StorageReview.com

Supermicro's new Kubernetes Edge AI Appliances drop the per-site storage array: Red Hat OpenShift plus Portworx aggregate local storage inside the edge servers into a software-defined layer that keeps running through network outages. Turnkey for retail, factories, and remote sites. storagereview.com/news/supermicr… #EdgeAI #Kubernetes #Storage @Supermicro @portwx @RedHat

WhiteFiber's Project Redwood linked two NVIDIA H200 clusters 83 km apart into one logical supercluster, hitting 111.2 Tbps and 0.9 ms round-trip latency (within 8% of fiber's physical limit) on only part of the spectrum. Built with DriveNets and WEKA. storagereview.com/news/whitefibe… #AI #Networking #GPU @WhiteFiber_ @drivenets @weka

More B70 photos including the guts!

Four Intel Arc Pro B70s = 128GB of pooled VRAM for ~$3,800, enough to host a 120B MoE model, and that quad beat a single RTX Pro 6000 by 65% in Mistral Small 24B at full concurrency. The card ($949 list, 32GB) is great. The catch is Intel's software stack. storagereview.com/review/intel-a… #Intel #ArcPro #LocalAI @intel

Samsung's new 990 is a DRAM-less Gen4 value drive for pricey NAND times: 7,250 MB/s reads on the 2TB and a claimed 38% better power efficiency than the 990 PRO. $269.99/1TB, $529.99/2TB. storagereview.com/review/samsung… #SSD #Samsung #PCBuild @SamsungSemiUS

Real FSC walnut, brown accents, zero RGB. The Antec Flux Pro Noctua Edition ships six Noctua G2 fans at offset speeds to kill resonance, rated up to 8 dB(A) quieter than the stock Flux Pro. A full tower built around silence instead of lighting. #Antec #Noctua #PCBuild @Noctua_at @antec

YouTube full podcast with HPE is here - youtube.com/watch?v=EJvqUb…

New podcast: Brian talks with HPE's Dan Gardner about how the Alletra Storage MP B10000 meets ransomware head-on: built-in behavioral detection, auto-preserved immutable snapshots, and a lab that tests against ~200 ransomware variants. 40 min, on YouTube and everywhere podcasts are found. #Ransomware #Storage #HPE @HPE

At RAISE Summit, DDN and European sovereign-cloud provider Nebul validated KV cache acceleration on NVIDIA infrastructure, reporting early measurable gains in time to first token and completing RoCE validation. The focus: cost per token and GPU utilization for inference. storagereview.com/news/ddn-and-n… #AI #Inference #Storage @DDNintelligence

DeepInfra opened a 1.7MW Toronto data center with 1,000+ NVIDIA Blackwell B300 GPUs, its first site outside the US and ninth overall. The inference-focused cloud says it now processes nearly 5 trillion tokens a week across 200+ open-source models. storagereview.com/news/deepinfra… #AI #Inference #NVIDIA @DeepInfra @NVIDIAAI

Scality and WEKA are expanding in France with a joint support deal giving customers a single in-country point of sale and French tier-one support. Their WEKA NeuralMesh + Scality ADI stack claims up to 10x faster performance and 20% lower infrastructure cost. storagereview.com/news/scality-a… #AI #Storage #Sovereignty @scality @weka

More photos of the Noctua case! Review live!

The Antec Flux Pro Noctua Edition packs six Noctua G2 fans (3x NF-A14x25, 2x NF-A12x25, 1x NF-A14x25 rear), FSC-certified walnut trim, and up to 8 dB(A) less noise than the stock Flux Pro. LED-neutral, mature, and $399.90. storagereview.com/review/antec-f… #PCBuild #Noctua #Antec @Noctua_at @antec

@IDFkushoverlord @IntelBusiness Asus Pro WS W890E-SAGE SE storagereview.com/review/intel-x…

@storagereview @IntelBusiness what mobo?

Intel B70 review is almost done. A few photos until then. #intel #gpu @IntelBusiness

@lupi_arsene @IntelBusiness Asus Pro WS W890E-SAGE SE storagereview.com/review/intel-x…

@storagereview @IntelBusiness What's the board under the B70s?

@fbwalker4 @IntelBusiness We're still wrapping the review but essentially the same as every other Intel GPU - good potential value but software ecosystem is a very distant third place behind NVIDIA and AMD.

@storagereview @IntelBusiness So in you tldr, thumbs up for running local models or things down? Or some part of the continuum?

@MatthewNDodd @IntelBusiness Model support is the problem. If Intel gets serious about software they’re in the game.

@storagereview @IntelBusiness They are interesting cards. Been having fun optimizing their llama.cpp performance.

NVIDIA is now pairing revenue sharing with credit support to help AI clouds finance GPU buildouts, earning from both hardware sales and a cut of cloud revenue. First partners: Sharon AI (up to 40,000 GB300 GPUs) and Firmus (a 360MW, 170,000-GPU campus in Indonesia). storagereview.com/news/nvidia-pa… #NVIDIA #AI #DataCenter @nvidia

@spiralout112 @SamsungSemiUS Will be a bit before they ship AND we get a new Gen6 platform. Dell seems closest, we saw early models based on AMD at DTW. Expect more at the end of this month, AMD is hosting their annual AI event.

@storagereview @SamsungSemiUS Be interesting to see latency figures.

Samsung's PM1763, its first PCIe Gen6 enterprise SSD, is in mass production: 28,400 MB/s reads (1.96x the Gen5 PM1753), 21,000 MB/s writes, and 6.8M random read IOPS on a new 4nm controller and 9th-gen V-NAND. storagereview.com/news/samsung-p… #PCIeGen6 #SSD #Samsung @SamsungSemiUS