00:00:05
Nvidia has just announced the GeForce 50
00:00:08
Series and dlss 4 technology I'm sure
00:00:10
many of you were watching the live
00:00:11
stream so it's time to give my thoughts
00:00:13
on what has been shown right here at CES
00:00:16
2025 in addition to all the specs
00:00:18
pricing performance claims and there was
00:00:20
even a lot of stuff that wasn't in that
00:00:22
live stream that you might have missed
00:00:24
as expected Nvidia focused on high-end
00:00:26
models for this launch announcing the
00:00:28
GeForce RTX 5090 the RTX 580 the RTX 570
00:00:33
TI and RTX 570 and it won't be long
00:00:36
until you can get your hands on one with
00:00:38
the RTX 580 and 590 launching first on
00:00:41
January 30th this will be followed by
00:00:43
the RTX 570 and 570 TI in February with
00:00:47
more 50 Series models coming later I'm
00:00:49
assuming 50 60s that sort of thing will
00:00:51
be announced down the track the flagship
00:00:53
model the RTX 590 packs a significant
00:00:56
increase in Hardware over its
00:00:57
predecessor the RTX 490 not only does
00:01:01
this GPU use nvidia's new Blackwell
00:01:03
architecture it also packs significantly
00:01:05
more cuticles more memory bandwidth and
00:01:07
a higher vram capacity the SM count has
00:01:10
increased from 128 with the 490 to a
00:01:13
whopping 170 with the RTX 590 a 33%
00:01:17
increase in the size of the core but
00:01:20
there's decreased GPU clocks going from
00:01:22
a boost of 2.5 GHz down to 2.4 GHz now
00:01:26
the memory subsystem is overhauled Now
00:01:28
featuring gddr 7ch technology on a huge
00:01:31
512-bit bus with this gddr 7 memory
00:01:34
clocked at 28 gbits per second memory
00:01:37
bandwidth comes out to
00:01:39
1,792 GB per second and the 80% increase
00:01:43
on the 409s bandwidth this is attached
00:01:45
to 32 gig of vram as well the most
00:01:47
Nvidia have provided on a consumer GPU
00:01:50
with consumer Blackwell still built on
00:01:52
tsmc 4 NM technology the board power
00:01:54
rating has risen to accommodate these
00:01:56
cores and memory rising to a mammoth 5
00:02:00
75 wats 28% higher than the 490 the RTX
00:02:03
580 uses a much smaller GPU
00:02:06
configuration in fact an entirely
00:02:08
different GPU die with just 84 SMS four
00:02:11
more than last gen's RTX 480 super
00:02:14
pretty much everything about this model
00:02:15
is half of the RTX 590 just under half
00:02:18
the Cuda cores half the memory bus width
00:02:21
to 256 bits half the vram capacity at
00:02:24
just 16 GB a little over half the memory
00:02:26
bandwidth at 960 GB per second through
00:02:29
the use of gddr 7 clocked at 30 GB per
00:02:32
second normally this is the
00:02:34
configuration you'd get with an RTX 5070
00:02:37
type of GPU which is historically where
00:02:39
Nvidia have positioned the model that's
00:02:41
half the size of the flagship but these
00:02:42
days it's a 80 in I guess a bit of a
00:02:45
disappointing move for gamers the RTX 57
00:02:48
TI is cut down further it falls to 70 SM
00:02:50
4 SM more than the last Generations 47
00:02:54
super however boost clocks are down from
00:02:56
2.61 GHz to 2.45 GHz now the memory bus
00:03:00
width Remains the Same as the RTX 580
00:03:02
256bit meaning it also has 16 GB of gddr
00:03:06
7 memory this time it's 28 GB per second
00:03:08
memory leading to 896 GB of bandwidth
00:03:12
the 570 TI is rated at 300 wats while
00:03:14
the 580 is rated at 360 WS then there's
00:03:18
the RTX 570 we're looking at 48 SMS here
00:03:21
which is just 2 sms more than the RTX
00:03:23
470 from the previous generation it also
00:03:26
has just a 192-bit memory bus and 12 GB
00:03:29
vram which I think is insufficient for
00:03:32
this tier of product in 2025 boost
00:03:34
clocks are only slightly higher than the
00:03:36
4070 and there's a total power rating of
00:03:39
250 wat all new GeForce 50 Series gpus
00:03:42
come with new connectivity including PCI
00:03:44
5.0 time6 for these announced models and
00:03:47
display port 2.1b uhb 20 a new revision
00:03:51
to display port that adds support for
00:03:53
active cables there's also HDMI 2.1 and
00:03:56
a single 16 pin power connector on each
00:03:58
model before getting into to even more
00:04:00
on these new Blackwell gpus a word from
00:04:02
our CES sponsor the harbor unboxed CES
00:04:05
coverage is brought to you by gigabyte
00:04:07
and the new range of x870 and z890
00:04:10
motherboards powering the latest AMD and
00:04:12
Intel processors gigaby z890 range
00:04:14
supports Intel's new core Ultra
00:04:16
processors while the x870 range supports
00:04:19
amd's ryzen 9000 series including the
00:04:21
gaming Flagship ryzen 7 9800 x3d all
00:04:24
models offer robust power delivery with
00:04:27
optimal cooling screwless m.2 storage
00:04:29
and aied overclocking software to ensure
00:04:32
you get the most out of your CPU the
00:04:34
Stacked IO offers various connectivity
00:04:36
options including Wi-Fi 7 up to 10 GB
00:04:39
Lan Thunderbolt 4 and plenty of USB
00:04:42
ports now a new feature of the AO top
00:04:44
models is the included utility that
00:04:46
allows you to train your own AI models
00:04:48
at home or for small/ large scale
00:04:51
businesses gigabytes x870 and Z 890
00:04:54
motherboards are available right now
00:04:56
from your favorite online retailers so
00:04:57
for more information please check the
00:04:59
links in the video description now let's
00:05:01
talk about the most important aspect of
00:05:03
Blackwell its price performance and
00:05:06
price to Performance ratio there's some
00:05:08
good stuff and some bad stuff in here
00:05:10
the GeForce RTX 590 is disgustingly
00:05:12
expensive as many people expected priced
00:05:15
at $2,000 us however the RTX 580 with
00:05:19
half the GPU Hardware is priced at
00:05:21
$1,000 us which is less ridiculous than
00:05:24
many of the rumors were expecting but
00:05:26
obviously still quite expensive I think
00:05:28
the surprises for me were the 570 TI
00:05:30
priced at
00:05:32
$750 and the 570 priced at
00:05:35
$550 both $50 lower than the same models
00:05:38
at launch in the previous generation for
00:05:41
performance claims Nvidia are using a
00:05:42
lot of their usual BS to obfuscate
00:05:45
actual gaming performance because RTX 50
00:05:47
gpus support dlss 4 multiframe
00:05:50
generation whereas RTX 40 gpus do not so
00:05:54
RTX 50 Series gpus are capable of
00:05:56
creating double the amount of frames
00:05:58
that previous gen models can
00:06:00
meaning any dlss supported games could
00:06:02
instantly be up to twice as fast on a 50
00:06:06
Series GPU than a 40 series GPU if we're
00:06:09
talking about DS S4 games and I'm using
00:06:11
air quotes here because the FPS number
00:06:14
will go up with dss4 but not necessarily
00:06:16
the latency and feel of the game that we
00:06:18
normally associate with performance in
00:06:21
the charts we did get Nvidia showed four
00:06:24
games and apps supporting dss4 cyberpunk
00:06:26
277 Allen Wave 2 Blackmouth Wukong and
00:06:29
called D5 render so I'll ignore those
00:06:32
because we have no idea how those games
00:06:34
actually perform in an Apples to Apples
00:06:36
sort of scenario so it's the other games
00:06:38
that matter immediately this makes the
00:06:40
claim that the RTX 5070 matches the RTX
00:06:43
490 in performance highly dubious
00:06:46
probably wrong most of the time Asar
00:06:48
claims it's twice as fast as a 4070
00:06:51
that's just not going to happen across
00:06:53
most games maybe you'll get that in
00:06:55
terms of frame rate when using dss4 but
00:06:58
certainly not when decide not to use
00:07:00
frame generation in some of the examples
00:07:02
the 5070 is around 20 to 30% faster than
00:07:05
the 4070 at 1440p when we do not include
00:07:09
multiframe generation this would make
00:07:11
the RTX 5070 a little faster than the
00:07:13
4070 super for about $50 less or
00:07:16
alternatively 20 to 30% faster than the
00:07:18
4070 at the same price depending on how
00:07:21
it Stacks up to the 470 super this could
00:07:23
be a reasonable though a little
00:07:25
underwhelming GPU 10% faster than the
00:07:27
470 super and around 9% % cheaper would
00:07:30
be a roughly 20% Improvement in cost per
00:07:33
frame Nvidia made similar claims for the
00:07:35
RTX 570 TI versus the 47 ETI at 1440p
00:07:39
somewhere in the 20 to 30% performance
00:07:42
Improvement range if we ignore
00:07:43
misleading deals S4 numbers those
00:07:46
numbers would be reduced a bit compared
00:07:48
to the 470 TI super which is 10% faster
00:07:51
than a 470 TI as it's $50 cheaper than
00:07:54
those gpus we're looking at a decent
00:07:56
cost per frame increase than the 470 TI
00:07:59
likely over 30% if nvidia's numbers are
00:08:01
accurate but that will fall back closer
00:08:03
to 20% relative to the newer 470 TI
00:08:05
super which was the much better graphics
00:08:08
card similar stuff for the RTX 580
00:08:11
relative to the RTX 480 around 20 to 30%
00:08:14
faster in those non dss4 games and given
00:08:16
the 480 super was only about 5% faster
00:08:18
than the 480 we're again going to get
00:08:20
about a 20% increase in cost per frame
00:08:23
going on nvidia's NDS S4 numbers it's
00:08:26
hard to get a good read on the
00:08:27
performance of the RTX 50 9 again the
00:08:30
non-dealers S4 numbers show a pretty
00:08:32
underwhelming gain 20 to 40% relative to
00:08:35
the RTX 490 the 90 is 25% more expensive
00:08:39
than the 490 so you'd want most games to
00:08:41
be closer to a 50% gain to get around a
00:08:44
20% Improvement in cost per frame
00:08:47
however nvidia's testing was done at 4K
00:08:49
using dlss performance mode in other
00:08:52
words rendering natively at 1080p I
00:08:54
suspect that feeding a GPU as huge as
00:08:57
the 5090 will be difficult at a 1080p
00:08:59
render resolution causing CPU
00:09:02
bottlenecks and underutilization of the
00:09:04
GPU in general this could be why we are
00:09:06
seeing more substantial improvements in
00:09:08
those path traced dss4 games relative to
00:09:12
the other RTX 50 Series gpus we've
00:09:14
looked at so far because I do expect
00:09:16
this card to be probably the largest
00:09:18
performance game gen over gen but to be
00:09:20
fair this could also be an issue for
00:09:21
gamers actually buying this GPU using
00:09:24
them in realistic conditions if you're
00:09:25
going to be playing at 1440p or 1080p
00:09:27
might be a bit harder to feed that GPU
00:09:29
so well as always we'll have to wait for
00:09:31
reviews to see how that plays out my gut
00:09:34
feel here and initial reaction is that
00:09:35
Nvidia seemed to be delivering around a
00:09:37
20% cost per frame Improvement and
00:09:39
therefore 20% value increase across the
00:09:42
board when comparing to RTX 40 super
00:09:44
gpus that is certainly better than the
00:09:46
RTX 40 series which in many situations
00:09:49
offered absolutely nothing for gamers in
00:09:51
terms of value but it also feels a bit
00:09:54
underwhelming when coming off such a bad
00:09:56
generation the value increase relative
00:09:58
to RTX 30 series gpus I think a lot of
00:10:01
people will be considering an upgrade
00:10:02
from that sort of GPU and if we compare
00:10:04
it to those gpus at the
00:10:06
MSRP the value increase isn't going to
00:10:08
be that large in many scenarios
00:10:09
considering it's a two Generation 4 and
00:10:12
a halfe step it's better than nothing
00:10:15
but to correct the Flop that was the RTX
00:10:17
40 series a value increase in the 30 to
00:10:20
40% range was probably needed 20% as it
00:10:23
stands is on the lower end of value
00:10:25
increases looking throughout nvidia's
00:10:27
history I don't want to critize ize it
00:10:29
too much because it could come down to
00:10:31
the actual margins and we don't have
00:10:33
enough performance data to go on right
00:10:35
now hopefully this apparent 20% value
00:10:37
increases more like 30% in reality in
00:10:39
which case this looks like a good
00:10:41
generation if the card's value Falls
00:10:43
closer to 20% or Worse even lower than
00:10:45
that it will be a disappointment but so
00:10:48
far there is enough to get at least a
00:10:49
little bit excited about and certainly
00:10:51
those prices aren't as horrendous as a
00:10:53
lot of people are expecting one of my
00:10:55
concerns with this generation is the
00:10:57
absolute Chasm Nvidia has created
00:10:58
between the highest tier model and the
00:11:01
next model down it's pretty absurd that
00:11:03
the RTX 5090 is basically double the
00:11:06
hardware of the RTX 580 it's almost like
00:11:08
we're back in the days of dual gpus
00:11:11
except this is a massive single GPU die
00:11:14
this makes the RTX 580 feel like quite a
00:11:17
weak product in comparison even though
00:11:19
there is a substantial price difference
00:11:21
to account for that gone of the days of
00:11:23
the ad0 class being just a small
00:11:25
reduction in capabilities relative to
00:11:27
the best model that Gap is growing
00:11:30
ensuring that gamers are forced to pay a
00:11:32
ludicrous amount of money to access
00:11:34
performance that's materially better
00:11:36
than the best of the 40 series this has
00:11:38
flowing effects for the rest of the
00:11:40
product stack which also become
00:11:41
comparatively weak relative to the best
00:11:43
GPU however there's at least some good
00:11:45
news here because in the previous
00:11:47
generation the 480 had 26% more cicor
00:11:50
than the 470 TI in this generation the
00:11:52
580 has 20% more criticals than the 570
00:11:55
TI as well as much closer memory specs
00:11:58
so the 570 TI isn't cut down as much
00:12:00
relative to the 5080 when compared to
00:12:02
what Nvidia was doing with the RTX 40
00:12:05
series launch let's move now to talking
00:12:07
about nvidia's Blackwell architecture
00:12:09
now this is an iteration on previous
00:12:11
architectures improving various aspects
00:12:13
that existed in previous generations
00:12:15
like rate tracing units and tensor
00:12:17
Nvidia spoke for a ridiculous amount of
00:12:20
time about AI so I won't bore you with
00:12:23
all that stuff but basically Blackwell
00:12:25
is powerful for AI in addition to
00:12:27
regular processing I know shocker and
00:12:29
new Nvidia gpus being advertised to be
00:12:31
the best for AI yet in more interesting
00:12:34
things the architecture has tighter
00:12:35
integration with the tensor cores inside
00:12:37
the SM improved Shader execution
00:12:39
reordering which is twice as efficient
00:12:41
and RT cores with twice the ray triangle
00:12:44
intersection rate and enhanced
00:12:45
compression to reduce the memory
00:12:47
footprint while rate tracing Nvidia
00:12:49
claims this will allow for more geometry
00:12:51
as well while R tracing Blackwell also
00:12:53
improves video encoding and decoding to
00:12:55
a new generation of the hardware engine
00:12:57
this is the first time we are getting a
00:12:59
video decoding engine Improvement since
00:13:01
the RTX 20 to 30 series upgrade the new
00:13:04
display engine is also much better and
00:13:06
allows for new multim monitor
00:13:07
configurations up to four 4K 165 HZ
00:13:10
monitors without DSC or 2 4K 360 HZ
00:13:14
monitors with DSC and of course various
00:13:16
combinations of those sorts of things
00:13:17
you can even do two 8K 100 HZ monitors
00:13:20
with dsse for example Nvidia are
00:13:22
releasing Founders editions for all the
00:13:25
models except the RTX 570 TI this Fe
00:13:28
model looks pretty similar to the 4090
00:13:30
FV cards in the general design except
00:13:33
now both fans are on the same side
00:13:36
amazingly Nvidia are using just a two-
00:13:38
slot design for the 5090 despite the
00:13:40
card being rated for 575 wats it'll be
00:13:43
very interesting to see how hot that
00:13:44
thing runs the RTX 580 appears to use a
00:13:47
very similar Founders Edition design
00:13:49
while the 5070 uses a scale down model
00:13:52
that isn't as long 242 mm long for that
00:13:55
version versus 304 mm for the higher
00:13:58
versions going on their spec
00:14:00
moving on to Features now and as part of
00:14:02
the presentation Nvidia announced dlss 4
00:14:04
the next generation of their upscaling
00:14:06
technology the headlining feature here
00:14:08
is frame generation with its ability to
00:14:10
generate three frames for every one real
00:14:12
frame now instead of generating one
00:14:14
frame for every one real frame I guess
00:14:17
this is something but let's be honest
00:14:19
there are significant question marks
00:14:21
here over latency and image quality when
00:14:23
generating one frame every second frame
00:14:26
50% of the time you were seeing a real
00:14:28
frame but when you're generating three
00:14:30
frames only 25% of the time you are
00:14:33
seeing a real frame 75% of the time it's
00:14:35
those generated frames so these
00:14:37
generator frames had better be amazing
00:14:39
quality otherwise you will see a ton of
00:14:41
artifacts a lot of the time something
00:14:43
that will be an issue unless the quality
00:14:45
of generated frames is significantly
00:14:47
overhauled relative to dss3 and they are
00:14:50
making some claims that it is the other
00:14:52
issue is using multiframe rendering to
00:14:54
hack our way to high frame rate gaming
00:14:57
in video showed a demo where without DLS
00:14:59
S 4 a gamer is running at 30 FPS but
00:15:01
then over 240 FPS with dss4 enabled now
00:15:05
of course not even dlss upscaling was
00:15:07
enabled in the 30fps example if you turn
00:15:09
on just upscaling the frame rate was
00:15:12
reasonable but I really don't want to
00:15:13
see any situations where generating
00:15:15
three frames is used to get a 30 FPS
00:15:18
game up to 120 FPS that would be
00:15:20
absolutely horrific in terms of input
00:15:22
latency dss4 is exclusive to Blackwell
00:15:26
gpus however only for this multiframe
00:15:28
rendering component deals SV also
00:15:31
features enhancements to regular frame
00:15:33
generation super resolution and Ray
00:15:35
reconstruction with the same support for
00:15:37
those features as we currently have the
00:15:40
restriction on multiframe rendering is
00:15:42
due to what Nvidia calls enhanced
00:15:44
Hardware flip metering capabilities in
00:15:46
the Blackwell architecture which moves
00:15:48
some of the things that were done on the
00:15:50
CPU for frame generation on the RTX 40
00:15:52
series into the GPU on Blackwell Nvidia
00:15:55
says this is required for a smooth
00:15:57
experience with this Tech which I hope
00:15:59
is true otherwise RTX 40 series owners
00:16:01
are getting screwed the quality
00:16:03
enhancements to upscaling and other
00:16:04
aspects to dss4 are significant
00:16:06
according to Nvidia thanks to the use of
00:16:08
what they say are Transformers Nvidia
00:16:11
claims this will improve image quality
00:16:13
with improved temporal stability less
00:16:15
ghosting and higher detail in motion
00:16:18
that sounds pretty exciting to me and
00:16:19
actually I'm more interested in the
00:16:21
upscaling quality improvements in DS S4
00:16:24
than multiframe rendering Nvidia also
00:16:27
say the new frame gen model is 40%
00:16:29
faster and uses 30% less vram for
00:16:32
generating multiple frames the AI
00:16:34
algorithm only needs to be run once per
00:16:36
rendered frame Optical flow in the frame
00:16:38
gen component is replaced with an
00:16:40
efficient AI Hardware model which throws
00:16:42
up questions as to why frame generation
00:16:44
is still exclusive to the 40 series
00:16:46
given it was Hardware restricted due to
00:16:48
Optical flow accelerators and a love
00:16:50
lace maybe that AI model could now run
00:16:52
on older gpus but no still restricted
00:16:56
what's not clear is whether the new
00:16:57
Transformer model for dlss has a
00:16:59
performance impact and whether this
00:17:01
performance impact is the same for all
00:17:02
gpus for example they claim frame
00:17:04
generation is now faster but is that on
00:17:07
both RTX 50 and RTX 40 gpus is there a
00:17:10
performance impact from using the trans
00:17:12
performer model for upscaling especially
00:17:14
on those lower tier cards many questions
00:17:16
to be answered in testing the fact that
00:17:19
75 games and apps are coming at launch
00:17:21
with dss4 support is quite impressive
00:17:23
showing Nvidia has learned that you need
00:17:26
widespread game support to make features
00:17:28
worthwhile I don't know if this means
00:17:30
there will be significant game updates
00:17:31
at launch for all of those Tiles at once
00:17:33
or whether Nvidia has found some way to
00:17:35
automatically add dss4 to those games
00:17:37
but a quick look through the list it
00:17:38
isn't every DS S3 game and some games
00:17:41
are only adding in DS S4 support later
00:17:43
so it does suggest a widespread effort
00:17:46
from game developers to add in this
00:17:47
feature in addition Nvidia are bundling
00:17:50
in an override feature in the Nvidia app
00:17:52
which will act like a dll swap allowing
00:17:54
upgrades or overrides as they call them
00:17:57
to the latest DSS models for frame
00:17:59
generation super resolution and Ray
00:18:01
reconstruction it will also allow frame
00:18:03
generation to turn into multiframe
00:18:05
generation for 50 Series owners Nvidia
00:18:08
has announced reflex 2 as well which is
00:18:10
essentially bringing frame warp from VR
00:18:12
Gaming to regular PC gaming for the
00:18:15
first time basically the idea here is
00:18:17
instead of showing the rendered frame
00:18:19
based on Old input reflex 2 will take a
00:18:22
more recent input sample do some
00:18:25
calculations on the CPU and warp / shift
00:18:28
the rendered frame
00:18:29
based on that input in a sense this
00:18:32
fakes a frame update and applies it to
00:18:34
the rendered frame reducing perceived
00:18:36
latency for actions that shift the
00:18:38
camera because real input is being taken
00:18:41
into account to warp the frame by the
00:18:43
correct amount it will not reduce
00:18:45
latency for game actions like firing
00:18:47
your gun for example because that cannot
00:18:48
be warped but for camera actions you get
00:18:51
I don't even know if faked is the right
00:18:52
word to use for that it it could reduce
00:18:54
latency if it works well it's a cool
00:18:57
concept I guess it remains to be seen
00:18:58
how it actually works in games is it
00:19:00
going to feel nice to play a game with
00:19:02
reflex 2 on or will it feel a bit floaty
00:19:04
and weird if the warp calculations
00:19:06
aren't quite right it's hard to say at
00:19:08
this point although it's good to see
00:19:09
Nvidia claiming gains for both CPU and
00:19:11
GPU limited scenarios because previous
00:19:13
versions of reflex didn't really do much
00:19:15
if you were CPU limited there's also the
00:19:17
question of image quality and while
00:19:19
Nvidia claimed they are using inpainting
00:19:21
to solve issues with gaps in the Warped
00:19:23
render every single frame is affected by
00:19:26
this warp so if the quality of the warp
00:19:28
is bad bad this will make the entire
00:19:30
game look bad all of the time a little
00:19:32
different to what you see from frame gen
00:19:34
where that only affects a every second
00:19:36
frame or now three out of four frames I
00:19:38
guess reflex 2 is first coming to the
00:19:40
finals and valerant soon with support
00:19:42
for RTX 50 Series gpus initially and
00:19:45
other RTX gpus later so it doesn't sound
00:19:48
like this will be something restricted
00:19:49
to new Blackwell models however it's
00:19:51
important to note here that frame warp
00:19:53
in reflex 2 will not solve latency
00:19:55
issues with dlss frame generation
00:19:57
especially multiframe gener because
00:19:59
right now there is no indication it will
00:20:01
be integrated into games that support
00:20:03
dss4 multi or single frame generation it
00:20:06
might not even work together for all I
00:20:07
know Blackwell also supports RTX neural
00:20:10
shaders and RTX neural faces though I'm
00:20:13
not sure how soon we'll actually be
00:20:14
seeing those in games so it seems more
00:20:16
of a future looking thing for now Nvidia
00:20:19
claims that RTX neural shaders can be
00:20:20
used to compress textures by up to 7
00:20:22
times saving memory which is perhaps why
00:20:25
they're still being a bit stingy with
00:20:26
vram this generation but of course this
00:20:28
feature is absolutely useless in games
00:20:30
that don't use it making this not an
00:20:32
actual excuse for low vrm amounts on the
00:20:35
RTX 570 in particular so that is pretty
00:20:38
much everything Nvidia has announced at
00:20:40
CES 2025 the new lineup of GeForce RTX
00:20:43
50 Series graphics cards looks
00:20:45
reasonable at first glance but certainly
00:20:47
not to the level that blew me away
00:20:49
performance gains look reasonable when
00:20:50
you remove all the BS marketing stuff
00:20:52
and you features the value increase
00:20:54
looks modest to okay and the hardware
00:20:56
itself looks impressive the exception to
00:20:59
this would be the RTX 5070 which has
00:21:01
just 12 GB of vram at $550 us too low
00:21:04
for that class of graphics card in 2025
00:21:07
even though Nvidia are claiming reduced
00:21:09
memory usage through some of these new
00:21:11
features there are some games now where
00:21:14
12 GB is insufficient even without dlss
00:21:17
applied and most of the vram savings
00:21:19
come through dlss for model updates if
00:21:22
the base game still doesn't work with 12
00:21:24
GB of vram it doesn't matter if frame
00:21:27
generation is a little more memory
00:21:29
efficient this card really needed 16 gig
00:21:31
what's good to see is that pricing isn't
00:21:33
completely ridiculous or outrageous this
00:21:35
doesn't appear to be a repeat of the RTX
00:21:37
40 series where Gamers were treated to
00:21:39
little to no increase in value and
00:21:41
shockingly high prices for some models
00:21:43
it was a pleasant surprise to see that
00:21:45
in the same tier at least the RTX 570 TI
00:21:48
is now $50 cheaper and the RTX 570
00:21:51
matches the updated price of the 4070
00:21:54
after the 4070 super was launched of
00:21:57
course we still need to see how
00:21:58
performance games from those models to
00:22:00
get the required value increase but
00:22:02
Nvidia hasn't just jacked up the prices
00:22:04
like a lot of people are expecting I was
00:22:06
also impressed with dss4 claims but more
00:22:09
so with the model updates to improve the
00:22:11
quality of upscaling at least in theory
00:22:13
I haven't tested it yet the ability to
00:22:16
upgrade the dlss model in existing games
00:22:18
through an Nvidia app Feature is also
00:22:20
quite neat given many gamers were
00:22:22
performing manual DL swaps as for
00:22:24
multiframe Generation not solding that
00:22:27
feature yet given the implication for
00:22:28
image quality and latency especially if
00:22:31
it's used as a crutch to allow for
00:22:33
Native rendering at an even lower frame
00:22:35
rate previously I found with frame
00:22:36
generation at least for single frame
00:22:38
generation that you really need a
00:22:40
certain type of game for frame
00:22:42
generation to work well Works a little
00:22:44
better in third person games a little
00:22:46
worse in firstperson games so is that
00:22:48
still going to be the case with
00:22:49
multiframe generation is it going to
00:22:50
make it even worse for firstperson games
00:22:53
a lot of question marks there not
00:22:54
totally sold on it really have to test
00:22:56
that one personally and I absolutely
00:22:59
hate that we're now in a situation where
00:23:01
Nvidia can just claim ridiculous
00:23:03
performance games by generating more and
00:23:05
more frames maybe the next Generation
00:23:07
they'll be generating 100 frames so will
00:23:09
now be whatever it will be 20 times
00:23:11
faster or something which will be just
00:23:12
just silly so anyway that's it lot of
00:23:15
stuff that Nvidia announced here at CES
00:23:18
2025 feels like I've been standing here
00:23:20
forever talking about GeForce features
00:23:22
but yeah lots of exciting stuff to be
00:23:24
testing the performance the features and
00:23:26
it's not coming too far from now so
00:23:28
Steve had better get busy updating his
00:23:29
benchmarks to test all these new gpus uh
00:23:32
if you want to support our testing here
00:23:34
at Hardware unboxed and support the
00:23:36
channel we've got our patreon page links
00:23:38
to that is in the description below if
00:23:39
you sign up you get access to our
00:23:40
Discord chat which like with the am
00:23:42
announcement earlier today lots of
00:23:43
discussion about the announcement of the
00:23:45
GeForce 50 Series in there while it was
00:23:47
going on was checking in there seemed
00:23:49
like a bit of fun was happening so yeah
00:23:51
anyway that's it for this one thanks to
00:23:52
gigabyte for bringing us to CES and I'll
00:23:55
catch you in the next one
