DF Direct Weekly #286: Xbox Ally X20 Tested, DLSS Frame Gen/DLSS 5 Perf Boost, Epic All-In On AI
Digital Foundry Direct Weekly
October 19, 2026
A year on from its initial launch, Xbox Ally gets a brilliant OLED upgrade for the new X20 edition but are we any closer to an actual Xbox-like experience? Meanwhile, Nvidia delivers faster frame-gen and DLSS 5 while Epic acts like a lightning rod for all the AI hate with its latest pivot.
Speakers Richard Leadbetter, Oliver McKenzie
Richard Leadbetter (0:02)
Hello there, and a warm welcome. Sincerely, to this, the latest edition of the DF Direct podcast, DF Direct Weekly. And yes, it's our 286th edition, not including specials and other things. Lots to discuss this week.
Plenty of actual news for once. Joining me first of all, John Lineman. Hi.
Oliver McKenzie (0:21)
Hey Rich, still loving your background. I'm seeing those Radeon 7s back there. I'm keeping an eye on them just to make sure that they don't go anywhere though.
Richard Leadbetter (0:29)
Now, they're there to stay. Maybe I can find some more. Who knows?
Also joining me, it's Oliver McKenzie. Hi.
Oliver McKenzie (0:39)
Yes, Rich. I invested in a couple of things for my background, including a hopefully plausible looking fake plant back there, and a couple of lights, but we'll see how it goes. This is a better week in terms of my workload. I'm feeling pretty good actually.
Richard Leadbetter (0:54)
Good stuff. Well, your workload is always tremendous, so that's pretty good news.
Good stuff. Before we go on, a quick shout out to our website presence, digitalfoundry.net. Will Judd is in command there. He's got the con and just wants to point out some of the stuff that we've been doing this week. I guess there was the Asus ROG Xbox Ally X20, which we'll talk about later in the show. I did a report on that, which was half talking about how great the HDR screen is, the OLED screen, and half just bemoaning Windows as a hand held. So yeah, things have changed, but not that much in the last year since the Ally X came out. Other things, well apparently Valve is actually calling time on the Index VR headset. Obviously everything is all in on the Steam Frame now. So the existing VR solution from Valve is being sold off at a discount. Will loves his peripherals, we've got a review of the Razer Basilisk V4 Pro. So lots going on, including lots of news and stuff. So if you want a more immediate Digital Foundry experience rather than waiting a week for a podcast, head over there. But let's crack on with the show. We had two pieces of news this week. First of all, that DLSS Frame Generation has been updated. There's now a new preset that you can enable with the Nvidia app, and it's called Preset C. And effectively, it's an alternative to Preset B. There's a lot of presets going on here. And it's effectively what you might call a quote unquote performance alternative for Frame Generation. That's interesting. Talk about that in a moment. There's also been, and this one caught my eye. I mean, it was kind of like projected when DLSS 5 was launched with NBA 2K27.
They were suggesting that the DLSS 5 processing time is going to be considerably reduced ergo a lower performance, ergo higher frame rates.
Oliver, that's kind of like interesting to me because I'm wondering to myself, if you've got this like massive optimization that's coming literal weeks after the launch, why didn't you just wait?
Oliver McKenzie (3:10)
Yeah, that's a good question.
Richard Leadbetter (3:13)
Because the overwhelming negative feedback to DLSS 5 wasn't actually the quote unquote AI slop because it was actually quite tastefully done in NBA, right? It was the fact that you were looking at, well, an RTX 5080, it was like 13 milliseconds of processing time. In a world where a 60 FPS budget is 60 milliseconds, that's kind of crazy.
Obviously, it's scaled according to resolution and whatnot. But yeah, I'm kind of a bit of a loss. And to launch with just one game, almost like a beta, which typically Nvidia doesn't do with major product launches.
Oliver McKenzie (3:55)
Yeah, yeah, yeah. I think there's a bit of a disconnect maybe between some people who are more familiar with how models tend to get optimized and more familiar with just neural networks and machine learning in general, not on a technical level, but just have seen progress in the past in these fields, where we kind of expect that if you have a model that is on a pretty good track in terms of optimization, like they say that at launch it was five times faster than it was when we saw it back in March, Rich.
Richard Leadbetter (4:26)
Right.
Oliver McKenzie (4:27)
You kind of assume that trend would continue and they would find more optimization space in the model, and that optimization would be more meaningful than your typical optimization of a lighting system in a video game or something, from like a production-ready level to a more optimized state. That would not be totally analogous there. So I do think that at least that element of it is something that, I kind of expected, and obviously Nvidia had their projections, this actually beats their projections in terms of what they expected.
53 more minutes of transcript below
Thousands of transcripts fetched by people building searchable podcast archives
Fetch the whole transcript
The demo key returns a sample episode in full, no card needed:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000797095378