Yeah, I think that’s the running theory of the corporate backup plan if/when AI falls through and they need to find a use for all the data center resources.
You’re right, but I think you underestimate the intersection of what’s theoretically possible and desperation.
We’ll probably see multiple different things. Some of the equipment will be repurposed for other cloud GPU workloads. We’ll probably see some companies try to make adapters and drivers to get the server hardware compatible with consumer desktops (there’s someone who did a writeup, either on hacker news or hack a day who got older server RAM working on a desktop with an adapter and Linux tomfoolery). Buy up the stuff being sold at a loss, strap their adapter on it, and sell for profit.
But on a low level, all this shit is turing complete. There will be performance losses and considerable overhead, but in theory translation layers and emulation should be able to be written to effectively convert GPU compute to whatever other format is needed.
If it’s taking a total loss vs squeezing the last squirts of profit out, I think there’s going to be attempts to squeeze.
Let’s assume a modern AAA game streamed to a dumb terminal: what are the minimum specs for that terminal to run the game without hiccups (controlling for network bandwidth, of course)?
The requirements for cloud gaming revolve around having the hardware to drive the local monitor, and having fast enough internet with low enough latency.
Notably, the game data is not loaded into RAM. In case it isn’t clear, those requirements are very low, by 2026 standards. If it can display video on the monitor you have, then it can probably stream a game.
Why not rent and stream the games for only $99.99 per month. Subscribe now!
Well… Okay. But only if I can keep paying the subscription fee for the rest of my life in order to maintain access.
That’s why we’ve created the GenPro tier of subscriptions, the first generational loan. Get it now and your kids can help pay of back!
Yeah, I think that’s the running theory of the corporate backup plan if/when AI falls through and they need to find a use for all the data center resources.
An AI data center cannot be used to run virtual machines. Cannot be converted into a gaming streaming service. They don’t work that way
You’re acting like the BC-250 doesn’t exist. If there’s a monetary opportunity, someone is going to try it.
You’re right, but I think you underestimate the intersection of what’s theoretically possible and desperation.
We’ll probably see multiple different things. Some of the equipment will be repurposed for other cloud GPU workloads. We’ll probably see some companies try to make adapters and drivers to get the server hardware compatible with consumer desktops (there’s someone who did a writeup, either on hacker news or hack a day who got older server RAM working on a desktop with an adapter and Linux tomfoolery). Buy up the stuff being sold at a loss, strap their adapter on it, and sell for profit.
But on a low level, all this shit is turing complete. There will be performance losses and considerable overhead, but in theory translation layers and emulation should be able to be written to effectively convert GPU compute to whatever other format is needed.
If it’s taking a total loss vs squeezing the last squirts of profit out, I think there’s going to be attempts to squeeze.
since its mostly use for AI, it will likely be left to “rot” as there are no use for it anymore.
You still need to run streamed games in RAM. A lot of RAM.
Well…we’ll go back to Zork and Venture.
No you don’t. I can run moonlight on a potato. All the processing is server side.
Processing, yes. Game data needs to be loaded in RAM.
I suppose games of the future will be optimized to require less RAM while streaming.
Even modern web browsers require at least 8gb.
No, it doesn’t. The client receives the video + audio and sends control inputs to the server. The game data doesn’t touch the client device.
It’s the same concept as RDP and VNC. With fiber being rolled out in more and more places, input lag will become a thing of the past.
Source: I used to be a VDI Engineer. I designed and maintained the infrastructure for 1k zero clients that had lower specs than a Pi 3.
I’ll defer to your expertise.
Let’s assume a modern AAA game streamed to a dumb terminal: what are the minimum specs for that terminal to run the game without hiccups (controlling for network bandwidth, of course)?
The requirements for cloud gaming revolve around having the hardware to drive the local monitor, and having fast enough internet with low enough latency.
For example, here are Geforce Now’s minimum requirements: https://www.nvidia.com/en-us/geforce-now/system-reqs/
There is some overhead involved in streaming, but the RAM requirements are low and certainly don’t change based on the game.
Ok interesting. 4gb RAM. But they’re also requiring a GPU that supports direct x v11. So there’s some rendering happening at the client?
That’s still going to make a dumb terminal quite expensive.
Notably, the game data is not loaded into RAM. In case it isn’t clear, those requirements are very low, by 2026 standards. If it can display video on the monitor you have, then it can probably stream a game.
But Zork needs 64Kb ram!
Streaming only replaces storage, you actually need more RAM than running it from fixed memory.
Nah, you just need to show it like a video basically.
There is always a latency issue.
https://www.pcgamer.com/heres-how-stadias-input-lag-compares-to-native-pc-gaming/
That has nothing to do with ram consumption.
FYI IIRC gta 5 had 13 (or 14?) frames of latency when it came out (controller => gfx). Still a wild success.
Was really just talking about whole remote vs local gaming thing. Latency does depend on the type of game. Some it matters less than others.
Not when the processing happens in one of the many data centers that’ll be sold off when the AI bubble bursts.
I fully expect them to start heavily advertising VDI in the near future.