BRN Discussion Ongoing

[Space Cadet]

Probably already posted, but link: Appear to be testing/using Brainchip to assist with contraband ID etc in US. Could be huge?????

Robust Classification of Contraband Substances using Longwave Hyperspectral Imaging and Full Precision and Neuromorphic Convolutional Neural Networks

Several agencies such as the US Department of Homeland Security (DHS) seek to improve the detection of illegal threats and materials passing through P…

www.sciencedirect.com

From the other XXX.
Robust Classification of Contraband Substances using Longwave Robust Classification of Contraband Substances using Longwave Hyperspectral Imaging and Full Precision and Neuromorphic Hyperspectral
Abstract
Several agencies such as the US Department of Homeland Security (DHS) seek to improve the detection of illegal threats and materials passing through Ports of Entry (POE). A combined hardware/software solution that is portable, non-ionizing, handheld, low cost, and fast would represent a significant contribution towards that goal as existing systems do not fulfil many or all of these requirements. To design such a system, Quantum Ventura partnered with Bodkin Design and Engineering to combine long-wave infrared (LWIR) hyperspectral imaging (HSI) with convolutional neural networks (CNNs), implemented on full precision GPUs and neuromorphic computing modules.
Neuromorphic processors implement CNNs with dramatically reduced size, weight, power and cost (SWaP-C) compared to GPU versions. Here we describe converting the 3D CNN into a format that can be run on neuromorphic platforms. We had early access to BrainChip’s software developer kit (SDK) and simulator thus we focused our efforts
on this. We now have access to Intel Neuromorphic Research Consortium and are using it for other projects [11]. BrainChip can support many features of CNNs but not all. For example, it can only accept grayscale or RGB images, not hyperspectral images (HSIs) for convolutional input layers. (For regular input layers, it may be possible to input HSIs but only 4-bit precision can be used at this time.) Because of this, we had to remap the 61 bands of the HSI image into separate “grayscale” input channels and then fuse across input channels in groups. Furthermore, the skip connections in the original 3D CNN are implemented by copying activation values from one neural processor unit (NPU) to another, and then copying them to the original NPU with identical weights of the value 1. This was the recommendation from BrainChip. The AkidaTM chip has 80 NPUs so using a handful of extra NPUs to implement the skip connections would not prevent neuromorphic implementation [12]. In Figure 5, we show the translated CNN
compatible with the BrainChip hardware.

If you take that one step further Bodkin seem to looking at Hyperspectral Imaging from drones in Agriculture as well. Wonder where they could get a chip for that? Looking at the size of the camera they may have a way to go.

https://www.bodkindesign.com/products-page/hyperspectral-imaging/precision-agriculture/

SC

 

[rgupta]

These are very attractive prices to buy IMO.

The fact is many will not buy here phycologicly people are afraid it will go lower. going lower. But when news comes and we are trading at 1.00s then there will be truck load of buyers.

You need to look to the future and learn from the past it's not comming back.

This is what the short would win.
Say you shorted 100k shares at 1.80$ you now have mad 130000$ nice we know for a fact that that is true.

That is only a few shorts too way less.

Today the shorter will say short 400k shares worth about 200000$ to make the same return the price needs to go 0.125 cents think about that would that happen?

How many do they need to cover 4 times more.

In short they have very little fire to play with now very little. The volume of actual real shares on the market is very very low the trading volume is a strong indication of that.

So for you to go long with 50 k you have a better chance of earning 130k profit with the companies current direction then the shorter does and your only risking 1/3 the amount of dollars the shorter has to.

Logically long is the favour based on the current value with news flow and company activity. The shorter is just playing up the fear still stopping potential buyers buy playing the short and arguing the same same issue revenue which is true but we also know revenue shoildbincreas in the next 24 months not decrease what it increases we dont know.

I did some research on semiconductor stocks there are companies that are valued between 10 a the way to 986 p/e.

This is a hot hot market at the moment for the next decade or 2 really.

For the record I have never shorted this stock or would plan to and you don't lose if you don't sell is true. All this is my opinion.

All I know it's getting harder to follow the news of what's happening with BRN that to me means lots is going on.

Revenue is inevitable in my opinion.

Also for a fact US spends 1 trillion on defence annually maybe even more that we don't know if and al l the related work around that. They seam to be really studying and working on the SNN side of stuff. Weather we like it or not this space will provide $$$ To BRN. IMO

A good analysis.
The other thing I want add here is,
Who is the biggest holders of brn?
Insiders!
If we look back 2 years company had issued shares and options around 500 million which is owned by brainchip staff.
If the problem here is so deep rooted brainchip should not able to retain old staff, leave alone hiring new ones.
So again my idea is, it is time to accomulate than to be fearful.
It is once a while opportunity to fry the shorts in there own game.
Dyor

 

[Boab]

Probably already posted, but link: Appear to be testing/using Brainchip to assist with contraband ID etc in US. Could be huge?????

Robust Classification of Contraband Substances using Longwave Hyperspectral Imaging and Full Precision and Neuromorphic Convolutional Neural Networks

Several agencies such as the US Department of Homeland Security (DHS) seek to improve the detection of illegal threats and materials passing through P…

www.sciencedirect.com

From the other XXX.
Robust Classification of Contraband Substances using Longwave Robust Classification of Contraband Substances using Longwave Hyperspectral Imaging and Full Precision and Neuromorphic Hyperspectral
Abstract
Several agencies such as the US Department of Homeland Security (DHS) seek to improve the detection of illegal threats and materials passing through Ports of Entry (POE). A combined hardware/software solution that is portable, non-ionizing, handheld, low cost, and fast would represent a significant contribution towards that goal as existing systems do not fulfil many or all of these requirements. To design such a system, Quantum Ventura partnered with Bodkin Design and Engineering to combine long-wave infrared (LWIR) hyperspectral imaging (HSI) with convolutional neural networks (CNNs), implemented on full precision GPUs and neuromorphic computing modules.
Neuromorphic processors implement CNNs with dramatically reduced size, weight, power and cost (SWaP-C) compared to GPU versions. Here we describe converting the 3D CNN into a format that can be run on neuromorphic platforms. We had early access to BrainChip’s software developer kit (SDK) and simulator thus we focused our efforts
on this. We now have access to Intel Neuromorphic Research Consortium and are using it for other projects [11]. BrainChip can support many features of CNNs but not all. For example, it can only accept grayscale or RGB images, not hyperspectral images (HSIs) for convolutional input layers. (For regular input layers, it may be possible to input HSIs but only 4-bit precision can be used at this time.) Because of this, we had to remap the 61 bands of the HSI image into separate “grayscale” input channels and then fuse across input channels in groups. Furthermore, the skip connections in the original 3D CNN are implemented by copying activation values from one neural processor unit (NPU) to another, and then copying them to the original NPU with identical weights of the value 1. This was the recommendation from BrainChip. The AkidaTM chip has 80 NPUs so using a handful of extra NPUs to implement the skip connections would not prevent neuromorphic implementation [12]. In Figure 5, we show the translated CNN
compatible with the BrainChip hardware.

There was a post from a couple of weeks ago that I recall talking abut spotting concealed weapons/guns/knives etc.
I don't recall the company.
This one has much more detail and looks very positive.
Thank you for sharing.

 

[IloveLamp]

Always-On Smart Sensors: Solving the Connectivity Challenge

IoT isn’t really about the internet. It’s not even about […]

tealcom.io

 

[Stable Genius]

There was a post from a couple of weeks ago that I recall talking abut spotting concealed weapons/guns/knives etc.
I don't recall the company.
This one has much more detail and looks very positive.
Thank you for sharing.

Hi @Boab

That was a different device, from memory using a hand held radar to scan people and use ML to identify objects on peoples bodies which were a threat, e.g. a gun/knife. It was a law enforcement company creating it. It’d be a bit like the TASER. Once one state has it they will all be wanting it. If it helps reduce the issues with shoot/no shoot scenario’s, and stop, search & detain laws to make things safer for the public/police then all the better!


Very interesting article that was shared. One of the reasons Brainchip will be successful is the price. How good it is that a $50 item is being compared with Nvidia’s $30K device.

Robust Classification of Contraband Substances using Longwave Hyperspectral Imaging and Full Precision and Neuromorphic Convolutional Neural Networks

Several agencies such as the US Department of Homeland Security (DHS) seek to improve the detection of illegal threats and materials passing through P…

pdf.sciencedirectassets.com

 

[Fullmoonfever]

Just noticed role ad by BAE which I mentioned in a previous post that BRE had won a subaward from for some project work.

Not saying related just that BAE into the neuromorphic sphere as well. Few things there that our partners like Intellisense, ISL and BRE who deal with DOD ecosystem, work on with Akida.

FAST Labs - Portfolio Lead, Digital Microelectronics Technology Development (Hybrid) in Burlington, Massachusetts, United States | Engineering & Technology at BAE Systems

Apply for FAST Labs - Portfolio Lead, Digital Microelectronics Technology Development (Hybrid) job with BAE Systems in Burlington, Massachusetts, United States. Engineering & Technology at BAE Systems

jobs.baesystems.com

FAST Labs - Portfolio Lead, Digital Microelectronics Technology Development (Hybrid)​

Clearance Level - Must Be Able To ObtainTop Secret
ITAR: U.S. Citizen Or Green Card RequiredYes
Travel Percentage10%
Clearance Level - Must Currently PossessNone
Career LevelExperienced
Is Relocation AvailableYes

Job Description
Because this role involves a combination of collaborative/in-person and independent work, it will take the form of a hybrid work format, with time split between working onsite and remotely.

BAE Systems FAST Labs™ Advanced Electronics (AE) Tech Area is growing rapidly. AE architects novel and beyond-next-generation RF, mixed-signal, and digital integrated circuits and chipsets, and delivers those new custom microelectronics capabilities into systems. With this rapid growth, AE is seeking a technology leader to help establish and drive its strategy toward a future expanded state. The candidate for this position will lead AE’s Digital Microelectronics portfolio. The trajectory of this portfolio includes algorithmic hardware accelerators, secure microprocessors and secure microsystems, digital signal processor SOCs and ASICs, radiation hard microelectronics, and new high-risk/high-reward novel compute technologies.

The candidate will lead others in engagement with external and internal customers to build, win, and execute programs that support the research and development of digital microelectronics technology. They are expected to leverage their position, reputation, and technical expertise to build and maintain relationships with government R&D organization such as DARPA, AFRL, OUSD(R&E), etc. Additionally, the candidate is expected to define and advance the Digital microelectronics technical and programmatic roadmaps while maintaining alignment across the broader Advanced Electronics (AE) tech area.

In this job role, qualified candidates can expect to:

• Lead relevant research programs of 5-25 people as principal investigator
• Provide business and execution oversight of programs valued at $1-50M
• Contribute to and lead proposals
• Build and maintain strategic partnerships with government agencies, external companies, and other BAE Systems groups and business areas
• Implement a programmatic strategy to fund technology development
• Pursue personal and portfolio growth to target standing up, managing, and supporting a team of scientists, researchers, and technology developers

This position can be based out of our Merrimack, NH; Burlington, MA; or Arlington, VA facilities, though it will require collaboration with staff across our business areas and facilities in the Northeast.

Please note that pursuant to a government contract, this specific position requires US citizenship status and eligibility to obtain a US security clearance.

Required Education, Experience, & Skills
The ideal candidate will possess the following:

• Excellent written and oral communication skills
• Experience capturing and leading projects at the cutting edge of technology
• Track record of technical innovation as evidenced by journal and conference publications or patent filings
• Demonstrable entrepreneurial drive
• Experience in at least one of the following areas:
  
  • Digital Signal Processing
  • AI/ML Hardware Acceleration
  • Digital system architecture
  • Secure microprocessor architecture
  • Neural networks or neuromorphic engineering

Ability to obtain clearance at the Top Secret level or higher

Preferred Education, Experience, & Skills

• Masters or PhD in electrical engineering or related field
• Prior experience as principal investigator on DoD R&D programs

About BAE Systems Electronic Systems
BAE Systems, Inc. is the U.S. subsidiary of BAE Systems plc, an international defense, aerospace and security company which delivers a full range of products and services for air, land and naval forces, as well as advanced electronics, security, information technology solutions and customer support services. Improving the future and protecting lives is an ambitious mission, but it’s what we do at BAE Systems. Working here means using your passion and ingenuity where it counts – defending national security with breakthrough technology, superior products, and intelligence solutions. As you develop the latest technology and defend national security, you will continually hone your skills on a team—making a big impact on a global scale. At BAE Systems, you’ll find a rewarding career that truly makes a difference. Electronic Systems (ES) is the global innovator behind BAE Systems’ game-changing defense and commercial electronics. Exploiting every electron, we push the limits of what is possible, giving our customers the edge and our employees opportunities to change the world. Our products and capabilities can be found everywhere – from the depths of the ocean to the far reaches of space. At our core are more than 14,000 highly talented Electronic Systems employees with the brightest minds in the industry, we make an impact – for our customers and the communities we serve.

 

[Stable Genius]

I’m dropping this report here for those new to Brainchip querying defence revenue. I’ve posted it a few times however in light of some of the recent news it’s worth a reminder there is a roadmap to success: and we’re firmly on it; in pole position!

The National Security Commission on Artificial Intelligence

www.nscai.gov

The easiest way to navigate around it is to hit the ”View the interactive report” button I circled in yellow (On the link supplied).

There are a plethora of chapters we are perfect for. I’m sure once the scientists/engineers get their head around how to use Akida it will be become ubiquitous throughout Defence. It’s just time to go from proof of concept to operational/commercial production!

Enjoy

 

[Bravo]

Qualcomm's Snapdragon 8 Gen 3 Rumored For A Massive GPU Uplift For Android Devices​

by Aaron Klotz — Tuesday, April 11, 2023, 05:20 PM EDT

A new rumor suggests that Qualcomm’s upcoming Snapdragon 8 Gen 3 SoC will feature a whopping 50% greater graphical uplift over its already super strong predecessor, the Snapdragon 8 Gen 2. This could enable future Gen 3-equipped devices to really excel in gaming performance, and far outpace even the likes of Apple's A16 bionic chipset.
The performance claim comes to us by way of 数码闲聊站 (translated to "Digital chatter") on Weibo. We don’t know how true this 50% performance upgrade could be, but it's not completely unreasonable. Qualcomm’s current Snapdragon Gen 2 is consistently 30% faster than the Snapdragon 8 Gen 1, in most graphical applications, and slightly less compared to Qualcomm’s mid-cycle refresh, the Snapdragon 8+ Gen 1, based on our review of the Samsung Galaxy S23 Ultra.

​

A 50% performance lead would make the Snapdragon 8 Gen 3 the dominant SoC for gaming and any GPU applications by a long shot. It would also solve any niche performance issues found in highly demanding mobile games and emulators, like the notoriously demanding Genshin Impact which can struggle to hit a steady 60FPS even on the Gen 2 equipped-Galaxy S23 Ultra. Apple’s closest competitor to the Gen 3 will be its upcoming A17 bionic, but current rumors slate that chipset to only have a 35% efficiency uplift over the previous generation. So at best, the A17 could be 35% faster in the same power envelope, if the rumors are true.

The Snapdragon 8 Gen 3 is rumored to arrive as soon as Q4 2023, and is rumored to feature a unique (1+5+2) core configuration, consisting of 1 speedy X4 mega-core clocked at 3.75GHz, five A720 performance cores clocked at 3GHz and two A520 efficiency cores clocked at 2 GHz. The Gen 3 chipset is also rumored to be manufactured on TSMC’s more efficient N4P node which promises up to 11% greater performance and 22% higher efficiency compared to the standard N4 node.

Gen 3’s rumored upgrades are shaping up to be a massive performance uplift over the Snapdragon Gen 2 and any of Apple’s current bionic chipsets. Hopefully, Qualcomm can accomplish this 50% performance target without killing the chip’s thermal performance as it did with the Snapdragon 8 Gen 1

Qualcomm's Snapdragon 8 Gen 3 Rumored For A Massive GPU Uplift For Android Devices

A new rumor suggests that Qualcomm’s upcoming Snapdragon 8 Gen 3 SoC will feature a whopping 50% greater graphical uplift over its already super strong predecessor the Snapdragon 8 Gen 2.

hothardware.com

 

[Bravo]

Snapdragon 8 Gen 3 To Feature Cortex-X4 Running At A 15 Percent Higher Clock Speed Than Snapdragon 8 Gen 2’s Cortex-X3

The Snapdragon 8 Gen 3 is said to be equipped with a faster Cortex-X4, with one tipster highlighting its expected clock speed

wccftech.com

 

[jtardif999]

What Is Few Shot Learning? (Definition, Applications) | Built In

Few-shot learning is a subfield of machine learning that aims to teach AI models how to learn from only a small number of labeled training data.

buff.ly

If this was anything to do with Akida (and I haven’t looked at it) then there would be no need to teach it to learn, it would just learn in the course of events. Nobody with a brain has to be taught to learn which suggests it is describing some computer process with no real learning outcome. Am I right?, I should read it I guess but I felt compelled by the wording ‘taught to learn’.

 

[FrederikSchack]

Probably already posted, but link: Appear to be testing/using Brainchip to assist with contraband ID etc in US. Could be huge?????

Robust Classification of Contraband Substances using Longwave Hyperspectral Imaging and Full Precision and Neuromorphic Convolutional Neural Networks

Several agencies such as the US Department of Homeland Security (DHS) seek to improve the detection of illegal threats and materials passing through P…

www.sciencedirect.com

From the other XXX.
Robust Classification of Contraband Substances using Longwave Robust Classification of Contraband Substances using Longwave Hyperspectral Imaging and Full Precision and Neuromorphic Hyperspectral
Abstract
Several agencies such as the US Department of Homeland Security (DHS) seek to improve the detection of illegal threats and materials passing through Ports of Entry (POE). A combined hardware/software solution that is portable, non-ionizing, handheld, low cost, and fast would represent a significant contribution towards that goal as existing systems do not fulfil many or all of these requirements. To design such a system, Quantum Ventura partnered with Bodkin Design and Engineering to combine long-wave infrared (LWIR) hyperspectral imaging (HSI) with convolutional neural networks (CNNs), implemented on full precision GPUs and neuromorphic computing modules.
Neuromorphic processors implement CNNs with dramatically reduced size, weight, power and cost (SWaP-C) compared to GPU versions. Here we describe converting the 3D CNN into a format that can be run on neuromorphic platforms. We had early access to BrainChip’s software developer kit (SDK) and simulator thus we focused our efforts
on this. We now have access to Intel Neuromorphic Research Consortium and are using it for other projects [11]. BrainChip can support many features of CNNs but not all. For example, it can only accept grayscale or RGB images, not hyperspectral images (HSIs) for convolutional input layers. (For regular input layers, it may be possible to input HSIs but only 4-bit precision can be used at this time.) Because of this, we had to remap the 61 bands of the HSI image into separate “grayscale” input channels and then fuse across input channels in groups. Furthermore, the skip connections in the original 3D CNN are implemented by copying activation values from one neural processor unit (NPU) to another, and then copying them to the original NPU with identical weights of the value 1. This was the recommendation from BrainChip. The AkidaTM chip has 80 NPUs so using a handful of extra NPUs to implement the skip connections would not prevent neuromorphic implementation [12]. In Figure 5, we show the translated CNN
compatible with the BrainChip hardware.

Very interesting! If Akida 1 could do an nVidia A100's job reasonably well at a fraction of the cost/size/price, then we're very competitive.

It also sounds like that they may benefit greatly from Akida 2's raw sensor input, TENN and 8 bit weights. Akida-P's 50 TFLOPS and probably much larger neuron capacity could probably help a lot.

 

[FrederikSchack]

Oh, AI-stralia, you keep giving. Our family is humbly thanking you for handing over another 220.000 shares.

 

[jtardif999]

If this was anything to do with Akida (and I haven’t looked at it) then there would be no need to teach it to learn, it would just learn in the course of events. Nobody with a brain has to be taught to learn which suggests it is describing some computer process with no real learning outcome. Am I right?, I should read it I guess but I felt compelled by the wording ‘taught to learn’.

I am right, this is just another article about extending generative AI.

 

[Steve10]

Crucial moment for the markets tonight.

US CPI due at 10.30pm AEST. Forecast to have big drop down to 5.2-5.3% YOY from 6% YOY.

However, core CPI forecast to remain flattish or go up slightly to 5.5-5.6% YOY from 5.5% YOY.

If market rises it will rise fast as shorts will get sizzled. April is seasonally a very good month.

 

[Taproot]

Hi @Boab

That was a different device, from memory using a hand held radar to scan people and use ML to identify objects on peoples bodies which were a threat, e.g. a gun/knife. It was a law enforcement company creating it. It’s be a bit like the TASER. Once once state has it they will all be wanting it. If it helps reduce the issues with shoot/no shoot scenario’s, and stop, search & detain laws to make things safer for the public/police then all the better!


Very interesting article that was shared. One of the reasons Brainchip will be successful is the price. How good it is that a $50 item is being compared with Nvidia’s $30K device.

View attachment 34104

Robust Classification of Contraband Substances using Longwave Hyperspectral Imaging and Full Precision and Neuromorphic Convolutional Neural Networks

Several agencies such as the US Department of Homeland Security (DHS) seek to improve the detection of illegal threats and materials passing through P…

pdf.sciencedirectassets.com

Meet the $10,000 Nvidia chip powering the race for A.I.

The $10,000 Nvidia A100has become one of the most critical tools in the artificial intelligence industry,

www.cnbc.com

Nvidia CEO Jensen Huang couldn’t stop talking about AI on a call with analysts on Wednesday, suggesting that the recent boom in artificial intelligence is at the center of the company’s strategy.
“The activity around the AI infrastructure that we built, and the activity around inferencing using Hopper and Ampere to influence large language models has just gone through the roof in the last 60 days,” Huang said. “There’s no question that whatever our views are of this year as we enter the year has been fairly dramatically changed as a result of the last 60, 90 days.”
Ampere is Nvidia’s code name for the A100 generation of chips. Hopper is the code name for the new generation, including H100, which recently started shipping.

More computers needed​

Nvidia A100 processor
Nvidia
Compared to other kinds of software, like serving a webpage, which uses processing power occasionally in bursts for microseconds, machine learning tasks can take up the whole computer’s processing power, sometimes for hours or days.
This means companies that find themselves with a hit AI product often need to acquire more GPUs to handle peak periods or improve their models.
These GPUs aren’t cheap. In addition to a single A100 on a card that can be slotted into an existing server, many data centers use a system that includes eight A100 GPUs working together.
This system, Nvidia’s DGX A100, has a suggested price of nearly $200,000, although it comes with the chips needed. On Wednesday, Nvidia said it would sell cloud access to DGX systems directly, which will likely reduce the entry cost for tinkerers and researchers.
It’s easy to see how the cost of A100s can add up.
For example, an estimate from New Street Research found that the OpenAI-based ChatGPT model inside Bing’s search could require 8 GPUs to deliver a response to a question in less than one second.
At that rate, Microsoft would need over 20,000 8-GPU servers just to deploy the model in Bing to everyone, suggesting Microsoft’s feature could cost $4 billion in infrastructure spending.
“If you’re from Microsoft, and you want to scale that, at the scale of Bing, that’s maybe $4 billion. If you want to scale at the scale of Google, which serves 8 or 9 billion queries every day, you actually need to spend $80 billion on DGXs.” said Antoine Chkaiban, a technology analyst at New Street Research. “The numbers we came up with are huge. But they’re simply the reflection of the fact that every single user taking to such a large language model requires a massive supercomputer while they’re using it.”
The latest version of Stable Diffusion, an image generator, was trained on 256 A100 GPUs, or 32 machines with 8 A100s each, according to information online posted by Stability AI, totaling 200,000 compute hours.
At the market price, training the model alone cost $600,000, Stability AI CEO Mostaque said on Twitter, suggesting in a tweet exchange the price was unusually inexpensive compared to rivals. That doesn’t count the cost of “inference,” or deploying the model.

 

[FrederikSchack]

Meet the $10,000 Nvidia chip powering the race for A.I.

The $10,000 Nvidia A100has become one of the most critical tools in the artificial intelligence industry,

www.cnbc.com

Nvidia CEO Jensen Huang couldn’t stop talking about AI on a call with analysts on Wednesday, suggesting that the recent boom in artificial intelligence is at the center of the company’s strategy.
“The activity around the AI infrastructure that we built, and the activity around inferencing using Hopper and Ampere to influence large language models has just gone through the roof in the last 60 days,” Huang said. “There’s no question that whatever our views are of this year as we enter the year has been fairly dramatically changed as a result of the last 60, 90 days.”
Ampere is Nvidia’s code name for the A100 generation of chips. Hopper is the code name for the new generation, including H100, which recently started shipping.

More computers needed​

Nvidia A100 processor
Nvidia
Compared to other kinds of software, like serving a webpage, which uses processing power occasionally in bursts for microseconds, machine learning tasks can take up the whole computer’s processing power, sometimes for hours or days.
This means companies that find themselves with a hit AI product often need to acquire more GPUs to handle peak periods or improve their models.
These GPUs aren’t cheap. In addition to a single A100 on a card that can be slotted into an existing server, many data centers use a system that includes eight A100 GPUs working together.
This system, Nvidia’s DGX A100, has a suggested price of nearly $200,000, although it comes with the chips needed. On Wednesday, Nvidia said it would sell cloud access to DGX systems directly, which will likely reduce the entry cost for tinkerers and researchers.
It’s easy to see how the cost of A100s can add up.
For example, an estimate from New Street Research found that the OpenAI-based ChatGPT model inside Bing’s search could require 8 GPUs to deliver a response to a question in less than one second.
At that rate, Microsoft would need over 20,000 8-GPU servers just to deploy the model in Bing to everyone, suggesting Microsoft’s feature could cost $4 billion in infrastructure spending.
“If you’re from Microsoft, and you want to scale that, at the scale of Bing, that’s maybe $4 billion. If you want to scale at the scale of Google, which serves 8 or 9 billion queries every day, you actually need to spend $80 billion on DGXs.” said Antoine Chkaiban, a technology analyst at New Street Research. “The numbers we came up with are huge. But they’re simply the reflection of the fact that every single user taking to such a large language model requires a massive supercomputer while they’re using it.”
The latest version of Stable Diffusion, an image generator, was trained on 256 A100 GPUs, or 32 machines with 8 A100s each, according to information online posted by Stability AI, totaling 200,000 compute hours.
At the market price, training the model alone cost $600,000, Stability AI CEO Mostaque said on Twitter, suggesting in a tweet exchange the price was unusually inexpensive compared to rivals. That doesn’t count the cost of “inference,” or deploying the model.

We definitely need another chip in the server room.

 

[FrederikSchack]

Are there anybody who has any idea how many neurons Akida 2.0 (Akida-P) can emulate at 8-bit weights?

I know about the 50 TOPS.

 

[cassip]

Hi to all,

I came over a company "Masimo". @Tothemoon24 postet their watch. They produce a lot of products and sensors for medical applications. Maybe their patents are worth a watch, they have many partners/licensees for these solutions:

Masimo - OEM Partners

Leading patient monitoring equipment manufacturers offer Masimo SET pulse oximetry. Find out more about our licensees.

www.masimo.de

Masimo - Patent Information

Masimo - Patents

www.masimo.com

Masimo - News & Media

www.masimo.com

Masimo Unveils the Future of Health Tracking and Smart Wearables with the Pre-Sale Launch of the Freedom™ Smartwatch

First of Its Kind Smartwatch Combines Masimo’s Accurate, Continuous Health Tracking Technology and Breakthrough Features in a Premium, Elegant Watch

Irvine, California - March 28, 2023 - Masimo(NASDAQ: MASI) a global leader in pulse oximetry and innovative noninvasive monitoring technologies, unveiled the newest addition to its wearable product line: the Masimo Freedom™ smartwatch. Developed to revolutionize the wearable technology industry, the Masimo Freedom smartwatch offers you the freedom to take control of your personal health and privacy through accurate and continuous health tracking, including a novel hardware feature designed to reduce radiation and free you from privacy infringement.


Masimo Freedom™
The Masimo Freedom smartwatch builds on the success of the Masimo W1™ advanced biosensing watch by leveraging its state-of-the-art sensor and digital signal processing technology, enabling it to provide continuous, accurate, and reliable readings of key health data, like arterial blood oxygen saturation (SpO2), hydration index (Hi™), pulse rate, heart rate, and respiration rate. The Freedom smartwatch seamlessly integrates Masimo’s cutting-edge health tracking technology into an elegant, user-friendly wearable with premium connectivity through Bluetooth®, Wi-Fi, and LTE technologies, enabling the everyday conveniences of texting, calling, music, and third-party app compatibility. Masimo Freedom is an ideal companion for active individuals looking for better control of their day-to-day health and wellness data. Unlike other products on the market, Masimo Freedom features a physical privacy switch that instantly stops all data sharing beyond the device, including user data, as well as microphone, location, and metadata.

“We are thrilled to finally unveil Freedom,” said Joe Kiani, Founder and CEO of Masimo. “We believe that this groundbreaking new product will revolutionize wearable technology and health tracking. We are allowing people to take control of their health with continuous and accurate actionable biosensing information along with the convenience of being connected, but without compromising their freedom.”

The Masimo Freedom smartwatch utilizes technology based on Masimo SET® pulse oximetry to optimize the capture of health data on the wrist. Masimo SET®is the primary pulse oximetry technology used at 9 of the top 10 hospitals, as ranked in the 2022-23 U.S. News & World Report Best Hospitals Honor Roll,1 and has been shown in over 100 studies to outperform other pulse oximetry technologies in clinical settings.2

In addition to Masimo Freedom, Masimo is also introducing a sleek, screenless band built on the innovations developed for Masimo W1 and Freedom. The band is designed to work in tandem and seamlessly synchronize with Freedom, so that you can wear one while charging the other – a 24/7 continuous wear ecosystem maximizing your ability to track your health both during the active day and at night while you sleep.

Consumers can pre-order Masimo Freedom now by placing a fully refundable $100 deposit, with expected shipping this fall. Masimo W1 is currently shipping. The Masimo band is expected to be ready for sale this summer. Those who would like to begin taking advantage of Masimo’s advanced biosensing technology now can purchase a Masimo W1 and automatically reserve their place in line to purchase Freedom with a $400 discount off the retail price. For more information about the Masimo family of wearables, and to order Masimo W1 and reserve your Masimo Freedom, visit freedom.masimoconsumer.com.

Masimo Freedom is not cleared for use in medical applications in the U.S.

@Masimo || #Masimo

About Masimo

Masimo (NASDAQ: MASI) is a global medical technology company that develops and produces a wide array of industry-leading monitoring technologies, including innovative measurements, sensors, patient monitors, and automation and connectivity solutions. In addition, Masimo Consumer Audio is home to eight legendary audio brands, including Bowers & Wilkins, Denon, Marantz, and Polk Audio. Our mission is to improve life, improve patient outcomes, and reduce the cost of care. Masimo SET® Measure-through Motion and Low Perfusion™ pulse oximetry, introduced in 1995, has been shown in over 100 independent and objective studies to outperform other pulse oximetry technologies.2 Masimo SET® has also been shown to help clinicians reduce severe retinopathy of prematurity in neonates,3 improve CCHD screening in newborns,4 and, when used for continuous monitoring with Masimo Patient SafetyNet™ in post-surgical wards, reduce rapid response team activations, ICU transfers, and costs.5-8 Masimo SET® is estimated to be used on more than 200 million patients in leading hospitals and other healthcare settings around the world,9 and is the primary pulse oximetry at 9 of the top 10 hospitals as ranked in the 2022-23 U.S. News and World Report Best Hospitals Honor Roll.1 In 2005, Masimo introduced rainbow® Pulse CO-Oximetry technology, allowing noninvasive and continuous monitoring of blood constituents that previously could only be measured invasively, including total hemoglobin (SpHb®), oxygen content (SpOC™), carboxyhemoglobin (SpCO®), methemoglobin (SpMet®), Pleth Variability Index (PVi®), RPVi™ (rainbow® PVi), and Oxygen Reserve Index (ORi™). In 2013, Masimo introduced the Root® Patient Monitoring and Connectivity Platform, built from the ground up to be as flexible and expandable as possible to facilitate the addition of other Masimo and third-party monitoring technologies; key Masimo additions include Next Generation SedLine® Brain Function Monitoring, O3® Regional Oximetry, and ISA™ Capnography with NomoLine® sampling lines. Masimo’s family of continuous and spot-check monitoring Pulse CO-Oximeters® includes devices designed for use in a variety of clinical and non-clinical scenarios, including tetherless, wearable technology, such as Radius-7®, Radius PPG®, and Radius VSM™, portable devices like Rad-67®, fingertip pulse oximeters like MightySat® Rx, and devices available for use both in the hospital and at home, such as Rad-97®. Masimo hospital and home automation and connectivity solutions are centered around the Masimo Hospital Automation™ platform, and include Iris® Gateway, iSirona™, Patient SafetyNet, Replica®, Halo ION®, UniView®, UniView :60™, and Masimo SafetyNet®. Its growing portfolio of health and wellness solutions includes Radius T°®and the Masimo W1™ watch. Additional information about Masimo and its products may be found at www.masimo.com. Published clinical studies on Masimo products can be found at www.masimo.com/evidence/featured-studies/feature.

ORi and RPVi have not received FDA 510(k) clearance and are not available for sale in the United States. The use of the trademark Patient SafetyNet is under license from University HealthSystem Consortium.

 

[Derby1990]

That line's getting vertical real fast.
I reckon BRN SP will follow suit.

 

[Tothemoon24]

Apologies if posted previously.
This is 6 weeks old , interesting ??

Northrop Grumman Touts New Multimode Sensor As Quicker, Cheaper Upgrade | Air & Space Forces Magazine

Northrop is moving into test and integration with a new software-controlled, multimode, “ultra-wideband” sensor that it said can simultaneously conduct radar operations, communications, and electronic warfare.

www.airandspaceforces.com