Kevin Culp’s Post

MEGWARE now supporting 5th Gen AMD EPYC processors ensuring unmatched performance and density to meet the evolving needs of AI-enabled data centers. Built on the AMD “Zen 5” architecture, these processors deliver impressive performance and energy efficiency for a wide range of workloads, from #AI and cloud infrastructure to demanding business-critical applications. Test #AMDEPYC 9005 processors in the MEGWARE #BenchmarkCenter and convince yourself of the exceptional performance! More information at https://lnkd.in/ewUgi7jw #AMDTurin #TestDrive #Benchmarking

Go beyond the 8-GPU server limit with SuperNODE, built for massive AI workloads by integrating up to 32 #MI300X accelerators into one unified system. Learn how the GigaIO #SuperNODE maximizes GPU utilization and efficiency, simplifies your infrastructure, and effortlessly scales with #GenAI workloads. https://bit.ly/3zMW2FZ #supercomputer #AIworkloads #acceleratedcomputing

Hello #HPC community! Today during his keynote at #MSIgnite, our CEO Satya Nadella announced the newest addition #Microsoft #Azure H-series VMs, HBv5. This supercomputing class VM is designed to dramatically accelerate memory bandwidth-bound HPC workloads, driving performance per dollar up and overall HPC spend down. HBv5 is based our close partnership with #AMD to develop a custom #EPYC processor only available in Azure. The AMD EPYC 9V64H processor uses high frequency “Genoa” cores coupled with on-package HBM3. With four of these processors per VM, each HBv5 virtual machine delivers 6.9 TB/s of memory bandwidth, which is up to 8x higher than alternatives found elsewhere, and up to 35x higher than end of lifecycle hardware deployments on-premises. Each HBv5 VM also features 800 Gb/s of #NVIDIA InfiniBand connectivity enabling HPC applications to scale up to hundreds of thousands of cores. For memory-bandwidth-bound workloads like #CFD, #weather, #energy, quantum chemistry, and more, this will be a game-changer. We’re sharing more information about HBv5 VMs at https://aka.ms/AzureHBV5, and you can sign up for our preview, which will be opening in the first half of next year, at https://lnkd.in/eeuKs52s. If you’re in Atlanta for #SC24 this week, be sure to stop by the Microsoft booth # 1905 to see an HBv5 server, and be sure to catch one of my talks introducing HBv5 this week. I’ll be presenting in the Microsoft booth Tuesday at 12:10 p.m. and Wednesday at 2:10 p.m., and in AMD’s booth on Thursday at 10 a.m. #AzureHPCAI

New Mbv3 series (Mbsv3 and Mbdsv3), featuring both storage-optimized and memory-optimized, are now available in public preview 👉 Based on the 4th generation Intel® Xeon® Scalable processors, scale for workloads up to 4TB, and deliver higher remote disk storage performance 👉 Up to 650,000 IOPS, providing a 5x improvement in network throughput over the previous M-series families. 👉 Up to 10GBps of remote disk storage bandwidth, providing a 2.5x improvement in remote storage bandwidth over the previous M-series families. 👉 Increased remote storage performance for storage throughput-intensive workloads such as relational databases and data analytics applications. 👉 Designed from the ground up for increased resilience against failures in memory, disks, and networking based on intelligence from past generations. https://lnkd.in/dJUVaeTx #Microsoft #Azure #MicrosoftAzure #AzureCloud #VM

6.9 TB/S of memory bandwidth per node!! 35x more than what most HPC users have in their own datacenter! If you are doing CFD, FEA, crash, weather, anything that depends on memory bandwidth, this is a game changer!

The new Mbv3 storage and memory optimized virtual machines are now in public preview! #Azure #Compute #Microsoft

#DigitalOcean Bare Metal GPUs are now available: https://lnkd.in/d5XuiHJ2 🚀 Purpose-built to handle the most demanding #AI/ML workloads, these servers offer dedicated hardware for unparalleled performance & complete control over your infrastructure. Why choose Bare Metal #GPUs? 💡 → Train at scale: Process massive datasets with optimized performance. → Fine-tune models: Adapt pre-trained models to specialized use cases with precision. → Run high-speed inference: Deliver real-time predictions with unmatched responsiveness. → Build custom solutions: Configure your environment for advanced orchestration or unique workloads.

🚀 Ready to transform your data center's storage performance? Meet the TS70A-B8056: a 2U1S storage server powered by #AMD #EPYC™ 9005 series processor built for high IOPS and demanding workloads.   🔑 Key Highlights: ✔️26 NVMe U.2 hot-swap trays for ultra-fast, tool-less storage. ✔️3,072GB DDR5-5200 memory capacity across 24 DIMM slots. ✔️Expandable with PCIe 5.0 OCP v3.0 mezzanine for flexible networking. ✔️Energy-efficient 1,600W CRPS power supply with 80 Plus Titanium.   Optimized for: Databases, real-time analytics, and hybrid cloud storage. Streaming, #AI/ #ML workloads, and #HPC environments. Boost your data center's performance, scalability, and efficiency today.   👉 Learn more: https://pse.is/TS70AB8056   #DataStorage #HighIOPS #ServerInnovation

Following the KISS rule 1 of you can avoid going DMP via faster memory and so keep in a single node that's your 1st interest. Particularly for engineering applications where the biggest cost is the application licence costs more than the infrastructure costs.

“Microsoft and AMD have worked together to develop a custom 4th Generation EPYC™ processor with high bandwidth memory (HBM).” #azure #amd #epyc #hpc