Skip to content

Updates2 #2

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
mktip merged 2 commits into from
Oct 8, 2025
Merged

Updates2 #2

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f99a514
Added info to Javid's and Sinan's works.
hanZaq25 Oct 8, 2025
f81818b
Edited Dogan project description.
hanZaq25 Oct 8, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Binary file added docs/assets/CG.png
View file
Open in desktop
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
31 changes: 24 additions & 7 deletions docs/index.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -490,7 +490,7 @@ a {
<div class="card grid grid-cols-4 justify-center items-center">
<img class="shadow rounded-full max-w-full h-auto align-middle border-none" src="./team-images/abdullah.png" width="100px" />
<p class="grid-colspan-3">
<b>Alumni:</b> Muhammad Abdullah Soyturk
<b>Alumni:</b> Muhammed Abdullah Soytürk
<br>
<b>Research Focus:</b> Scalable deep learning.
</p>
Expand Down Expand Up @@ -531,6 +531,7 @@ BeyondMoore Software Ecosystem
**_Profiling Tools_**
* [Snoopie](#SNOOPIE): A Multi-GPU Communication Profiler and Visualiser
* [PES AMD vs Intel](#PRECISE-EVENT-SAMPLING): A Precise Event Sampling Benchmark Suite
* [aCG](#ACG): CPU- and GPU-initiated Communication Strategies for CG Methods

</div>

Expand Down Expand Up @@ -591,10 +592,7 @@ BeyondMoore Software Ecosystem
<img width="32" src="./assets/git.webp" />
<a href="https://github.com/ParCoreLab/Uniconn" class="text-xl font-semibold font-sans visited:text-teal-700">Unified Communication Library</a>
</div>
<p class="text-lg">We're undertaking the design of an API for a unified communication library to streamline device-to-device communication within the CPU-free model by aiming to optimize communication efficiency across diverse devices. We are also investigating how the available communication libraries for a system perform under different
message sizes and communication patterns. Thus, we ex-
tensively benchmark current communication methods for
single-process, multi-threaded, and multi-process codes. More details about the project will be available soon. The related paper is under preparation.</p>
<p class="text-lg">Modern HPC and AI systems increasingly rely on multi-GPU clusters, where communication libraries such as MPI, NCCL/RCCL, and NVSHMEM enable data movement across GPUs. While these libraries are widely used in frameworks and solver packages, their distinct APIs, synchronization models, and integration mechanisms introduce programming complexity and limit portability. Performance also varies across workloads and system architectures, making it difficult to achieve consistent efficiency. These issues present a significant obstacle to writing portable, high-performance code for large-scale GPU systems. We present Uniconn, a unified, portable high-level C++ communication library that supports both point-to-point and collective operations across GPU clusters. Uniconn enables seamless switching between backends and APIs (host or device) with minimal or no changes to application code. We describe its design and core constructs, and evaluate its performance using network benchmarks, a Jacobi solver, and a Conjugate Gradient solver. Across three supercomputers, we compare Uniconn's overhead against CUDA/ROCm-aware MPI, NCCL/RCCL, and NVSHMEM on up to 64 GPUs. In most cases, Uniconn incurs negligible overhead, typically under 1 % for the Jacobi solver and under 2% for the Conjugate Gradient solver.</p>

<p>
<a href="https://github.com/ParCoreLab/Uniconn" class="text-xl font-semibold font-sans visited:text-teal-700">More details and git repository of the project.</a>
Expand All @@ -606,13 +604,32 @@ single-process, multi-threaded, and multi-process codes. More details about the
</div>


<div id="ACG" class="h-auto bg-gray-100 rounded-s p-4 border-solid border-1 border-gray-200 flex flex-row justify-start items-start gap-5 transform transition-all hover:shadow-gray-100 hover:shadow-lg shadow-none">
<div clas="flex flex-col justify-start">
<div class="flex flex-row gap-2 justify-start items-center flex-shrink">
<img width="32" src="./assets/git.webp" />
<a href="https://github.com/ParCoreLab/aCG" class="text-xl font-semibold font-sans visited:text-teal-700">CPU- and GPU-initiated Communication Strategies for CG Methods</a>
</div>
<p class="text-lg">This work revisits Conjugate Gradient (CG) parallelization for large-scale multi-GPU systems, addressing challenges from low computational intensity and communication overhead. We develop scalable CG and pipelined CG solvers for NVIDIA and AMD GPUs, employing GPU-aware MPI, NCCL/RCCL, and NVSHMEM for both CPU- and GPU-initiated communication. A monolithic GPU-offloaded variant further enables fully device-driven execution, removing CPU involvement. Optimizations across all designs reduce data transfers and synchronization costs. Evaluations on SuiteSparse matrices and a real finite element application show 8–14% gains over state-of-the-art on single GPUs and 5–15% improvements in strong scaling tests on over 1,000 GPUs. While CPU-driven variants currently benefit from stronger library support, results highlight the promising scalability of GPU-initiated execution for future large-scale systems.</p>

<p>
<a href="https://github.com/ParCoreLab/aCG" class="text-xl font-semibold font-sans visited:text-teal-700">More details and git repository of the project.</a>
</p>
</div>
<div class="grid h-auto justify-center place-items-center">
<img width="400px" src="./assets/CG.png" />
</div>
</div>

<div id="CPU-FREE-MODEL-COMPILER" class="h-auto bg-gray-100 rounded-s p-4 border-solid border-1 border-gray-200 flex flex-row justify-start items-start gap-5 transform transition-all hover:shadow-gray-100 hover:shadow-lg shadow-none">
<div clas="flex flex-col justify-start">
<div class="flex flex-row gap-2 justify-start items-center flex-shrink">
<img width="32" src="./assets/git.webp" />
<a href="https://github.com/ParCoreLab/" class="text-xl font-semibold font-sans visited:text-teal-700">CPU Free Model Compiler</a>
</div>
<p class="text-lg">We're actively crafting a compiler to empower developers to write high-level Python code that compiles into efficient CPU-free device code. This compiler integrates GPU-initiated communication libraries, NVSHMEM for NVIDIA and ROC_SHMEM for AMD, enabling GPU communication directly within Python code. With automatic generation of GPU-initiated communication calls and persistent kernels, we aim to streamline development workflows. Our prototype will be available soon.</p>
<p class="text-lg">We're actively crafting a compiler to empower developers to write high-level Python code that compiles into efficient CPU-free device code. This compiler integrates GPU-initiated communication libraries, NVSHMEM for NVIDIA and ROC_SHMEM for AMD, enabling GPU communication directly within Python code. With automatic generation of GPU-initiated communication calls and persistent kernels, we aim to streamline development workflows.</p> <p>
<a href="https://github.com/ParCoreLab/CPU-Free-model" class="text-xl font-semibold font-sans visited:text-teal-700">More details and git repository of the project.</a>
</p>
</div>
<div class="grid h-auto justify-center place-items-center">
<img width="300px" src="./assets/dace-compiler.png" />
Expand Down Expand Up @@ -689,7 +706,7 @@ Graphs</a> <a href="https://docs.google.com/presentation/d/1po87zQeUQb5l12AXB5RM
<div class="card text-lg">Tugba Torun, Ameer Taweel, Didem Unat (2024) <a href="https://arxiv.org/pdf/2405.04944">A Sparse Tensor Generator with Efficient Feature Extraction</a>. <span class="italic">Accepted for publication; online release pending</span>. <a class="italic" href="https://arxiv.org/pdf/2405.04944">preprint pdf</a> </div>

<div class="card text-lg"> Javid Baydamirli, Tal Ben Nun, Didem Unat (2024) <a href="https://ieeexplore.ieee.org/abstract/document/10820747">Autonomous Execution for Multi-GPU Systems:
Compiler Support</a>. In SC24-W: Workshops of the International Conference for High Performance Computing, Networking, Storage and Analysis. <a class="italic" download href="./assets/preprint-pdfs/P3HPC_____Autonomous_Execution_for_Multi_GPU_Systems__Compiler_Support-2 (1).pdf">preprint pdf</a>
Compiler Support</a> <a href="https://docs.google.com/presentation/d/1nBsANrcLh0Tnc2qqqDL_-6khqo-Y-_mX5kfJbmRwawE/edit?slide=id.p#slide=id.p">(presentation)</a>. In SC24-W: Workshops of the International Conference for High Performance Computing, Networking, Storage and Analysis. <a class="italic" download href="./assets/preprint-pdfs/P3HPC_____Autonomous_Execution_for_Multi_GPU_Systems__Compiler_Support-2 (1).pdf">preprint pdf</a>
</div>
<div class="card text-lg"> Javid Baydamirli, Tal Ben Nun, Didem Unat (2024) <a href="https://sc24.supercomputing.org/proceedings/workshops/workshop_pages/ws_p3hpc108.html">Autonomous Execution for Multi-GPU Systems:
Compiler Support</a> <a href="https://sc24.conference-program.com/presentation/?id=ws_p3hpc108&sess=sess751">(presentation)</a>. In the 2024 International Workshop on Performance, Portability, and Productivity in HPC. <a class="italic" download href="./assets/preprint-pdfs/sc24-workshop-autonomous-execution-for-multi-gpu-systems-compiler-support.pdf">preprint pdf</a>
Expand Down