GPU Benchmarks NVIDIA H100 NVL (SXM5) vs. NVIDIA RTX 4090
NVIDIA H100 NVL (SXM5)
NVIDIA RTX 4090
Quick links:
- Benchmarks
- Specifications
- Deep Learning GPU Benchmarks 2024–2025
- 3D, GPU Rendering Benchmarks 2024–2025
- RELION Cryo-EM Benchmarks 2024-2025
- Llama3 70B Inference Benchmark 2024–2025
Best GPUs for deep learning, AI development, compute in 2023–2024. Recommended GPU & hardware for AI training, inference (LLMs, generative AI). GPU training, inference benchmarks using PyTorch, TensorFlow for computer vision (CV), NLP, text-to-speech, etc.
We benchmark NVIDIA H100 NVL (SXM5) vs NVIDIA RTX 4090 GPUs and compare AI performance (deep learning training; FP16, FP32, PyTorch, TensorFlow), 3d rendering, Cryo-EM performance in the most popular apps (Octane, VRay, Redshift, Blender, Luxmark, Unreal Engine, Relion Cryo-EM).
Our benchmarks will help you decide which GPU (NVIDIA RTX 4090/4080, H100 Hopper, H200, A100, RTX 6000 Ada, A6000, A5000, or RTX 6000 ADA Lovelace) is the best GPU for your needs. We provide an in-depth analysis of the AI performance of each graphic card's performance so you can make the most informed decision possible. We offer deep learning and 3d rendering benchmarks that will help you get the most out of your hardware.
Looking for a GPU workstation or server for AI/ML, design, rendering, simulation or molecular dynamics?
Explore BIZON AI workstations or GPU servers.
Contact us today or explore our various customizable AI solutions.
Featured GPU benchmarks:
- NVIDIA H100 vs. A100 vs. RTX 6000 Ada
- NVIDIA RTX 4090 vs. RTX 4080 vs. RTX 3090
- NVIDIA RTX 3080 vs. 3080 Ti vs. RTX 3090
- NVIDIA RTX 3090 vs. NVIDIA A6000
- NVIDIA A6000 vs. A5000 vs. NVIDIA RTX 3090
- NVIDIA A100 vs. NVIDIA A6000
- NVIDIA RTX 2080 Ti vs. Titan RTX vs Quadro RTX8000
Benchmarks
Deep Learning GPU Benchmarks 2024–2025 [Updated]↑
As of May 2025
Resnet50 (FP16)
1 GPU
NVIDIA H100 NVL (SXM5)
4197 points
NVIDIA RTX 4090
1720 points
4 GPU
NVIDIA H100 NVL (SXM5)
15945 points
NVIDIA RTX 4090
5934 points
8 GPU
NVIDIA H100 NVL (SXM5)
41496 points
NVIDIA RTX 4090
n/a
higher is better
Resnet50 (FP32)
1 GPU
NVIDIA H100 NVL (SXM5)
1903 points
NVIDIA RTX 4090
927 points
4 GPU
NVIDIA H100 NVL (SXM5)
7828 points
NVIDIA RTX 4090
1715 points
8 GPU
NVIDIA H100 NVL (SXM5)
20616 points
NVIDIA RTX 4090
n/a
higher is better
Resnet152 (FP16)
1 GPU
NVIDIA H100 NVL (SXM5)
1701 points
NVIDIA RTX 4090
n/a
4 GPU
NVIDIA H100 NVL (SXM5)
6461 points
NVIDIA RTX 4090
n/a
8 GPU
NVIDIA H100 NVL (SXM5)
17128 points
NVIDIA RTX 4090
n/a
higher is better
Resnet152 (FP32)
1 GPU
NVIDIA H100 NVL (SXM5)
733 points
NVIDIA RTX 4090
n/a
4 GPU
NVIDIA H100 NVL (SXM5)
3015 points
NVIDIA RTX 4090
n/a
8 GPU
NVIDIA H100 NVL (SXM5)
8331 points
NVIDIA RTX 4090
n/a
higher is better
Inception V3 (FP16)
1 GPU
NVIDIA H100 NVL (SXM5)
2532 points
NVIDIA RTX 4090
n/a
4 GPU
NVIDIA H100 NVL (SXM5)
9618 points
NVIDIA RTX 4090
n/a
8 GPU
NVIDIA H100 NVL (SXM5)
21842 points
NVIDIA RTX 4090
n/a
higher is better
Inception V3 (FP32)
1 GPU
NVIDIA H100 NVL (SXM5)
1204 points
NVIDIA RTX 4090
n/a
4 GPU
NVIDIA H100 NVL (SXM5)
4952 points
NVIDIA RTX 4090
n/a
8 GPU
NVIDIA H100 NVL (SXM5)
13771 points
NVIDIA RTX 4090
n/a
higher is better
Inception V4 (FP16)
1 GPU
NVIDIA H100 NVL (SXM5)
1170 points
NVIDIA RTX 4090
n/a
4 GPU
NVIDIA H100 NVL (SXM5)
4446 points
NVIDIA RTX 4090
n/a
8 GPU
NVIDIA H100 NVL (SXM5)
9411 points
NVIDIA RTX 4090
n/a
higher is better
Inception V4 (FP32)
1 GPU
NVIDIA H100 NVL (SXM5)
535 points
NVIDIA RTX 4090
n/a
4 GPU
NVIDIA H100 NVL (SXM5)
2202 points
NVIDIA RTX 4090
n/a
8 GPU
NVIDIA H100 NVL (SXM5)
6294 points
NVIDIA RTX 4090
n/a
higher is better
VGG16 (FP16)
1 GPU
NVIDIA H100 NVL (SXM5)
1497 points
NVIDIA RTX 4090
n/a
4 GPU
NVIDIA H100 NVL (SXM5)
5687 points
NVIDIA RTX 4090
n/a
8 GPU
NVIDIA H100 NVL (SXM5)
21953 points
NVIDIA RTX 4090
n/a
higher is better
VGG16 (FP32)
1 GPU
NVIDIA H100 NVL (SXM5)
1095 points
NVIDIA RTX 4090
n/a
4 GPU
NVIDIA H100 NVL (SXM5)
4503 points
NVIDIA RTX 4090
n/a
8 GPU
NVIDIA H100 NVL (SXM5)
12267 points
NVIDIA RTX 4090
n/a
higher is better
3D, GPU Rendering Benchmarks 2024–2025 [Updated]↑
As of May 2025
V-Ray
1 GPU
NVIDIA H100 NVL (SXM5)
n/a
NVIDIA RTX 4090
5556 points
higher is better
Octane
1 GPU
NVIDIA H100 NVL (SXM5)
n/a
NVIDIA RTX 4090
1445 points
higher is better
Redshift
1 GPU
NVIDIA H100 NVL (SXM5)
n/a
NVIDIA RTX 4090
1.16 minutes
lower is better
Blender
1 GPU
NVIDIA H100 NVL (SXM5)
n/a
NVIDIA RTX 4090
12123.96 score
higher is better
Luxmark
1 GPU
NVIDIA H100 NVL (SXM5)
n/a
NVIDIA RTX 4090
158815 points
higher is better
Unreal Engine
1 GPU
NVIDIA H100 NVL (SXM5)
n/a
NVIDIA RTX 4090
92.1 FPS
higher is better
RELION Cryo-EM Benchmarks 2024-2025 [Updated]↑
As of May 2025
Total run time
1 GPU
NVIDIA H100 NVL (SXM5)
n/a
NVIDIA RTX 4090
105.2 Min
4 GPU
NVIDIA H100 NVL (SXM5)
n/a
NVIDIA RTX 4090
53.2 Min
lower is better
Llama3 70B Inference Benchmark 2024–2025 [Updated]↑
As of May 2025
Eval rate
1 GPU
NVIDIA H100 NVL (SXM5)
n/a
NVIDIA RTX 4090
9.95 tokens/s
2 GPU
NVIDIA H100 NVL (SXM5)
n/a
NVIDIA RTX 4090
19.99 tokens/s
higher is better
Technical Specifications
Board Design
| NVIDIA H100 NVL (SXM5) | NVIDIA RTX 4090 | Difference | |
|---|---|---|---|
| Outputs | No outputs | 1x HDMI, 3x DisplayPort | |
| Power Connectors | None | 1x 16-pin | |
| Slot width | N/A | Triple-slot | |
| TDP | 700 W | 450 W | -250 W (-36%) |
| Length | - | 13 in / 336 mm |
Clock Speeds
| NVIDIA H100 NVL (SXM5) | NVIDIA RTX 4090 | Difference | |
|---|---|---|---|
| Boost Clock | 1837 MHz | 2520 MHz | 683 MHz (37%) |
| GPU Clock | 1665 MHz | 2235 MHz | 570 MHz (34%) |
| Memory Clock | 5300 MHz | 21200 MHz | 15900 MHz (300%) |
Graphics Card
| NVIDIA H100 NVL (SXM5) | NVIDIA RTX 4090 | Difference | |
|---|---|---|---|
| Bus Interface | PCIe 5.0 x16 | PCIe 4.0 x16 | |
| Generation | Server Hopper (Hxx) | GeForce 40 |
Graphics Features
| NVIDIA H100 NVL (SXM5) | NVIDIA RTX 4090 | Difference | |
|---|---|---|---|
| OpenCL | 3 | 3 | - |
| CUDA | 9 | 8.9 | |
| DirectX | - | 12 Ultimate (12_2) | |
| OpenGL | - | 4.6 | |
| Shader Model | - | 6.7 |
Graphics Processor
| NVIDIA H100 NVL (SXM5) | NVIDIA RTX 4090 | Difference | |
|---|---|---|---|
| Architecture | Hopper | Ada Lovelace | |
| Die Size | 814 mm2 | 608 mm2 | -206 mm2 (-25%) |
| GPU Name | GH100 | AD102-300-A1 | |
| Process Size | 5 nm | 5 nm | - |
| Transistors | 80000 million | 76300 million | -3700 million (-5%) |
Memory
| NVIDIA H100 NVL (SXM5) | NVIDIA RTX 4090 | Difference | |
|---|---|---|---|
| Bandwidth | 3360 GB/s | 1018 GB/s | -2342 GB/s (-70%) |
| Memory Bus | 5120 bit | 384 bit | -4736 bit (-92%) |
| Memory Size | 96 GB | 24 GB | -72 GB (-75%) |
| Memory Type | HBM3 | GDDR6X |
Render Config
| NVIDIA H100 NVL (SXM5) | NVIDIA RTX 4090 | Difference | |
|---|---|---|---|
| ROPs | 24 | 192 | 168 (700%) |
| Shading Units/ CUDA Cores | 16896 | 16384 | -512 (-3%) |
| TMUs | 528 | 512 | -16 (-3%) |
| Tensor Cores | 528 | 512 | -16 (-3%) |
| RT Cores | - | 128 |
Theoretical Performance
| NVIDIA H100 NVL (SXM5) | NVIDIA RTX 4090 | Difference | |
|---|---|---|---|
| FP16 (half) performance | 248.3 TFLOPS | 82.58 TFLOPS | -165.72 TFLOPS (-67%) |
| FP32 (float) performance | 62.08 TFLOPS | 82.58 TFLOPS | 20.5 TFLOPS (33%) |
| FP64 (double) performance | 31040 GFLOPS | 1290 GFLOPS | -29750 GFLOPS (-96%) |
| Pixel Rate | 44.09 GPixel/s | 483.8 GPixel/s | 439.71 GPixel/s (997%) |
| Texture Rate | 969.9 GTexel/s | 1290 GTexel/s | 320.1 GTexel/s (33%) |
Price
| NVIDIA H100 NVL (SXM5) | NVIDIA RTX 4090 | Difference | |
|---|---|---|---|
| Release Date | Mar 21st, 2023 | Oct 12th, 2022 | |
| MSRP | - | $1,599.00 |
Test bench configuration
| NVIDIA H100 NVL (SXM5) | NVIDIA RTX 4090 | Difference | |
|---|---|---|---|
| Hardware | BIZON X5000 More details | BIZON X5500 More details | |
| Software | 3D Rendering: VRay Benchmark: 5 Octane Benchmark: 2020.1.5 Redshift Benchmark: 3.0.28 Demo Blender: 2.90 Luxmark: 3.1 | 3D Rendering: Nvidia Driver: VRay Benchmark: Octane Benchmark: Redshift Benchmark: Blender: Luxmark: |
Recommended hardware
| NVIDIA H100 NVL (SXM5) | NVIDIA RTX 4090 | Difference | |
|---|---|---|---|
| Best GPU workstations | - | BIZON X5500 |
