so as you've seen in our FPS charts for
different reviews video cards CPUs so
forth we report FPS in three metrics
there's the average FPS 1% lows and 0.1%
lows it could also be called 99th
percentile 99.9 percent I'll but what do
those things actually mean I've
discussed this kind of briefly in an
asked GN episode we in every single
article with review we post this
information and that's always in the in
the link in the description but just to
kind of make a dedicated video to this
one percent lows and 0.1% blows that's
our way of conveying frame times that a
means that that I think is a little bit
easier to read in comparison to the
average FPS so what are frame times
frame times and frame pacing this is
something that Scott Watson formerly of
tech report really heavily drove as a
new metric to review products and when I
saw the post years ago it really made a
lot of sense to me so we adopted it what
he was doing is sort of frame time
testing he called it and this looks at
the average or overall millisecond
disparity between frame a to frame B so
if you're looking at say frame number
one versus frame number two maybe
there's an eight millisecond difference
between them if you're on crossfire or
something like that and then maybe frame
number two to frame number three there's
a 20 millisecond difference this creates
a very perceptible sort of latency to
the user that you can pick up in your
game play it'll feel like something has
slowed down you might use colloquial
terms like frame drops or stutters it
can introduce tearing it really it just
depends on the game and what's going on
but that's kind of what frame times are
it's the milliseconds between frames and
this number is not necessarily shown in
average FPS numbers that's why we've
moved away from a pure average and to
that end using just an average will sort
of obfuscate these occasional low dips
in your frame rate that may be seen in
real gameplay as a frame drop so this a
really good example of this this is a
we've got a tray of CPUs this is AG
three two five eight Intel Pentium CPU
and the G three two five eight and some
of our reviews using this product it
does pretty well until you really throw
a heavily multi-threaded game at it and
as an example in GTA 5 the and the seven
sixty K will actually perform on average
worse than the g3 two five eight it's
about 10 FPS lower average so in our
testing I don't remember the exact
numbers off my head but it was something
like 76 vs. 65 FPS average something
like that for 1080p with the settings we
used but the three to five eight is
actually worse or was worse at that time
with those settings and that's because
the the three to five eight would
introduce the stuttering as you play so
as you're playing GTA with this
particular processor there will be very
jarring and hard sort of stops and the
frame output its a latency and this
happens because the frame time in
milliseconds between one frame and the
next will occasionally spike maybe 50
milliseconds something like that and
what matters here is more the offset
than the actual absolute value so if
you've got really consistent frame time
say every frame is twenty milliseconds
apart that's really it's not fantastic
but it's not bad and as long as everyone
is twenty milliseconds apart then it's
really not that big of an issue because
it appears fluid but when you start
having disparity in your frame times
maybe one frame is 16 milliseconds the
next one is 24 milliseconds now we get
some stuttering and it gets worse as you
increase that disparity the percentage
difference between frame times so with
the three to five eight in that example
what we're seeing is a high average but
it's actually kind of unplayable because
occasionally there'll be a really hard
hit to your frame rate where it might
stutter or frame drop or whatever word
you want to use lag for a fraction of a
second and it's long enough to really if
you're trying to aim at someone and
shoot them you get that stutter suddenly
you're not aiming at the right
anymore so that's where this testing
methodology came from but what are the
1% lows and point 1 percent low as well
we take this the frame times we take the
overall a test period which is 30
seconds we look at 30 seconds of frame
data we look at every single frame
output versus the frame time and we
create an average fps and I pull from
that one percent of the the slowest 1%
of frames and then we pull again the
slowest point one percent of frames and
the reason this is done is to show very
clearly what the converted FPS value is
of those very slow two frames and that's
because we don't want to use a minimum a
lot of older testing methodology would
use a minimum and a maximum frame rate
which really very honestly here is
totally useless because if it is
actually a minimum and not some sort of
normalized value the actual minimum is
an outlier and may not be representative
of actual gameplay if we ran a game with
10 test passes we might see minimums
that have a range of 100 FPS which is a
massive range totally unreliable and
unusable as a metric but if we average
the slowest 1% of those frames suddenly
we have something that is more normal
repeatable representative of a real
everyday gaming experience so that's
where the 1% in point one percent loads
come from what do they mean well
now generally if you have say the fury X
is another good example a product like
the fury X performs pretty well in
average frame rates but in some games
and we have charts to show this in some
games the point 1 percent will be so low
that we might actually recommend
something like even a 390 X instead and
in the case of that particular example
it may be because the 4 gigabyte versus
8 gigabyte disparity we saw this in our
our X 480
our X 480 4 gigabyte vs. 8 gigabyte test
and that test shows that
yeah the averages can be pretty close
sometimes maybe worst case they're a
couple percent difference but in some
games like Assassin's Creed or black ops
or Mirror's Edge occasionally
we'll see such bad low frame rates the
0.1% value that it really it might be
like 6 FPS or 20 FPS as opposed to the 8
gigabyte card doing a double value to
that or 60% higher value to that and and
that is where you would really clearly
see an impact in gameplay and where we
as a review outlet would make the
recommendation to buy the one that has a
more stable frame output or more
reliable repeatable frame output than
just an average so the average values
don't tell the whole story if you see a
high average value on a product you
really need to look at the low values as
well to determine if it's actually any
good for what you're doing because just
a high average averages will by nature
of an average collect all the data
divide it and that can kind of smooth
over any imperfections that as a user
would be perceptible but as a metric
might not be and then if you're looking
at frame times in milliseconds and we
show this for ashes of a singularity
now frame times the thing to pay
attention to is once there's more than
an 8 millisecond swing between one frame
to the next and frame time values that
begins to become perceivable so if you
have 16 milliseconds to 20 milliseconds
really not super noticeable in fluidity
of the the sort of frame rate output but
when you have 8 milliseconds to 24 it's
starting to become noticeable or sorry
16 milliseconds to 24 8 milliseconds 24
definitely is noticeable but that's
that's your value you're looking at 8
milliseconds swing to kind of become
perceptible to the human eye on average
and that's a number we've gotten from
talking to GPU manufacturers so that's
the basics if you want to learn more
about this I've written about it in the
past but the best article in the
industry is by scott watson and there's
there's no reason to kind of try and
push all that info out in a video so if
you if you really want to go deep on
this
I believe the article is called inside
the second that's that's the article
name couple pages long and it explains
what all this stuff is
we've taken that core concept over the
last few years and refined it into
something that I think for our audience
is easier to to read and understand and
that's by converting the sort of frame
rate versus time chart where you see
this spiky output that tech report and a
couple other sites will do we take that
chart and we instead convert it into
what we call one-percent lows and point
one percent lows and then we put it out
and have in a normal bar graph against
the average of yes that makes a little
bit easier to kind of comprehend what's
exactly going on so hopefully that
answers the question we will be pointing
everyone to this video in the future
when the question in counts is asked
what is 1% lows that that's your answer
so as always thank you for watching
check out the channel subscribe to get
more information and patreon link the
postal video if you wanna help out
directly I'll see you all next time
you
you
We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for us to earn fees by linking to Amazon.com and affiliated sites.