NVIDIA H200 NVL vs NVIDIA A40 PCIe
We compared two Desktop platform GPUs: 141GB VRAM H200 NVL and 48GB VRAM A40 PCIe to see which GPU has better performance in key specifications, benchmark tests, power consumption, etc.
Main Differences
NVIDIA H200 NVL 's Advantages
Released 4 years and 1 months late
Boost Clock has increased by 3% (1785MHz vs 1740MHz)
More VRAM (141GB vs 48GB)
Larger VRAM bandwidth (4.89TB/s vs 695.8GB/s)
6144 additional rendering cores
NVIDIA A40 PCIe 's Advantages
Lower TDP (300W vs 600W)
Score
Benchmark
FP32 (float)
H200 NVL
+61%
60.32 TFLOPS
A40 PCIe
37.42 TFLOPS
Graphics Card
Nov 2024
Release Date
Oct 2020
Server Hopper
Generation
Tesla
Desktop
Type
Desktop
PCIe 5.0 x16
Bus Interface
PCIe 4.0 x16
Clock Speeds
1365 MHz
Base Clock
1305 MHz
1785 MHz
Boost Clock
1740 MHz
1593 MHz
Memory Clock
1812 MHz
Memory
141GB
Memory Size
48GB
HBM3e
Memory Type
GDDR6
6144bit
Memory Bus
384bit
4.89TB/s
Bandwidth
695.8GB/s
Render Config
-
-
-
132
SM Count
84
16896
Shading Units
10752
528
TMUs
336
24
ROPs
112
528
Tensor Cores
336
-
RT Cores
84
256 KB (per SM)
L1 Cache
128 KB (per SM)
50 MB
L2 Cache
6 MB
-
-
-
Theoretical Performance
42.84 GPixel/s
Pixel Rate
194.9 GPixel/s
942.5 GTexel/s
Texture Rate
584.6 GTexel/s
241.3 TFLOPS
FP16 (half)
37.42 TFLOPS
60.32 TFLOPS
FP32 (float)
37.42 TFLOPS
30.16 TFLOPS
FP64 (double)
584.6 GFLOPS
Board Design
600W
TDP
300W
1000 W
Suggested PSU
700 W
No outputs
Outputs
3x DisplayPort 1.4a
8-pin EPS
Power Connectors
8-pin EPS
Graphics Processor
GH100
GPU Name
GA102
-
-
-
Hopper
Architecture
Ampere
TSMC
Foundry
Samsung
5 nm
Process Size
8 nm
80 billion
Transistors
28.3 billion
814 mm²
Die Size
628 mm²
Graphics Features
N/A
DirectX
12 Ultimate (12_2)
N/A
OpenGL
4.6
3.0
OpenCL
3.0
N/A
Vulkan
1.3
9.0
CUDA
8.6
N/A
Shader Model
6.6
