Ramith Hettiarachchi ramithuh

Compiling DGL v2.4.0 on Modern GPUs (Blackwell/Hopper)

This guide provides a comprehensive walkthrough for compiling the Deep Graph Library (DGL) version 2.4.0 from source on modern hardware, specifically targeting systems with NVIDIA Blackwell (e.g., GB10) or Hopper GPUs and a recent CUDA toolkit (13.0+).

The standard build process for this older DGL version fails due to outdated build scripts that are incompatible with new GPU architectures and compilers. The following steps include the necessary patches to overcome these issues and create a fully optimized build.

Key Lessons from this Build

This process highlights several common challenges when compiling older source code on new hardware:

Cheat Sheet: Copying Folders Between Remote Servers

This guide uses rsync and SSH Agent Forwarding to securely and efficiently copy folders between two servers.

Method 1: Push from Source Server 📤

Use this when you can SSH into the source server.

On Your Local PC (Prep Agent): Load the key for the destination server.

Conda Environment and Package Setup for ESMFold

1. Create and Activate Conda Environment

conda create -n esm2 python=3.10
conda activate esm2

	Bootstrap: docker
	From: nvidia/cuda:12.6.3-base-ubuntu24.04
	Stage: spython-base

	%files
	docker/jackhmmer_seq_limit.patch /hmmer_build/
	. /app/alphafold
	%post
	# Copyright 2024 DeepMind Technologies Limited
	#

	import jax
	import jax.numpy as jnp

	jax.config.update("jax_default_device", jax.devices("cpu")[0])

	#runs on cpu
	long_vector = jnp.arange(int(1e7))
	%timeit jnp.dot(long_vector, long_vector).block_until_ready()

	jax.config.update("jax_default_device", jax.devices("gpu")[2])

	for f in "$@"
	do
	/usr/local/bin/pngquant 32 --skip-if-larger --strip --ext=.png --force "$f"
	/usr/local/bin/zopflipng -y "$f" "$f"
	/Users/ramith/Documents/ss/impbcopy "$f"
	done

	on run {input, parameters}

	do shell script "open /Users/ramith/Documents/SEM4_TRONIC_ACA"

	return input
	end run

	num=[0.9955 -1.8936 0.9955];
	den=[1 -1.8936 0.9911];

	zplane(num,den)
	grid on;

	%freqz(num,den,100);

	import wandb


	wandb.init(id="31cnstm6", project="test", resume="must")

	wandb.log({"this was added later": 1241241})