Microsoft is Developing Custom Networking Solutions to Break Free from NVIDIA's Grip

[AleksandarK]

According to sources close to The Information, Microsoft is developing a new custom networking card that could lessen the grip of NVIDIA's networking division, formed by the acquisition of Mellanox, on Microsoft server equipment. The card is reportedly intended to resemble NVIDIA's ConnectX-7 model, which supports 400 Gb Ethernet at maximum bandwidth. However, given that the chip is not out yet and availability is still far off, the final design could aim at a higher bandwidth (800 GbE, for example). Leading the network card project is Pradeep Sindhu, who previously co-founded Juniper Networks. Microsoft acquired Sindhu's latest startup, Fungible, last year, bringing him on board to head up this effort. By creating its own high-speed networking hardware tailored for AI workloads, Microsoft aims to accelerate and lower the cost of model training for AI ambitions, as well as reduce reliance on NVIDIA as the primary supplier of high-performance networking cards.

The Information's sources say developing the new network card will take over a year. But if successful, Microsoft's infrastructure could get a performance uplift from specific optimizations needed in its hyperscale data centers. The card may also enable energy savings compared to NVIDIA's networking gear. This would aid Microsoft in keeping its AI training costs in check even as model sizes rapidly scale up. In the long term, having custom networking technology purpose-built for AI could give Microsoft a competitive advantage, granting more granular control over its data center hardware for evolving AI workloads from OpenAI and other partners. Though NVIDIA GPUs currently dominate AI acceleration, alternatives such as Microsoft's Maia chip and future networking cards could loosen its grip. As AI permeates more business and consumer applications, Microsoft is making big bets on next-gen infrastructure for an AI-centric future.



View at TechPowerUp Main Site | Source

 

[mechtech]

Get a bunch of intel i225v rev0 and glue them together.

 

[ThrashZone]

Hi,
Will be funny when hopefully most people will reject AI as manipulated BS just like fact checker is.

 

[cvaldes]

Will be funny when hopefully most people will reject AI as manipulated BS just like fact checker is.

Highly unlikely.

People have been using AI/machine learning successfully for years. It just depends on the usage case. You can see a bunch of them under Solutions on the Nvidia website. And this field of computing is still very much in its infancy.

Remember that AI/ML isn't static, the models are constantly evolving and improving. Even your smartphone has likely been taking baby steps for several years, things like text recognition in photographs.

What is likely to happen is people will have a more nuanced understanding on where and when AI/ML can work better than previous conventional tools and where it still falls short. No one sane thinks AI can write Beethoven's Ninth Symphony or paint the ceiling of the Sistine Chapel.

However could AI set FedEx delivery routes? Program the MLB regular season schedule for all 30 teams? Pick stocks for a short-term investment strategy better than a fund manager? Create your fantasy sports team? Analyze protein structures to create better pharmaceuticals?

AI is already way more than the crappy coverage (like deepfake videos) mainstream media provides.

And AI can probably write Q&A forum comments just as comical and silly as humans can.





Ahahahahahahahahahahaha!!!!!

Anyhow, it's better for the world not to rely so heavily on one producer of anything, whether it be almonds, routers, copper, cotton, tulip bulbs, batteries, whatever.

Sticking your head in the sand isn't going to make AI go away.

 

[TechLurker]

Idly, I wonder whatever happened to AMD's own plans to get into Networking via Xilinx?

There was a story awhile back of AMD looking to make their own wired and wireless adapters to reduce reliance on Intel and offer better integration with their SoCs and semi-custom solutions, and later to offer a potentially cheaper alternative to Intel or MediaTek adapters on AMD mobos.

 

[trsttte]

Hi,
Will be funny when hopefully most people will reject AI as manipulated BS just like fact checker is.

This post is not about AI, it's about networking interfaces (nvidia bought mellanox which was a major provider of those things)

But truth be told, you could replace the mentions of networking with ai accelerators and it would still be true, they're developing their own npu for similar reasons (not just break free of nvidia but also increase efficiency)

 

[Minus Infinity]

And here I'd be happy if Synology could even discover 2.5GbE.

 

[LabRat 891]

And here I'd be happy if Synology could even discover 2.5GbE.

We're just 'consumers' and plebs; what makes you think us worthy of the tools of our betters?
(I'm joking but, this is the generalized feeling I get from Industry Professionals when talking about SOHO/consumer wired networking lagging behind so very very far)

 

[dir_d]

Idly, I wonder whatever happened to AMD's own plans to get into Networking via Xilinx?

There was a story awhile back of AMD looking to make their own wired and wireless adapters to reduce reliance on Intel and offer better integration with their SoCs and semi-custom solutions, and later to offer a potentially cheaper alternative to Intel or MediaTek adapters on AMD mobos.

AMD owns Pensado as their DPU. I'm pretty sure that division is working on some networking solutions (for enterprise). I dont think any of this will make its way back down to consumer.

 

[Arcdar]

Highly unlikely.

People have been using AI/machine learning successfully for years. It just depends on the usage case. You can see a bunch of them under Solutions on the Nvidia website. And this field of computing is still very much in its infancy.

Remember that AI/ML isn't static, the models are constantly evolving and improving. Even your smartphone has likely been taking baby steps for several years, things like text recognition in photographs.

What is likely to happen is people will have a more nuanced understanding on where and when AI/ML can work better than previous conventional tools and where it still falls short. No one sane thinks AI can write Beethoven's Ninth Symphony or paint the ceiling of the Sistine Chapel.

However could AI set FedEx delivery routes? Program the MLB regular season schedule for all 30 teams? Pick stocks for a short-term investment strategy better than a fund manager? Create your fantasy sports team? Analyze protein structures to create better pharmaceuticals?

AI is already way more than the crappy coverage (like deepfake videos) mainstream media provides.

And AI can probably write Q&A forum comments just as comical and silly as humans can.





Ahahahahahahahahahahaha!!!!!

Anyhow, it's better for the world not to rely so heavily on one producer of anything, whether it be almonds, routers, copper, cotton, tulip bulbs, batteries, whatever.

Sticking your head in the sand isn't going to make AI go away.

Thanks. Thanks so many times for a sane, informed, neutral and qualitative posting that highlights a few of the key parts most people overlook for the "shiny-tingy"-topics.

Also nice that you didn't join in on his argumentation. Well handled, my kudos to you, sir

 

[ThrashZone]

Highly unlikely.

People have been using AI/machine learning successfully for years. It just depends on the usage case. You can see a bunch of them under Solutions on the Nvidia website. And this field of computing is still very much in its infancy.

Remember that AI/ML isn't static, the models are constantly evolving and improving. Even your smartphone has likely been taking baby steps for several years, things like text recognition in photographs.

What is likely to happen is people will have a more nuanced understanding on where and when AI/ML can work better than previous conventional tools and where it still falls short. No one sane thinks AI can write Beethoven's Ninth Symphony or paint the ceiling of the Sistine Chapel.

However could AI set FedEx delivery routes? Program the MLB regular season schedule for all 30 teams? Pick stocks for a short-term investment strategy better than a fund manager? Create your fantasy sports team? Analyze protein structures to create better pharmaceuticals?

AI is already way more than the crappy coverage (like deepfake videos) mainstream media provides.

And AI can probably write Q&A forum comments just as comical and silly as humans can.





Ahahahahahahahahahahaha!!!!!

Anyhow, it's better for the world not to rely so heavily on one producer of anything, whether it be almonds, routers, copper, cotton, tulip bulbs, batteries, whatever.

Sticking your head in the sand isn't going to make AI go away.

Hi,
Man that's some good stuff you're smoking I need a card to lol

Bottom line this will be used as a toy for children hell just look at tiktok content should open peoples eye of kids using it to conceal faces.... if that doesn't impress you also refer to Phub lol

Plus a tool to commit fraud more easily in very compromising ways hell I saw AI asking user to share your voice/ videos/... and AI can replicate and create more content so sure make it easy for the bad guys out there to duplicate your content for you hehe

More recently with celebs which have always been targeted but now it makes it easy to just use photos.. of people online from facebook/...... to blackmail in new awesome AI ways

ID theft will flourish and of course a political tool for extremist so nothing new there.

Yeah don't worry MS/ NV/.... will all save us all against deepfake

 

[A Computer Guy]

That picture of the NVIDIA NIC looks like it could use a 12VHPWR connector.

​