Good morning Guys! So I promised to curate a well structured path for learning as beginners and intermediates. I’d be posting that soon, so watch out for my tweet. 👍🏾

Real life, baby

Introducing Griffin, the first model to pass the video Turing test. 48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video. It’s the first Human Interaction Model (HIM).

Write a statement not too long but very powerful.

Day 2/365 — I went inside the GPU today. And honestly… I think I may have opened a door I wasn't supposed to open. 😂, I entered into so many rabbit holes yesterday and I actually enjoyed it. (Y'all are missing 😉) Yesterday was about understanding why GPUs exist. Today, I wanted to understand: “Okay, but what is actually inside this thing?” And that's where GPU architecture gets crazy. Think of an NVIDIA GPU like a massive company. At the top, you have Graphics Processing Clusters (GPCs). Inside the GPCs, you have Texture Processing Clusters (TPCs). And inside the TPCs? Streaming Multiprocessors (SMs). The SM is basically a self-contained workshop where the actual computation happens. And inside this workshop, things get even more interesting: • CUDA Cores • Tensor Cores • RT Cores • Warp Schedulers • Registers • Shared Memory / L1 Cache • Texture units and other supporting hardware So I started thinking: CUDA Cores = the regular workers. Tensor Cores = the specialists. RT Cores = the guys doing one very specific job. 😂 CUDA Cores handle general-purpose arithmetic operations. Tensor Cores are specialized for matrix operations — which is VERY important for AI workloads. And RT Cores are specialized for ray tracing — calculating things like how light interacts with objects in 3D scenes. Then I got stuck on Tensor Cores. Because this is where things start getting REALLY interesting for AI. Neural networks are basically drowning in matrix multiplication. And NVIDIA has spent multiple GPU generations making Tensor Cores increasingly specialized for these workloads. Volta → Turing → Ampere → Hopper → Blackwell → Rubin. And Rubin is where my brain started buffering. 😂 NVIDIA's Rubin GPU has: 336 billion transistors. 🤯 224 Streaming Multiprocessors. 🤯 896 Tensor Cores. 🤯 288GB of HBM4. 🤯 And up to 22 TB/s of memory bandwidth. 🤯 TERABYTES/Per second. (I don't think y'all are clocking how insane this is!!!) At this point I stopped studying and started staring at the numbers. 😂 But here's the part I found even more interesting: Compute is useless if you can't feed the compute fast enough. You can have incredibly powerful Tensor Cores sitting there ready to work… …but if the data isn't arriving quickly enough, your expensive compute resources can sit around waiting. This is why memory architecture matters so much. Rubin combines its compute with high-bandwidth HBM4, and NVIDIA says the GPU can have up to 288GB of HBM4 delivering up to 22 TB/s of bandwidth. And then there's TMA — Tensor Memory Accelerator. TMA isn't new to Rubin; it was introduced with Hopper. Its job is to make moving multidimensional blocks of data between global memory and shared memory more efficient, while taking some of that data-movement/address-calculation work away from the SM's normal instruction pipeline. So my mental model today became: GPU architecture isn't just about having more compute. It's about: Compute + Memory + Data Movement + Scheduling All working together. And I'm beginning to understand why AI infrastructure is so much more than: “Just give me a GPU.”😂 Day 2 complete. Tomorrow, I'm going deeper into Tensor Cores. Because apparently I enjoy suffering. #GPU #AIInfrastructure #CUDA #NVIDIA #MLOps #DevOps #LearningInPublic

BREAKING: Anthropic is reportedly planning to spend $518,000,000,000.00 on cloud and infrastructure in the coming years.

Day 1/365 of learning GPU & AI Infrastructure (Let's see if I can be consistent enough to post for 1 year, y'all need to hold me accountable🥲) For years, I’ve worked with CPUs, Servers. Kubernetes. Containers. Databases. Networking. Cloud infrastructure. Today was less about GPUs themselves and more about understanding why GPUs became necessary in the first place. I started with something fundamental: How do we increase the processing power of a chip? One approach was increasing the number of processor cores. As the industry moved from single-core to multi-core processors, we were able to execute more tasks concurrently. But then something bigger happened: Parallelism became increasingly important. The semiconductor industry began moving toward two broad directions: → Multi-core processors → Many-core processors And this is where GPUs became particularly interesting. The gaming industry became one of the biggest forces pushing GPU development. Graphics workloads require enormous amounts of computation to happen simultaneously. Rendering millions of pixels, processing textures, lighting, physics, and other visual operations created a demand for massive parallel processing. And eventually, the same characteristics that made GPUs powerful for graphics became incredibly useful for another field: AI and scientific computing. But one thing I found particularly interesting: The rise of GPUs did NOT mean the death of CPUs. Instead, CPUs and GPUs became complementary. The CPU is generally better suited for sequential, general-purpose tasks and coordinating workloads. The GPU excels when a problem can be broken down into huge numbers of operations that can run in parallel. So rather than: CPU vs GPU the more interesting architecture is: CPU + GPU And this combination eventually led to another major development: CUDA — Compute Unified Device Architecture. NVIDIA introduced CUDA as a platform/programming model that enables CPU and GPU execution to work together, making it possible to leverage GPU acceleration for workloads beyond traditional graphics. I also started getting familiar with some terminology that I'll be seeing A LOT during this journey: • CPU • GPU • Streaming Multiprocessor (SM) • CUDA Core • Tensor Core • VRAM • HBM • Memory Bandwidth • PCIe I also came across some of the factors that matter when evaluating processors, including: → Performance → Marketplace presence → Accessibility → Support for IEEE floating-point standards What I'm realizing already is that learning AI infrastructure isn't just about learning how to deploy an AI model. There is an entire hardware and software stack underneath it. And I want to understand that stack. Day 1 complete. we go deeper. 🚀 #GPU #AIInfrastructure #CUDA #NVIDIA #MLOps #DevOps #Cloud #LearningInPublic

really interesting how CUDA brings CPUs and GPUs together. that combo is powerful for AI tasks

I’m starting a new 12-month learning journey: GPUs + AI Infrastructure. I’ve spent years building and operating cloud infrastructure, Kubernetes platforms, CI/CD pipelines, and production systems. Now I want to understand the infrastructure underneath modern AI — from GPU architecture and CUDA to distributed training, inference, Kubernetes GPU orchestration, optimization and AI platforms. For the next 12 months, I’ll be learning in public. I’ll share: • What I’m learning • Labs & experiments • Architecture decisions • Things that broke • Cost/performance lessons • Projects I build • Resources that actually helped No pretending to be an expert. Just 12 months of consistent learning and building. If you’re interested in GPUs, AI infrastructure, ML platforms, cloud, Kubernetes or just learning something difficult alongside others: Join me. Day 1 starts now. 🚀 #GPU #AIInfrastructure #MLOps #DevOps #Kubernetes #Cloud

If you’re a very senior level Cyber security/Security engineer to spearhead a team for $1500. Familiar with - Linux - Kubernetes/OpenShift - Terraform - Cloud infrastructure - AWS/Azure/GCP - OpenStack - Networking - IAM - Vault - PKI - CI/CD - Infrastructure security - Detection engineering - Incident response - SIEM/observability - Python/Bash/PowerShell or similar automation - Vulnerability management GPU infrastructure Send me a DM asap!!

The reason behind Lagos Traffic in the past 24 hours is so complex. From inadequate road infrastructure to unworthy heavy duty vehicles and impatience. Hope we do better as a people.

4 pictures of you in 4 different countries?