Cloud gaming did not have its greatest week this week as Google decided to shut down its cloud gaming service Stadia by January.
But as it did so, the company said its service had proven itself over the past few years, and it would continue providing that technology. On top of that, other rivals such as Nvidia’s GeForce Now, Microsoft’s Xbox Cloud Gaming, and Amazon’s Luna are carrying on with their own services.
That’s encouraging to Jeff McVeigh, vice president and general manager of Intel’s Super Compute Group. The company has launched its new datacenter graphics processing units (GPUs), dubbed the Flex Series. I talked with McVeigh about this at the Intel Innovation event this week. He sees new markets, like providing cloud gaming to hotel rooms so gamers can play while they’re traveling.
Intel is rolling out GPUs for the datacenter and its $329 Intel Arc 770 for mid-range gaming computers as well. Its timing isn’t the greatest, as there is a glut of GPUs in the market now thanks to a sudden crash in Chinese market orders, a crypto crash and changes to mining, and a general global economic slowdown.
Still, McVeigh is optimistic about the future and Intel’s long-term plans to battle Nvidia and Advanced Micro devices in both datacenter and consumer graphics markets. McVeigh’s role is to focus on the high-performance computing and datacenter markets, where the success of applications such as cloud gaming will determine Intel’s own path forward in graphics.
Here’s an edited transcript of our interview.
VentureBeat: And you’ve got new chips coming out.
Jeff McVeigh: Chips coming out! Different products. This is our Flex 140 and 170. Focused around flexible use cases, which is why we came up with the name. It allows us to build our value add around media processing, cloud gaming, AI inference. We’ll enable new use cases in the future. We’ve gone from PowerPoint to reality.
VentureBeat: Where are you finding the opportunity in the market? What’s the opening?
McVeigh: We initially started with media delivery and cloud gaming. We’re seeing some good traction for both of those. We’re not ready to reveal some of the customers, but some that are in active deployment, and others that are deep in evaluations on those use cases. This is where we’re rolling out our AI enablement solution. We think that’s the next area you’ll see a lot of pull from. But we’re seeing good traction on these first use cases.
VentureBeat: Where do you come in on price and performance compared to AMD and Nvidia?
McVeigh: Quite well. In terms of density, as an example, for media transfer performance, 1080p, we’re talking about 36 streams versus Nvidia can do seven on an A10. Now, something like an L40, they’re going to have more streams. But you’re looking at 300 watts versus 75 watts. From a TCO standpoint, it’s not only the cost of the capital, but also your operational cost. Power consumption is quite good there. We feel we have a strong offering around media.
Our gaming stack, we continue to optimize. I’d say we’re higher density, but maybe not as far ahead. On the AI side we have more software optimization to do, but I feel like we’re going to be very competitive versus an A10. We don’t know what H10 and others will look like because those aren’t in the market yet, but that’s where we stand right now.
Right now we have two versions. One for Android gaming, where we see a lot of adoption in China, for example, where that use case is very popular. Our Windows cloud gaming stack is in the beta stage right now. We have more optimization to do, game compatibility, support for more legacy games, things we need to optimize for DX11 and so forth. That’s where we aren’t quite so far along versus the Android stack.
VentureBeat: It feels like cloud gaming has an interesting opportunity. The Samsung gaming hub TVs–in a time when there was a shortage of consoles, getting a TV that could get you into gaming without a console seemed like a pretty good value proposition. Has that changed already, though, as the market has changed?
McVeigh: There have been market changes when it comes to supply versus demand. Regulations in China as far as the amount of time you can spend gaming, that’s also another damper. But let’s take another example. Consider a hotel room. Every hotel room has a TV. A lot of them have smart TVs. Most of them don’t have game consoles, but I can play a game with a smart TV that is streaming from a service. It could be local to the hotel or something from a CSP and so forth. We see interesting opportunities there that go beyond the traditional console in the home environment.
VentureBeat: Is there still a good entry point into the market, even though these shortages have turned into a glut? Is it harder to break into the market now than it was just a few months ago?
McVeigh: I don’t think it’s better. There are probably more headwinds versus anything making it easier. But this is a long-term play. How much are you enabling those experiences on any device? We were showing cloud gaming and Android gaming streamed from a data center that was in the other room, but onto a laptop, an Android tablet, and an iPad. It doesn’t matter what device you’re on. You can play that same game, the same experience. Pause it on one, pick it up on the other, and continue on. It’s not always a console-based experience, but more on the go, wherever I am, being able to experience that with that type of quality.
The glut of supply I think would be a difficulty if we were just going after console opportunities. But we have more of these use cases. I’m on the go. I want to be between devices. I want to be in environments where I don’t lug around my console with me. That opens it up considerably. Now, you have some down sides, obviously. You might not get the highest performance. You might have some latency issues. But the class of games you can target is pretty wide. It has a different value proposition.
VentureBeat: You still come in pretty inexpensive relative to the competition. Is that also part of the strategy?
McVeigh: Right, it is. We’re not trying to be–like Pat said, Moore’s Law is alive. We can use volume, both across the client as well as the data center, to have very attractive price points. We can have it packed into very power-efficient solutions that allow you to do it at scale in the data center.
VentureBeat: But these aren’t hardware loss leaders. I’m not sure how to interpret the comments about Moore’s Law that Pat made, and then that [Nvidia CEO] Jensen [Huang] made last week. I guess if you’re Jensen you might just be saying that because it explains your high price. But how do you look at the contrast?
McVeigh: No. We’re not giving these things away. We’re still making a good profit on them. But it’s not obscene profit like maybe the competition has started to try to go after. They’re trying to have excuses for why the price needs to go up. For us, Moore’s Law continues. We have a road map where we’re accelerating, and we can leverage that to deliver value for our customers.
VentureBeat: Mining seems to be maybe disappearing from the market. I wonder if everybody sees that as a good thing, that these products are now going where they were intended to go.
McVeigh: Going back to supply versus demand, it’s creating some of the rebalancing there. There’s also the inefficiency, if you’re only doing mining, in the power consumption. We’ve announced our Blockscale-based solutions that are more power-efficient for that dedicated workload. That’s adjusting how GPUs are being used in that environment.
VentureBeat: Mining made the market a bit more unpredictable. Maybe you could call it volatile. If you remove that from the market, it feels like it’s a good thing for that predictability. It’s now a more understandable market.
McVeigh: It’s a very speculative environment. People are spending money based on a market that’s highly volatile. That makes it hard to predict. How much demand is required? Well, today it’s massive. Tomorrow it’s really low. The day after? It’s not easily predictable.
VentureBeat: The timing with the CPU seems closely coordinated to have a superior gaming solution out there across the board.
McVeigh: You’re talking about the Raptor Lake generation? Yeah. Combining the CPU and GPU we have some additive capabilities. That’s great. And then on the data center side, obviously we’re coming out with Sapphire Rapids. Those will be paired with our data center GPUs to get the benefits, how we have software that goes between the CPU and GPU. You can load balance appropriately. It’s not like every part of the workload should go to the GPU. Some should stay on the CPU. We’ll have the right balance.
VentureBeat: As far as things that drive the demand on the datacenter side, are you seeing anything related to the metaverse?
McVeigh: Some of the cloud gaming environments that we’re engaged in — they’re precursors to metaverse. They have a clear use case if you want to do gaming, but then enable metaverse opportunities. I think they view some of these opportunities as stepping stones to get there.
VentureBeat: I was thinking in general that it’s a bad idea for Moore’s Law to end if we’re just getting started on the metaverse.
McVeigh: It’s almost shooting yourself in the foot. If Moore’s Law is dead, then some companies don’t make sense, like GPU companies. That’s kind of important to the business model, that we’ll continue to have more performance.
VentureBeat: I guess their argument is that architecture has never been more important. Clever design, smart design.
McVeigh: Right. There’s some agreement there. That’s why we’re doing GPUs. We’re doing GPUs. We’re doing dedicated accelerators. We have FPGAs. We build all of those architecturally. It is the right time to find the right balance of those architectures. But all those are enabled by silicon scaling. Not only the process itself, the wafer boundaries, but how you package that together. Moore’s Law, maybe in the strictest sense, is about transistor density, but there are many dimensions to it in my mind. It’s really around how you pack more performance into a certain cost envelope and power envelope. You do that with different things. You do it with architecture. You do it with 3D stacking. You do it with new materials. All those things come together.
VentureBeat: I remember Raja Koduri said we might need 1,000 times more computing power for the metaverse, or a real time metaverse for billions of people. It sounds like that might be many years away.
McVeigh: It’s not as if one day we’ll just suddenly cross over to the metaverse. It’ll be a continuum. I think he was painting a picture of something indistinguishable from the physical world. That said, we have a road map that will get us many orders of magnitude within the next decade. Process, technology, architectural changes. How much do we integrate? And then software. Sometimes you get 10, 100 times better just from tuning the software for the architecture you created. That gets us pretty close to 1,000 times.
VentureBeat: Does it mean we’ll have a 3D internet? As far as just interpretations of what qualifies as the metaverse.
McVeigh: I don’t know if I have a canonical definition. My view is, some of the things around simulations of the world for manufacturing, for safety, for the digital twins are out there. That’s one aspect of it. One is 3D representation for entertainment and communication, feeling like we’re in the same room together without having to get on a plane to make that happen. All those things, in my mind, constitute the metaverse.
VentureBeat: Do you have something that might be the equivalent of what Nvidia has been calling the Omniverse?
McVeigh: We do have some work in that space. We probably haven’t done as good a job of branding it and bringing it together. But we have a number of things around advanced rendering technologies that deal with the same data formats and how those come together. Like I said, probably not as well-packaged as what they have, but we have some things in that space.
VentureBeat: It feels like that’s a place where some kind of leadership is possible. If you’re pushing the simulation software ecosystem forward on top of the hardware, then that feels like something Intel would do.
McVeigh: We want to work with the ecosystem to enable an open version of that, as opposed to, “Here’s a proprietary version. You have to use ours. Good luck.” That’s why we work with game engine providers. We work with others that have defined the data formats and so forth so they can all interact with each other. Now, the positive side of having a complete vertical solution is it’s more turnkey. Everything is already there. The downside is you’re locked into a proprietary environment. We’re trying to give the same experience as the turnkey, but with interoperability of components so multiple people can participate. It’s not just one company driving it.
VentureBeat: A familiar way of how things play out would be pushing your software ecosystem forward so you can sell more of these. It sounds like you would think that if, say, the metaverse is open and the ecosystem is open, you would sell more of these.
McVeigh: Exactly. It’s the “raise all boats” kind of strategy. Make it open so that everyone can participate. There’s more demand for it, more demand for compute. Then let’s make sure our hardware and our systems are highly competitive. Pat’s example is USB. He always tells the joke about his granddaughter thanking him whenever she plugs it in. That’s good for Intel, because now PCs are a key part of that ecosystem.
VentureBeat: Are we heading that way, to an open ecosystem and metaverse, or do you think we have some things to worry about?
McVeigh: I think it’s still early. It could go either way at this stage. We’re still in the earliest stages. Just like our tagline for innovation: “Open, choice, and trust.” That’s how we prefer it. We think the ecosystem benefits from that as well, so we’ll keep going down that path. I think Nvidia is going to focus on a proprietary solution. So I think it’s open. It’s not decided yet.
The post Intel plunges into consumer and datacenter GPUs amid market uncertainty appeared first on Venture Beat.