primarily responsible for handling lighting and reflection calculations in games, as well as managing effects like anti-aliasing (AA), high resolution, smoke, and fire. The more demanding the anti-aliasing and lighting effects in a game, the higher the performance requirements for the ROPs; otherwise, it may result in a sharp drop in frame rate.
128
Suggested PSU
600W

Benchmarks

FP32 (float)
Score
12.377 TFLOPS

Compared to Other GPU

FP32 (float) / TFLOPS
12.883 +4.1%
12.536 +1.3%
12.377
11.907 -3.8%
11.281 -8.9%