the graphics “pipeline.” It indicates “the smoothness of the
animation delivered to the GPU,” in Nvidia’s words. Displayed
FP S, in contr ast, is w hat you ac tually see on screen in your game.
If there’s a noticeable difference between the two, it indicates the
game is stuttering.
>> If both rendered and displayed FPS are enabled in
FrameView’s settings, you’ll see two columns of data on the in-
game overlay. The left-hand column shows rendered FPS, while
the right-hand column is for displayed FPS.
>> Our readings for rendered and displayed FPS are very
close [Image B], right through from the average FPS to the 99th
percentile. This means that we’re getting minimal stuttering.5
UNDERSTANDING FPS PERCENTILES
FrameView enables you to record FPS for the 90th, 95th,
and 99th percentiles. If your 99th percentile score is
70fps, for example, that means only one frame out of every 100
is below this rate. All three percentiles are useful, but the 99th
percentile is the best indicator of your GPU’s performance.
>> Like rendered and displayed FPS scores, 99th percentile
results can determine if you’re suffering inconsistent frame
rates under load. The closer your 99th percentile results are to
your average frame rate, the less micro stutter you’re getting.
And if you’re getting a lot of micro stutter and inconsistent frame
rates, it might be time to upgrade your graphics card.
>> As you can see [Image C], our average displayed FPS (the
top row) is 64, while our 99th percentile displayed FPS rating
is 62. This means that we suffered very little stuttering, as our
performance hit 62fps or higher 99 percent of the time.6
OVERLAY MODE TAGS
FrameView uses a few shorthand annotations to denote
various things; these are known as the “overlay mode
tags.” At the bottom of the overlay, one or more of the following
letters may be displayed: F, W, I, T, and V.
>> “F” denotes that the game is running in full-screen mode.
This is important, because it ensures the performance is properly
measured at the resolution specified in-game. In contrast, “W”
indicates that you’re running the game in windowed mode. If you
see “I,” it means iFlip (or Independent Flip) mode is running,
whereby the game is simulating Full Screen Exclusive mode.
>> The other tags, “T” and “V,” relate to VSync. “T” means
“tearing” and indicates that VSync is disabled. This is better for
performance analysis, because your frame rate isn’t being tied
to your monitor’s refresh rate, but the upshot is you’re more
likely to experience screen tearing. If you see “V,” that means
VSync is enabled; if you want the most accurate testing results,
it’s best to disable this.
>> For example, the tags “F,” “I,” and “T” are showing for us
[Image D]. This means we’re running the game in full-screen
iFlip mode with VSync disabled.C7
ANALYZING YOUR RESULTS
When you finish benchmarking, FrameView
saves the results in two spreadsheets, each with
the game name and timestamp in the file name. The
information here can seem overwhelming, but there are
a few key things to pick out.
>> The “Scan_Log” file is fairly straightforward, giving
a quick overview of things such as the average number
of frames rendered (“RenderAvg”), the 99th percentile of
frames displayed (“Display99”), and more.
>> The “Scan_Report” file is more complicated,
showing data in millisecond intervals. The key things
here to look out for are the “MsBetweenPresents” and
“MsBetweenDisplayChangeActual” columns, which
represent rendered FPS and displayed FPS respectively.
>> The “Scan_Report” file is also useful because it
can help you chart power draw and performance-per-
watt data over time. If you see sudden drops or spikes in
these numbers, you may have a problem.DIf you have an AMD graphics card, you’ll notice that
FrameView’s overlay has a reading for “AMDPWR” instead
of performance-per-watt, chip power, and total power
consumption. What does this mean for you?
Nvidia says that AMD’s power consumption API only
gives a value for something between chip power and total
power, rather than separate readings for each. Without
knowing the GPU’s total power consumption, FrameView
can’t calculate your performance-per-watt (PPW) rating,
because you work out PPW by dividing your FPS by total
power consumed.
In practice, this probably won’t affect you too much—
you may find FrameView’s FPS readings more useful, and
the program accurately reports those for AMD cards. But if
you really want to dive into your card’s power consumption
readings, it’s less than ideal.
Ultimately, this would require AMD changing how its API
works in order to give separate readings for chip power and
total board power. You therefore shouldn’t use FrameView
to compare power consumption between Nvidia and AMD
graphics cards, due to its inaccuracies measuring the
power used by the latter.MEASURING AMD
PERFORMANCE
maximumpc.com OCT 2019 MAXIMUMPC 61