NVLink

roadmap.

the Ascend 960 series and their unified bus interconnect, their answer to NVLink, with company-reported claims of superclusters up to a million processors.

NVIDIA's advantage was never only per chip.

It was making thousands of chips behave like one machine.

Right.

This is where the research highlights NVIDIA NVLink.

Break down the mechanics of NVLink for us.

Why does it justify building an on-premises super POD rather than just spinning up a thousand random cloud GPUs?

Okay.

The GPUs process data incredibly fast, but they starve waiting for the next batch of data to arrive through that bottleneck.

Exactly.

NVLink bypasses the CPU entirely.

Now, one technical wrinkle here, networking.

Apple has discussed using NVIDIA's NVLink Fusion to connect those M8 chips, including NVIDIA switches, chiplets, and software.

Apple engineers reportedly believe that NVIDIA may currently offer the best connectivity technology.

And this kind of fits into NVIDIA's broader strategy with this NVLink.

It seems like they're positioning themselves to increasingly want it to become the interconnect standard even when an accelerator isn't an NVIDIA GPU.

The information article notes that Amazon has increasingly announced plans to use NVLink Fusion with its next generation AI servers.

So NVIDIA's goal essentially is if you don't buy our GPU, that's fine.

We'd still like you to buy our networking.

And now to your point, Jensen no longer talks about GPU roadmap alone.

He starts off with NVLink cluster.

He talks about NVLink as a separate business unit that is generating certain revenue.

And that has been growing 200% year over year.

And the networking component of the business is outgrowing the processor component of the business in certain companies.

It's $8.5 million.

That's one GPU, all connected with NVLink, two million parts, right? 250,000 kilowatts.

That's a GPU, and we ship thousands of them." Huang added that orders of their rack scale systems are currently growing an astonishing 27% month over month.

And overall, the core idea that Huang tried to communicate was that NVIDIA is still at the center of the AI build-out, despite headlines about new competition every other week.

That was Nvidia CEO Jensen Huang speaking to me after the deal with MediaTek on XPUs last month.

Today, startup dMatrix is the latest to partner with Nvidia to plug its specialized inference XPUs into Nvidia's platforms, targeting ultra-low latency AI workloads, where its tech works alongside Nvidia GPUs using NVLink Fusion.

dMatrix CEO Sid Chatterjee joins us now.

Very interesting.

Uh, so the biggest opportunity here is really about getting that product in customers' hands because we have got a lot of customer pull for that specific product in the Nvidia ecosystem.

Right.

So on your newest chip or your newest platform, I should say, is there any data point you can give me on how the partnership with Nvidia tangibly allows you to move faster? In other words, will this deploy into the real world quicker than it otherwise would had you not struck a deal with Nvidia for NVLink Fusion?

Yeah, no, 100%.

And now to your point, Jensen no longer talks about GPU roadmap alone.

He starts off with NVLink cluster.

He talks about NVLink as a separate business unit that is generating certain revenue.

And that has been growing 200% year over year.

And the networking component of the business is outgrowing the processor component of the business in certain companies.