--- title: DigitalOcean patches AMD's Safe RET flaw fleet-wide,... description: DigitalOcean rebooted 1,600 AMD hypervisors ahead of a public CVE disclosure using automation built during a prior incident three weeks earlier. Meta... canonical: https://daily.dev/posts/digitalocean-patches-amd-s-safe-ret-flaw-fleet-wide-meta-ships-custom-silicon-for-agentic-ai-cl6kusdms twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: DigitalOcean patches AMD's Safe RET flaw fleet-wide, Meta ships custom silicon for agentic AI | daily.dev og:description: DigitalOcean rebooted 1,600 AMD hypervisors ahead of a public CVE disclosure using automation built during a prior incident three weeks earlier. Meta... og:url: https://daily.dev/posts/digitalocean-patches-amd-s-safe-ret-flaw-fleet-wide-meta-ships-custom-silicon-for-agentic-ai-cl6kusdms og:image: https://api.daily.dev/og/posts/CL6kUSDMS.png og:image:alt: DigitalOcean patches AMD's Safe RET flaw fleet-wide, Meta ships custom silicon for agentic AI og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # DigitalOcean patches AMD's Safe RET flaw fleet-wide, Meta ships custom silicon for agentic AI **[Backend Digest](https://daily.dev/sources/backend_digest)** · 4 min read · 1 upvotes · 0 comments ## Summary DigitalOcean rebooted 1,600 AMD hypervisors ahead of a public CVE disclosure using automation built during a prior incident three weeks earlier. Meta introduced two new hardware pieces for AI-scale networking: MetaRoCE, a lossy RDMA transport, and MTIA 300, a training chip with NICs built directly onto the die. NVIDIA countered with its own agentic-AI infrastructure stack spanning Vera Rubin GPUs, a Vera CPU, and BlueField-4 DPUs. Elsewhere, PostgreSQL's prepared statement planning quirk and a fresh SharePoint RCE chain are worth your attention today. ## Content **TLDR:** DigitalOcean rebooted 1,600 AMD hypervisors ahead of a public CVE disclosure using automation built during a prior incident three weeks earlier. Meta introduced two new hardware pieces for AI-scale networking: MetaRoCE, a lossy RDMA transport, and MTIA 300, a training chip with NICs built directly onto the die. NVIDIA countered with its own agentic-AI infrastructure stack spanning Vera Rubin GPUs, a Vera CPU, and BlueField-4 DPUs. Elsewhere, PostgreSQL's prepared statement planning quirk and a fresh SharePoint RCE chain are worth your attention today. --- ## DigitalOcean closes two fleet-wide security incidents three weeks apart DigitalOcean patched a KVM guest-to-host escape (CVE-2026-53359) across its hypervisor fleet within eight days via livepatching, with a few hundred older-kernel hosts needing full rebuilds. Three weeks later, a privately disclosed AMD Zen 'Safe RET' speculative-execution flaw forced a full reboot of roughly 1,600 AMD hypervisors across twelve regions, since the fix was compiler-inlined and couldn't be livepatched. They finished a day before AMD's public bulletin, reusing the capacity-management playbook from the first incident, and reported zero confirmed customer impact on both. [Read more](https://daily.dev/posts/Enoi72tn8) ## Meta builds its own RDMA transport and training chip for million-GPU scale MetaRoCE ditches standard RoCE's in-order delivery and PFC-based losslessness for out-of-order delivery, multipathing, and loss tolerance handled entirely on the NIC. On a 64-node AMD GPU cluster it held about 86% throughput at 1% packet loss, well past where standard RoCEv2 falls apart. Separately, MTIA 300 puts 12 custom 800 Gbps RDMA NICs and 16 dedicated communication engines directly on-chip, keeping AllReduce and AllToAll off the compute path entirely and cutting compute degradation from over 20% on GPUs to under 0.5%. Meta is contributing MetaRoCE's spec and reference implementation to the Open Compute Project ahead of a full release in October. [Read more](https://daily.dev/feed-by-ids?id=4RzMHAXu4&id=JPXE0EdSp) ## NVIDIA answers with a full agentic-AI hardware stack NVIDIA published three pieces of infrastructure aimed squarely at agent workloads: Spectrum-X Ethernet, which cuts failover time from 1.08s to 2.68ms during link failures; a new Vera CPU claimed to deliver up to 1.5x the agentic performance of AMD's Venice chip based on telemetry from 163,000 real agentic sessions; and BlueField-4 DPUs offloading security and storage off host CPUs with 4x the memory bandwidth of BlueField-3. On the SemiAnalysis AgentX benchmark (built from replayed Claude Code sessions), Vera Rubin NVL72 previews show up to 30x higher throughput per megawatt than GB300 NVL72. [Read more](https://daily.dev/feed-by-ids?id=bOjae9thx&id=JhkHnzlI9&id=ZVsUY0ROB&id=R1iqkQTkN) ## PostgreSQL's sixth-execution plan switch quietly tanks query performance Prepared statements get a custom plan for the first five executions, then PostgreSQL builds a generic plan on the sixth and switches permanently if its estimated cost looks cheaper, even though that decision never checks actual runtime. Generic plans use blind selectivity estimates that fall apart on skewed data, which is why some queries mysteriously slow down right around execution six, or ten to eleven if your driver auto-promotes to named statements first. Fixes include plan_cache_mode=force_custom_plan, disabling driver-side prepare, or partial indexes on the skewed column. [Read more](https://daily.dev/posts/A6UMIbUt5) --- ## Also notable - **SharePoint RCE chain lets attackers execute OS commands via .NET type instantiation:** CVE-2026-63520 exploits unrestricted Type.GetType() calls in SharePoint's BDC subsystem via a malicious .bdcm file, and when chained with auth-bypass CVE-2026-55040 becomes unauthenticated RCE. [Read more](https://daily.dev/posts/5LEsi7Ws4) - **Stripe acquires OpenRouter as agent token usage overtakes humans:** Agentic token consumption grew 14x since February 2026 to 7.3 trillion tokens weekly versus 2.8x growth for human usage, with OpenRouter now routing over 10 trillion tokens daily across 400+ models. [Read more](https://daily.dev/posts/ndkbAe9Zp) - **Kafka's real cost driver is cross-AZ network traffic, not brokers:** Replication factor 3 across three AZs multiplies produced data into roughly 4.7x cross-AZ traffic, turning a 100 MB/s cluster's $2,500/month broker bill into $24,000/month in transfer fees. [Read more](https://daily.dev/posts/OVhcnP0sB) - **vLLM's DFlash speculative decoding nearly quadruples CPU token throughput:** On Qwen3.5-9B with Intel Xeon 6, DFlash in vLLM v0.25.0 delivered a 3.92x average speedup and 74% lower cost per token at concurrency 1 across HumanEval, GSM8K, and MT-Bench. [Read more](https://daily.dev/posts/SUiryHqim) - **MinIO's end of life leaves a billion-pull project without security patches:** MinIO was archived upstream on February 13, 2026, and Docker's Extended Lifecycle Support now backports CVE fixes and ships SBOMs and SLSA Level 3 provenance for up to five years past EOL. [Read more](https://daily.dev/posts/Ychzc2oun) ## Similar posts on daily.dev - [Patching at Fleet Scale, Twice: How DigitalOcean Closed Januscape and the AMD Safe RET Issue Without Customer Impact](https://daily.dev/posts/patching-at-fleet-scale-twice-how-digitalocean-closed-januscape-and-the-amd-safe-ret-issue-without-enoi72tn8) · DigitalOcean · 1 upvotes · 1 comments - [MetaRoCE: A New RDMA Transport Built for AI-Scale Ethernet](https://daily.dev/posts/metaroce-a-new-rdma-transport-built-for-ai-scale-ethernet-4rzmhaxu4) · Facebook Engineering · 0 upvotes · 0 comments - [AMD Sends Out Latest Linux Patches For RMPOPT Optimization](https://daily.dev/posts/amd-sends-out-latest-linux-patches-for-rmpopt-optimization-lkh6zfsjl) · Phoronix · 1 upvotes · 1 comments --- Tags: [#security](https://daily.dev/tags/security), [#nvidia](https://daily.dev/tags/nvidia), [#digitalocean](https://daily.dev/tags/digitalocean) [View this post on daily.dev](https://daily.dev/posts/digitalocean-patches-amd-s-safe-ret-flaw-fleet-wide-meta-ships-custom-silicon-for-agentic-ai-cl6kusdms) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/digitalocean-patches-amd-s-safe-ret-flaw-fleet-wide-meta-ships-custom-silicon-for-agentic-ai-cl6kusdms","headline":"DigitalOcean patches AMD's Safe RET flaw fleet-wide, Meta ships custom silicon for agentic AI","text":"DigitalOcean rebooted 1,600 AMD hypervisors ahead of a public CVE disclosure using automation built during a prior incident three weeks earlier. Meta introduced two new hardware pieces for AI-scale networking: MetaRoCE, a lossy RDMA transport, and MTIA 300, a training chip with NICs built directly onto the die. NVIDIA countered with its own agentic-AI infrastructure stack spanning Vera Rubin GPUs, a Vera CPU, and BlueField-4 DPUs. Elsewhere, PostgreSQL's prepared statement planning quirk and a fresh SharePoint RCE chain are worth your attention today.","url":"https://daily.dev/posts/digitalocean-patches-amd-s-safe-ret-flaw-fleet-wide-meta-ships-custom-silicon-for-agentic-ai-cl6kusdms","datePublished":"2026-08-25T04:17:48.748Z","dateModified":"2026-09-13T20:57:56.767Z","author":{"@type":"Organization","name":"Backend Digest","logo":"https://media.daily.dev/image/upload/s--9uCSrsAb--/f_auto,q_auto/v1773839398/logos/backend_digest?_a=BAMAMiiu0","url":"https://daily.dev/sources/backend_digest"},"interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"isPartOf":{"@type":"WebPage","url":"https://daily.dev/sources/backend_digest","name":"Backend Digest"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Backend Digest","item":"https://daily.dev/sources/backend_digest"},{"@type":"ListItem","position":3,"name":"DigitalOcean patches AMD's Safe RET flaw fleet-wide, Meta ships custom silicon for agentic AI"}]} ```