Latest / The Edge Computing Podcast with Fexingo: Local Compute, CDNs, and Distributed Infrastructure / How Edge Computing Eliminates Latency in Cloud Gaming
Transcript
- Lucas: So cloud gaming has been the 'next big thing' for, what, a decade now? Google Stadia tried it, Amazon Luna tried it, and most of them fizzled out because the user experience just wasn't there. Luna: Right — the lag. You press a button and your character moves half a second later. That's a dealbreaker for any competitive game. Lucas: Exactly. And the fundamental problem is distance. If your game is rendering on a server two thousand kilometers away, even with a fiber connection, the round-trip time for a single frame is going to be at least 80 to 150 milliseconds. That's before the GPU even does its work. Luna: So the fix has to be moving the compute closer. That's where edge computing comes in. Lucas: Right. And the company that's been quietly making this work is NVIDIA — not with a flashy game-streaming service of their own, but with GeForce NOW, which is essentially a platform that lets you stream games you already own on Steam or Epic. They've been deploying gpu powered edge nodes inside telco data centers for the last two years. Luna: Can you give me a concrete example of where they've placed these nodes? Lucas: Sure. They've partnered with SoftBank in Japan and Deutsche Telekom in Germany. Instead of pulling from a regional cloud region like AWS's Frankfurt zone, the rendering happens inside the telco's own network — sometimes within the same city. So the data path goes from your controller to a local edge node, and back, in under 20 milliseconds. Luna: Twenty milliseconds is competitive. That's about the same latency you'd get from a console plugged into a TV. Lucas: Exactly. And it's not just about distance. The architecture matters. A standard CDN like Cloudflare or Akamai is great for caching static video files, but it can't render a frame. GeForce NOW's edge nodes are essentially headless gaming PCs — they have a GPU, they run the game executable, they encode the video output in real time, and they stream it to you. That's a fundamentally different workload. Luna: So the edge node isn't just a cache; it's a compute engine. How do they handle virtualization? I mean, you can't just run one game per GPU card — that would be insanely expensive. Lucas: They use GPU virtualization — specifically, NVIDIA's own virtual GPU technology. A single A16 GPU can be split into up to 32 virtual instances, each serving a different user. So one physical card handles 32 concurrent game streams. That brings the per-user cost way down. Luna: And I assume they're using the latest generation GPUs? Like the Blackwell architecture? Lucas: Actually, for edge deployments they're using the A series — A16 and A40 — because those are designed for data center density and power efficiency. Blackwell is more focused on AI training. But the A16s have enough punch to run AAA titles at 1080p 60 frames per second. For 4K, they step up to the A40. Luna: So what's the business model here? Is it subscription, like Netflix for games? Lucas: GeForce NOW has a subscription tier — about ten dollars a month for priority access to the edge nodes — but they also sell a per-session model. You can pay for a single day pass for about three bucks. That's smart because it lowers the barrier. You don't have to commit to a monthly fee if you only want to play for a weekend. Luna: Interesting. And the telcos get a new revenue stream, too. They're selling rack space and power to NVIDIA, and in some cases they get a cut of the subscription. Lucas: Right. SoftBank actually markets it as part of their 5G gaming bundle. The tagline is basically: 'Low latency network plus low latency compute.' And that's the edge computing value prop in a nutshell. Luna: You know, this conversation reminds me of something. If you find these deep dives into tech infrastructure useful — and it sounds like you do — there's a small way you can help keep the show ad-free. We're listener-supported, and every bit helps. You can buy us a coffee at buy me a coffee dot com slash fexingo. Lucas: Yeah, honestly, it's what lets us spend time on stories like this — digging into the real architecture instead of just reading press releases. And we really appreciate it. So, back to GeForce NOW — they're now in over 30 edge locations globally. But not all games work well on edge. Luna: Which games don't work? And why? Lucas: The biggest issue is latency-sensitive competitive shooters like 'Counter-Strike' or 'Valorant'. Even 20 milliseconds is noticeable for top-tier players. They want under 5. So GeForce NOW doesn't even market those titles. They focus on narrative-driven games, RPGs, strategy — anything where 20 milliseconds is fine. Luna: So it's not a universal solution. But for the vast majority of gamers, it's good enough. And it means you don't need a thousand-dollar GPU in your living room. Lucas: Exactly. And that's the bigger picture. Edge computing is democratizing access to high-end gaming. Someone in a dorm room with a cheap laptop can play 'Cyberpunk 2077' at high settings because the compute is happening at the edge. The laptop is just a screen and a keyboard. Luna: What about the data caps? Streaming games at 1080p uses about 10 to 15 gigabytes per hour. That adds up fast. Lucas: It does. And that's a real barrier. But telcos are starting to offer unlimited data plans specifically for gaming. In Japan, SoftBank has an unlimited 5G plan that whitelists GeForce NOW traffic — it doesn't count against your cap. That's a clever network management move. Luna: So edge computing isn't just solving the latency problem; it's creating new business models for ISPs. They can differentiate on quality of service for specific applications. Lucas: Right. And we're seeing the same pattern in other latency-sensitive apps — autonomous vehicles, remote surgery, industrial robotics. But gaming is a nice beachhead because it's a massive consumer market with clear pain points. Luna: One last thing: Do you think we'll ever get to a point where edge nodes are as ubiquitous as cell towers? Like, every neighborhood has one? Lucas: Not every neighborhood, but every major metro area, yes. We're already seeing that with 5G edge. The key enabler is the telco edge — these nodes sit right at the aggregation point where 5G traffic meets the fiber backbone. That's within 50 kilometers of most users in dense cities. So for cloud gaming, that's the sweet spot. Luna: It feels like we're at an inflection point. The hardware is cheap enough, the networks are fast enough, and the business models are starting to click. Lucas: Yeah. And the best part is, you don't need to buy a new console to experience it. That's the edge computing promise — compute where you need it, when you need it.