There’s no doubt that when it comes to large-scale GPU cluster deployment, almost everyone including Microsoft is siding with Jensen and NVIDIA. With Team Green’s latest Blackwell-based B200 GPUs as well as Grace CPU-integrated GB200 Superchip hitting the market soon, there’s no doubt certain mega corp will be able to get “early access” to the portfolio so that they can plan, design, prepare, whatever they need to get new systems up and running within a certain timeframe.
Yet there’s one issue when one that is as strong as NVIDIA who’s capable of providing a full stack solution of hardware and software does business with partners and whatnot – the level of control. One might think “Sure, why not just use everything from NVIDIA and follow their instructions clearly” but the problem for today isn’t just about that.
It is their level of involvement with a given project that can be described as “intrusive” (at least for me) where not even Microsoft is safe from such a thing.
It came from The Information‘s writeup of a piece stating that both companies are currently at a standoff because of Microsoft’s intention on how the B200 GPUs were to be installed in their server room.
Yes, NVIDIA is apparently not comfortable with said proposal.
Just for some context, customers were the ones responsible for building/buying server racks to house the hardware but with all the DGX and SuperPods reveals done during GTC 2023 and the most recent COMPUTEX 2024, it is clear that NVIDIA wants its clients to have a shift in their purchase habit by buying “a whole lot more GPUs at once and we’ll just give you the required racks, connection hardware, etc.”. Well, because, “Buy More, Save More”.
That’s just part of the story though because according to the post, NVIDIA VP Andrew Bell directly asked Microsoft to buy a new server rack design for B200 specifically because their existing server racks currently in active cycle were “a few inches off”.
And as expected, Microsoft retaliated, stating that it would “make them difficult to switch between NVIDIA and rival AMD’s MI300X GPUs” (And this would probably violate some sort of anti-competitive laws and such, I don’t know, you tell me) and NVIDIA backed off with the recommendation.
And knowing the current ongoing shift to AI-based accelerated computing, there is still going to be a lot of existing infrastructure online to keep the whole Internet up and running, and Microsoft clearly doesn’t want to spend their bucks on something this trivial (Hey, remember the 700$ Mac Pro Wheel? I’d be like Microsoft in this case too).
Additional Sources: Tom’s Hardware
Like our Facebook Page here at NasiLemakTech.com for more news and in-depth reviews! Also, join our Facebook Group for insightful information and memes!
Subscribe to our YouTube channel too!
