# Incorrect Peak Performance Boundaries in Nsight Compute Roofline Charts

**URL:** <https://forums.developer.nvidia.com/t/incorrect-peak-performance-boundaries-in-nsight-compute-roofline-charts/218636>\
**Category:** Nsight Compute\
**Created:** [June 23, 2022, 12:04pm UTC](https://forums.developer.nvidia.com/t/incorrect-peak-performance-boundaries-in-nsight-compute-roofline-charts/218636 "2022-06-23T12:04:32Z")\
**Posts on this page:** 5\
**Page:** 1

<div class="post-metadata">

**Author:** ![DiegoJimenez](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@DiegoJimenez](https://forums.developer.nvidia.com/u/DiegoJimenez)\
**Post date:** [June 23, 2022, 12:04pm UTC](https://forums.developer.nvidia.com/t/incorrect-peak-performance-boundaries-in-nsight-compute-roofline-charts/218636/1 "2022-06-23T12:04:32Z")

</div>

Hello

I’m having trouble understanding why when plotting the roofline chart on compute for a kernel running on an A100, the FP64 peak performance boundary is set at around 7.5 TFLOP/s (V100 peak) and not around the actual 9.5 TFLOP/s it should use as limit. I’m attaching a screenshot from an analysis on my kernel. You can see the tool correctly identifies the A100 GPU but the peak performance as I hover over the plateau is wrong. I’m using Nsight compute version 2021.3. I tried this on version 2022.2 but when I hover over the boundaries the pop-up tooltip is not working (Linux version).

 ![roofline_possible_bug](https://global.discourse-cdn.com/nvidia/original/3X/2/3/236332c790028362b12e5aee768a461f520b61c2.png)

Is there something wrong here or does it set the peak performance as 7.5 because that’s the nominal peak and not the theoretical?

Thanks for your help.

---

<div class="post-metadata">

**Author:** ![Sanjiv.Satoor](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/sanjiv.satoor/32/14043_2.png) [@Sanjiv.Satoor](https://forums.developer.nvidia.com/u/Sanjiv.Satoor)\
**Post date:** [June 28, 2022, 4:04pm UTC](https://forums.developer.nvidia.com/t/incorrect-peak-performance-boundaries-in-nsight-compute-roofline-charts/218636/2 "2022-06-28T16:04:11Z")

</div>

The roofline is constructed based on the clock rate at which the application was run (because that sets the upper limit).

What is the _clock\_rate_ for your profiling run? You can find this under _Device Attributes_ on the _Session_ page in the Nsight Compute UI.

---

<div class="post-metadata">

**Author:** ![DiegoJimenez](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@DiegoJimenez](https://forums.developer.nvidia.com/u/DiegoJimenez)\
**Post date:** [June 29, 2022, 3:41pm UTC](https://forums.developer.nvidia.com/t/incorrect-peak-performance-boundaries-in-nsight-compute-roofline-charts/218636/3 "2022-06-29T15:41:43Z")

</div>

Thanks for the answer @Sanjiv.Satoor , I’m really interested in correctly interpreting this chart. The reported clock\_rate for my A100 was: 1305000

![clock_rate](https://global.discourse-cdn.com/nvidia/original/3X/e/0/e0504fd33f5676f3ee6da6a03bea3ab2cd7cf548.png)

How is that clock\_rate used to compute the FP boundary? Could you share that formula? I’m guessing the 9.5 TFLOP/s peak would then require some GPU Boost Clock?

---

<div class="post-metadata">

**Author:** ![DiegoJimenez](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@DiegoJimenez](https://forums.developer.nvidia.com/u/DiegoJimenez)\
**Post date:** [June 29, 2022, 5:31pm UTC](https://forums.developer.nvidia.com/t/incorrect-peak-performance-boundaries-in-nsight-compute-roofline-charts/218636/4 "2022-06-29T17:31:46Z")

</div>

I’ve figured it out. This has to do with the clock control option on Nsight Compute and the default application clock that’s set on the A100 I’m running on (1305 MHz). I should then set the clock rate through nvidia-smi and then profile the application with no clock control.

> **[Kernel Profiling Guide :: Nsight Compute Documentation](https://docs.nvidia.com/nsight-compute/ProfilingGuide/index.html#clock-control)**
>
> Nsight Compute profiling guide.

However, this still doesn’t add up the 7.5 TFLOP/s boundary that plot is showing. How is that peak boundary computed using the clock rate?

---

<div class="post-metadata">

**Author:** ![Sanjiv.Satoor](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/sanjiv.satoor/32/14043_2.png) [@Sanjiv.Satoor](https://forums.developer.nvidia.com/u/Sanjiv.Satoor)\
**Post date:** [July 5, 2022, 9:00am UTC](https://forums.developer.nvidia.com/t/incorrect-peak-performance-boundaries-in-nsight-compute-roofline-charts/218636/5 "2022-07-05T09:00:35Z")

</div>

Please refer the reply posted here: [About the flops in ncu report - #6 by Sanjiv.Satoor](https://forums.developer.nvidia.com/t/about-the-flops-in-ncu-report/215350/6)