00:00:02
okay so obviously one of the big um
00:00:04
drawers of CES 2025 is going to be the
00:00:07
reveal of the black World line of RTX uh
00:00:09
gpus from Nvidia and uh well what could
00:00:12
I say there's been a massive uh array of
00:00:15
uh Tech spec leaks uh almost the entire
00:00:18
stack part from the 50 60 at this point
00:00:20
seems to be pretty much locked in you
00:00:22
can find those specs all over the
00:00:24
internet now what's interesting here is
00:00:27
well we can sum it up with this question
00:00:29
from uh d hard x86 he says hello DF I've
00:00:32
been seeing chatter from PC Gamer via
00:00:34
Ino 3D and mad Reddit posts about neural
00:00:38
rendering being the next big step in
00:00:40
nvidia's technology road map have you
00:00:42
seen or heard Nvidia mention neural
00:00:44
rendering or is this news to you as well
00:00:47
the thought of machine learning models
00:00:49
not only to upscale but to create 3D
00:00:51
environments in real time exciting stuff
00:00:54
now let's talk about the Ino 3D leak
00:00:57
where basically there seems to be uh a
00:01:00
range of bullet points which I suspect
00:01:02
that Nvidia isn't particularly happy to
00:01:04
have leaked however with that said they
00:01:07
are still sufficiently vague enough to
00:01:11
act almost as a teaser of sorts uh
00:01:14
they're talking about advanced dlss
00:01:16
technology well you know we would expect
00:01:19
a new generation of cards to come with a
00:01:21
new dlss enhanced rate racing ditto um
00:01:24
AI accelerated Graphics well what's that
00:01:26
all about neural rendering capabilities
00:01:30
um this is all new and interesting stuff
00:01:33
um they've also talked about well
00:01:35
there's actually a some sort of
00:01:37
screenshot of an in3d PR blast here uh
00:01:41
which which seems to be offering up uh
00:01:43
some further details newal rendering
00:01:46
capabilities revolutionizing how
00:01:48
Graphics are processed and
00:01:51
displayed I've got to go to you for your
00:01:53
your thoughts on this one Oliver yeah I
00:01:56
mean it could mean a lot of things I
00:01:57
don't really know it could mean
00:01:59
integration of AI models at different
00:02:01
steps of the rendering process uh it it
00:02:04
could mean something like certain
00:02:07
postprocess techniques brought in but
00:02:09
given that they're talking about a
00:02:10
revolution here doesn't seem like a
00:02:13
revolution to me to throw in AI as a
00:02:15
postprocess potentially the other things
00:02:17
here I mean Advanced dlss Tech I think
00:02:20
that could just be some improvement to
00:02:22
image upscaling and frame gem
00:02:24
potentially um that's sort of what's
00:02:26
implied here because they're talking
00:02:27
about better image quality better frame
00:02:29
rates um that would make more sense
00:02:31
there I am really curious about the AI
00:02:33
video up stealing that they're talking
00:02:34
about here because could that be
00:02:36
something similar to Topaz for instance
00:02:38
but faster possibly potentially in real
00:02:41
time that would be very exciting
00:02:43
potentially you know you're streaming a
00:02:46
game let's just say and you're getting
00:02:48
this filter this AI filter that's
00:02:50
running on top of it on your RTX GPU and
00:02:53
it's improving the video quality of that
00:02:54
stream substantially in a way that we
00:02:55
don't see with nvidia's current solution
00:02:57
there where it's a more minor
00:02:58
enhancement in line with other kinds of
00:03:01
video enhancement techniques that we see
00:03:03
currently um or maybe just in general
00:03:05
with internet stream content that could
00:03:07
be quite nice I'd also be kind of
00:03:09
curious to see how much this ends up
00:03:11
being supported by prior RTX cards older
00:03:14
40 series 30 series 20 series um nothing
00:03:17
here like
00:03:19
screams uh 50 Series or newer
00:03:22
potentially for that extra functionality
00:03:24
again I don't know but with the 40
00:03:25
series you had dss3 which for the
00:03:27
extreme generation component required
00:03:30
um more intensive uh higher-end Optical
00:03:33
flow acceleration which was provided by
00:03:35
a dedicated unit on that card so um that
00:03:38
necessity presumably maybe wouldn't be
00:03:41
here for some of these other things
00:03:42
maybe they would apply to older RTX
00:03:43
series cards as well so yeah I I'm not
00:03:47
quite too sure about that neural
00:03:48
rendering thing I'm I'm just very
00:03:49
curious like the more I think about it I
00:03:51
am really curious to see what Nvidia
00:03:53
does and I think it could be you know it
00:03:56
could go in any number of different
00:03:57
directions but I think what they will
00:03:58
demo with it will probably be very
00:04:00
interesting and very
00:04:01
compelling I mean from my perspective um
00:04:05
AI Hardware on gpus has basically been
00:04:07
used for vendor specific stuff like dlss
00:04:11
right um I think the way forward is just
00:04:12
to open it all up we've had Direct ml
00:04:15
which basically doesn't seem to have
00:04:17
done anything doesn't seem to have been
00:04:19
any impact whatsoever I think there
00:04:21
probably needs to be some kind of new
00:04:23
API to actually access this hardware and
00:04:26
be able to uh get more out of it so I
00:04:29
suspect expect maybe something along
00:04:30
those lines something to actually uh
00:04:33
make this Hardware more useful to
00:04:35
developers you know as opposed to
00:04:37
effectively you know operating almost
00:04:40
like a fixed function block really you
00:04:42
know you've got all of these tensor
00:04:43
cores in there and they're being used
00:04:45
for dlss and dlss only I'd love to see
00:04:49
um all of that Hardware being made
00:04:51
available to developers to use as they
00:04:54
wish uh that would be fantastic I mean
00:04:56
let's take a look at some of the other
00:04:57
bullet points here I agree with you
00:04:59
Advanced de DS technology well you know
00:05:02
um dlss has basically become a musthave
00:05:04
feature for um people buying graphics
00:05:07
cards right so what do you do you don't
00:05:09
sit on your lws you you basically you
00:05:11
push harder you do better that's
00:05:14
basically been what Nvidia has done
00:05:16
across you know ever since during came
00:05:18
out you know they've they've iterated
00:05:19
they've
00:05:20
improved is this going to be an
00:05:22
iteration or is it going to be something
00:05:23
much much bigger I can't wait to find
00:05:25
out enhanced rate racing um it's really
00:05:28
interesting um the backlash against raid
00:05:31
facing continues but then you know you
00:05:33
listen to The Mark cery talk for
00:05:36
PlayStation 5 professional in the
00:05:37
interview that you did Oliver we we'll
00:05:39
talk a bit more about that later you
00:05:41
know basically the people that matter
00:05:43
the The Architects of the hardware of
00:05:45
the future realizing that rasterization
00:05:48
is kind of reaching its limits and R
00:05:50
tracing is you know the Wild West it's
00:05:52
the new frontier and we've kind of just
00:05:55
started that Journey
00:05:57
remarkably um you know AI accelerated
00:06:00
Graphics again I I still think it's
00:06:03
about giving that Hardware to developers
00:06:06
um yeah so you know there's a lot to go
00:06:08
a lot to to to what can I say there's a
00:06:11
lot of potential here I just really want
00:06:13
to see what the presentations are going
00:06:15
to come uh going to deliver in uh in
00:06:17
January there um John any thoughts on
00:06:20
this yeah when talking about neural
00:06:23
rendering capabilities and additional
00:06:25
uses of the AI
00:06:27
acceleration you you really have to
00:06:29
start thinking the different pipelines
00:06:30
going into what what makes up a frame
00:06:33
right we're already seeing dlss
00:06:35
intercept certain things and taking
00:06:37
control of that final output to deliver
00:06:39
improved image quality but what if you
00:06:41
insert it earlier into the process or
00:06:43
for some other sorts
00:06:45
of sort of rendering tasks that are
00:06:47
happening and so this just a random
00:06:49
example and with their focus on rate
00:06:51
tracing wouldn't even necessarily make
00:06:53
sense here but imagine if you were able
00:06:55
to intercept the uh the pipeline for
00:06:57
displaying screen space ref flection
00:06:59
right and use machine learning to sort
00:07:01
of extend out where what it's actually
00:07:04
capable of drawing so it's no longer
00:07:06
just like SSR it's actually like drawing
00:07:08
information from somewhere else using
00:07:10
machine learning to sort of extrapolate
00:07:12
what may or may not be outside the frame
00:07:14
you know or inserting it into like a
00:07:16
pixel Shader of some sort and sort of
00:07:19
like interpretate interpreting the
00:07:21
results and maybe enhancing it you know
00:07:24
basically finding ways to like stick it
00:07:25
into the various like rendering pipeline
00:07:27
bits to improve the end result I mean
00:07:30
we've seen them dabbling this with
00:07:31
things like Omniverse and RTX remix I
00:07:34
believe right so that was kind of a
00:07:36
pitch that we've seen before but
00:07:38
applying this at a larger scale to
00:07:40
shipped games and such could be very
00:07:43
interesting I think for the display of
00:07:45
graphics and would be a cool way to use
00:07:47
these sort of AI features that have
00:07:49
largely gone mostly untapped I would say
00:07:52
at this point like the specifications
00:07:55
and what it's capable of doing s of
00:07:56
outstrip what's actually being used in
00:07:58
games at this point
00:08:00
and with the Next Generation coming up
00:08:01
there's going to be even more Headroom
00:08:02
there so it makes sense to leverage that
00:08:04
in new ways uh which I guess we'll find
00:08:08
out what that is soon enough mhm the
00:08:10
only question mark for me really is the
00:08:12
extent to which these car uh these
00:08:13
features are going to actually be uh
00:08:15
capable of running on older Hardware um
00:08:18
it sounds like there's a lot of
00:08:19
state-of-the-art Technology coming right
00:08:22
and um I think the danger would be in um
00:08:26
basically only allowing it to run on the
00:08:28
new car I suspect maybe you know some
00:08:31
stuff possibly will take some time to
00:08:35
proliferate into games and I guess that
00:08:37
might be okay if it only ran on 50
00:08:39
Series and above but you know a new dlss
00:08:43
for example it would be a tough sale for
00:08:45
that not to be available on 40 series
00:08:48
especially bearing in mind how
00:08:49
contentious 40 series was across the
00:08:51
board in terms of its pricing so I think
00:08:55
some sort of level of balancing between
00:08:57
what you can run on 50 Series and what
00:08:59
you you can run on anything up to the
00:09:00
old 20 series is going to be required
00:09:02
there uh to to to make a sort of
00:09:05
crowd-pleasing pitch from Nvidia uh in
00:09:08
that respect um I don't think you could
00:09:11
have a dss4 that only ran and again I
00:09:14
was going to say that only ran on 50
00:09:16
Series when you know maybe that's not
00:09:19
the case bearing my dss3 only ran on 40
00:09:22
series but that was introducing a
00:09:24
fundamentally very different the frame
00:09:27
component of D lss3 yeah frame component
00:09:30
mhm I mean with this I mean it's all
00:09:33
about increasing I would imagine
00:09:34
efficiency to produce better looking
00:09:36
imagery at a lower performance cost
00:09:38
right yeah and you'd expect that model
00:09:40
to be sort of backported to uh to older
00:09:43
N Video Hardware as well that's the only
00:09:46
sort of slightly worrying thing that I'm
00:09:48
thinking about at the moment with that
00:09:50
said you know 40 series hugely
00:09:53
controversial in terms of um you know
00:09:56
user reaction online however it's sold
00:09:59
massive amounts of units which is is
00:10:01
quite
00:10:02
remarkable um but yeah well what can I
00:10:05
say um I think the Halo product
00:10:08
continues to work by the way when you
00:10:10
have highend products like the 4090 even
00:10:12
though the price was ridiculous people
00:10:14
look at that and they strive for that
00:10:16
they try to they buy up as close as they
00:10:18
can get to it well it works I actually
00:10:21
had a DM from one of the supporters
00:10:23
saying that the whole economics of
00:10:25
buying a graphics card um he buys the
00:10:28
new gen generation every time it comes
00:10:30
out and sells the old one because you
00:10:33
know what what he's suggesting is that
00:10:35
while the prices of the newer products
00:10:37
have gone up or you know or best the
00:10:40
same um he's suggesting that what this
00:10:43
is uh causing is um higher residual
00:10:47
prices for the older generation cards so
00:10:50
he's actually suggesting a different
00:10:51
economics model you know typically uh a
00:10:54
user is expected to buy a graphics card
00:10:56
and keep it for like you know three or
00:10:57
four years let's say but he's saying he
00:10:59
buys every single generation simply
00:11:02
because the price uh that he gets back
00:11:04
from his existing GPU is is
00:11:07
significantly higher therefore the
00:11:10
fundamental cost of the new GPU is quote
00:11:12
unquote lower which is certainly an
00:11:14
interesting way to look at
00:11:16
things but anyway uh CES 2025 Nvidia uh
00:11:21
they're going to be running the keynote
00:11:22
there we're expecting to see some
00:11:24
amazing stuff I hope to see some amazing
00:11:26
stuff obviously we've had that slight
00:11:28
tease
00:11:30
um of The Witcher 4 cinematic which is
00:11:32
apparently running um pre-
00:11:35
rendered doesn't say much to me CLE but
00:11:39
I guess we'll get to see some surprises
00:11:40
on that front too hopefully but only
00:11:43
time will tell
