Hey what’s up guys, TECHOSAUR here. So AMD has just released FSR 3 As an open-source project for DirectX 12
and Unreal Engine 5. This move could significantly Make it easier for developers
to integrate it into more games. And just a day
after, a great title from SEGA Like a Dragon Gaiden
The Man Who Erased His Name, Has received an
update adding FSR 3 upscaler. So I wanted to do an
in-depth performance analysis With it ON versus OFF along
with Frame Generation at the lowest TDP necessarily to maintain 60fps
on both the LEGION GO and the ROG ALLY. However, the catch
is that during my testing I encountered Some mysterious and cool findings, which I really wanted to share with you guys, So I hope you are as excited as I am. So in this video, we are going to have a set of different tests, but for the first one, I had to set the LEGION GO memory speed at 6400MT/s To match it with the ALLY’s. To have an adequate testing environment for accurate results, I used the exact most recent 780m GPU driver version for both devices. And regarding BIOS, It's updated on both of them
with the official latest versions. And I've set the dedicated GPU
VRAM as 4G, Which is quite sufficient to run the game
without any issues or stuttering. Finally, for APU TDP,
I’ll set it as 15w. That's because at lower than that I noticed some issues like having
a cracked sound or some stuttering, So I opted to use 15w, which is a good
all-around balance between Optimal performance with good power saving. In-game graphics settings, I've set the resolution As 1080p and selected
the pre-defined Medium setting preset, To have a good balance
between image fidelity and performance. So that will be the base
graphics settings across all tests We are about to see. But in this initial one specifically, I'm
going to switch between different configs Of FSR BALANCE ON versus OFF
along with Frame Generation To get a better understanding
of the difference. So with everything
set up, Let's dive right in. We see here that at native resolution, Both are quite struggling to hit 30fps
but with FSR 3 in BALANCE mode We got over 50fps and adding
Frame Generation We get over 60fps easily
while still being at 15w TDP. Surprisingly though, even with the same
memory speed and drivers, The LEGION GO got a better performance in all three different scenarios. As we can clearly see here,
there is a gap of up to 7 to 8 fps In the frame generation comparison. At this point, they have the same APU And drivers and I even have a fresh Windows installation On both as well as the same running TDP. Moreover, we have set the LEGION GO
memory speed at 6400MT/s To match it with the ALLY’s, yet
we see it performing better. That got me really wondering
what's going on, So I started to dig into this matter
by first checking the LEGION GO Default Memory timings for the 6400MT/s mode,
and here we can see On the highlighted timings that LENOVO or the memory vendor has defined Some Timings a little bit tighter
than how it is on the ALLY. Though that is easily possible to modify
on the ALLY by using SMOKELESS TOOl. I have an explanation Video guide on
how to do it step-by-step That you can check it in here. But now I'm showing you the SPD and the non-SPD timings That I had to change to match it with the LEGION GO ones, So here is the result. I tried as much as possible,
but some are not available to modify In the SMOKELESS TOOL, At least they are pretty close now. Still in SMOKELESS TOOL, by navigating
in the SMU common options menu, I found a way to activate
STAPM thermal policy, which is an option That is a present on the LEGION GO
default BIOS menu, But absent on the ALLY’s one, so
I could at least activate it in this way. I've previously covered the difference
between STAPM and STT in another video So you can watch it in here, Usually, STAPM got the better performance over STT. In addition to all of that, Found in CPU common options,
I disabled the global C-state To see if it could potentially close the gap for the ALLY as well. But despite all those changes, it's
still better on the LEGION GO, We may have got one or two fps increase In best case scenario, But there is still a gap
between the LEGION GO and the ALLY
at the same power TDP. On this same scene,
running at the same settings, It's clear there's like 7 to 8 fps difference
on average When using FSR 3
and Frame Generation ON. But it's also in this instance, Where I noticed that the LEGION GO GPU Clock is a little bit higher on average. The ROG ALLY seem to have a higher
CPU clock but rather a lower GPU one, Which seems like the opposite case
on the LEGION GO. Now we know that usually GPU clock
quite matter in most modern games, Especially like 400MHz
difference could make a good change. So maybe that could be the root cause
of the performance gap we are getting. To confirm that, I went ahead
and installed on my ALLY, The Ryzen Universal Tuning Tool Found in GitHub, and here you can
adjust both the CPU and GPU clock. In this case I wanted to mimic the LEGION GO
GPU clock on the ALLY, So I tuned the GPU minimum clock slider to 1400MHz
and a maximum to something like 1500MHz But as you can see as I do that, The GPU VRAM clock went down as the GPU core started to eat power from the VRAM In order
to reach our defined minimum clock, Which is a quite interesting Given that on the LEGION GO,
it's boosting to 1400MHz Yet the VRAM isn't being affected at all. Even FPS graph shows weird
fluctuations and stuttering as opposed to Before when it was at least stable. I tried different combinations
and you see that at 1100MHz It's the sweet spot for the GPU clock,
to not eat from the VRAM one at 15w TDP. Another answer to this
could be behind the different memory types. I did some research about both models,
used my ALLY and LEGION GO units. And we all know that the LEGION GP has LPDDR5X
that could reach up to 7500MT/s While the ALLY is equipped with LPDDR5
at a rated speed of 6400MT/s. Both my units memory manufacturer
is Samsung and based on their claims, LPDDR5X is more Power Efficient
in comparison to LPDDR5 by 20%. So maybe that gap we are seeing
is for the fact That the memory on the LEGION
GO requires less power and thus it's Allowing the GPU clock to boost higher
at the same TDP. That could theoretically be the cause,
to get the LEGION GO fps on the ALLY, I had to dial up the TDP to 17w
and you can see that in this way, It did not reduce the VRAM clock
while still having over 1400MHz GPU clock And a great CPU clock as well. That made me think that the LEGION GO
engineers made the power management In a way to prioritize the GPU core clock
along with the VRAM over the CPU, Or maybe it’s the ALLY in need of a BIOS update
or something that could allocate More GPU power budget at lower TDP,
like the scaling between 1100 and 1400MHz Translate to about 8 fps difference
in this example. Let's now shift our focus
to another set of testing, But this time with only the LEGION GO That's because I want to
test the difference we get between using. A memory speed of 6400MT/s and 7500MT/s
at 15w TDP. Like one might wonder
why would Lenovo implement A feature in BIOS that allows you
to downgrade memory speed, To answer that, I previously made another video comparing both in a set of games. I came to understand that, for example,
at TDP of 15w, we usually start To lose performance when we use memory
in 7500MT/s in comparison to 6400MT/s Maybe because memory start
to require more power And that effects
in reducing the GPU clock. I found that this get solved at higher TDP like 20w or more. Even in this test, We see similar results. At 15w, we got better performance on 6400MT/s
rather than 7500MT/s, by about three fps on average. Which all align
with previous results we got. Now, finally for the final test,
I also wanted to compare the two Thermal Policies We got on the LEGION GO,
STT versus STAPM. Also previously in my other video testing this, STAPM got the upper hand with the cost
of a minimal skin temperature increase. Also in this test, STAPM performs
a tad higher than STT. So in conclusion, for the LEGION GO 6400MT/s with STAPM
as the Thermal Policy, Actually gives
the higher performance possible at 15w TDP. Now even though SEGA was generous enough
and updated the game to support FSR 3, It still needs better optimization because
sometimes with Frame Generation ON, It crashes randomly, especially in cutscenes, that could be addressable With game patches, so I'm hopeful
they fix it as soon as possible. Now finally here, I'll show you some
gameplay demonstration on both devices. With the new AMD HYPR-RX Eco mode,
that basically reduces input lag and limits fps to 60 For a consistent good performance
while potentially saving power. So overall thanks to FSR 3
with Frame Generation enabled, Both the ALLY and the LEGION GO Could now reach a new level of smoother
gameplay while running at Lower TDP, Which ultimately gives you the possibility
to extend your enjoyment Without sacrificing your handheld
battery life. So I'm really excited for more developers
to start including this set of tech Into more games. I really hope Nobody thinks I'm siding with one product
or claiming one is better than the other, I'm just trying To tell the difference based on my test
and to help you guys know that as well. In the end, I really hope
you found this video useful and enjoyable. If you liked it, give it a thumbs up And subscribe to TECHOSAUR to be up to date with all my latest content. Thanks for watching
and see you in the next one.
