Cloud Hosting

Threat actors have found a lucrative new attack vector that hijacks legitimate proxyware services, whichallow people to sell portions of their Internet bandwidth to third parties.In large-scale attacks that exploit cloud-based systems, cybercriminals can use this vector dubbed "proxyjacking" to earn potentially hundreds of thousands of dollars per month in passive income, researchers from Sysdig Threat Research Team (TRT) have found.

In a February blog post, Kaspersky researchers describedproxyware services like this: "[Users install a client that creates a]proxy server. Installed on a desktop computer or smartphone, it makes the device's Internet connection accessible to an outside party." That outside party the proxyware service then resells an agreed-upon portion of the user's bandwidth to other people.

"Depending on how long the program remains enabled and how much bandwidth it is permitted to use, the client accumulates points [for the user]that can eventually be converted into currency and transferred to a bank account," according to researchers at Kaspersky.

In one attack that the Sysdig researchers observed, threat actors compromised a container in a cloud environment using the Log4j vulnerability, and then installed a proxywareagent that turned the system into a proxy server without the container-owner's knowledge, the researchers revealed in a blog post on April 4.

This allowed the attacker to "sell the IP to a proxyware service and collect the profit," in an unusual type of Log4j exploit. Usually, Log4j attacks involve anactor dropping a backdoor or cryptojacking payload on the device, Crystal Morin, Sysdig threat research engineer, wrote in the post. "While Log4j attacks are common, the payload used in this case was uncommon," she wrote.

Proxyjacking shares characteristics of cryptojacking in that both profit off the bandwidth ofa victim and bothare about equally profitable for the attacker, Morin said. However, these attacks differ in that attackers typically install CPU-based miners to extract maximum value from compromised systems, while proxyjacking mainly uses network resources leaving a minimal CPU footprint, she wrote.

"Nearly every piece of monitoring software will have CPU usage as one of the first (and rightfully most important) metrics," she wrote. "Proxyjacking's effect on the system is marginal: 1 GB of network traffic spread out over a month is tens of megabytes per day very likely to go unnoticed."

Proxyjacking is a relatively new phenomenon spurred by the growth and use of proxyware services in the last couple of years, the researchers said. As mentioned, these services, such as IPRoyal, Honeygain, and Peer2Profit, are installed as apps or software on Internet-connected devices that, when running, allow someone to share Internet bandwidth by paying to use the IP address of the app users.

Proxyware comes in handy for people who want to use someone else's IP address for activity such as watching a YouTube video that isn't available in their region, conducting unrestricted Web scraping and surfing, or browsing dubious websites without attributing the activity to their own IP, the researchers said. According to the service, people pay for each IP address that someone shares via proxyware based on the number of hours they run the application.

In the attack investigated by Sysdig researchers, attackers targeted an unpatched Apache Solr service running in Kubernetes infrastructure to take control of a container in the environment, and then downloaded a malicious script from a command-and-control server (C2), which they placed in the /tmp folder to have privileges to perform their activity.

"The attacker's first execution was downloading an ELF file renamed /tmp/p32, which was then executed with some parameters, including the email address [emailprotected][.]com and the associated password for their pawns.app account," Morin wrote in the post.

Pawns.app is a proxyware service that has been seen sharing IPs from IPRoyal's proxy network. Indeed, Sysdig TRT correlated the binary downloaded and executed in the malicious script to the command-line interface version of the IPRoyal Pawns application from GitHub, which uses the same parameters, researchers said. In this way, attackers began using the compromised pod to earn money on the service, they said.

Attackers covered their tracks by cleaning the compromised system of their activity, clearing the history, and removing the file they dropped in the containers and the temp files, the researchers added.

While the list of proxyware services reported as being used for proxyjacking is small right now, Sysdig researchers believe that this attack vector will continue to grow and eventually "defenders will uncover more nefarious activities," Morin wrote."This is a low-effort and high-reward attack for threat actors, with the potential for far-reaching implications."

Researchers estimate that in 24 hours of activity for one proxyjacked IP address, an attacker can earn $9.60 per month. In a modest compromise of, say, 100 IP addresses then, a cybercriminal could net passive income of nearly $1,000 per month from this activity, they said.

When exploiting Log4j on unpatched systems, this figure can climb even higher, as millions of servers are still running vulnerable versions of the logging tool, and more than 23,000 of them can be reached from the Internet, according to Censys, the researchers said. "This vulnerability alone could theoretically provide more than $220,000 in profit per month" for an attacker, Morin wrote.

To avoid "receiving potentially shocking usage bills" due to proxyjacking activity, organizations need to take actions to mitigate potential attacks, the researchers said. They recommended that organizations set up billing limits and alerts with their respective cloud service provers, which can be an early indicator that something is amiss, Morin wrote.

Morin advised that organizations should also have threat-detection rules in place to receive alerts on any initial access and payload activity preceding the installation of a proxyware service application on your network.

Originally posted here:
'Proxyjacking' Cybercriminals Exploit Log4j in Emerging, Lucrative ... - Dark Reading

Read More..

When a man arrived in the middle of the night at a North London hospital and was emotionally upset, distressed, with seizure-like movements and unable to speak, Isabel Straw, an NHS emergency doctor, first struggled to find the reason because all the tests her team performed on him did not reveal any issues.

That is until she realized the man had a brain stimulator implanted inside his head and its malfunctioning was probably the reason for his pain.

Straw, also the director of the non-profit bleepDigital, urged decision-makers at all levels to start investigating further the cybersecurity risks of medical devices, from the consumer ones through the implanted and ingested technologies.

In the past 10 years, weve seen a lot of advances in these technologies, which has opened up new vulnerabilities, she said during a presentation at UK Cyber Week, on April 4, 2023.

The Internet of medical things (IoMT), as all these devices have come to be called, is increasingly used in healthcare settings and at home, both outside and inside the body, and is ever more interconnected, and so the security threats the IoMT poses are becoming more concerning and can have a significant impact on patients health.

The fear that these devices could start malfunctioning, or even get hacked, is real, and examples of cyber incidents involving IoMT devices are growing. As a result, there needs to be increased coordination between manufacturers and governments to implement more safeguards against security incidents and more capabilities to operate digital forensics, Straw said.

She also insisted that healthcare professionals should be trained on technical issues they could encounter with IoMT devices and on as many models as possible.

With the patient I mentioned, we had to go through his bag, where we found a remote control for the brain stimulator, which no doctor at the hospital knew about. So, I took a photo of it, did a reverse Google image search and found the manual online after a few hours. We realized the device was just desynchronized, but it took us 13 hours to find someone to reset it. If this happened again tomorrow, we would still not know how to treat him, she explained.

To this date, we still dont know why it malfunctioned. Often, these medical devices dont have the memory space or the ability for digital forensics, she added.

These devices can process increasing amounts of data, posing a security risk and data privacy concerns.

Since 2013, the electrodes in brain stimulators have started to be able to read more data, on top of just delivering a voltage. This allowed us to get more data from the patients brain activity and read it externally, which can be used to personalize the data youre analyzing to the patients disease. But streaming peoples brain data also brings a confidentiality issue, Straw highlighted.

In that case, not only does the brain stimulator needs to be secure, but also the communication streams with the health center, the system the health professional is using, and the cloud servers as health professionals increasingly use cloud services to process and analyze data.

Another challenge is what to do when someone dies because of a medical device. If this man had died, what would have happened with his device? Should we bury it with him, or dispose of it? Does it go to the general waste? And what do you mention on the death certificate? These questions are still unanswered, and we dont get training on those issues, Straw noted.

See the article here:
Securing Medical Devices is a Matter of Life and Death - Infosecurity Magazine

Read More..

When Twitter changed hands last November I switched to Mastodon; ever since Ive enjoyed happier and more productive social networking. To enhance my happiness and productivity I began working on a Mastodon plugin for Steampipe. My initial goal was to study the fediverse writ large. Which people and which servers are powerful connectors? How do moderation policies work? Whats it like to join a small server versus a large one?

These are important questions, and you can use the plugin to begin to answer them. But I soon realized that as a newcomer to a scene thats been evolving for six years, and has not welcomed such analysis, I should start by looking for ways to enhance the experience of reading Mastodon. So I began building a set of dashboards that augment the stock Mastodon client or (my preference) elk.zone. And Ive narrated that project in a series of posts.

Last week we released the plugin to the Steampipe Hub. If youve installed Steampipe, you can now get the plugin using steampipe plugin install mastodon. The next phases of this project will explore using the plugin and dashboards in Steampipe Cloud, and speeding up the dashboards by means of persistent Postgres tables and Steampipe Cloud snapshots. Meanwhile, heres a recap of what Ive learned thus far.

While the dashboards use charts and relationship graphs, they are mainly tables of query results. Because Steampipe dashboards dont (yet) render HTML, these views display plain text onlyno images, no styled text. Ive embraced this constraint, and I find it valuable in two ways. First, Im able to scan many more posts at a glance than is possible in conventional clients, and more effectively choose which to engage with. When I described this effect to a friend he said: Its a Bloomberg terminal for Mastodon! As those of us who rode the first wave of the blogosphere will recall, RSS readers were a revelation for the same reason.

Second, I find that the absence of images and styled text has a calming effect. To maintain a healthy information diet you need to choose sources wisely but, no matter where you go, sites deploy a barrage of attention-grabbing devices. I find dialing down the noise helpful, for the same reason that I often switch my phone to monochrome mode. Attention is our scarcest resource; the fewer distractions, the better.

Theres a tradeoff, of course; sometimes an image is the entire point of a post. So while I often read Mastodon using these Steampipe dashboards, I also use Elk directly. The Steampipe dashboards work alongside conventional Mastodon clients, and indeed depend on them: I click through from the dashboards to Elk in order to boost, reply, or view images. That experience is enhanced by instance-qualified URLs that translate foreign URLs to ones that work on your home server.

The ability to assign people to lists, and read in a list-oriented way, is a handy Twitter affordance that I never used much because it was easy to let the algorithms govern my information diet. Because Mastodon doesnt work like that, lists have become the primary way I read the fediverse flow. Of the 800+ people I follow so far, Ive assigned more than half to lists with titles like *Climate* and *Energy* and *Software*. To help me do that, several dashboards report how many of the people I follow are assigned to lists (or not).

I want as many people on lists as possible. So I periodically review the people I follow, put unassigned people on lists, and track the ratio of people who are, or arent, on lists. Heres the query for that.

When you read in a list-oriented away, as is also true when you read by following hashtags, there are always people whose chattiness becomes a distraction. To control that Ive implemented the following rule: Show at most one original toot per person per list per day. Will I miss some things this way? Sure! But if youve said something that resonates with other people, Im likely to hear about it from someone else. Its a tradeoff thats working well for me so far.

Heres the SQL implementation of the rule.

On the home timelines dashboard Ive made it optional to include or hide boosts, which can be the majority of items. On the list-reading dashboard Ive opted to always exclude them, but the SQL idiom for doing soselect distinct on (person, day)is simple, easy to understand, and easy to change.

Ive so far found three ways in which relationship graphs can make Mastodon more legible. First, in Mastodon relationship graphs, I showed how to use SQL-defined nodes and edges to show boost relationships among people and servers. In another article I used the same tools to map relationships among people and tags. And most recently I used them to explore server-to-server moderation.

In all three cases the format conveys information not directly available from tabular views. Clusters of interesting people pop out, as do people who share tags. And when I graphed servers that block other servers I discovered an unexpected category: some servers that block others are themselves also blocked, like infosec.exchange in this example.

The Steampipe combo of SQL-oriented API access and dashboards as code is a uniquely productive way to build relationship graphs that can unlock insights in any domain. As weve seen with Kubernetes, they can help make cloud infrastructure more legible. The Mastodon graphs suggest that the same can happen in the social networking realm.

When you append .rss to the URL of a Mastodon account, or tag, you produce an RSS feed like https://mastodon.social/@judell.rss or https://mastodon.social/tags/steampipe.rss. These feeds provide a kind of auxiliary API that includes data not otherwise available from the primary API: related tags, which appear in the feeds as RSS category elements. Steampipe really shines here thanks to the RSS plugin which enables joins with the primary Mastodon API. This query augments items in accounts feed with tags that appear in each item.

A similar query drives the graph discussed in Mapping people and tags on Mastodon.

In that example, surfacing the connection between a user, @themarkup, and a pair of tags, scotus and section230, was useful in two ways. First, it helped me instantly spot the item that I most wanted to read, which was buried deep in the search results. Second, it helped me discover a source that Ill return to for guidance on similar topics. Of course I added that source to my Law list!

Everyone who comes to Mastodon appreciates not having an adversarial algorithm control what they see in their timelines. Most of us arent opposed to algorithmic influence per se, though; we just dont like the adversarial nature of it. How can we build algorithms that work with us, not against us? Weve already seen one example: The list-reading dashboard displays just one item per list per person per day. Thats a policy that I was able to define, and easily implement, with Steampipe. And in fact I adjusted it after using it for a while. The original policy was hourly, and that was too chatty, so I switched to daily by making a trivial change to the SQL query.

In News in the fediverse I showed another example. The Mastodon server press.coop aggregates feeds from mainstream news sources. I was happy to have those feeds, but I didnt want to see those news items mixed in with my home timeline. Rather, I wanted to assign them to a News list and read them only when I visit that list in a news-reading mindset. The fediverse offers an opportunity to reboot the social web and gain control of our information diets. Since our diets all differ, it ought to be possibleand even easyfor anyone to turn on a rule like *news only on lists, not timelines*. Steampipe can make it so.

When you ask people on Mastodon about these kinds of features, the response is often Have you tried client X? It offers feature Y. But that solution doesnt scale. It would require massive duplication of effort for every client to implement every such policy; meanwhile, people dont want to switch to client X just for feature Y (which might entail losing feature Z). Could policies be encapsulated and made available to any Mastodon client? Its interesting to think about Steampipe as a component that delivers that encapsulation. A timeline built by SQL queries, and governed by SQL-defined policies, is a resource available to any app that can connect to Postgres, either locally or in Steampipe Cloud.

If youre curious about the Steampipe + Mastodon combo, install the plugin, try out the sample queries, then clone the mod and check out the dashboards. Do they usefully augment your Mastodon reader? What would improve them? Can you use these ingredients to invent your own customized Mastodon experience? Join our Slack community and let us know how it goes!

See the article here:
The Mastodon plugin is now available on the Steampipe Hub - InfoWorld

Read More..

Business leaders are facing the challenges of addressing sustainability goals including reducing carbon footprint and managing energy consumption costs, whilst also ensuring that they position their firm to take advantage of the rapid pace of change and new business opportunities that advancing technology, in particular AI, is enabling across every sector of the economy.

As an Intel Ambassador, I am delighted to continue my collaboration with Intel in relation to the 4th Generation of Intel Xeon Scalable Processors and the potential to scale AI across the economy whilst also helping meet sustainability objectives.

With built-in accelerators and software optimizations, 4th Gen Intel Xeon with built-in Accelerators have been shown to deliver leading performance per watt on targeted real-world workloads. This results in more efficient CPU utilization, lower electricity consumption, and higher ROI, while helping businesses achieve their sustainability goals.

One may add broad AI or Artificial Broad Intelligence (ABI) into the categories on the lower left side from the image above. We are now in the era of ABI as Multimodal, Multitasking Transformers from the likes of Microsoft, Google, OpenAI, and others enable certain Deep Learning algorithms to perform both vision and natural language processing (NLP) tasks, albeit such powerful algorithms require capable Central Processing Units (CPUs) and Graphical Processing Units (GPUs) that scale to perform well.

Intel 4th Generation Xeon Scalable Processors enable accelerated AI workloads 3x to 5x for Deep Learning inference on SSD- ResNet34 and up to 2x for training on ResNet50 v1.5 with Intel Advanced Matrix Extensions (Intel AMX) compared with the previous generation. Furthermore, in terms of AI performance the 4th Gen IntelXeon Scalable Processors deliver up to 10X higher PyTorch performance for both real-time inference and training with built-in AMX (BF16) vs prior generation. (FP32).

As we enter an era of ever more powerful AI algorithms such as Transformers with Self-Attention and Generative AI and the rise of AI meets the IoT (AIoT) well need the kind of capability that the 4th Generation Gen IntelXeon Scalable Processors deliver with more efficient and powerful CPUs that allow for AI to scale and process large volumes of data very rapidly in low latency use cases, and yet at the same time to do so with energy efficiency and reduced carbon footprint as a key objective too.

Microsoft commissioned a report from PWC entitled How AI can deliver a sustainable future in relation to the potential for AI across four sectors of the global economy:

Energy;

Agriculture;

Water;

Transportation.

The results from the report demonstrated the potential of AI to drive a reduction in emissions, whilst also increase jobs and economic growth across the four sectors explored in the report:

Reduction of CO2 emissions by up to 4% globally;

GPD growth of 4.4% amounting to a vast $5.2 trillion;

Employment growth amounting to a net 38 million jobs created.

The potential for the reduction in GHG emissions (up to 4% globally) is based upon assumptions applied across all four sectors (water, energy, agriculture and transportation) and the role that AI may play across those sectors including but not limited to precision agriculture, precision monitoring, fuel efficiencies, optimising use of inputs, higher productivity.

Furthermore, the resulting gains from Standalone 5G networks was set out by the US EPA and BCG (see right side of infographic above) whereby the resulting ability of SA 5G networks to enable massive scaling of the AIoT (AI applied onto IoT devices and sensors) and the increased automation flows with machine-to-machine communications may result in both a jobs gain and potential to reduce GHG emissions.

The latest Intel Accelerator Engines and software optimizations help improve power efficiency across AI, data analytics, networking and storage. Organizations can achieve a 2.9x average performance per watt efficiency improvement for targeted workloads utilizing built-in accelerators compared with the previous generation. This leads to more efficient

CPU utilization, lower electricity consumption and higher return on investment, while helping businesses achieve their sustainability and carbon reduction goals.

The 4thGeneration of IntelXeon Scalable Processors provide energy efficiency improvements achieved due to innovations within the design of the built-in accelerators. This allows for particular workloads to consume less energy whilst running at faster speeds.

The result per watt (on average) is 2.9X over the 3rd Gen Intel Xeon Processors whilst also allowing for massive scaling of workloads that will needed in the new era of the AIoT that we are entering into, for example inferencing and learning increased by 10X, 2X for improved compression, 3X for data analytics all achieved with 95% less cores. [1]

Another innovation is the Optimized Power Mode feature that, when enabled, provides 20% energy savings (up to 140 Watts on a dual socket system) while only minimally impacting performance (2-5% on select workloads).

The convergence of Standalone (SA) 5G Networks that will allow for a massive increase in device connectivity and ultra-low latency environments will allow for a massive scaling of the Internet of Things (IoT) with internet connected devices and sensors communicating with human users and each other (machine to machine). Increasingly these IoT devices will have AI embedded onto them (on the edge of the network).

Furthermore, Statista forecast that by 2025 there will be a staggering 75 billion internet connected devices, or over 9 per person on the planet! And IDC Seagate forecast that the volume of data generated will increase from 64 Zetabytes in 2020 (when we talked about the era of big data) to almost three times the volume amounting to 175 Zetabytes in 2025 with 1/3rd of this data consumed in real-time! Applying AI will be essential to efficiently manage networks and also to make sense of the data and provide near real-time responses to users.

Furthermore, this new era will allow us to measure, analyse (evaluate) and respond dynamically to our environment (whether that be healthcare, energy, smart cities with traffic, manufacturing, etc). AI capabilities and inference performance will be key to succeed in this era that we are entering into.

A world where Machine-to-Machine Communication reduces risk (broken down car is detected by the red car that then broadcasts to other vehicles around it who also then broadcast and thereby also avoid traffic jams where emissions can increase due to slow moving traffic) as shown in the illustration below.

IntelXeon Scalable processors provide for more networking compute at lower latency while helping preserve data integrity. Achieve up to 79% higher storage I/O per second (IOPS) with as much as 45% lower latency when using NVMe over TCP, accelerating CRC32C error checking with Intel Data Streaming Accelerator (Intel DSA), compared to software error checking without acceleration.

BCG in an article entitled Reduce Carbon and Costs with the Power of AI forecast that AI technology applied towards corporate sustainability goals may yield reductions of 2.6 to 5.3 gigatons or 1 to 3 USD trillion in value added.

The process for achieving this entails:

Monitoring emissions;

Predicting emissions;

Reducing emissions.

BCG believes that the sectors with the greatest potential for reductions of GHGs due to application of AI include: Industrial goods, transportation, pharmaceutical, consumer packaged goods, energy and utilities.

Intels vision is to accelerate sustainable computing, from manufacturing to products to solutions, for a sustainable future. Organizations can help reduce their scope 3 GHG emissions by choosing 4th Gen Gen IntelXeon Scalable Processors, which are manufactured with 90-100% renewable energy at sites with state-of-the-art water reclamation facilities that in 2021 recycled 2.8 billion gallons of water. For the avoidance of doubt, it is noted that the statistics provided in this paragraph entail Scope 3 emissions related to embodied carbon that don't impact the operational emissions of carbon, however, Scope 3 also includes operational carbon within which servers form a larger part of the equation.

Use case examples of applying the AIoT towards sustainability include the following:

Sensors that may detect that a no person is present in a room and hence switch off the lights and turn the heating (or if summer the air conditioning) off or to a lower level;

Sensors that may realise that a window is open whilst the heating is running and close it;

Predicting issues before they occur such as bust water pipes, unplanned outages in monitoring, traffic congestion spots and trying to reroute traffic or amend the traffic light sequencing to reduce the jams;

In relation to agriculture applying computer vision on a drone to determine when the crops are ripe for harvesting (so as to reduce wasted crops) and also to check for signs of drought and insect infestations;

Deforestation near real-time analytics of illegal logging.

Renewable energy drones applying Computer Vision from Deep Learning algorithms that may inspect the blades of wind turbines and solar panels on solar farms for cracks and damages thereby improving asset life and enhancing amount generated.

Energy storage optimisation with Machine Learning algorithms applied towards maximising the operational performance and return on investment for battery storage.

Rolnick et al. (2019) published a paper entitled Tackling Climate Change with Machine Learning (co-authored by leading AI researchers including Demis Hassabis (Co-founder of DeepMind), Andrew Y Ng, and Yoshua Bengio set out the potential to reduce emissions by applying AI across the manufacturing operations of a firm all the way from the design stage with generative design and 3D printing, supply chain optimization with a preference for low greenhouse gas emissions options, improving factory energy consumption with renewable supplies and efficiency gains (including predictive maintenance) through to detection emissions with the follow up action of abating emissions from heating and cooling and optimizing transport routes.

The 4th Generation of Intel Xeon Scalable Processors also have power management tools to enable more control and greater operational savings. For example, new Optimized Power Mode in the platform BIOS can deliver up to 20% socket power savings with a less than 5% performance impact for selected workloads.

Furthermore, the paper by Rolnick et al. sets out how firms may deal with the unsold inventory problem for retailers with some estimates placing the annual costing the fashion industry $120 billion a year! This is both an economic and an environmental wastage. Targeted recommendation algorithms to match supply with demand, and application for Machine Learning towards forecasting demand and production needs may also help reduce such wastage.

In the world of the AIoT a customer could be walking along the high street or the mall and a Machine Learning algorithm could offer them personalised product recommendations based upon the stores in close proximity to them.

Both the retail and manufacturing examples would require near real-time responses from the AI algorithms and hence a reason why accelerators within the CPU are important factors to deliver enhanced performance.

The world of the AIoT will require the ability to work within power constrained environments and respond to user needs in near-real time.

Intel enables organizations can make dynamic adjustments to save electricity as computing needs fluctuate. Gen Intel Xeon Scalable Processors have built-in telemetry tools that provide vital data and AI capabilities to help intelligently monitor and manage CPU resources, build models that help predict peak loads on the data centre or network, and tune CPU frequencies to reduce electricity use when demand is lower. This opens the door to greater electricity savings, the ability to selectively increase workloads when renewable energy sources are available and an opportunity to lower the carbon footprint of data centres.

In addition, only Intel offers processor SKUs optimized for liquid-cooled systems, with an immersion cooling warranty rider available, helping organizations further advance their sustainability goals.

AI will literally be all around us across the devices and sensors that we use, allowing for mass hyper- personalisation at scale with near real-time instant responses to the customer user. However, in order to avail these opportunities business leaders will need to ensure that they have invested in the appropriate technology that can meet the needs of the business and its customers.

We are entering an era where near immediate responses (often on the fly) will be necessary to engage with customers and also to respond dynamically in a world of machine-to-machine communication.

Intel Advanced Matrix Extensions (Intel AMX) allows for efficient scaling of AI capabilities to respond to the needs of the user and the network.

Significantly accelerate AI capabilities on the CPU with Intel Advanced Matrix Extensions (Intel AMX). Intel AMX is a built-in accelerator that improves the performance of Deep Learning training and inference on 4th Gen Intel Xeon Scalable Processors, ideal for work-loads like natural language processing, recommendation systems, and image recognition.

4th Gen Intel Xeon Scalable Processors have the most built-in accelerators of any CPU on the market to deliver performance and power efficiency advantages across the fastest growing workload types in AI, analytics, networking, storage, and HPC. With all newaccelerated matrix multiply operations, 4th Gen Intel Xeon Scalable Processors have exceptional AI training and inference performance.

Other seamlessly integrated accelerators speed up data movement and compression for faster networking, boost query throughput for more responsive analytics, and offload scheduling and queue management to dynamically balance loads across multiple cores. To enable new built in accelerator features, Intel supports the ecosystem with OS level software, libraries, and APIs.

Performance gains from the 4th Gen Intel Xeon Scalable Processors include the following (source 4th Gen Intel Xeon Scalable Processors perf index):

Run cloud and networking workloads using fewer cores with faster cryptography. Increase client density by up to 4.35x on an open-source NGINX web server with Intel QuickAssist Technology (Intel QAT) using RSA4K compared to software running on CPU cores without acceleration.

Improve database and analytics performance with 1.91x higher throughput for data decompression in the open source RocksDB engine, using Intel In MemoryAnalytics Accelerator (Intel IAA) compared to software compression on coreswithout acceleration solutions with 8.9x increased memory tomemory transfer using Intel Data Streaming Accelerator (Intel DSA), versusprevious generation direct memory access.

For 5G vRAN deployments, increase network capacity up to 2x with new instruction set acceleration compared to the previous generation.

Security is a key issue in the era of the AIoT as SA 5G networks expand and scale.

Businesses need to protect data and remain compliant with privacy regulations whether deploying on premises, at the edge, or in the cloud. 4th Gen Intel Xeon Scalable processors unlock new opportunities for business collaboration and insights even with sensitive or regulated data. Confidential computing offers a solution to help protect data in use with hardware-based isolation and remote attestation of workloads. Intel Software Guard Extensions (Intel SGX) is the most researched, updated, and deployed confidential computing technology in data centres on the market today, with the smallest trust boundary of any confidential computing technology in the data centre today. Developers can run sensitive data operations inside enclaves to help increase application security and protect data confidentiality.

Intels Bosch case study provides an example of an application of security in the IoT sector.

The case study observed that access to raw data sets is ideal for the development of analytics based on Artificial Intelligence. The example sets out how Boschs autonomous vehicles unit reduced risks associated with data or IP leakage using the open source project Gramine, running on Intel SGX. For more details, please refer to Implementing Advanced Security for AI and Analytics.

By the end of this decade, we may experience a substantial increase in the number of advanced Electric and Autonomous Vehicles (EVs and AVs) on the road and a world where

battery storage will be of greater importance as more renewable energy scales across the grid (following the Inflation Reduction Act in the US, and the continued policies of the UK and EU towards reducing carbon emission targets). Powerful CPUs with built in accelerators can help Machine Learning techniques scale across battery storage facilities to optimise the availability of energy and battery performance. This is relevant for edge and network scenarios with power and battery constraints, such as EVs and power-optimized devices in smart homes and manufacturing facilities.

In this world mass hyper-personalisation at scale enabled by the AIoT will allow for both near real-time engagement with the customer on the fly as well as greater efficiency and hence less wastage as Machine Learning and Data Science will enable superior prediction of customer needs from the vast amount of data that will be created.

One may imagine the users engaging in retail or entertainment on their way into and back home from work and the EV/AV recognising the passengers with Computer Vision from Deep Learning and personalising the environment of the car (entertainment, etc) to the user profile. The AV/EVs will go on one journey to another adjusting to different passengers and allowing the user to utilize their time efficiently and as they wish (engaging with brands, working, entertainment). However, even before more advanced EV/AVs arrive, there are many opportunities for firms to avail in the era of the AIoT for near real-time engagement with the customer whilst also reducing wastage (for example better matching supply and demand, improved demand forecasting, identifying and matching supply chain and manufacturing processes).

The 4th Gen Intel Xeon Scalable processors enable a more secure environment for developing IoT services and applications across the edge of the network, in turn enabling businesses to create new opportunities with greater confidence around security.

This vision of scaling and enabling a secure AIoT aligns with my own personal vision of applying AI and related data analytical and digital technology to deliver on sustainability objectives whilst also delivering world of genuine mass hyper-personalisation at scale whereby firms can truly respond to their customers needs in real-time conditions and further tailor their offerings to the individual customers needs.

Were entering an exciting new era from this year and across the rest of this decade whereby AI will scale rapidly across the devices and sensors around us as well as the remote cloud servers that will continue to remain important for training algorithms, acting as data lakes and enabling analytics on historic data to improve learning outcomes for AI and improve the personalisation of service or identify the opportunities to further enhance operational efficiencies across organisations.

Well be able to measure and evaluate emissions and energy consumption around us and identify wastage and reduce inefficiencies.

The AI algorithms across the Edge of the network will require energy efficient CPUs to operate across power constrained environments and to achieve reduced carbon footprint. The 4th Gen Intel Xeon Scalable Processors allow organisations to scale AI capabilities, provide hyper-personalisation at scale, and manage their internal operations at the Edge more efficiently whilst also helping enable security and sustainability goals to be met.

Imtiaz Adam

Data Scientist

Postgraduate in Computer Science with research in AI, Sloan Fellow

More here:
The Future of Applied AI: Towards A Hyperpersonalised & Sustainable World - BBN Times

Read More..