Jensen Huang: And Stacy, if I could just add something. You say it’s H100 and I know you know what your mental image in your mind. But the H100 is 35,000 parts, 70 pounds, nearly 1 trillion transistors in combination. Takes a robot to build – well, many robots to build because it’s 70 pounds to lift. And it takes a supercomputer to test a supercomputer. And so these things are technology marvels, and the manufacturing of them is really intensive. And so I think we call it H100 as if it’s a chip that comes off of a fab, but H100s go out really as HGXs sent to the world’s hyperscalers and they’re really, really quite large system components, if you will.
Operator: Next, we go to Mark Lipacis with Jefferies. Your line is now open.
Mark Lipacis: Hi. Thanks for taking my question and congrats on the success. Jensen, it seems like a key part of the success — your success in the market is delivering the software ecosystem along with the chip and the hardware platform. And I had a two-part question on this. I was wondering if you could just help us understand the evolution of your software ecosystem, the critical elements. And is there a way to quantify your lead on this dimension like how many person years you’ve invested in building it? And then part two, I was wondering if you would care to share with us your view on the — what percentage of the value of the NVIDIA platform is hardware differentiation versus software differentiation? Thank you.
A – Jensen Huang: Yeah, Mark, I really appreciate the question. Let me see if I could use some metrics, so we have a run time called AI Enterprise. This is one part of our software stack. And this is, if you will, the run time that just about every company uses for the end-to-end of machine learning from data processing, the training of any model that you like to do on any framework you’d like to do, the inference and the deployment, the scaling it out into a data center. It could be a scale-out for a hyperscale data center. It could be a scale-out for enterprise data center, for example, on VMware. You can do this on any of our GPUs. We have hundreds of millions of GPUs in the field and millions of GPUs in the cloud and just about every single cloud.
And it runs in a single GPU configuration as well as multi-GPU per compute or multi-node. It also has multiple sessions or multiple computing instances per GPU. So from multiple instances per GPU to multiple GPUs, multiple nodes to entire data center scale. So this run time called NVIDIA AI enterprise has something like 4,500 software packages, software libraries and has something like 10,000 dependencies among each other. And that run time is, as I mentioned, continuously updated and optimized for our installed base for our stack. And that’s just one example of what it would take to get accelerated computing to work. The number of code combinations and type of application combinations is really quite insane. And it’s taken us two decades to get here.
But what I would characterize as probably our — the elements of our company, if you will, are several. I would say number 1 is architecture. The flexibility, the versatility and the performance of our architecture makes it possible for us to do all the things that I just said, from data processing to training to inference, for preprocessing of the data before you do the inference to the post processing of the data, tokenizing of languages so that you could then train with it. The amount of — the workflow is much more intense than just training or inference. But anyways, that’s where we’ll focus and it’s fine. But when people actually use these computing systems, it’s quite — requires a lot of applications. And so the combination of our architecture makes it possible for us to deliver the lowest cost ownership.
