on one data point and a drastically different execution of render resources. Bubsy 3D would render at 100+ FPS before DLSS. 18ms added on is literally nothing
Dynamic 4K isn't an unrealistic expectation for games like Super Mario Bros. Wonder. In fact I might say it's likely to get an NG patch;
LiC and others already looked at Nvidia data and Linux commits. T239 does not feature DLA and I'm puzzled as why he added that to his video (As, iirc, Oldpuck asked the same question here ages ago based on his conversation with Rich and we were able to see that T239 lacked DLA). Guess it's just something he didn't could cut from his video.looks like t239 has a DL accelerator
(i hope)
I mean, BoTW is a Wii U game. It shouldn't be difficult for Nintendo to run it at 4K60 DLSS on the new hardware. Although the reports of BoTW 4K60 goes against Rich finding of 18ms 4K DLSS frametime cost, so that's something to keep in mind.
BOTW runs at 900p on switch, not a crazy thought that it could just be native 4k60, no DLSS involved.
you're right, I miss read that.
Does commits indicating DL accelerator show up for other chips? Just wanted to make sure it is something we would/should find in Linux commits.
Easy findings to disregard when things don't, in fact, point to 8nm.
I mistyped, I meant to put 3.45TF. Which, no, isn't expecting clock speeds to exceed what we saw in the leak. In fact, speeds of up to 4TF were explicitly present, I just doubt that's the target for production. Mockery is not, I think, productive.It seems the more the time passes without Nintendo announcing the Switch 2, the higher the clocks go
Handheld is already at ~720MHz, and docked [to reach 4.35TF] has even passed the ultimate best (and very unlikely) scenario, with that "stress test clock" at 1.38GHz.
If the announcement is planned to 2024, then we'll get to PS5 clocks before the year ends
This mirrors my take away. I basically believe his results show the ground floor.
The DLA mention is a technical misstep of mine. I couldn't locate the DLA stuff, and told Rich that it wasn't eliminated
This is something it could be done, yes. Either a spatial pass after the render pipeline or some simple integer upscaling.
Ah, that explains it. Thank you for info.
If the guideline exists, it's evident that it is relatevely new and we will see more games following it with the passing of time.
Yeah, this is actually an incredible sign for what we're about to get. This little chip actually does put up with these games just fine without any actual optimizations on the table, and of course... The comments on the video are full of people in denial, how charming.
The 720p and above results were already good enough for them, I suppose. I really hope they make another video with Alan Wake 2 soon, perhaps upscaling from there...
I've never seen anyone say that 4K 60 fps would be anything close to a standard for current-gen games/ports, so I don't think the DF video really disproved any expectations here (tenuous PC comparison aside). 1440p 30 fps versions of games will still be very impressive (and some heavier titles will likely dip lower). Instead I'm seeing people go too far in the other direction and conclude that there won't be 4K games at all, which is just wrong.
DLSS 720p>4K is like twice as expensive in frametime as DLSS 720p>1440p going by NVIDIA's documentation so I would expect 720p>1440p to be the standard.
But 18ms sounds incredibly high compared to NVIDIA's documentation.
Obviously these chips are all much stronger than the Switch 2, but the gap shouldn't be this large unless DLSS starts scaling really badly at low spec.
If it does, hopefully NVIDIA can develop a pruned DLSS neural network for the Switch 2... We'll see if they're interested as pruned neural networks are really expensive (in terms of dollars and manpower) to create and the results are less good (obviously)
It’s not so surprising from a raw computational point of view. Just to pick a GPU from that table, say the 4090: it has 128 SMs and is clocked, in the Lovelace white paper, at 2520 Hz. By comparison, the underclocked 2050 in the video has 16 SMs and is clocked at 750 MHz. The 0.51 ms in the table for the 4090 would scale to 13.7 ms on this card, and that’s not the only consideration: memory bandwidth, post-processing at output resolution, unknown clock speeds in the DLSS programming guide table - pick your poison. 18 ms seems perfectly reasonable to me.
yeah, after reading through more of the replies in this thread, it does seem like the consensus after DF's analysis is more positive than i thought. maybe read too many YT comments lol.
I recommend taking what Tech_Reve posts with a healthy grain of salt, since Tech_Reve did posted rumours with factually incorrect and likely incorrect information. I've put a couple of examples below.
One of the slides from the Microsoft leaks explicitly mentioned co-designing the GPU with AMD or licencing the Navi 5x GPU IP from AMD.
And considering AMD mentioned being able to design Arm based SoCs for AMD's customers, and AMD rumoured to design Arm based SoCs for PCs, the rumours are probably not accurate with respect to Microsoft.
It’s convolutional, not multilayer perceptrons.
Eh, we got some Holidays next week. That should keep us occupied. I doubt they would preclude an announcement, since these are US Holidays.
Oh, we're the same age.
