Aethir’s Post

In my recent article on hashnode, I have highlighted the significant milestones reached at CUDOS in Q2. I have covered the upgrades made at CUDOS Intercloud, additional GPUs to boost supply, the stretegic partnerships, Notable recognitions CUDOS earned and the expansions in a bit to expand the reach of sustainable computing. Q2 in CUDOS has been spectacular. Dive deep into the details in this article:

I've just published a detailed blog post on Hashnode highlighting the remarkable achievements of CUDOS in Q2, 2024. From strategic partnerships and technological advancements to impressive growth metrics and community engagement, this quarter has been transformative. Check out the full details here👇 https://lnkd.in/d4Z3NWyP #CUDOS #Intercloud

Compute as Rewards! Dive into the future of decentralized computing with Fetch.ai exciting initiative, Fetch Compute. With an electrifying $100 million investment in their infrastructure, they are unleashing the power of AI and blockchain like never before! What’s in It for You? • Exclusive GPU Access: Get your hands on the elite Nvidia H200, H100, and A100 GPUs, fueling your projects with unmatched computing power. • Innovate and Build: With Fetch Compute, your dreams have no limits. Explore complex models, develop innovative solutions, and push the boundaries of what’s possible. • Earn While You Support: Stake Fetch.ai’s native coin, $FET, and earn Fetch Compute Credits. Use these credits to access their top-tier GPU resources on the Fetch Compute network, starting March 7. How Does It Work? 1. Stake Your $FET: Join the program by staking Fetch.ai’s native coin. It’s your ticket to the world of unlimited computing potential. 2. Earn Compute Credits: Your staked $FET not only supports the network but rewards you with Compute Credits, the key to unlocking our powerful GPUs. 3. Power Your Projects: Use your credits to tap into our cutting-edge computing resources. Whether it’s for AI development, complex simulations, or innovative applications, the power is now in your hands. Amazing initiative Humayun Sheikh ! Read more - https://lnkd.in/gQ_jDNtp #compute #fetchai #rewards

https://lnkd.in/gHjd_8u4 Always a bummer when Satya & Sam realize that some smart squatter has already booked the domain names "stargate.ai" & "stargate.com"... The choice is to spend some of that $100B buying these names off the squatter or go with the next best options: 1) star-gate.ai 2) star-gate.com #stargate #microsoft #openAI

Aethir Launches Groundbreaking Decentralized AI and Gaming GPU Node Sale Aethir's Node Sale introduces enterprise-grade AI and Gaming GPU-as-a-service, powered by NVIDIA's H100 chips. Explore how Aethir's decentralized approach is democratizing high-performance computing and reshaping the future of cloud services

We're happy to be featured on DePINs.blog, in collaboration with DePIN Connect, as a case study for "Decentralizing Computational Power for #AI and Cloud Gaming". Learn more about our Web3 GPU #DePIN protocol solves the bottlenecks of centralized computing below🔥 https://lnkd.in/g8-nTmAM

🚀 GPU.net : Decentralized GPU Marketplace - Revolutionary Solution🚨: Break free from traditional cloud services with transparent, secure, and fair GPU access. - Smart Contract Power📄: Solidity-backed architecture ensures seamless, secure transactions. - User-Friendly Onboarding🫂: Simple setup and registration for providers and renters. - Top-Tier Security🔐: `Ownable` patterns and cryptographic operations ensure safe transactions.

Talking with Olivier Avaro of Blacknut we unearthed pivotal insights into the evolution and hurdles present within the cloud gaming infrastructure: 1. Strategic Value of Concurrent Sessions: The capability of servers to accommodate multiple users simultaneously is a critical performance and economic KPI. Enhancements in server capacity for concurrent gaming sessions are closely linked to operational cost reductions, emphasizing the significance of scalability in cloud gaming infrastructures. 2. Limitations of Traditional CPU+GPU Architectures: The conventional approach of segmenting a GPU to support numerous gaming instances faces inherent limitations, resulting in bottlenecks that affect both user concurrency and cost-efficiency. NVIDIA GPU’s are capable of supporting 24 sessions, 48 at max on a single server, this shows the need for innovation in architectural design. 3. Innovative Integration of VPU in Architecture: Blacknut is leading the charge in circumventing GPU constraints by integrating an architecture comprising CPU, GPU, and VPU. The incorporation of a Video Processing Unit (VPU), such as those developed by NETINT Technologies Inc. for video encoding alleviates the GPU’s workload, augmenting server capacity. 4. Enhanced Density and Flexibility through VPU Utilization: Assigning video encoding tasks to VPUs can result in a tenfold increase in the capacity for hosting concurrent game sessions, enabling a server to support as many as 200 sessions simultaneously. This marks a substantial advancement from the previous limitation of 24 to 48, illustrating the potential with VPU integration. 5. Economic Implications of VPU Integration: The integration of a VPU into the server architecture fundamentally shifts the economic landscape of cloud gaming. With the enhancement of server capabilities, the operational costs associated with cloud gaming services decrease, fostering a more economical provision for service providers and making cloud gaming more accessible to a wider audience. 6. Fostering a Sustainable Cloud Gaming Ecosystem: The synthesis of CPU, GPU, and VPU architectures is instrumental in creating a resilient and scalable cloud gaming ecosystem. This innovative approach ensures the long-term viability and economic sustainability of cloud gaming infrastructures. I invite the community to delve into these advancements further by viewing the discussion on YouTube and joining us at #GDC2024. https://lnkd.in/g_JzyABx

There is big evolution in AI and LLM models currently and to deploy most of these models you'll need GPU resources. One of the tools I've been used recently and saw good potential in is akash network https://akash.network/ you can offer your unused GPUs, CPUs, Memory to a decentralised network and someone else will rent it.

I’m curious, will Google decide to sell their server grade TPUs in addition to renting (via GCP) them out? I think maybe yes because, TPUs especially the latest ones are pretty good (V5p has 459 TFLOPS/chip and has 95GB High Bandwidth memory and is better than H100s) and they have already proven that they can be used for both #machinelearning #inference and #training at #google scale. So for people looking for alternatives to #nvidia, TPUs might be a better bet than going to other unproven competitors in the server grade AI accelerator space like AMD , Groq (I know Groq’s latest LPU is supposedly much faster for LLM inference) etc. Maybe Google can eat some of NVIDIA ‘s lunch 🥪 🤓. What do you all think? A #DGX equivalent from Google powered by TPUs will be interesting