Jake Tech Lab

NVIDIA H100 vs AMD MI300X

AMD MI300X comes out ahead of NVIDIA H100 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 H100
NVIDIA H100
6.0
Fair
1 of 7 rounds
vs
AMD MI300XWinner
AMD MI300X
6.8
Fair
6 of 7 rounds

The verdict

AMD MI300X takes it, 6–1 on rounds.

On the published figures, AMD MI300X 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 H100 if…

  • • You want NVIDIA H100's edge on how much power it draws

Choose AMD MI300X if…

  • • You want AMD MI300X's edge on how much memory it has and how fast its memory is

Round by round

  1. Round 1How much memory it hasAMD MI300X wins
    NVIDIA H1005/10

    80 GB HBM3

    gpu memory

    • 80 GB HBM3, against 192 GB HBM3

    Source: nvidia.com ↗

    AMD MI300X8/10

    192 GB HBM3

    dedicated memory size

    Source: amd.com ↗

    AMD MI300X wins this round on what the company itself publishes.

  2. Round 2How fast its memory isAMD MI300X wins
    NVIDIA H1005/10

    3.35 TB/s

    memory bandwidth

    • 3.35 TB/s, against 5.3 TB/s

    Source: nvidia.com ↗

    AMD MI300X8/10

    5.3 TB/s

    peak memory bandwidth

    Source: amd.com ↗

    AMD MI300X wins this round on what the company itself publishes.

  3. Round 3How fast it runs FP8 AI mathAMD MI300X wins
    NVIDIA H1005/10

    1,979 TFLOPS

    fp8 tensor core

    • 1,979 TFLOPS, against 2.61 PFLOPs

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

    AMD MI300X8/10

    2.61 PFLOPs

    peak fp8 performance

    Source: amd.com ↗

    AMD MI300X wins this round on what the company itself publishes.

  4. Round 4How fast it runs FP16 AI mathAMD MI300X wins
    NVIDIA H1005/10

    989 TFLOPS

    fp16 tensor core

    • 989 TFLOPS, against 1.3 PFLOPs

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

    AMD MI300X8/10

    1.3 PFLOPs

    peak fp16 performance

    Source: amd.com ↗

    AMD MI300X wins this round on what the company itself publishes.

  5. Round 5How much power it drawsNVIDIA H100 wins
    NVIDIA H1008/10

    700 W

    max tdp

    Source: nvidia.com ↗

    AMD MI300X5/10

    750W

    typical board power (peak)

    • 750W, against 700 W

    Source: amd.com ↗

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

  6. Round 6How it handles scientific FP64 mathAMD MI300X wins
    NVIDIA H1005/10

    67 TFLOPS

    fp64 tensor core

    • 67 TFLOPS, against 163.4 TFLOPs

    Source: nvidia.com ↗

    AMD MI300X8/10

    163.4 TFLOPs

    peak fp64 matrix

    Source: amd.com ↗

    AMD MI300X wins this round on what the company itself publishes.

  7. Round 7How new it isAMD MI300X wins
    NVIDIA H1005/10

    2022

    year announced

    • 2022, against 12/06/2023

    Source: nvidianews.nvidia.com ↗

    AMD MI300X8/10

    12/06/2023

    launch date

    Source: amd.com ↗

    AMD MI300X wins this round on what the company itself publishes.

NVIDIA H100NVIDIA H100
AMD MI300XAMD MI300X

Side by side

Overall score
6.0
6.8
Best
Price
See price
See price
How fast its memory is
5.7
7.0
How much memory it has
5.4
7.0
How much power it draws
7.1
6.5
How fast it runs FP16 AI math
5.9
6.5
How new it is
5.4
6.6
Head-to-head findings
How fast its memory is
3.35 TB/s
5.3 TB/s
How much memory it has
80 GB HBM3
192 GB HBM3
How much power it draws
700 W
750W
How fast it runs FP16 AI math
989 TFLOPS
1.3 PFLOPs
How new it is
2022
12/06/2023
Which memory generation it uses
HBM3
HBM3
How fast it runs FP8 AI math
1,979 TFLOPS
2.61 PFLOPs
How it handles scientific FP64 math
67 TFLOPS
163.4 TFLOPs
Record
Head-to-head record
2–7
4–2
Head-to-heads won
22 %
67 %
Rounds won
31 %
56 %
Pros
      Cons
      • 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
      • 1.3 PFLOPs, against 1,678 TFLOPS
      • 12/06/2023, against September 24, 2024
      • 192 GB HBM3, against 288 GB HBM3E
      • 5.3 TB/s, against 8 TB/s
      Add a third option to this comparison →

      More AI accelerators match-ups