The seventh generation PCIe specification draft has been released, with the full spec scheduled for publication sometime in 2025.
The Peripheral Component Interconnect Express (PCIe) bus is used by PCs and servers to connect CPUs, networking, graphics cards, and storage devices. It is used to transmit data between them. As with preceding PCIe generations, this one doubles the PCIe bus transfer speed, from gen 6’s 64 Gbps link bandwidth to 128 Gbps. That’s 512 GBps bandwidth across 16 lanes.
A blog by PCI-SIG president Al Yanes says: “Progress continues on the PCI Express (PCIe) 7.0 specification, which PCI-SIG announced at US DevCon in June 2022. Thanks to the hard work of our technical work groups, we are pleased to announce that version 0.5 is now available for member review. This is the official first draft of the specification, incorporating all the feedback we received from members after the release of Version 0.3 in June 2023.”
Yanes says gen 7 PCIe is aimed at the high bandwidth needs of 800G Ethernet, artificial intelligence/machine learning, hyperscale datacenters, HPC, quantum computing, and the public cloud. The intended feature list includes:
Delivering 128 GT/s raw bit rate and up to 512 GB/s bi-directionally via x16 configuration
Utilizing PAM4 (Pulse Amplitude Modulation with 4 levels) signaling
Focusing on the channel parameters and reach
Continuing to deliver low-latency and high-reliability targets
Improving power efficiency
Maintaining backwards compatibility with all previous generations of PCIe technology
PCIe gen 6 enables CXL v3.0 memory pooling, sharing, and multi-level switching support. PCIe 7 will increase the bandwidth available for such external memory sharing and pooling. A program to verify compliance with the PCIe 7 standard is expected to be available in 2027 with products released after that.
PCI-SIG members can access the v0.5 PCIe 7.0 spec on the members workspace on Causeway. Non-members can join here.
Media search and management company Axle AI has joined forces with object storage and data lake vendor Cloudian to offer a hybrid system that “ensures data sovereignty, reduces costs and simplifies media workflows.”
Axle AI’s customer roster includes Turner, the BBC, Paramount, Facebook, Patagonia, Madison Square Garden, NBC, Amazon, Canon, and Radio City.
With the combined solution, users can store their data either on-prem or in the cloud, while maintaining single-point management and unified media search. Using AI-powered tools, the solution catalogs, manages, and searches media across on-prem and cloud-based storage for “simple and immediate access,” the companies said in a joint statement.
“As users seek to optimize data sovereignty, security, and storage costs, they are increasingly looking to balance their use of on-prem and cloud-based compute and storage platforms,” they added.
Axle AI face identification
Cloudian’s says its hybrid data lake offers limitless on-prem media capacity that can be situated anywhere, plus cloud integration that facilitates policy-driven data management across both cloud and on-prem platforms.
Cloudian’s full S3 API compatibility is meant to simplify the process of migrating workflows between cloud and on-prem environments. This allows users to leverage existing S3 API-based media, tools, and applications while refining their infrastructure strategy in a bid to reduce costs.
Sam Bogoch
Axle AI’s media asset management software generates searchable metadata based on advanced AI techniques including speech transcription, face recognition, and object/logo identification. Integrated review and approval workflows in the Axle AI browser frontend, as well as a panel for Adobe’s Premiere Pro application, enable “seamless” post-production workflows.
With “military-grade” security, Cloudian ensures the protection of media assets from unauthorized access and ransomware threats. And an intuitive interface across both Axle AI and Cloudian systems “significantly reduces the learning curve,” said the companies, the aim being to let all levels of staff to quickly adapt and enhance productivity.
“The power of AI is transforming the way media professionals interact with their content,” said Sam Bogoch, CEO of Axle AI. “By integrating our AI-driven metadata tagging and search capabilities with Cloudian’s robust data lake, we are simplifying access and management of media files to a degree that was previously unimaginable.”
Michael Tso
Michael Tso, CEO of Cloudian, added: “Our hybrid data lake allows media assets to be located anywhere, while remaining fully interoperable with existing S3-compatible tools. By collaborating with Axle AI, we now make those assets instantly searchable, both on-prem and in the cloud, with the power of AI.”
The complete combined system is now available through Axle AI, Cloudian, and authorized resellers.
Last month, Cloudian announced support for AWS Mountpoint for Amazon S3, an open source file client. The enhancement allows data stored on Cloudian data lakes to be mounted as local file systems with high-throughput access, enabling Cloudian users to leverage object storage system Cloudian HyperStore through a familiar file interface.
Common use cases for AWS Mountpoint include large-scale machine learning, autonomous vehicle simulation, genomic analysis, data ingest (ETL), and image rendering.
N2WS, a provider of enterprise cloud data protection solutions, has hired Alon Maimoni as chief revenue officer to spearhead the company’s growth aspirations.
As a cloud-native backup, disaster recovery, and archiving tool, N2WS’s pitch is that it protects data and cuts storage costs with customizable backup policies. It targets Fortune 500 companies and managed service providers (MSPs) operating large-scale production environments on Amazon Web Services and Microsoft Azure.
Alon Maimoni
In his new role, Maimoni will be charged with the establishment of a revenue organization within N2WS, encompassing marketing and sales, and the creation of a new customer success division.
In his last role as vice president of marketing for Komodor, a Kubernetes reliability and debugging platform, we’re told Maimoni helped the company grow by more than 400 percent in under two years. Before that Maimoni worked at NetApp from 2018 to 2022 as a director for marketing and sales at its Cloud Innovation Center in Tel Aviv.
The company says it currently has more than 1,000 customers across multiple cloud platforms.
“Alon’s track record in scaling tech organizations, along with his strategic vision and hands-on approach, align perfectly with N2WS’s mission to empower businesses with top-tier data protection,” said CEO Ohad Kritz in a statement.
“I’m joining a company with a stellar reputation for innovation and customer-centric solutions,” said Maimoni in a prepared remark. “I plan to build on N2WS’s strong foundation, leveraging our cross/multi-cloud capabilities and industry-leading backup recovery tools to propel the company into a new era of growth.”
Israel-based N2WS was founded in 2012 and, according to Crunchbase, received an undisclosed amount from a venture funding round in 2017, with Insight Partners and Microsoft ScaleUp Tel Aviv contributing. Pitchbook says it received a $573,000 paycheck protection program loan in 2020.
Veeam bought N2WS in 2018 for $42.5 million in cash, but it later decided to sell it back to the original owners as some US public sector N2WS customers did not want their backup supplier to be owned by the two Russians who then owned Veeam.
That sale was completed in October 2019. Veeam is now owned by US investment firm Insight Partners, and runs its own US public sector/federal business unit.
Storage manager and protector Quantum has added a Z200 all-flash product to its ActiveScale object storage line with a focus on AI and data lake workloads.
The ActiveScale products have until now been disk-based and positioned as object storage suitable for high-performance computing secondary storage workloads. They accept S3 protocol requests and can frontend a Scalar tape library which provides virtually unlimited storage for low access rate object data. There is a single namespace across disk and tape, and the ActiveScale software provides erasure coding and geo-spreading across three sites to increase data resiliency.
Bruno Hald, Secondary Storage VP for Quantum, said in a statement: “With the introduction of the ActiveScale Z200, we are meeting the need for a fully integrated enterprise object storage solution that merges high-performance scale-out flash and scale-out tape librariesto build data lakes and storage clouds of outstanding scalability, simplicity, performance, and cost efficiency, without compromising data availability and long-term durability.”
ActiveScale product comparison table. The Z200 uses TLC flash
Quantum claims that, with its high throughput and flexibility to densely store metadata for billions of objects in flash, plus its integrated ability to durably store massive amounts of data in nearline tape storage, the ActiveScale Z200 reduces the overall cost of data lakes, storage clouds, and long-term archives, and enables easy data access to massive data sets for additional analysis, model recalibration, and re-monetization.
It is, the company claims, “the industry’s only object storage platform architected for both active and cold data.” This makes it suitable for “building massive data lakes, storage clouds, and long-term archives,” with a smaller overall footprint and cost than competing systems.
Performance
A Quantum spokesperson told Blocks and Files: “We have optimized the Z200 for price/performance to meet a variety of use cases and lower cost entry points rather than absolute performance, i.e. each Z200 is 32 core, 256 GB, 25GbE, 15 TB SSD configs as compared to 48-64 core, 100GbE systems being sold at a premium today. On a per core basis, we achieve double the performance of other vendors and scale linearly.
“It’s important to recognize that our stated performance reflect real-world achievable performance. We have seen recently some real gamesmanship in object storage performance benchmarks which include three-way mirrored config for GETs, 4+2 EC config for PUTs and only four drives configured per server, all in the same benchmark. Grossly expensive and unachievable in a real-world config.
“Our roadmap includes beefier configs, higher density drives, and QLC drives as well, so stay tuned. Performance has scaled well on these beefier configs.”
Competition
NetApp has an all-flash StorageGRID object storage product. Scality added all-flash object storage support in 2020 and object storage startup rival Cloudian also has all-flash support. IDrive added all-flash support to its e2 object storage last year.
Ravit Sadeh, Senior Director of Product Management at Israel-based CTERA
Comment War isn’t a snapshot in time. It’s a relentless storm that tests every part of a person, every part of a company. Weeks, months, sometimes years of uncertainty make “war-life balance” a whole new battleground. This article explores how Israeli tech companies like CTERA keep functioning through it all, focusing on the deeply personal impact on those facing this struggle every day. We’ll examine the mix of professional duty and the raw reality of conflict, how traditional roles shift in times of war, and how women and men adapt to new challenges. This isn’t about politics or picking sides – it’s about understanding the human toll and the incredible strength that keeps things moving.
The reality of a ‘war-life balance’
When war crashes into your world, everything you know about normal life gets thrown out the window. It’s not just about work anymore, it’s about staying safe and keeping your loved ones safe too. You switch between being an employee, a parent, and a worried citizen, all on high alert.
Ravit Sadeh
The morning starts the same, but it’s not. The news – it punches you in the gut. So many lives lost, families ripped apart, and the haunting worry about those still missing. Coffee – a bitter comfort, a small luxury in a world turned upside down. You’re lucky you have it, lucky you have a roof over your head.
Rousing the kids feels different. It’s your job to get them to their virtual classroom, to keep them safe, yet every school day feels like a risk now. Three people crammed in the house, each with their own battles. Kids focused on school, you focused on work, but it’s all a tense juggling act.
Then the alarm. That is a sound you’ll never get used to. Rushing to the shelter with fear tight in your throat. The neighbor, his face etched with worry – his daughter, a soldier, no word for 30 endless days. Back home, and you can’t focus. Not right away. You stare at the fridge, the same one you’ve opened a hundred times. More food? For what? What kind of days lie ahead?
Work grinds on, the kids finish their tasks, but you keep them inside. No playing out today. You try to explain what they saw on the news, those terrible images they shouldn’t have seen. A knock at the door. It’s the neighbors. One of those fallen soldiers is from this very street. We go to the family with flags in hand. A silent procession of grief.
Back home, work again. Another day ends, exhausting and filled with a dread you can’t shake. And amid it all, there’s also a pang of guilt: I’m privileged. This is my day, and so many have it far worse.
Navigating work under wartime conditions
Things are tough in Israel right now. We’re in the middle of a conflict, and it’s affecting everyone. Individuals from all teams. These days they are not only employees but also participants in the conflict, men and women serving together on the front lines. The absence of these individuals is palpably felt across the organization, yet it fuels a collective resolve to support them and their families and ensure the company’s strength and continuous delivery. Now, more than ever, the real job extends beyond daily tasks to keeping the company stable and ensuring unwavering service to customers and partners. “Business as usual” transcends being merely a phrase – it encapsulates our core mission during these trying times.
Management plays a critical role, motivating teams to persevere in their work amid the chaos. This involves a meticulous balance of understanding each team member’s location and safety, offering flexibility in work hours amid the constraints, yet still maintaining high expectations. The essence of our work has shifted – we’re not just focusing on tasks but ensuring there’s a stable environment to return to, contributing to the stability of the Israeli economy. The concept of solidarity has never been more relevant or deeply felt.
As a community, we’ve adapted to the circumstances with resilience and innovation. Remote work has become the norm, with office days being carefully considered. We recognize the paradox of continuing work and delivery when hearts and minds are elsewhere. Initiatives like fundraising for families affected by the war, organizing social activities to support soldiers, and providing personal equipment for those on the battlefield exemplify our commitment to going beyond our professional duties. These actions not only boost morale but also reinforce the importance of each employee’s contribution, motivating them to be a part of a global effort that might seem disconnected from the immediate crisis.
Our commitment extends to our customers and partners as well. We’ve adapted by pre-recording training and demonstrations to avoid interruptions from alarms and reorganizing support teams across Israel and globally to ensure 24/7 availability. Our engineering team has worked tirelessly, continuing to release versions and major features, demonstrating our dedication to innovation and service despite the circumstances.
In ensuring that employees have the support they need, we embody the principle of “business as usual” in the truest sense. It’s a stance we adopt with pride, not as a cliché but as a testament to our resilience and solidarity. Through these efforts, we not only sustain our operations but also contribute to a sense of normalcy and stability during tumultuous times, proving that our work, our contributions, indeed matter.
140 days of war – are we business as usual?
It’s been 140 days since the war began. The daily toll is heavy, with countless lives lost, a grim reminder of the war’s brutality. Somehow, we’ve found a way to get used to the stream of bad news, bracing ourselves for a conflict with no end in sight. Many of us have returned to work, and schools have reopened their doors to students, trying to find a semblance of normality amid chaos. Yet the constant updates of tragedy weigh heavily on us, a stark reality that routine cannot erase. Despite this, there’s a palpable sense of unity and strength among us, a shared resolve that we’ll emerge from this stronger. This sentiment of solidarity keeps us going. “We’ll get through this; we’ll be better,” I reassure my kids and myself.
The filing reveals details of the cybersecurity and data protection business’s growth since it was founded in 2014. With the help of six funding rounds, Rubrik has raised more than $552 million to this end. Rubrik has more than 6,100 customers, up 22 percent year-on-year, and $784 million of subscription annual recurring revenue (ARR), which has risen 47 percent year-on-year. Total revenue in the year ending January 31, 2024, was $627.9 million, an increase of just 4.7 percent year-on-year. There was a net loss of $354.2 million compared to a loss of $277.7 million the year before. Rubrik is still spending heavily to grow its customer base and expand its business.
Bipul Sinha
In an introductory letter, Rubrik CEO and co-founder Bipul Sinha writes: “When market transitions occur, opportunity arises to transform and massively expand existing markets. These disruptions allow new products to gain market relevance quickly and set the agenda of the new market order. This is precisely why I left my venture partnership to start Rubrik. The legacy backup and recovery market was ripe for such a transformation into data security based on cloud and cybersecurity trends.”
In its fiscal 2024, Rubrik’s operating cash flow was -$4.5 million, compared to $19.3 million in 2023. The free cash flow numbers were -$15 million in fiscal 2023 and -$24.5 million in 2024.
Rubrik believes its total addressable market (TAM) will be approximately $36.3 billion by the end of calendar 2024 and approximately $52.9 billion by the end of 2027, based on Gartner research, representing an average 13 percent compound annual growth rate (CAGR).
The company’s revenue growth rate has shrank due to a switch from perpetual license sales of its Cloud Data Management product to Rubrik Security Cloud (RSC) subscription-based revenues. This transition started around fiscal 2022 and makes Rubrik’s recent revenue history chart resemble that of a low-growth business:
Rubrik notes: “We expect new and existing customers to increasingly purchase subscriptions of RSC, which is recognized ratably over the term of the subscription. Our revenue will fluctuate based on the timing for transitioning our existing customers to RSC and when qualified customers choose to exercise or forfeit their customer options … The combination of both of these factors will limit and cause fluctuations in our subscription revenue growth through fiscal 2027, depending in part on the timing of our existing customers’ transition to RSC.”
As classic revenue growth has slowed, Rubrik has concentrated on measuring its ARR growth rate as that, it says, is a better indicator of its actual expansion:
As of January 31, 80 percent of total ARR comes from customers with $100,000 or more in subscriptions
The S-1 doc says of Rubrik profitability: “While we have experienced rapid revenue growth in recent periods, we are not certain whether or when we will obtain a high enough volume of sales to achieve or maintain profitability in the future.”
This market has several large and strongly competing players such as Veeam, Cohesity (which is buying most of Veritas’s data protection business), Commvault, Dell, Druva, IBM, and myriad smaller players. Rubrik’s S-1 points out: “Many of our current and potential competitors have longer operating histories and have substantially greater financial, technical, sales, marketing, and other resources than we do, as well as larger installed customer bases, greater name recognition, lower labor and development costs, and broader product solutions, including servers … We expect that competition will increase as a result of future software industry consolidation.”
The S-1 form does not specify the price range for Rubrik’s Class A common stock shares, nor the number that will be offered to the public. Earlier reports estimate the cash raised will be between $500 million to $700 million.
The bankers identified in the filing are Goldman Sachs, Barclays, Citigroup, Wells Fargo Securities, Guggenheim Securities, Mizuho Securities, Truist Securities, BMO Capital Markets, Deutsche Bank Securities, KeyBanc Capital Markets, Cantor CIBC Capital Markets, Capital One Securities, Wedbush Securities, SMBC, and Nikko.
Global digital consulting firm JK Tech is seeking to improve enterprise AI data insights with the launch of its Google-driven JIVA platform.
The generative AI (Gen AI) technology promises improved data accessibility, while “notably enhancing” precision and minimizing hallucinations – a common challenge when using traditional large language models. JK Tech has implemented Gen AI in the automotive insurance sector where, for example, Gen AI chatbots can help handle initial claim registration, which reduces claim handling time by up to 50 percent. This helps both the insurer, which can handle more claims in less time, and the policyholder, who receives their payout faster.
Using “advanced” Gen AI technologies, JIVA uses Google’s Knowledge Graph and other tools to construct a dynamic interconnected graph that fosters traceability and contextual comprehension.
Dipankar Ganguly.
The in-context training and reasoning engine within JIVA facilitates “deeper understanding,” leading to more accurate conclusions. The Nodal Level Security is a “unique feature” of JIVA, claims the Noida, Uttar Pradesh, India-headquartered provider. It integrates enterprise access management, ensuring secure data access, and mitigating the risk of data misinterpretation.
“As organizations continue to navigate the complexities of digitization, data has become a critical asset for strategic decision-making and driving innovation,” explained Dipankar Ganguly, chief technology officer of JK Tech. “We believe in unleashing the full potential of Gen AI in the digital transformation journey of businesses. Through in-depth discovery sessions, we work with clients to identify key areas where Gen AI can deliver tangible value.”
The offer includes customization and integration tailored to specific needs, end-to-end training, and “seamless” implementation.
“Through our ongoing support and optimization services, we remain committed to keeping our customers’ Gen AI initiatives innovative, and leading the way in the evolving business environment,” declared Ganguly.
JK Tech’s market sweet spots are retail and consumer products, insurance, financial services, and healthcare industries – all verticals that are big on trying to make AI work in a meaningful way.
Bloomberg reported over Easter that Rubrik is prepping an IPO and could file papers with the SEC this week.
This follows a Reuters report in February that the Bipul Sinha-led business was planning an April 2024 IPO once a fraud investigation completed. Bloomberg previously suggested that Rubrik could IPO by the end of last year.
Bipul Sinha.
Rubrik started up in 2014 as a backup and restore-based data protector, and has evolved to add cyber security and resilience. This has become its main focus. The founders were ex-venture capitalist and CEO Bipul Sinha, CTO Arvind “Nytro” Nithrakashyap, VP engineering Arvind Jain and Soham Mazumdar. It has raised in excess of $550 million – possibly as much as $1 billion – with a valuation thought to be in the $4 billion area when Microsoft invested in 2021. The firm reckons it has more than 5,000 customers and in excess of $600 million in annual recurring revenue. Previous reports have identified Goldman Sachs, Barclays and Citigroup as its IPO bankers.
The Bloombergers guess that Rubrik could raise $500 million to $700 million from this IPO. A Rubrik IPO would represent a pinnacle of achievement for Bipul Sinha, who was born into poverty in India.
The Information reported two other cyber security vendors – Snyk and Cato Networks – could also run IPOs this year.
Arch rival Cohesity, which filed for an IPO in December 2021, is acquiring the bulk of the business from legacy data protector Veritas and plans for an IPO to follow that in due course. This acquisition could make it a larger business – in terms of revenue and customer count – than Rubrik.
Social media platform Reddit and semiconductor supplier Astera have both successfully launched IPOs recently, providing a generally positive investor environment.
Enterprise data management vendor Starburst, which bills itself as “the open lakehouse,” has brought in a number of seasoned data veterans to help it scale up.
Steven Chung has been named as president, Tobias Ternstrom is now chief product officer, and Adam Ferrari has been brought in as senior vice president of engineering.
Starburst’s exec trio of hires. Left to right: Adam Ferrari, Steven Chung, and Tobias Ternstrom.
“Steven, Tobias, and Adam have joined our team, inspired by our vision to help organizations unlock greater business value from their data through the Starburst Lakehouse platform,” declared Justin Borgman, co-founder and CEO of Starburst.
“The appointment of these industry leaders will help scale our business globally,” added Mike Volpi, Starburst board member and founding investor from Index Ventures. “Steven, Tobias, and Adam bring proven experience to Starburst as we power new opportunities with AI, ML, and analytics in cloud, on-prem, and in hybrid environments.”
Prior to Starburst, Chung served as president at Delphix, the enterprise DataOps biz that was recently acquired by Perforce Software. In addition, he has held senior executive roles at PagerDuty, Demandware, Symantec, Microstrategy, and PwC Consulting.
Ternstrom joins the party from Nutanix, where he served as vice president of platform services, bringing database-as-a-service to hybrid multi-cloud environments. Ternstrom’s other previous roles were at Amazon Web Services, Google Cloud Platform, and Microsoft. At AWS, Ternstrom led product management for Amazon Aurora and Amazon Relational Database Service (RDS).
Ferrari brings 20 years of engineering leadership, where he most recently served as SVP of engineering at Salsify – a leading product information management (PIM) provider. He has also led product and engineering teams at Oracle and Endeca (prior to its acquisition by Oracle in 2011).
Starburst says it offers a “full-featured,” open data lakehouse platform, built on open source Trino. Its end-to-end analytics include the capabilities needed to discover, organize, consume, and share data, with “industry-leading price-performance” for both cloud and on-premise workloads. The technology is used by end customers like Comcast, Grubhub, and Priceline.
Data protection outfit Rubrik is feeding data to CrowdStrike in a bid to improve its attack detection and response.
CrowdStrike’s Falcon XDR product ingests data from various security tool repositories across a customer’s IT estate and uses AI to help it hunt threats and respond to attacks. It has been involved in investigations of several high-profile cyberattack investigations, such as the 2014 Sony Pictures breach and the cyberattacks and email leaks at the Democratic National Committee.
The more data it has about IT estate activities, particularly those involving critical data, the better. Rubrik’s integration may well remove a blindspot from CrowdStrike’s threat hunting activities.
Anneka Gupta
Anneka Gupta, Rubrik chief product officer, said: ”The gap between threat detection, data discovery, and classification creates significant visibility challenges for security teams defending critical data. With CrowdStrike, we are helping our customers up the ante against cyber adversaries, allowing security teams to identify and defend against attacks swiftly – and ultimately boost cyber resilience.”
CrowdStrike’s chief business officer, Daniel Bernard, added: “Our partnership with Rubrik strengthens CrowdStrike’s data gravity, unifying threat detection with data discovery, classification and backup. Through this partnership, we’re delivering the visibility and context security teams need to prioritize and accelerate response actions required to stop breaches of sensitive information – all from a single platform.”
The company says it generates security telemetry and enriches it with adversary intelligence and human expertise via its Falcon platform. The Rubrik data feed to Falcon’s LogScale functionality provides data target indicators, particularly for critical data, aiding attack behavior characterization.
With the integration, Rubrik says customers can optimize security and IT operations, reduce alert fatigue, and better focus efforts on stopping data breaches.
Arch Rubrik rival Cohesity set up an integration with CrowdStrike and its Falcon LogScale dashboard in November last year as part of Cohesity’s Data Security Alliance activities. The Falcon LogScale and Cohesity DataHawk combo helps provide faster correlation, investigation, and response to incidents in one location. Cohesity said the CrowdStrike integration provides closed-loop detection and response for attacks directly within the CrowdStrike Falcon platform.
Rubrik Security Cloud for Falcon Logscale is available from today in the CrowdStrike Marketplace. Learn more at an upcoming webinar on Wednesday, April 24, at 2pm ET.
It says DBRX lets customers build, train, and serve their own custom LLMs more cheaply, without having to rely on a small set of closed models. Closed LLMs, like ChatGPT and GPT-3.5, are based on private model weights and source code whereas open source ones such as LlaMa, Dolly, and DBRX have publicly available source code and model weights.
Developers can inspect their model architecture and training data and customize the source code, according to Databricks.
Databricks co-founder and CEO Ali Ghodsi said: “We’re excited about DBRX for three key reasons: first, it beats open source models on state-of-the-art industry benchmarks. Second, it beats GPT-3.5 on most benchmarks, which should accelerate the trend we’re seeing across our customer base as organisations replace proprietary models with open source models. Finally, DBRX uses a mixture-of-experts architecture, making the model extremely fast in terms of tokens per second, as well as being cost effective to serve.”
A Databricks chart shows DBRX delivering results faster than other open source LLMs like Llama 2 70B and Mixtral-8x7B on industry benchmarks such as language understanding, programming, maths, and logic.
The MMLU (Massive Multitask Language Understanding) is a benchmark designed to measure knowledge acquired during pre-training by evaluating models exclusively in zero-shot (no labeled data available for new classes) and few-shot (limited number of labeled examples for each new class) settings. The benchmark covers 57 subjects across STEM, humanities, social sciences, and more. Programming (Human Eval) is a programming challenge dataset. Math GSM8K is a dataset of 8.5K high quality linguistically diverse grade school math word problems created by human problem writers.
The performance differences are minor on the MMLU test, significant on the Programming (Human Eval) run, and less significant on the Match (GSM 8K) benchmark/
DBRX also goes faster than GPT-3.5, according to claims from Databricks:
The pattern of performance differences is the same as with the open source chart above; minor on the MMLU test, significant on the Programming (Human Eval) run, and less significant on the Match (GSM 8K) benchmark.
Databricks says it has optimized DBRX for efficiency and with a mixture-of-experts (MoE) architecture, built on the MegaBlocks open source project. The MoE architecture is said to enable trillion parameter-class training with billion parameter-class compute resources. A Hugging Face blog explains the concept.
DBRX is claimed to be up to twice as compute-efficient as other available LLMs. Databricks says it can be used, when paired with Databricks’ Mosaic AI tooling, to build and deploy safe, accurate, and governed production-quality GenAI applications without customers giving up control of their data and intellectual property.
Mike O’Rourke, head of AI and Data Services at Nasdaq, said in a canned remark issued by Databricks: “Databricks is a key partner to Nasdaq on some of our most important data systems … and we are excited about the release of DBRX. The combination of strong model performance and favorable serving economics is the kind of innovation we are looking for as we grow our use of Generative AI at Nasdaq.”
DBRX is freely available on GitHub and Hugging Face for research and commercial use. Databricks is running a DBRX webinar on April 25, 1500 UTC. Check out a DBRZ article at its Mosaic AI Research blog.
Comment
Relying on a particular GenAI model’s speed at this early stage in development, on both hardware and software fronts, doesn’t appear to be a sustainable strategy unless you can afford to keep developing it at the same rate as the extremely well-heeled closed model developers.
Startup Observe has raised a $115 million B-round for its SaaS-based observability service following 171 percent ARR growth in its latest fiscal year.
Jeremy Burton
Observability and application performance monitoring (APM) software looks at applications – particularly cloud-native applications – and collects events and log data, activity metrics, component service invocations, and results. Observe adds meaningful business context to the raw basic data – such as customers and users, a shopping cart, a Kubernetes pod, a software build identifier, a problem ticket, and so forth. It brings this data to a single Snowflake repository and uses elastic cloud features for its analysis compute and capacity, helping issues get identified, understood, and fixed faster.
CEO Jeremy Burton explained: “Legacy monitoring and APM players, shackled by outdated architectures, are dead companies walking. As private equity or strategic acquirers strip them down for parts, Observe is taking a new approach designed for today’s modern distributed applications and massive data volumes. We’re thrilled to have investors who are thinking big and validating Observe’s approach in one of the fastest-growing segments in tech.”
A Gartner magic quadrant named the leading observability and APM vendors in 2023 as Dynatrace, Datadog, New Relic, Splunk, and Honeycomb. There were 14 other vendors mentioned in the MQ, but not startup Observe – it hadn’t registered with the Gartner analysts. Cisco has acquired Splunk while private equity buyers took Sumo Logic and also New Relic. The latter was bought by Francisco Partners and TPG at an equity valuation of about $6.5 billion.
Observe’s new investors include Sutter Hill Ventures, which led the round, and existing investors Capital One Ventures and Madrona plus a marquee new investor: Snowflake Ventures. They all put in the cash valuing Observe ten times higher – $325 million as we understand it – than the company’s $35 million Series A round four years ago. That included contributions from Michael Dell, Snowflake CEO Frank Slootman, and Pure CEO Scott Dietzen.
Observe funding history:
2017 – founded
2020 – $35 million A-round
2021 – $7 million venture round
2022 – $70 million A2 round
2024 – $115 million B-round
Total funding: $227 million
Recent growth stats are what pleased investors:
New average contract value (ACV) increased 176 percent year-over-year in FY24
Total Contract Value (TCV) increased 194 percent in FY24
Net Retention Revenue (NRR), which indicates stickiness of product, is 174 percent (industry-leading is considered 130 percent)
Mike Speiser, MD at Sutter Hill Ventures and a founder of Observe, elaborated: “Observe has … delivered a product that is architecturally different to everyone else. The incredible growth in ARR and NRR is testament to the fact that this new architecture is now paying off for their customers.”
Customers include Capital One, Reveal, and Top Golf. Mark Cauwels, managing VP, enterprise platforms technology at Capital One, provided a canned blurb: “Like many cloud-first organizations, our data volume continues to expand. Observe provides a centralized and pre-correlated data layer that meaningfully organizes telemetry data from many sources at scale, helping drive faster response times.”
Observe’s headcount increased more than 50 percent year-over-year and the biz is growing its sales organization, planning to expand its market presence in North America over the coming year. It expects to more than double the size of its business.
