Sam Serrels

Act III Ltd
Studio Gobo
- LEGO Horizon Adventures
- Hogwarts Legacy
- Synced: Off-Planet
Supermassive Games Ltd
- The Quarry
- The Dark Pictures Anthology: Little Hope.
- Little Nightmares 2
Edinburgh Napier University
PhD researcher & Graduate Teaching Assistant, 2016 - ongoing
Currently working towards a PhD on methods for visualizing and improving the performance of heterogeneous and parallel systems .
BSc (Hons), 2012 - 2016
Graduated with a First Class Degree with HonoursRegularly organise and help with various events and groups to boost the local Edinburgh Video Games/Software Development, and creative technologies community. Including running exhibitions at the Science festival, maker faire, and independent events such "Games are for Everyone"
Organised and ran the Edinburgh chapter of the 48 hour Global Game Jam in 2016,17,18.
Founded and ran the NVGS over the period of my degree. The society was developed to over a hundred members, with maintained relationships and partnerships with external organisations, and hosted Edinburgh wide events and charity fundraisers.
Involved in the Edinburgh group of a nation wide attempt to break the Guinness World Record for the largest number of people simultaneously learning to code an Android app.
You can drop me an email at:sam@samserrels.com
My GPG Key:samserrels.com/gpg.txt
I'm also available on:Started in 2016 alongside my teaching role. Unfortunately ended in 2019 before I could finish the work, due to external circumstances. The project involved looking at how to improve the process of coding for GPU's, by utilizing new visual techniques to aid programmers.
This started out by looking at methods for live-capture debugging of GPGPU applications on AMD's rocm stack. This worked as a prototype, apps could be paused and the entire state of the gpu could be read out and analysed. This was promising but would require significantly more man hours than available to turn into a usable or measurable process for end-users or even just study participants.
Simulation was looked into, but the complexity, black box nature, and fast moving target that are GPUs made the task unfeasible for a phd project.
A static analysis approach was taken as a compromise. The compiler output ISA codes for GPUs give a good picture of the runtime behavior. As of that point in time the tooling around inspecting ISA code was lacking or nonfunctional. I built a web-based process for taking in a source OpenCL program, compiling it and showing the resultant ISA with metrics and visual aids. This was to help users understand and play with compiler options and rearranging the source code.
The code for that is here: github.com/dooglz/gpuvisunfortunately taking this prototype to users for analysis and iteration was never achieved. I did learn a whole lot about the innerworkigns of GPUS and an appreciation for tool design. I look back at this project as a worthwhile research endeavour
Final Thesis (incomplete): link
Info About HPC2
I was put in charge of the University's Compute cluster project. I was Responsible for purchasing, installation and config of computer server hardware and networking gear. I built a containerised Slurm deployment on top of Ubuntu across the 50 node system. Also I built web interfaces, wrote documentation, and integrated it all into corporate LDAP and email system.
A fun project, I learnt a load about sysadmin and networking. Configuring ancient switches, diagnosing dead hardware, now I can't live without IPMI in all my gear.
Loved: Docker, Ubuntu MAAS, networked power distro gear (so cool)
Hated: MS AD, Harddrives (despicable creatures), Cisco's serial config (no-shutdown anyone? Don't even ge me started on spanning tree)
The live config is all firewalled and confidential, however I did but out a generic version of the config: hgithub.com/dooglz/slurm_docker
See the workbook and lectures here:
dooglz.github.io/set09121
psst, it's all generated as github pages, even the lecture slides!
Source for labs and material github.com/dooglz/set09121