Thursday, November 30th 2023
Dell Partners with Imbue on New AI Compute Cluster Using Nearly 10,000 NVIDIA H100 GPUs
Dell Technologies and Imbue, an independent AI research company, have entered into a $150 million agreement to build a new high-performance computing cluster for training foundation models optimized for reasoning. Imbue is one of the few independent AI labs that develops its own foundation models, and trains them to have more advanced reasoning capabilities—like knowing when to ask for more information, analyzing and critiquing their own outputs, or breaking down a difficult goal into a plan and then executing on it. Imbue trains AI agents on top of those models that can do work for people across diverse fields in ways that are robust, safe, and useful. Imbue's goal is to create practical tools for building agents that could enable workers across a broad set of domains, including helping engineers write new code, analysts understand and draft complex policy proposals, and much more.
"The purpose of technology is to drive human progress, and this often begins at the research level," said Jeff Boudreau, chief AI officer at Dell Technologies. "Dell technology will provide Imbue with the powerful engine to help unearth the next generation of impactful AI innovation."
Imbue is already using the cluster - powered by Dell PowerEdge XE9680 servers with NVIDIA H100 Tensor Core GPUs - to train AI models and develop early prototype agents that can correct bugs in code and analyze lengthy documents. Longer term, the company is designing more capable, trustworthy AI agents that don't require constant supervision from their users, opening the door to a future where, for example, agents can plan a vacation on their users' behalf, not simply generate travel ideas, freeing people up to spend their downtime however they'd like.
Imbue and Dell designed the system to include smaller clusters to support rapid experimentation on novel model architectures as well as rapid networking into a large cluster, to enable the efficient training of large-scale foundation models.
"Building a new generation of foundation models requires the very best IT infrastructure, and Dell Technologies has helped us deploy a custom cluster much more quickly than other providers could have," said Josh Albrecht, chief technology officer of Imbue. "Dell has been an invaluable collaborator as we pursue our work to create AI systems with much stronger reasoning abilities."
Built for extreme acceleration for AI, machine learning and deep learning training, the Dell systems are equipped to deploy AI computing initiatives with high GPU memory, bandwidth and security. The PowerEdge servers' Smart Cooling features sustain great performance more efficiently while reducing the data center's overall carbon footprint.
Imbue's system is managed by Voltage Park, a cloud compute provider that builds solutions for machine learning.
Source:
Dell
Imbue is already using the cluster - powered by Dell PowerEdge XE9680 servers with NVIDIA H100 Tensor Core GPUs - to train AI models and develop early prototype agents that can correct bugs in code and analyze lengthy documents. Longer term, the company is designing more capable, trustworthy AI agents that don't require constant supervision from their users, opening the door to a future where, for example, agents can plan a vacation on their users' behalf, not simply generate travel ideas, freeing people up to spend their downtime however they'd like.
Imbue and Dell designed the system to include smaller clusters to support rapid experimentation on novel model architectures as well as rapid networking into a large cluster, to enable the efficient training of large-scale foundation models.
"Building a new generation of foundation models requires the very best IT infrastructure, and Dell Technologies has helped us deploy a custom cluster much more quickly than other providers could have," said Josh Albrecht, chief technology officer of Imbue. "Dell has been an invaluable collaborator as we pursue our work to create AI systems with much stronger reasoning abilities."
Built for extreme acceleration for AI, machine learning and deep learning training, the Dell systems are equipped to deploy AI computing initiatives with high GPU memory, bandwidth and security. The PowerEdge servers' Smart Cooling features sustain great performance more efficiently while reducing the data center's overall carbon footprint.
Imbue's system is managed by Voltage Park, a cloud compute provider that builds solutions for machine learning.
7 Comments on Dell Partners with Imbue on New AI Compute Cluster Using Nearly 10,000 NVIDIA H100 GPUs
If I see how things work out when automated... its suitable for specific processes sure, but its also completely unsuitable for the overwhelming majority called the rest.
I've walked around in companies where a 'robot' suddenly nuked an entire insurance claim history across over 20000 customers. Over a year and an army of 40 workers was called in to fix that, by hand. A clear demonstration that if automation fucks up, and it inevitably will somewhere, your fix cost is going to pay that automation 3-4 times over. Now, enter AI: a principle that isn't as controlled as a robot, and might produce a new result ;) I think AI is mostly useful to keep us confused, its certainly not a tool to remain 'in control' if you can't predict every possible outcome.
If anything I think AI will push more workers to the jobs where they can make a difference. Not a bad thing. Also, I think IT is slowly moving more and more into the social and socio-economic domain. Software design is increasingly an influence on society and its workings. More people skills are going to enter IT. Its already happening. The line between digital and reality fades.
The world population grows, the IT crowd keeps growing. We automate more. The IT crowd keeps growing. We have shortages in other (education-required) sectors right now - the IT crowd keeps growing regardless. People are just going to move to more managerial and product design - type roles instead of bashing in code, but there will be growth. If anything AI will spur even more growth in jobs. Its what I do in my day to day... I'm on the edge of 'cold hard code', I write product designs, I manage getting that content through a test street, and I think along in requirements-sessions to devise new stuff. The coding? Anyone can do it if the design is written proper. You could AI-ify it. You could already automate it without AI (hell, a simple Excel with macros will get you there, ;)). That'd be a little bit like sending the US army to crush a rat infestation in your home...Way overkill. Getting the AI to the point of usability would already be higher in cost than just doing the work.
We're humans. For everything we managed to automate or 'not do anymore' we figured out something to replace it.
A very straightforward example wrt AI: we want to keep control. Do you know what's behind those high-tech social media feeds we see every day? People of flesh and blood. Not AI! People of flesh and blood are sifting through mountains of explicit content to keep it away from us. If something does leak through, you have a real-world media shitstorm (see recent X revelations). Good luck giving AI the benefit of the doubt there ;) You can only do that if you are 100%, or maybe 99,9% sure that it complies with the assignment it was given. But it is AI, right? So if you want to really control its output, aren't you really just building yet another algorithm? So, if you want AI to do its thing, you still need a moderator on top of it to check everything it does. And given the efficiency of AI, its scalability, you probably need a huge workforce to keep pace. And where are we then in the end? We've got a system that has significantly higher output, but its fueled by a similar, and likely a much bigger workforce. 'Pay more to get more', in a relative sense little changed.
www.washingtonpost.com/technology/2019/07/25/social-media-companies-are-outsourcing-their-dirty-work-philippines-generation-workers-is-paying-price/
Seriously, this whole AI affair is one big pile of utter big tech bullshit. We already have this, and we've had it for decades. No single company in their right mind wants real AI. AI does create new business, again, lots of Silicon Valley startups can rake in millions to make idiots rich and perhaps something thrown at the wall will stick. That's what we have today. In the meantime, the real world keeps turning.
Buzz Words <> Reality. Just apply common sense. We already figured out computing quite a while ago. All we have left is increasing complexity to the point we stop understanding it ourselves. We then call it AI, until we figure it out again. The eternal cat&mouse of human and computer. And the end result: an infinite amount of E-waste, and free Google Maps plus a ChatGPT that can make idiots look smart, until someone with real brains takes a look at them.
Yay!