FLOPS Performance Comparisons

Discussions center on comparing FLOPS/TFLOPS metrics across GPUs, CPUs, supercomputers, and AI chips, debating their impressiveness, units, precision, sparsity, efficiency like FLOPS per watt, and relevance to real performance.

➡️ Stable 0.7x Hardware
2,174
Comments
20
Years Active
5
Top Authors
#3669
Topic ID

Activity Over Time

2007
1
2008
11
2009
10
2010
29
2011
28
2012
34
2013
64
2014
52
2015
84
2016
145
2017
135
2018
119
2019
105
2020
187
2021
188
2022
157
2023
238
2024
315
2025
252
2026
20

Keywords

ATLAS FP32 e.g CPU AlphaZero GFLOPS PFLOPS TPU MMA PR gpu nvidia watt tensor compute supercomputer performance units number precision

Sample Comments

mzh0 • Jul 31, 2017 • View on HN

Do you have benchmark number like FLOPS compared to GPU / CPU?

agumonkey • Nov 28, 2016 • View on HN

Nobody needs more than 640TFLOPS.

sudosysgen • Mar 4, 2021 • View on HN

Is 900GFlops of GPU compute supposed to be impressive?

danielharan • Nov 19, 2013 • View on HN

While making GFlops/watt cheaper for all number crunching applications, no?

NkVczPkybiXICG • Aug 28, 2024 • View on HN

I think they mean TFLOPS, not FLOPS

l33tman • Jun 19, 2023 • View on HN

The RTX 4090 GPU (a single PCI card) hits 82k GFlops in FP32

AtHeartEngineer • Jun 23, 2020 • View on HN

Id be interested to see the wattage per teraflop difference between the two.

adastra22 • Dec 12, 2025 • View on HN

FLOPS are not what matters here.

lostmsu • Sep 29, 2022 • View on HN

2 TFlops or 5 TFlops does not matter much. 3090Ti does 160 TFlops, e.g. at least 30x (!) times faster.

telotortium • Jan 29, 2025 • View on HN

Maybe a little, but FLOPs and memory bandwidth don't lie.