Michael Fabian 'Xaymar' Dirks
28 lines
4.0 KiB
HTML
28 lines
4.0 KiB
HTML
---
|
||
title: 'FidelityFX Super Resolution: DLSS Killer, or not?'
|
||
category: Review
|
||
tags: [AMD, FidelityFX, FidelityFX Super Resolution, FSR, NVIDIA, DLSS, 'WordPress Archive']
|
||
---
|
||
|
||
{% capture content %}It’s been a day since the reveal of FidelityFX Super Resolution, and it appears to be a DLSS killer. But on closer inspection, it seems that AMD made a mistake in at least one of the examples, and forgot to show the FSR footage, instead only showing the native footage. Let’s go into the details on AMDs FidelityFX Super Resolution a bit.{% endcapture %}{% include blocks/paragraph.liquid content=content %}
|
||
|
||
{% capture content %}A mistake in the Presentation{% endcapture %}{% include blocks/heading.liquid level=2 content=content %}
|
||
|
||
{% capture content %}If you watched the AMD presentation only once, you would most likely not have spotted this – I’ve even had to watch it three times to spot it. There’s three demos in the presentation, but only one of them has this issue:{% endcapture %}{% include blocks/paragraph.liquid content=content %}
|
||
|
||
{% capture content %}<li>{% capture link %}The first demo highlights the Ultra Quality mode.{% endcapture %}{% include inline/link.liquid target="_blank" content=link url="https://youtu.be/eHPmkJzwOFc?t=84" %}</li>
|
||
<li>{% capture link %}Then we get a demo comparing each quality mode at the same time.{% endcapture %}{% include inline/link.liquid target="_blank" content=link url="https://youtu.be/eHPmkJzwOFc?t=109" %}</li>
|
||
<li>{% capture link %}And finally, an example of the Quality mode in motion.{% endcapture %}{% include inline/link.liquid target="_blank" content=link url="https://youtu.be/eHPmkJzwOFc?t=182" %}</li>{% endcapture %}{% include blocks/list.liquid content=content %}
|
||
|
||
{% capture content %}Going by AMDs patent registrations, FidelityFX Super Resolution is entirely Spatial, but in the second demo it has information that would be completely hidden without Temporal information. It is able to perfectly represent textures and surfaces, despite it lacking the information to do so. This can’t even be explained away by VP9, HEVC or H264 compression artifacts, as those are vastly different from what we end up seeing here.{% endcapture %}{% include blocks/paragraph.liquid content=content %}
|
||
|
||
{% capture content %}My conclusion is that AMDs marketing team made an honest mistake here. They seem to have forgotten to show the layers or even add them after testing the overlay. It is still impressive in the other demos for a Spatial upscaler. If AMD wishes to clarify this, please contact me at info(at)xaymar(dot)com.{% endcapture %}{% include blocks/paragraph.liquid content=content %}
|
||
|
||
{% capture content %}Update: Other sources in contact with AMD have confirmed that this is a Spatial upscaler, therefore the second demo should be physically impossible.{% endcapture %}{% include blocks/paragraph.liquid content=content %}
|
||
|
||
{% capture content %}Is it a DLSS killer?{% endcapture %}{% include blocks/heading.liquid level=2 content=content %}
|
||
|
||
{% capture content %}It depends on which version of DLSS. If you compare FidelityFX Super Resolution to DLSS 1.0 and 1.1, I’d take FidelityFX Ultra Quality over DLSS 1.0 and 1.1 any day. The temporal issues in DLSS 1.0 and 1.1 are horrible to look at, and make games unplayable because of heavy ghosting issues and similar. But if you compare it to DLSS 2.0 and 2.1, FidelityFX Super Resolution doesn’t hold up anymore. DLSS 2.0 and 2.1 look almost identical to the native resolution, while running up to 400% faster.{% endcapture %}{% include blocks/paragraph.liquid content=content %}
|
||
|
||
{% capture content %}AMD definitely has a strong contender, and if they can figure out how to add a Temporal compontent to the upscaler, they will be competing with DLSS 2.0 instead of DLSS 1.0. As an example of a great upscaler, Unreal Engine 5 recently unveiled Temporal Super Resolution which gives a slight improvement in performance, while losing nearly no quality.{% endcapture %}{% include blocks/paragraph.liquid content=content %}
|