Contents
Zotac GeForce GTX 1650 Super / p42
AMD Radeon RX 590 / p43 Nvidia GeForce GTX 1660 / p44
Zotac GeForce GTX 1660 Super / p46 Asus ROG Strix Radeon RX 5500 XT 8GB / p47
Nvidia GeForce GTX 1660 Ti / p48 Sapphire Pulse Radeon RX 5600 XT / p49
Nvidia GeForce RTX 2060 / p50 AMD Radeon RX 5700 / p51
AMD Radeon RX 5700 XT / p52 Zotac GeForce RTX 2060 Super Mini / p53
Results graphs / p54percentiles, so you can see what frame rate you’re likely to maintain
most of the time. Instead of the very minimum, we’re now reporting the
99th percentile frame rate for each benchmark.
This means that 99 per cent of the frame rates recorded in each
benchmark are at this figure or above. Occasional dropped frames are
discarded as part of the 1 per cent (unless there’s an awful lot of them),
but noticeable, prolonged drops in the frame rate are recorded.
There are other benefits to using FrameView over FRAPS. The latter
doesn’t work in all DirectX 12 titles, for example, and FrameView also
allows us to benchmark with Vulkan. As before, we run every test at
least three times at each resolution, discarding spurious results, and
then reporting the average.Red Dead Redemption 2
We wanted to replace our aging Deus Ex: Mankind Divided test with a
new game that used a different API, and Red Dead Redemption 2 fits the
bill perfectly. When you fire up the game, it automatically detects your
hardware and applies the sittings it thinks will work best.
That’s good for gamers, as it means you can get the game working
on a variety of different levels of hardware. As benchmarkers, however,it causes a few problems, because it also changes the definition of
the blanket presets – the High preset on one card will enable different
graphics settings than the same preset on another card.
For this reason, we have to use our own custom settings and apply
them again from scratch each time we install a new card. We couldn’t
just set every setting to Ultra, either, as it would then take up too much
graphics memory (there’s a handy chart in the game to show you how
much memory your chosen settings need).
As a compromise, we turn on nearly every advanced graphics
feature, set the individual settings to High and set the sliders around
three quarters of the way across. This enables us to compare GPUs
on a like-for-like basis in a demanding test. We run the game’s built-in
benchmark, and set FrameView to record the end portion, which is
based on real gameplay.Ray tracing and DLSS
Due to teething problems when Battlefield V was first released, we
used to run it in both DirectX 11 and 12, but we now just test in the latter
mode. We record a custom play-through in the Under No Flag War
Story with FrameView. If a card supports ray tracing, we also test it
in two modes. One with High DXR and TAA enabled, which shows
us how the GPU copes with ray tracing at the ‘true’ resolution, while
also enabling us to compare cards that don’t support DLSS, such as
Nvidia’s GeForce GTX 1660 series.
On Nvidia’s RTX cards, we also test with DLSS. This method of
super-sample anti-aliasing effectively renders at a lower resolution
than the one selected, and scales it up using Nvidia’s Tensor cores. It
enables you to switch on ray tracing while getting smoother frame
rates. It generally looks fine at 2,560 x 1,440 and above, but it can
look overly blurry at 1080p, as the resolution from which it’s scaling
is so much lower.
We’re also now incorporating ray tracing into our scores. It has a low
weighting (10 per cent), as only a few games support it and not everyone
is fussed about it. However, the list of supporting games is growing, and
CyberPunk 2077 will also support it when it’s released this year.
Our other tests are Total War: Warhammer II, using DirectX 11, as
the DirectX 12 beta has stuttering issues with Nvidia’s current drivers- we record the built-in Battle benchmark with FrameView. We also
record the built-in benchmark in Shadow of the Tomb Raider with
FrameView, running it at the Highest graphics preset with TAA. Finally,
we run Unigine’s Superposition benchmark to load the GPU, and
record the power draw at the mains using a Watts Up? Pro meter.
We now use Nvidia’s FrameView software to
measure the 99th percentile frame ratesContents
ZotacGeForceGTX 1650 Super/p42
AMDRadeonRX 590 / p43
NvidiaGeForceGTX 1660 / p44
ZotacGeForceGTX 1660 Super/ p46
AsusROGStrixRadeonRX 5500 XT8GB/ p47
NvidiaGeForceGTX 1660 Ti/ p48
SapphirePulseRadeonRX 5600 XT/p49
NvidiaGeForceRTX 2060 / p50 AMDRadeonRX 5700 / p51
AMDRadeonRX 5700 XT/ p52 ZotacGeForceRTX 2060 SuperMini/ p53
Resultsgraphs/ p54percentiles, so you can see what frame rate you’re likely to maintain
most of the time. Instead of the very minimum, we’re now reporting the
99th percentile frame rate for each benchmark.
This means that 99 per cent of the frame rates recorded in each
benchmark are at this figure or above. Occasional dropped frames are
discarded as part of the 1 per cent (unless there’s an awful lot of them),
but noticeable, prolonged drops in the frame rate are recorded.
There are other benefits to using FrameView over FRAPS. The latter
doesn’t work in all DirectX 12 titles, for example, and FrameView also
allows us to benchmark with Vulkan. As before, we run every test at
least three times at each resolution, discarding spurious results, and
then reporting the average.
Red Dead Redemption 2
We wanted to replace our aging Deus Ex: Mankind Divided test with a
new game that used a different API, and Red Dead Redemption 2 fits the
bill perfectly. When you fire up the game, it automatically detects your
hardware and applies the sittings it thinks will work best.
That’s good for gamers, as it means you can get the game working
on a variety of different levels of hardware. As benchmarkers, however,
it causes a few problems, because it also changes the definition of
the blanket presets – the High preset on one card will enable different
graphics settings than the same preset on another card.
For this reason, we have to use our own custom settings and apply
them again from scratch each time we install a new card. We couldn’t
just set every setting to Ultra, either, as it would then take up too much
graphics memory (there’s a handy chart in the game to show you how
much memory your chosen settings need).
As a compromise, we turn on nearly every advanced graphics
feature, set the individual settings to High and set the sliders around
three quarters of the way across. This enables us to compare GPUs
on a like-for-like basis in a demanding test. We run the game’s built-in
benchmark, and set FrameView to record the end portion, which is
based on real gameplay.Ray tracing and DLSS
Due to teething problems when Battlefield V was first released, we
used to run it in both DirectX 11 and 12, but we now just test in the latter
mode. We record a custom play-through in the Under No Flag War
Story with FrameView. If a card supports ray tracing, we also test it
in two modes. One with High DXR and TAA enabled, which shows
us how the GPU copes with ray tracing at the ‘true’ resolution, while
also enabling us to compare cards that don’t support DLSS, such as
Nvidia’s GeForce GTX 1660 series.
On Nvidia’s RTX cards, we also test with DLSS. This method of
super-sample anti-aliasing effectively renders at a lower resolution
than the one selected, and scales it up using Nvidia’s Tensor cores. It
enables you to switch on ray tracing while getting smoother frame
rates. It generally looks fine at 2,560 x 1,440 and above, but it can
look overly blurry at 1080p, as the resolution from which it’s scaling
is so much lower.
We’re also now incorporating ray tracing into our scores. It has a low
weighting (10 per cent), as only a few games support it and not everyone
is fussed about it. However, the list of supporting games is growing, and
CyberPunk 2077 will also support it when it’s released this year.
Our other tests are Total War: Warhammer II, using DirectX 11, as
the DirectX 12 beta has stuttering issues with Nvidia’s current drivers- we record the built-in Battle benchmark with FrameView. We also
record the built-in benchmark in Shadow of the Tomb Raider with
FrameView, running it at the Highest graphics preset with TAA. Finally,
we run Unigine’s Superposition benchmark to load the GPU, and
record the power draw at the mains using a Watts Up? Pro meter.
We now use Nvidia’s FrameView software to
measure the 99th percentile frame rates