The upcoming demise of Google Stadia dominates discussion, but we find plenty of time to discuss a range of topics: PS5’s new 6nm processor, the arrival of XeSS and the imminent deployment of Intel Arc, the quality of Ryzen 7000 CPUs and the challenges of CPU reviewing in general, DLSS 3 reaction and EVEN MORE.
Note: Stadia GPU power is rated at 10.7TF, not in the 9TF region as mentioned here.
Join the DF Patreon for EARLY ACCESS to every DF Direct Weekly, pristine video downloads, behind the scenes content and much, much more: https://bit.ly/3jEGjvx
Podcast RSS Feed: https://digitalfoundry.podbean.com/
Subscribe for more Digital Foundry: http://bit.ly/DFSubscribe
00:00:00 Introduction
00:00:48 News 01: Google Stadia is shutting down
00:20:00 News 02: PS5 gets a die-shrink!
00:23:31 News 03: XeSS debuts, Arc A750/A770 priced
00:37:13 News 04: New Ryzen 7000 CPUs reviewed
00:57:49 News 05: Sony PC game announcements, leaks
01:02:18 DF Content Discussion: DLSS 3 reactions and follow-up
01:07:48 DF Supporter Q1: Is latency a big issue with DLSS 3? Could a future version have unnoticeable lag?
01:10:22 DF Supporter Q2: Could tech similar to DLSS 3 come to consoles?
01:11:40 DF Supporter Q3: Would a slower version of DLSS 3 be able to run on Ampere cards?
01:13:48 DF Supporter Q4: Is vendor-agnostic frame generation tech similar to DLSS 3 possible?
01:16:40 DF Supporter Q5: How does DLSS 3 cope with RT effects?
01:19:14 DF Supporter Q6: Would DLSS 2 Quality mode work better as a base for DLSS 3 than Performance mode?
01:22:15 DF Supporter Q7: What do you expect from AMD’s new graphics cards?
01:28:47 DF Supporter Q8: Thoughts on EVGA leaving the GPU market?
01:32:45 DF Supporter Q9: Is 8K excessive? Shouldn’t we focus on visual features and not pixel count?
01:35:13 DF Supporter Q10: Could PCs move to unified memory instead of two discrete memory pools?
source
The intel GPU is still gonna have a bigger library of games to play than MacOS.
The majority of people already knew from the very start that Stadia will fail.
An inferior way to play games fails? Shocked.
@19:33
I live in Japan. I have the fastest optic fiber in the country, and LTE from Docomo, xCloud works really well tho.
Back before xCloud supports Japan, I used to VPN to S.Korea and it works fine too.
Yes, news from the nvidia marketing department. cool
is that a 4090 in the background?
Imagine if your team meetings at work were broadcast for the whole world to see.
If Stadia would’ve released with a sister PC client that would’ve allowed you to play all of your games locally it would’ve been great. A true possible competitor to Steam that would give you the added flexibility of streaming when you weren’t at home using your PC. If I had the option to buy a game on two platforms and one of them gives me the ability to stream my library when I’m away from my home PC, I’d choose that option almost every time.
Why's everyone talking about upcoming 30fps era around consoles? Sorry but I believe, it's never coming back. Heck even now, after 2 years, new games runs 120fps. 60fps is going to be option till these consoles dies…. (Exept series s of course)
No Intel Arc Review?
1:23:30 AMD proved that Nvidia is inflating the prices of their RTX 4000 unnecessarily much by selling their AMD Ryzen Zen4 CPU:s at a not as high margin in comparison!
I love the idea of John pouring one out and toasting the death of Stadia 🤣👍
DLSS 3 artifacts are nothing compared to the artifacts of lossy video codecs. Even bluray paused frames can look bad up close yet completely unnoticable in motion.
No linux gamers anyway
Gaming ownership is an illusion. Cloud only will become standard and everyone will forget about the complaining phase.
Xcloud may not ever obtain Stadia's quality.
I don't understand why Google requires the developers to port their game to Linux. Why not just run a customized version of Proton?
Hey guys, you should really open a Spotify account for all those great podcasts!
Yes good. Fuck stadia.
I'm not at all surprised by Stadia dying. When I spent a few weeks trying it out it never felt quite right and the quality was never on par with the direct experience. I do think there's a market out there but I didn't have faith in Google to take advantage of it.
Re your wishes for AMD's hardware: As an engineer, I always get confused when people want "AI" acceleration. What exactly is that supposed to mean? I suspect you mean acceleration of specific matrix operations (e.g. multiplication). I would love to see AMD implement those, but the case you make for it here is pretty weak.
You state that DLSS 2 kind of proved the use of those operations. I strongly disagree with that. FSR 2.1 *is* clearly inferior to DLSS 2.3, but that's not a sound basis to conclude that matrix-mul acceleration is a worthwhile hardware feature – because hardware features come at a cost. And if you look at how close FSR 2.1 comes to DLSS 2.3, and also how even DLSS 2.3 deemphasized the naive AI approach DLSS 1.0 used – I think the opposite is clear: in the final implementation, the "AI" acceleration was not worth it. You'd get a better result investing in e.g. better raytracing support, and accept the slightly lower quality of FSR yet achieve a net result that's better.
In short: there are tradeoffs, and DLSS 2.3 doesn't exactly convincingly show that investing in those hardware matrix-mul units at the cost of cutting back on e.g. raytracing units is a wise trade off. If anything, it demonstrates the reverse: that AI is a great tool to quickly come up with great algorithms, but actually running straightforward inferencing algorithms isn't always efficient.
To be clear: I do believe such matrix-mul operations are super-useful and will prove their value in the long run; but simply that when we analyze their cost/benefits we should be sufficiently critical, and not blinded by the hype.
Incidentally, there are machine learning techniques to encourage convergence towards mostly sparse matrices or representations that achieve similar degrees of freedom via a few vector ops – and while those regularization techniques can come at a cost of reduced fit, they allow learning of models that may be better suited to run on non-"AI" hardware. AI as a technique is flexible enough and frankly vague enough to be useful even when there are some onerous usage restrictions – such as lack of "AI" hardware. And given just how much throughput a GPU needs to do anything like this at the scale and latency required, it's quite conceivable that judiciously restricting complexity is the winning strategy, not simply sledgehammering "AI" hardware into everything. That's not free either.
the grim spectre of Phil Harrison signals the death knell of yet another otherwise promising tech initiative. le sigh.
I think Cloud gaming is a great way of playing games to be honest. I just never would ONLY pay for it like Stadia. I love having the OPTION of using Cloud gaming because of GamePass, I use it more often than I want to admit on my Steam Deck.
I understand John's sentiment that Alex mentioned though.
One place i see cloud gaming working is MMO, it could remove some limitations they've had with timings and delays etc, with a subscription base you get the funding required to keep the service up.
Great video! I am interested in what a proper CPU is for 1440pTriples (Sim-Racing) while streaming with Streamlabs on the same PC with 1440p@60 and 23MBit/s bandwidth
Only wishing those Stadia specific ports could be released in public for SteamOS/Linux, particularly the online MMO one like Destiny 2.
I do not need an RTX 4090. Or RTX 4080. I got a steam deck and I'm happy.
the 2060 KO was a 2080 die cut down but without compute TOO crippled (from 2080 levels), IIRC?
1:21:24 Please test DLSS 3 Quality Mode vs Performance Mode.
The experience you have with cloud gaming will completely depend on your internet connection, and you need a much better connection that all the services advertise if you want the best out of it. It's not until I got 900mb fibre direct to my home that I can finally say I get basically a native experience on geforce now 3080 tier. My ping to the service through WiFi is a solid 7ms and with ethernet 4ms. It never goes above that ever at any time of the day and the experience is amazing, even if multiple people in the house are using the broadband heavily. I always read about people saying the severs slow down at peak time, but they don't, it's people internet providers that are slowing down at peak times. You really need fibre all the way to the home to enjoy it properly, zero copper (it kills the ping and gets congested the moment all your neighbours start using their services)