NVIDIA GeForce RTX 4090 vs NVIDIA H100 SXM5 96 GB
AI GPU
We compared a Desktop platform GPU: 24GB VRAM GeForce RTX 4090 and a GPU: 96GB VRAM H100 SXM5 96 GB to see which GPU has better performance in key specifications, benchmark tests, power consumption, etc.
Main Differences
NVIDIA GeForce RTX 4090 's Advantages
Boost Clock has increased by 27% (2520MHz vs 1980MHz)
Lower TDP (450W vs 700W)
NVIDIA H100 SXM5 96 GB 's Advantages
Released 6 months late
More VRAM (96GB vs 24GB)
Larger VRAM bandwidth (3.36TB/s vs 1.01TB/s)
512 additional rendering cores
Score
Benchmark
FP32 (float)
GeForce RTX 4090
+23%
82.58 TFLOPS
H100 SXM5 96 GB
66.91 TFLOPS
Graphics Card
Sep 2022
Release Date
Mar 2023
GeForce 40
Generation
Server Hopper
Desktop
Type
AI GPU
PCIe 4.0 x16
Bus Interface
PCIe 5.0 x16
Clock Speeds
2235 MHz
Base Clock
1350 MHz
2520 MHz
Boost Clock
1980 MHz
1313 MHz
Memory Clock
1313 MHz
Memory
24GB
Memory Size
96GB
GDDR6X
Memory Type
HBM3
384bit
Memory Bus
5120bit
1.01TB/s
Bandwidth
3.36TB/s
Render Config
-
-
-
128
SM Count
132
16384
Shading Units
16896
512
TMUs
528
176
ROPs
24
512
Tensor Cores
528
128
RT Cores
-
128 KB (per SM)
L1 Cache
256 KB (per SM)
72 MB
L2 Cache
50 MB
-
-
-
Theoretical Performance
443.5 GPixel/s
Pixel Rate
47.52 GPixel/s
1290 GTexel/s
Texture Rate
1045 GTexel/s
82.58 TFLOPS
FP16 (half)
267.6 TFLOPS
82.58 TFLOPS
FP32 (float)
66.91 TFLOPS
1290 GFLOPS
FP64 (double)
33.45 TFLOPS
Board Design
450W
TDP
700W
850 W
Suggested PSU
1100 W
1x HDMI 2.1
3x DisplayPort 1.4a
Outputs
No outputs
1x 16-pin
Power Connectors
8-pin EPS
Graphics Processor
AD102
GPU Name
GH100
AD102-300-A1
GPU Variant
-
Ada Lovelace
Architecture
Hopper
TSMC
Foundry
TSMC
5 nm
Process Size
5 nm
76.3 billion
Transistors
80 billion
609 mm²
Die Size
814 mm²
Graphics Features
12 Ultimate (12_2)
DirectX
N/A
4.6
OpenGL
N/A
3.0
OpenCL
3.0
1.3
Vulkan
N/A
8.9
CUDA
9.0
6.8
Shader Model
N/A
