Category: networks

It seems Malaysia is getting a DC boom, but it based on coal???

This is a MS NVIDIA GB200 based rack. I am quite impressed with the cooling systems being twice as big as the compute rack! And yes, MS is sticking with IB for AI networking.

I didnt know that Oracle OCI was getting that big in the DC/AI business. And they were related to xAI. Their biggest DC is 800 megwatts… and a new one will have three nuclear reactors??

FuriosaAI: A new AI accelerator in the market. Good: cheap, less power. Bad: memory size.

OCP concrete: Interesting how far can go the OCP consortium.

IBM Mainframe Telum II: You think the mainframes business doesnt exist. Well, it is not. Honestly, at some point, I would like to fully understand the differences between a “standard” CPU and a mainframe CPU.

NotoebookLM: It seems it is possible to make summary of youtube videos! (and free)

EdgeShark: wireshark for containers. This has to be good for troubleshooting

22-bit RSA broken with Quantum computer: I think Quantum computing is the underdog in the current “all-is-AI” world. Schneier says we are ok.

OCP24 Meta AI: It is interesting comparing the Catalina rack with the one from MS above. The MS has the power rack next to it but FB doesnt show it, just mention Orv4 supports 140kW and it is liquid cooled. I assume that will be next to Catalina like MS design. And AMD GPU are getting into the mix with NVIDIA. It mentions Disaggregated Scheduled Fabric (DSF), with more details here. And here from STH more pictures.

Intro LLM

LLM Bootcamp 2023:

NVIDIA Computex 2024: It seems they are going to yearly cadence for networking kit. They showed plans for 2025 and 2026… I liked the picture of a NVLink spine and the huge heatsinks for B200….

UALink: The competition for NVLink. This is for GPU-to-GPU communication. UltraEthernet is for connecting pods.

Aurora supercomputer: Exascale broken. Based on HPE slingshot interconnect (nearly 85k endpoints) Everything else is Intel.

Arista AI Center: it seems they are going to team-up with NVIDIA. Some EOS running on the nics.

Kubenet: Seems interesting but only supporting Nokia SRLinux at the moment.

Nutrigenomia:

“Lo que hicimos fue un trabajo personalizado en el que cuidamos todos los aspectos de la nutrición y buscamos la regeneración y la correcta expresión de sus genes.”

“fisiogenómica: Yo lo llamo así porque mezcla fisioterapia, nutrición y nutrigenómica. En cada persona tenemos que buscar por síntomas, análisis e intervenciones qué alimentos limitar por producir una mala expresión genética, pero todas las pautas están basadas en la Pirámide de la Dieta Mediterránea”

Videos:

Bear Grylls: Be kind, never give up.

Several posts worth reading. There are plenty of things go over my knowledge. I already posted this, it is a good refresher.

GPU Fabrics: The first of the article is the one I am more lost as it about training and the communications between the GPU depending on the take to handle the models. There are several references to improvements as the use of FP8 and different topologies. As well, a bit more clear about NVLink (as internal switch for connecting GPUs inside the same server or rack)

When it moved to the inter-server traffic, I started to understand a bit more things like “rail-optimized” (it is like having a “plane” for my old job where the leaf only connects to a spine instead of all spines, in this case each GPU connects to just one leaf. If you cluster is bigger then you need spines). I am not keen of modular chassis from operations point of view but it is mentioned as an option. Fat-tree CLOS, Dragon-Fly: reminds me to Infiniband. Like all RDMA.

And Fabric congestion it is a big topic with many different approaches: adaptive LB (IB again), several congestion control protocols and mention to Google (CSIG) and Amazon (SDR) implementations.

In general I liked the article because I dont really feel any bias (she works for Juniper) and it is very open with the solutions from different players.

LLM Inference – HW/SW Optimizations: It is interesting the explanation about LLM inferencing (doubt I can’t explain it though) and all different optimizations. The hw optimization (different custom hw solutions vs GPU) section was a bit more familiar. My summary is you dont need the same infrastructure (and cost) for doing inference and there is an interest for companies to own that as it should be better and cheaper than hosting with somebody else.

Network Acceleration for AI/ML workloads: Nice to have a summary of the different “collectives”. “collectives” refer to a set of operations involving communication among a group of processing nodes (like GPUs) to perform coordinated tasks. For example, NCCL (Nvidia Collective Communication Library) efficiently implements the collective operations designed for their GPU architecture. When a model is partitioned across a set of GPUs, NCCL manages all communication between them. Network switches can help offload some or all of the collective operations. Nvidia supports this in their InfiniBand and NVLink switches using SHARP (Scalable Hierarchical Aggregation and Reduction Protocol – proprietary). This is call “in-network computing”. For Ethernet, there are no standards yet. The Ultra Ethernet Consortium is working on it but will take years until something is seen in production. And Juniper has the programmable architecture Trio (MX routers – paper) that can do this offloading (You need to program it though – language similar to C). Still this is not a perfect solution (using a switches). The usage of collectives in inference is less common than their extensive use during the training phase of deep learning models. This is primarily because inference tasks can often be executed on a single GPU

From a different topics:

Learning at Cambridge: Spend less hours studying, dont take notes (that’s hard for me), go wild with active learning (work in exercises until you fully understand them)

British Library CyberAttack: blog and public learning lesson. I know this is happening to often for many different institutions but this one caught my eye 🙁 I think is a recurrent theme in most government institutions were upgrading is expensive (because it is not done often), tight budgets and IT experts.

“Our major software systems cannot be brought back in their pre-attack form, either because they are no longer supported by the vendor or because they will not function on the new secure infrastructure that is currently being rolled out”

However, the first detected unauthorised access to our network was identified at the Terminal Services server. Likely a compromised account.

Personally, I wonder what you can get from “stealing” in a library ???