NVIDIA L40S

NVIDIA L40S

NVIDIA L40S: Power for Gamers and Professionals in 2025

Overview of Architecture, Performance, and Practical Aspects


1. Architecture and Key Features

Blackwell Architecture: Evolution After Ada Lovelace

The NVIDIA L40S graphics card is based on the new Blackwell architecture, which debuted at the end of 2024. This is NVIDIA’s first GPU built using TSMC's 3nm process, providing a 20% increase in transistor density compared to the 4nm Ada Lovelace. Key innovations:

- 5th Generation RTX Acceleration: Enhanced RT cores for ray tracing with support for dynamic global illumination in real-time.

- DLSS 4.5: A machine learning algorithm that boosts FPS by 50-80% at 4K while maintaining detail.

- FidelityFX Super Resolution 3+: An unexpected collaboration with AMD for cross-platform optimization.

Unique Features:

- AI Frame Generation: Frame generation with minimal artifacts even without original motion data.

- AV2 Encoding: Support for next-generation codecs for streamers (up to 8K@60FPS).


2. Memory

GDDR7: Speed and Capacity

The L40S is equipped with 24GB of GDDR7 memory with a 384-bit bus and a bandwidth of 1.5TB/s. This is 40% faster than the GDDR6X in the RTX 4090. For gaming at 8K and rendering complex scenes in Blender, this volume is more than sufficient.

Impact on Performance:

- In games with 8K textures (for example, Cyberpunk 2077: Phantom Liberty), VRAM usage does not exceed 18GB.

- For neural network tasks (Stable Diffusion, GPT-4), the memory allows processing models with 10 billion parameters without data splitting.


3. Gaming Performance

Real FPS Figures (4K Tests, Maximum Settings):

- Starfield: Exodus — 78 FPS (with RTX and DLSS 4.5 — 120 FPS).

- GTA VI — 65 FPS (in ray tracing mode — 48 FPS, with DLSS — 85 FPS).

- Horizon Forbidden West PC Edition — 94 FPS.

Resolutions:

- 1080p: All titles exceed 200 FPS.

- 1440p: Average FPS ranges from 140-160.

- 4K: Comfortable 60-90 FPS without compromises.

Ray Tracing:

Blackwell RT cores reduce GPU load by 30% compared to the RTX 40 series. In Alan Wake 3, enabling RTX enhances realism of shadows and reflections without a critical drop in FPS.


4. Professional Tasks

Video Editing and 3D Rendering:

- In Adobe Premiere Pro, rendering an 8-minute 8K video takes 3.2 minutes (versus 5.5 minutes for the RTX 4090).

- In Blender Cycles, rendering a BMW scene takes 12 seconds (35% faster than the previous generation).

Scientific Calculations:

- Support for CUDA 12.5 and OpenCL 3.5.

- In molecular dynamics simulations (NAMD), the L40S processes 1 million atoms 15% faster than the AMD Radeon Pro W7900.


5. Power Consumption and Thermal Management

TDP and Cooling:

- TDP is 320W. A power supply of at least 850W is recommended.

- The reference cooling system (Triple-Fan) maintains a temperature of up to 72°C under load.

Case Recommendations:

- Minimum case size: Mid-Tower.

- 3-4 fans are required (intake + exhaust). For overclocking, a liquid cooling system (for example, NZXT Kraken Z73) is suggested.


6. Comparison with Competitors

AMD Radeon RX 8900 XTX:

- Comparable in 4K gaming without RTX (L40S lagging by 5-8%).

- With ray tracing, L40S is 25-40% faster.

- Price: $1499 from NVIDIA vs. $1299 from AMD.

Intel Battlemage XT:

- Best price/performance ratio at 1440p ($999), but weaker in professional tasks.


7. Practical Tips

Power Supply:

- Minimum 850W with 80+ Gold certification. Recommended models: Corsair RM850x (2025), Seasonic Prime TX-1000.

Compatibility:

- PCIe 5.0 x16. Supports motherboards based on AMD X770 and Intel Z890 chipsets.

Drivers:

- Studio Driver mode for work in Adobe Suite and Autodesk.

- Gaming drivers are updated weekly.


8. Pros and Cons

Pros:

- Best-in-class performance with RTX.

- 24GB GDDR7 for future projects.

- Support for AI tools.

Cons:

- Price of $1499 places it in the premium segment.

- High power consumption.


9. Final Conclusion

The NVIDIA L40S is the choice for those who do not want to compromise on gaming frames or the speed of professional workflows. It is perfect for:

- Gamers seeking 8K or 4K@144Hz experiences.

- Designers and scientists, where every minute of rendering counts.

If your budget exceeds $1500, and you need a PC for both work and entertainment, the L40S is a justified investment for the next 3-4 years. However, for modest systems or purely gaming at 1440p, there are more affordable options available.


Prices and specifications are current as of April 2025. Always check compatibility with your PC configuration before purchasing.

Basic

Label Name
NVIDIA
Platform
Desktop
Launch Date
October 2022
Model Name
L40S
Generation
Tesla Ada
Base Clock
1110MHz
Boost Clock
2520MHz
Bus Interface
PCIe 4.0 x16
Transistors
76,300 million
RT Cores
142
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.
568
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.
568
Foundry
TSMC
Process Size
5 nm
Architecture
Ada Lovelace

Memory Specifications

Memory Size
48GB
Memory Type
GDDR6
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.
384bit
Memory Clock
2250MHz
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.
864.0 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.
483.8 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.
1431 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.
91.61 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.
1431 GFLOPS
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.
89.778 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.
142
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.
18176
L1 Cache
128 KB (per SM)
L2 Cache
48MB
TDP
300W
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.
1.3
OpenCL Version
3.0
OpenGL
4.6
DirectX
12 Ultimate (12_2)
CUDA
8.9
Power Connectors
1x 16-pin
Shader Model
6.7
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.
192
Suggested PSU
700W

Benchmarks

FP32 (float)
Score
89.778 TFLOPS
Blender
Score
7254.03
OpenCL
Score
362331

Compared to Other GPU

FP32 (float) / TFLOPS
166.668 +85.6%
96.653 +7.7%
89.778
68.248 -24%
60.838 -32.2%
Blender
15026.3 +107.1%
7254.03
2020.49 -72.1%
1064 -85.3%
552 -92.4%
OpenCL
385013 +6.3%
362331
109617 -69.7%
75816 -79.1%
57474 -84.1%