# Is there any report on DSMEM bandwidth on H100 or specific usage examples?

**URL:** https://forums.developer.nvidia.com/t/is-there-any-report-on-dsmem-bandwidth-on-h100-or-specific-usage-examples/316440
**Category:** CUDA Programming and Performance
**Created:** [December 11, 2024, 6:40am UTC](https://forums.developer.nvidia.com/t/is-there-any-report-on-dsmem-bandwidth-on-h100-or-specific-usage-examples/316440 "2024-12-11T06:40:45Z")
**Posts on this page:** 5
**Page:** 1

<div class="post-metadata">

### Author: ![202476410arsmart](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@202476410arsmart](https://forums.developer.nvidia.com/u/202476410arsmart)
#### Post date: [December 11, 2024, 6:40am UTC](https://forums.developer.nvidia.com/t/is-there-any-report-on-dsmem-bandwidth-on-h100-or-specific-usage-examples/316440/1 "2024-12-11T06:40:45Z")

</div>

Is there any detailed report or data on DSMEM (Distributed Shared Memory) bandwidth on NVIDIA H100 GPUs? Additionally, are there any specific examples or case studies demonstrating its usage?

---

<div class="post-metadata">

### Author: ![202476410arsmart](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@202476410arsmart](https://forums.developer.nvidia.com/u/202476410arsmart)
#### Post date: [December 11, 2024, 6:41am UTC](https://forums.developer.nvidia.com/t/is-there-any-report-on-dsmem-bandwidth-on-h100-or-specific-usage-examples/316440/2 "2024-12-11T06:41:04Z")

</div>

specifically, for GEMM

---

<div class="post-metadata">

### Author: ![Curefab](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@Curefab](https://forums.developer.nvidia.com/u/Curefab)
#### Post date: [December 11, 2024, 12:05pm UTC](https://forums.developer.nvidia.com/t/is-there-any-report-on-dsmem-bandwidth-on-h100-or-specific-usage-examples/316440/3 "2024-12-11T12:05:25Z")

</div>

Perhaps best about bandwidth measurement would be

Luo et al: Benchmarking and Dissecting the Nvidia Hopper GPU Architecture:  
[https://arxiv.org/pdf/2402.13499v1](https://arxiv.org/pdf/2402.13499v1)

---

<div class="post-metadata">

### Author: ![202476410arsmart](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@202476410arsmart](https://forums.developer.nvidia.com/u/202476410arsmart)
#### Post date: [December 14, 2024, 6:44am UTC](https://forums.developer.nvidia.com/t/is-there-any-report-on-dsmem-bandwidth-on-h100-or-specific-usage-examples/316440/4 "2024-12-14T06:44:13Z")

</div>

Well, thanks! Just that is not for GEMM. But I get to know DSM’s bandwidth could be 3TB/s. Interesting!

---

<div class="post-metadata">

### Author: ![Curefab](https://developer.download.nvidia.com/images/forums/profile-default-devtalk-84.png) [@Curefab](https://forums.developer.nvidia.com/u/Curefab)
#### Post date: [December 14, 2024, 7:07pm UTC](https://forums.developer.nvidia.com/t/is-there-any-report-on-dsmem-bandwidth-on-h100-or-specific-usage-examples/316440/5 "2024-12-14T19:07:47Z")

</div>

You could also use cuBLAS functions and look into the performance with Compute NSight.