Articles by Suresh
-
A New Chapter for Sysdig and for Me
A New Chapter for Sysdig and for Me
Today marks a significant moment for me: after nearly seven incredible years of leading , I’ve decided to step down as…
291 Comments -
Beyond EDR: CDR for Mission-Critical Cloud Operations
Beyond EDR: CDR for Mission-Critical Cloud Operations
Real-time threat detection and response is not a luxury for organizations running mission-critical applications in the…
5 Comments
Activity
8K followers
-
Suresh Vasudevan reposted thisSuresh Vasudevan reposted thisAI's next bottleneck isn't chips, and it isn't datacenters. 𝗜𝘁'𝘀 𝗳𝗶𝗻𝗮𝗻𝗰𝗶𝗻𝗴. That's the question on the Master Stage at RAISE Summit Paris this Wednesday — here's why it matters: SemiAnalysis projects AI debt needs approaching $𝟳.𝟭 𝘁𝗿𝗶𝗹𝗹𝗶𝗼𝗻 by 2029 — on track to surpass every other US asset-backed market. And every one of those loans gets underwritten against a single question: how much revenue-generating compute does this cluster actually produce? That question changes what infrastructure means. When lenders size debt on a cluster's real output, every layer that determines that output becomes a credit variable: how fast storage feeds the GPUs, whether the fabric holds at scale, whether data pipelines keep up, whether a hardware failure costs minutes or days of paid compute. Infrastructure quality is becoming the difference between a cluster that's bankable and one that isn't. Infrastructure as destiny, quite literally. Nobody sits closer to that shift than SemiAnalysis. Their ClusterMAX rating system and GPU Rental Pricing Index are becoming the tools lenders use to price this market. On Wednesday, Jordan Nanos, lead author of ClusterMAX at SemiAnalysis, moderates our CEO, Suresh Vasudevan alongside Greg Matson (Solidigm), Stephanie Cohen (Cloudflare), Jeff Denworth (VAST Data), and Don Barnetson (Credo) — the storage, network, data, connectivity, and resilience layers that decide what a GPU dollar actually returns. If you finance, build, or run AI infrastructure, this is 40 minutes on what capital providers now scrutinize before a cluster gets funded — and what that means for how you build. At RAISE? Add it to your agenda: "Infrastructure as Destiny: The Compute-Capital-Cloud Trinity" · 𝗝𝘂𝗹𝘆 𝟴 · 𝟭𝟬:𝟰𝟬 𝗔𝗠 · 𝗠𝗮𝘀𝘁𝗲𝗿 𝗦𝘁𝗮𝗴𝗲. Not in Paris? Follow Clockwork.io — we'll share the takeaways after the session. #RAISESummit #AIInfrastructure #AIEconomics #GPU
-
Suresh Vasudevan shared thisAI networking is entering a period of rapid change and innovation - at a pace that seems unprecedented: RoCEv2 and InfiniBand; Adaptive Routing and/or Dynamic Load Balancing; and appearing on the horizon - MRC and UEC! For infrastructure leaders, the question is no longer just “which fabric is fastest?” It is: what should we adopt, when should we transition, and how do we preserve visibility and fault tolerance through the shift? On June 2, I’ll join Roy Chua and Balaji Prabhakar to discuss network observability, workload-aware fabric design, failure recovery, and the real goal: higher effective utilization of scarce GPU/XPU capacity. Join us.Suresh Vasudevan shared this𝐘𝐨𝐮𝐫 𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐣𝐨𝐛 𝐣𝐮𝐬𝐭 𝐬𝐭𝐚𝐥𝐥𝐞𝐝. 𝐄𝐯𝐞𝐫𝐲 𝐝𝐚𝐬𝐡𝐛𝐨𝐚𝐫𝐝 𝐬𝐚𝐲𝐬 𝐭𝐡𝐞 𝐧𝐞𝐭𝐰𝐨𝐫𝐤 𝐢𝐬 𝐟𝐢𝐧𝐞. 𝐍𝐨𝐰 𝐰𝐡𝐚𝐭? This is the failure mode that costs the most time — not because the outage is large, but because the diagnosis is slow. The network reports healthy. The GPUs report utilized. The job isn't moving. And somewhere in the gap between those three facts is the actual problem. It might be a hot spine absorbing traffic from a single flow. A silent bad NIC slowing one rank in a synchronized job. A congestion event in the scale-out fabric that the scale-up layer can't see. A coordination stall that looks like compute but traces back to a fabric event three layers down. 𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝑚𝑜𝑛𝑖𝑡𝑜𝑟𝑖𝑛𝑔 𝑤𝑎𝑠𝑛'𝑡 𝑏𝑢𝑖𝑙𝑡 𝑓𝑜𝑟 𝑡ℎ𝑖𝑠. 𝐼𝑡 𝑤𝑎𝑠 𝑏𝑢𝑖𝑙𝑡 𝑓𝑜𝑟 𝑓𝑎𝑖𝑙𝑢𝑟𝑒𝑠 𝑡ℎ𝑎𝑡 𝑎𝑛𝑛𝑜𝑢𝑛𝑐𝑒 𝑡ℎ𝑒𝑚𝑠𝑒𝑙𝑣𝑒𝑠. Gray failures in AI fabrics are the opposite — they're quiet, they're compounding, and they look like something else until you have the right instrumentation to see through them. The teams that recover fastest aren't the ones with the fastest GPUs. They're the ones who can close the gap between "the dashboard says fine" and "here's what actually happened." 🔗 Join this live panel discussion where Roy Chua, Suresh Vasudevan, and Balaji Prabhakar will discuss and debate three principles that will shape how operators think about AI fabric economics and observability over the next 18 months: 👇 #AIInfrastructure #DataCenterNetworking #GPUClusters #AIFabric #MLInfrastructure
-
Suresh Vasudevan reposted thisSuresh Vasudevan reposted thisLast week I joined a great discussion at the Scale Venture Partners RevOps Council with Craig Rosenberg, Andy Mowat and Kristina McMillan on what teams are actually starting, stopping, and continuing with AI. One of the ideas I shared (and have been seeing consistently across teams): STOP relying on legacy, rules-based systems. We've depended on them for years, but they’re: - brittle - hard to maintain - and can't adapt to context START moving toward reasoning-based systems. Not because they’re “AI,” but because they behave fundamentally differently: They handle edge cases instead of breaking on them They can pull in additional context (via tools, APIs, web search, etc.) when needed They’re configured in natural language, so updates aren’t bottlenecked by a systems admin. And importantly, they can actually explain WHY they did what they did Well-designed reasoning systems aren’t a black box, they’re actually more auditable than legacy tools. Curious how others are thinking about this: Where are rules still holding up well, and where are they starting to break? Sharing the full session in the comments if you're interested
-
Suresh Vasudevan shared thisClockwork Systems, Inc.' TorchPass has been featured in the SemiAnalysis ClusterMax 2.1 announcement, “How Much Do GPU Clusters Really Cost?”. Central to the announcement is a focus on fault tolerance and goodput (defined as the amount of useful work users can perform on their cluster), given the massive impact of poor resilience on overall GPU cluster utilization. The statement that TorchPass tested as the "only option that maintains the same performance as jobs without fault tolerance" is very gratifying and what we have been working towards over the past year! Thank you Jordan NanosSuresh Vasudevan shared thisSpent a few hours with the new SemiAnalysis piece on Cluster TCO. A few notes worth sharing for anyone currently sizing their next training cluster. The takeaway most people will latch onto is that $/GPU-hr accounts for only 60 to 70% of what you actually pay. The rest is storage, networking, control plane, support, setup, debugging, and goodput expense. That last one is quietly the biggest swing factor at scale, and it's the one most TCO spreadsheets don't have a row for. Their benchmark is worth walking through. 5,184 GB300 GPUs. A single 4,096-GPU pretrain using about 80% of the cluster. $4/GPU-hr held constant across provider tiers to isolate everything else. Even with pricing held constant, goodput expense ranges from 6.14% to 20.91% of total TCO. That spread is the compounding effect of two things: how often your provider's nodes fail, and how much your fault-tolerance stack bleeds when they do. Most operators think about the first. The second is where the money is. The three frameworks they benchmarked are a good lens: TorchFT (Meta, open source) works, but cross-replica allreduce falls back to GLOO over frontend TCP because NCCL doesn't handle dynamic world sizes gracefully. That means CPU, kernel path, NIC, switch firmware on every cross-replica step. On comparable HSDP jobs, SemiAnalysis measured a ~10% throughput hit. At 4,096 GPUs, that compounds step over step. HyperPod Checkpointless (AWS) holds redundant model replicas live on peer GPUs and recovers via RDMA over EFA. Very fast. But the redundant replicas cost device memory roughly equal to one DCP checkpoint, so you either run smaller batches or shift parallelism. AWS measured about 5%. Also: K8s-only, NeMo/Megatron-only, AWS-only. TorchPass is a scheduler plugin plus a Manager class in the training script. When the scheduler flags a degrading node, we JIT-checkpoint via get_state() and RDMA-transfer state to an idle spare. No GLOO fallback, no redundant replicas. The cost is the spare pool: 4 nodes in their benchmark, 0.62% of the cluster. The reason scheduler-level migration wins specifically at >1k GPUs is that soft failures dominate there. Accumulating ECCs, GPUs intermittently off the bus, power blips, link flaps. The scheduler sees those signals before NCCL stalls, so migration happens while the node is still healthy enough to hand over state cleanly. Reactive recovery has to wait for the failure to actually manifest. If you want to run this on your own numbers, SemiAnalysis released their TCO and Goodput calculators free at clustermax.ai/tco. Load "Large LLM Pretrain," drop in your actual $/GPU-hr and MTBF, swap the fault-tolerance framework. Maybe a 2 minute exercise, and the goodput line moves a lot. If you're sizing a >1k-GPU cluster and haven't explicitly priced fault tolerance as a line item, it's probably the single biggest TCO lever you're not tracking.
-
Suresh Vasudevan shared thisI keep asking engineering leaders the same question: what's your effective GPU utilization: not nvidia-smi utilization, but the percentage of paid GPU-hours that produce useful forward progress? Many don't know the number. The ones who do usually don't like it. This two-part blog series aims to fix that. Part 1 (https://shorturl.at/KBdN3) was about breaking down utilization into its underlying components. This latest installment, The Dirty Dozen, names the twelve specific leaks responsible, ranked by impact, with the team and metric that owns each fix. (https://lnkd.in/gMpA5e7S). I am curious as to which 1 or 2 or 3 of the dozen (or something I completely missed) are the most painful ones for you!?Suresh Vasudevan shared thisTwelve leaks. That's what sits between the GPU-hours you're paying for and the useful work you're actually getting. Part 2 of this series from our CEO Suresh Vasudevan names each one — four that hit both training and inference, four specific to training, four specific to inference — sized by impact, mapped to the team and metric that owns it. Read more about the 'Dirty Dozen' here: https://lnkd.in/gMpA5e7S What's the biggest bottleneck in your GPU stack right now? #GPU #AIInfrastructure #MachineLearning #GPUInfrastructure #EnterpriseAI #AI #MLOps
-
Suresh Vasudevan shared thisThe Big Tech earnings season just wrapped, and the numbers are staggering: Amazon, Alphabet, Meta and Microsoft - $650B in 2026 capex - up 70% from 2025, with roughly 75% tied to AI infrastructure. At this rate, global AI investment will comfortably exceed a trillion dollars. But here's the uncomfortable question: Who's underwriting all of this? Ultimately, enterprises are — and right now, only 12% of them are seeing both cost savings and revenue gains from AI (PwC 2026 CEO Survey). The gap between what's being built and what's being monetized is the defining tension of this infrastructure cycle. That's exactly what we're digging into, and I am looking forward to insights from Vinita Ananth (Nebius) and Chris Morgan - two operators who have been in the trenches scaling AI infrastructure across cloud, neocloud, and enterprise environments.Suresh Vasudevan shared thisEnterprise AI Infrastructure ROI: Training vs Inference How do the economics and operational risks differ? As enterprise AI moves from experimentation to mission-critical production, ROI is determined by infrastructure decisions—how performance, reliability, and cost hold up at scale. Join a live virtual panel with leaders who’ve built and operated AI infrastructure at scale, discussing how leading enterprises take AI into production with predictable performance and sustainable economics: 🎙️ Speakers Vinita Ananth — Senior Director of Product, Nebius (formerly AWS & Microsoft) Chris Morgan — CEO, 7sg.ai (former VP of AI Solutions, VAST Data) Suresh Vasudevan — CEO, Clockwork.io (Moderator) Topics: • How training and inference drive different cost and risk profile • Why mixed workloads expose utilization and architectural gaps • How full-stack control translates into stronger SLAs and faster recovery • How enterprises scale AI without over-provisioning or runaway spend 📅 Feb. 19, 2026 at 10am PST / 1pm EST 👉 Register to join: https://lnkd.in/gY_cY4DJ
-
Suresh Vasudevan shared thisVery grateful to Sandesh “Dish” Patel and Chris Brandt for the opportunity to reminisce for a bit and also reflect on some key trends in AI infrastructure. We at Clockwork Systems, Inc. are fixated on how to make AI efficient, reliable and performant - while building a fun organization and culture.Suresh Vasudevan shared thisBack in 1998, when I was fresh out of college at NetApp, my peers and I would see one person power walking across campus with a laptop in one hand: Suresh Vasudevan. His name was always tied to excellence with constant promotions and praise from NetApp executive leadership. Years later, I met him again as CEO of Nimble Storage during its IPO journey and later at Sysdig. Today, he’s leading Clockwork Systems, Inc., an AI startup focused on making the world’s largest GPU powered AI infrastructures more observable, reliable, efficient and performant. During our upcoming What the futr Podcast we spent time breaking down the hard problems in AI infrastructure. And as always, I walked away inspired and well informed. Suresh has this rare combination of brilliance, clarity and humility that makes people instantly respect him. Grateful for the conversation and excited for what Clockwork is building. Here’s the trailer: https://lnkd.in/gXqiAxKN #WhatTheFutr #FutrConnect #Leadership #TechLeadership #EnterpriseTech #AIInfrastructure #Founders #StartupEcosystem #Innovation #PodcastAnnouncement
-
Suresh Vasudevan shared thisI have had the pleasure of partnering with Tarun Thakur on both his startups. Even at the very outset when Tarun Thakur, Maohua Lu Ph.D, and Robert Whitcher conceived the brilliant idea of building an identity platform rooted in the principle of least privilege, it was clear that they were on to something big and special. Veza, in combination with ServiceNow's AI platform, will be unmatched in allowing organizations to understand and control who and what has access to their critical data, applications, systems, and AI environments. Congrats to the Veza founders to all Veza teams and ServiceNow teams on this milestone!Suresh Vasudevan shared thisThis morning at 6am PT, we have publicly announced that Veza has entered into a definitive agreement to join ServiceNow. This is a monumental event in our journey, and one that reflects the incredible dedication and belief each of #Vezanite has put into building Veza. When Maohua Lu Ph.D, Robert Whitcher, and I founded Veza, we did so with the strong conviction that identity would evolve to become a critical pillar of security and that the legacy players had only - and could only - scratch the surface of what needed to be done. Over the past 5 years, we’ve collectively disrupted a traditionally sleepy part of the cybersecurity — identity security — by bringing a fundamentally transformative approach to solving an increasingly complex and challenging problem: how to achieve least privilege. Together, we’ve built more than a product; we’ve initiated a category-defining transformation in how enterprises think about identity and identity security. Most importantly, our vision resonates with customers who rely on Veza to answer the core question of identity security: who has access to what data, and who should. This clarity is mission-critical, and we’re proud to support customers such as Blackstone, Capital One, Intuit, Sallie Mae and Acrisure who trust us to bring transparency and control to their most sensitive access decisions. It's an honor to be partnering with sharpest data, identity, and AI enthusiasts: the team at ServiceNow - Bill McDermott Gina Mastantuono Amit Zavery John Aisien Philip Kirk Chris Bedi Ben de Bont Bryan Casper Pablo Stern Lou Fiorello and teams. Thank you Amit Zavery for setting a big bold business vision for this partnership and bringing it to life. Your commitment to both elite level execution and culture—ensuring Veza and ServiceNow’s hungry, humble, and customer-obsessed values remain aligned—sets the foundation for an exciting next chapter. Thank you Chris Bedi and Philip Kirk for believing in Veza’s unique architecture and our mission to transform identity when we met for the first time in 2023! Thank you John Aisien for recognizing the power of Veza’s Access Graph - I’m especially grateful for your unwavering support. This all started on July 4th @ 6am PST -- your pure startup mindset is the kernel behind ServiceNow + Veza. To our Vezanites: this moment is a tribute to your hard work, your engineering excellence, and your belief in a world where identity is secure by default. This acquisition validates everything we’ve achieved so far. Because of your hard work, persistence, and belief in our mission, we are now ready for the next step: partnering with ServiceNow teams to unleash the power of Veza’s platform across enterprise customers. To our Board of Directors and Investors: “thank you” doesn’t feel like enough. I’m incredibly grateful for your trust, support, advice, feedback, and dedication. You’ve challenged us, pushed us, and helped us grow into the company that reached this milestone. Onward!
-
Suresh Vasudevan shared thisThank you very much Frederic Lardinois - I thoroughly enjoyed the discussion about how Clockwork Systems, Inc. approaches the topic of improving AI infrastructure efficiency by providing deep visibility into GPU fleets, enabling resilience to network failures and link flaps and accelerating performance in the face of network bottlenecks.Suresh Vasudevan shared thisAI training jobs are growing larger, more distributed, and more failure-sensitive. Clockwork.io CEO Suresh Vasudevan recently joined Frederic Lardinois on The New Stack podcast to discuss how Clockwork's ultra-precise timing data — originally used to sync clocks across distributed systems — started highlighting issues in large-scale systems that weren’t visible through traditional monitoring. That insight ultimately shaped Clockwork’s platform for real-time visibility, automated remediation, and more reliable AI training at scale. Take a listen to learn: • Why modern LLM training workloads push networking to its limits • How packet-timing data can expose hidden congestion and failures • What dynamic traffic control (like FleetIQ) can do to keep training jobs running smoothly • How integration with NCCL, TCP, and RDMA enables smarter GPU communication • Why real-time network telemetry and automated remediation are becoming foundational for next-gen AI infrastructure 🎧 Watch the episode on The New Stack: https://lnkd.in/gS2gn7mW #AIInfrastructure #TheNewStack #Networking #GPUs #LLMTraining #FleetIQ #Clockwork
-
Suresh Vasudevan liked thisSuresh Vasudevan liked this‘Model fatigue’ sets in as AI labs race to roll out new versions at frenetic pace Anthropic, OpenAI, Meta and Google all released model updates last week as the pace of new releases continues to accelerate. Even the Mohamed bin Zayed University of Artificial Intelligence in Abu Dhabi released its own K2 Horizon family of AI models to the open-source community, underscoring the global nature of AI research and investment. Not all model updates are created equal. Unlike OpenAI’s GPT-6 Astra, the various rollouts week from Anthropic, Meta and Google represented “point releases," instead of brand-new AI models. I've been hearing from a lot of folks about the challenges of keeping up with all of the various AI models and updates, and the term "model fatigue" kept on coming up. It's not that technologists dislike the updates, because depending on their use case, the upgrades could be beneficial. But, it's challenging keeping pace, and evaluating each new model update drains precious computing resources. Read more below from CNBC: https://lnkd.in/gfK654Ka‘Model fatigue’ sets in as AI labs race to roll out new versions at frenetic pace‘Model fatigue’ sets in as AI labs race to roll out new versions at frenetic pace
-
Suresh Vasudevan reacted on thisSuresh Vasudevan reacted on this𝐋𝐚𝐫𝐠𝐞 𝐝𝐢𝐬𝐭𝐫𝐢𝐛𝐮𝐭𝐞𝐝 𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐣𝐨𝐛𝐬 𝐟𝐚𝐢𝐥 𝐢𝐧 𝐰𝐚𝐲𝐬 𝐭𝐡𝐚𝐭 𝐚𝐫𝐞 𝐡𝐚𝐫𝐝 𝐭𝐨 𝐫𝐞𝐩𝐫𝐨𝐝𝐮𝐜𝐞 𝐚𝐧𝐝 𝐞𝐱𝐩𝐞𝐧𝐬𝐢𝐯𝐞 𝐭𝐨 𝐢𝐠𝐧𝐨𝐫𝐞. A single GPU, network, or process failure can stall every healthy resource in a synchronized job — and at scale, that failure isn't rare. That's the problem five interns spent their summer inside of this year — not as a thought experiment, but as the actual engineering work. 𝐅𝐚𝐮𝐥𝐭 𝐭𝐨𝐥𝐞𝐫𝐚𝐧𝐜𝐞 𝐢𝐬𝐧'𝐭 𝐚 𝐜𝐡𝐞𝐜𝐤𝐛𝐨𝐱 𝐨𝐫 𝐚 𝐬𝐢𝐧𝐠𝐥𝐞 𝐫𝐞𝐜���𝐯𝐞𝐫𝐲 𝐦𝐞𝐜𝐡𝐚𝐧𝐢𝐬𝐦. The right approach depends on what's actually failing, how often the state gets captured, how far the damage spreads when it does, and whether recovery changes training semantics along the way. Our interns worked those tradeoffs from different points in the infrastructure stack. They learned to investigate behavior across distributed environments, question whether their assumptions held up on real clusters, and connect component-level signals to job-level impact. They also saw how a design that looks straightforward on paper turns into a much harder problem once it has to run reliably on real hardware, at real scale. Che Hung Liao, Gabriela Miranda, Amin Mamandipoor, Erika Li, and Shashank Mysore Radheshyam each brought a different angle to that work — you can hear what they took away from it in the graphic below. Their projects varied, but the tradeoffs they wrestled with are ones every infrastructure team eventually runs into. What's the one piece of advice you'd give them as they head back to campus? #AIInfrastructure #DistributedSystems #FaultTolerance #Internship
-
Suresh Vasudevan liked thisSuresh Vasudevan liked thisOur paper, "Predicting the Future of Supercomputing," just got published in the Predictions column of the July issue of IEEE Computer. You can find it here: https://lnkd.in/gTsXUBQu. Thanks to co-authors Scott Atchley, Rosa M. Badia, Bronis de Supinski, Joshua Fryman, Dieter Kranzlmueller, Srilatha (Bobbie) M., Pekka Manninen, Satoshi Matsuoka, Galen Shipman, Eric Van Hensbergen, and Robert Wisniewski. Also thanks to Cullen Bash, Paolo Faraboschi, and Samantika S. Sury for valuable feedback.
-
Suresh Vasudevan liked thisSuresh Vasudevan liked thisClaude will NOT stop second guessing me!! 😂 Earlier this week we announced that Salesloft is sunsetting Drift Email and that allGood is the #1 recommended replacement. That said... Salesloft hasn't made their own announcement yet. Every. Single. Time. I sit down with Claude to write something about this partnership, it does the exact same thing: searches the web, finds nothing directly from Salesloft, and fact-checks me 😭 I swear, Claude, I literally just got off the phone with our Salesloft contact 5 minutes ago! Anyways, the cherry on top is that Claude always frames it as "flagging it, not second-guessing you." So... thanks for that, at least, Claude 🙃
Experience & Education
Other similar profiles
-
Jack Mulloy
Jack Mulloy
Jack is a seasoned enterprise software leader with over 20 years of experience in procurement and supply chain consulting and solutions. As the Managing Director at Nitor Partners, he oversees the strategy, operations, and growth of an SAP Gold Partner that delivers value to global clients across various industries. <br><br>Jack is also an active investor and board member at two SaaS companies, RELISH and CodeFluent, that leverage AI and SAP's BTP framework to create innovative applications that augment and extend the capabilities of enterprise software platforms. He is passionate about helping organizations optimize their processes, performance, and profitability through digital transformation and sustained growth.
Explore more posts
-
Salil Deshpande
As AI workloads eat up global computing supply, DRAM prices are surging like never before—creating major challenges for teams trying to plan out their data center spend. One of my portfolio companies, Mext, is addressing the largest cost component in the datacenter: server memory (DRAM). From being on the board of Redis for the last twelve years, I’ve learned a lot about the various aspects of this problem and approaches to solving them. Mext found a breakthrough, using new AI techniques, for dramatically reducing the amount of server-DRAM required to run applications, all while maintaining performance. It intelligently manages the server’s memory, keeping hot pages (i.e., those in use by applications) in fast DRAM and offloading cold pages (i.e., those less used) to a much less expensive memory tier (e.g., NVMe Flash). Key to the approach is ensuring cold memory pages that are about to be accessed by an application are back in DRAM before the application needs them or notices that they were gone. This is done without modifications to the application or the OS – so it can run in the cloud or on-premise. Mext thus allows applications to either run using less DRAM or keep their DRAM footprint but do more with it. They're hosting a webinar along with Fred Weber (former CTO of AMD) on Jan 22nd at 10am pacific time. You can register here: https://lnkd.in/gXe9GCeE
3 Comments -
Brittany Walker
Voice is a super interesting modality right now - maybe the first modality we're seeing move to open source models across a number of scale ups / enterprises. Reliability concerns, high costs, and open source model performance are pushing engineers to do their own fine tuning vs. relying on third-party vendors of proprietary models. Many of these orgs have already been collecting their own first-party data and now with third-party vendors like Extrian, David AI, etc they can train really high quality models. RL has been insanely hyped, but it's been unclear how long it will take scale ups and enterprises to actually lean in. Voice AI might be hitting that inflection point faster than expected.
7 Comments -
Derek Kerton
We chose a good subject for our march Autotech Council meeting, and after this year's CES, it seems even more obvious. Our meeting on the AI Defined Vehicle will show how it's a natural next step after the Software Defined Vehicle phase that we've been in Tesla launched the Model S in 2012. Now, most carmakers have SDVs. It's time to layer on the AI in user-facing roles, and in invisible functions managing the vehicle.
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top contentOthers named Suresh Vasudevan
109 others named Suresh Vasudevan are on LinkedIn
See others named Suresh Vasudevan