Intel courts devs with open arms and exotic hardware

Is Developer Cloud enough to steal Nvidia's thunder?

Interview Intel is attempting to woo developers to its cloud with early access to unreleased hardware and a born-again attitude to open source in a bid to differentiate itself from competitors.

Markus Flierl, corporate veep for Intel's Developer Cloud, is an IT veteran that survived both Sun's absorption by Oracle and a stint as head of GPU cloud infrastructure at Nvidia, before taking on the current role.

Flierl attended the recent Kubecon Europe 2024 event and kicks off the conversation with us by describing Intel's approach compared to AI darling Nvidia as "fundamentally different."

"Everything is open," he says. "And we very much encourage the participation of the community to help work with us, as opposed to the more proprietary strategy that Nvidia is pursuing."

The Reg feels duty bound - for the purposes of balance - to point out that Nvidia is a significant contributor to many open source projects, including Kubernetes, Docker, and the Linux kernel itself.

Intel recently open sourced the Continuous Profiler, developed by Intel Granulate, in a bid to up CPU performance. The theory goes that a development team can run the optimization agent to spot bottlenecks in code and, therefore, make apps more efficient.

We mention to Flierl that such generosity benefits Intel. Software will eventually be optimized for Intel hardware.

"That's right, yeah," he says. "And I think that's just the difference in approach that we have compared to some of our competitors is that we're just very open. And we believe in the community and the collaboration. And that's just one of those examples of us doing that. And we want people to be able to optimize the software, and to get the maximum out of our hardware."

Intel's Developer Cloud is intended to give devs access to its latest hardware, usually a good few months before its rivals in the cloud space can get their hands on the silicon. However, there is no escaping from the fact that Intel trails the cloud giants in datacenter provision by some distance.

"We have datacenters predominantly in North America right now," Flierl too us. "They are leased – it's just a timing issue. I just came on board two years ago [but] it takes three years to build a datacenter.

"It's getting increasingly hard for datacenters with all the demand for AI … these are very power-hungry workloads."

We ask if there might be a chip giant that could, you know, devise some silicon that can do more work with less power.

Flierl laughs. "Or use Granulate to optimize the performance."

Of Intel's next datacenter location, Flierl says: "It's probably not going to be in North America right now. I'm on the lookout for a datacenter here in Europe, and I'm also looking at APAC."

"We're also looking at partnerships with local providers that can help us operate this as a sovereign cloud."

Flierl describes Intel's Developer Cloud plans as "a two-way street.".

"On the one hand, we want to give early access to strategic customers; the other is by talking directly to the end customers, we get that feedback directly to us.

"Traditionally, our model is we sell to OEMs, we sell to CSPs, and then they sell to the end customer … the benefit here is that we can talk directly to the end customer.

"We're the only company in the world that fabricates its own chips, designs its own chips, and also makes it available as a cloud service."

Intel, however, is not the one building their own AI chips.

"Yeah, you're right," says Flierl. "Google builds its own TensorFlow chips, but I cannot buy them anywhere else. There are some unique benefits we can offer as a company."

Has Intel done enough to persuade developers to use its cloud rather than those of its rivals? Unsurprisingly, Flierl reckons that the flexibility, with everything from bare-metal Xeons through to managed Kubernetes, makes the service a tempting choice.

"You can come in at the different layers of the stack, depending on what you're trying to do ... and then those services are available across the different instance types, so you can see – this workload – how well does it run on CPU versus how well it runs on a Gaudi versus how fast it runs on a GPU, and based on the results you're seeing, you're going to optimize your workload." ®

More about


Send us news

Other stories you might like