Jake Tech Lab

NVIDIA A100 vs NVIDIA H100

NVIDIA H100 comes out ahead of NVIDIA A100 on 6 of the 7 things that decide this, going by what each maker publishes. The one that matters for you may be one the leader loses.

NVIDIA A100
NVIDIA A100
6.0
Fair
1 of 7 rounds
vs
NVIDIA H100Winner
NVIDIA H100
6.0
Fair
6 of 7 rounds

The verdict

NVIDIA H100 takes it, 6–1 on rounds.

On the published figures, NVIDIA H100 takes 6 of 7 rounds. Treat that as a starting point and check the current figures on both pages, because prices and plans move.

Choose NVIDIA A100 if…

  • • You want NVIDIA A100's edge on how much power it draws

Choose NVIDIA H100 if…

  • • You want NVIDIA H100's edge on how fast its memory is and how fast it runs fp16 ai math

Round by round

  1. Round 1How fast its memory isNVIDIA H100 wins
    NVIDIA A1005/10

    2,039 GB/s

    memory bandwidth

    • 2,039 GB/s, against 3.35 TB/s

    Source: nvidia.com ↗

    NVIDIA H1008/10

    3.35 TB/s

    memory bandwidth

    Source: nvidia.com ↗

    NVIDIA H100 wins this round on what the company itself publishes.

  2. Round 2How fast it runs FP16 AI mathNVIDIA H100 wins
    NVIDIA A1005/10

    312 TFLOPS

    fp16 tensor core

    • 312 TFLOPS, against 989 TFLOPS

    Source: nvidia.com ↗

    NVIDIA H1008/10

    989 TFLOPS

    fp16 tensor core

    Source: dam-cdn.nvd.orangelogic.com ↗

    NVIDIA H100 wins this round on what the company itself publishes.

  3. Round 3How much power it drawsNVIDIA A100 wins
    NVIDIA A1008/10

    400 W

    max tdp (sxm)

    Source: nvidia.com ↗

    NVIDIA H1005/10

    700 W

    max tdp

    • 700 W, against 400 W

    Source: nvidia.com ↗

    NVIDIA A100 wins this round on what the company itself publishes.

  4. Round 4How fast chips talk to each otherNVIDIA H100 wins
    NVIDIA A1005/10

    600 GB/s NVLink

    nvlink gpu-to-gpu

    • 600 GB/s NVLink, against 900 GB/s NVLink

    Source: nvidia.com ↗

    NVIDIA H1008/10

    900 GB/s NVLink

    nvlink gpu-to-gpu

    Source: nvidia.com ↗

    NVIDIA H100 wins this round on what the company itself publishes.

  5. Round 5How it handles scientific FP64 mathNVIDIA H100 wins
    NVIDIA A1005/10

    19.5 TFLOPS

    fp64 tensor core

    • 19.5 TFLOPS, against 67 TFLOPS

    Source: nvidia.com ↗

    NVIDIA H1008/10

    67 TFLOPS

    fp64 tensor core

    Source: nvidia.com ↗

    NVIDIA H100 wins this round on what the company itself publishes.

  6. Round 6Which memory generation it usesNVIDIA H100 wins
    NVIDIA A1005/10

    HBM2e

    memory type

    • HBM2e, against HBM3

    Source: nvidia.com ↗

    NVIDIA H1008/10

    HBM3

    memory type

    Source: dam-cdn.nvd.orangelogic.com ↗

    NVIDIA H100 wins this round on what the company itself publishes.

  7. Round 7How new it isNVIDIA H100 wins
    NVIDIA A1005/10

    2020

    year announced

    • 2020, against 2022

    Source: nvidianews.nvidia.com ↗

    NVIDIA H1008/10

    2022

    year announced

    Source: nvidianews.nvidia.com ↗

    NVIDIA H100 wins this round on what the company itself publishes.

NVIDIA A100NVIDIA A100
NVIDIA H100NVIDIA H100

Side by side

Overall score
6.0
6.0
Price
See price
See price
How fast its memory is
6.1
5.7
How much memory it has
6.3
5.4
How much power it draws
6.9
7.1
How fast it runs FP16 AI math
5.8
5.9
How new it is
5.4
5.4
Head-to-head findings
How fast its memory is
2,039 GB/s
3.35 TB/s
How much memory it has
80 GB HBM2e
80 GB HBM3
How much power it draws
400 W
700 W
How fast it runs FP16 AI math
312 TFLOPS
989 TFLOPS
How new it is
2020
2022
Which memory generation it uses
HBM2e
HBM3
How fast it runs FP8 AI math
—
1,979 TFLOPS
How it handles scientific FP64 math
19.5 TFLOPS
67 TFLOPS
Record
Head-to-head record
2–5–1
2–7
Head-to-heads won
29 %
22 %
Rounds won
31 %
31 %
Pros
      Cons
      • 80 GB HBM2e, against 128 GB HBM2E
      • 2,039 GB/s, against 3.7 TB/s
      • 312 TFLOPS, against 1,678 TFLOPS
      • 2020, against September 24, 2024
      • 80 GB HBM3, against 288 GB HBM3E
      • 3.35 TB/s, against 8 TB/s
      • 1,979 TFLOPS, against 5 PFLOPs
      • 989 TFLOPS, against 2.5 PFLOPs
      Add a third option to this comparison →

      More AI accelerators match-ups