算力商业
techspot.com ·
@Tim Schiesser
测试DLSS 5真实性能（无6倍帧生成）
本文在RTX 50系列上对DLSS 5进行基准测试，不使用6倍帧生成，结果显示Nvidia新神经渲染技术的性能需求较高。
AI推荐理由 对DLSS 5在NBA 2K27中的实际性能进行基准测试，涉及图形技术，对游戏和硬件爱好者有参考价值，但与AI核心领域关联较弱。
60
正文
服务科技爱好者超过25年。TechSpot意味着你可以信赖的技术分析和建议。
当你通过我们的链接购买时，我们可能会赚取佣金。
Reddit
Facebook
RSS
英伟达首个DLSS 5基准测试遗漏了我们真正想看到的一个数字：开启该技术会损失多少性能。那些“官方”NBA 2K27结果严重依赖6倍帧生成来夸大输出帧率，这是一种在每个图表和每种配置中都启用帧生成的可笑基准测试尝试。
我们没有得到的是适当的对比数据，显示仅仅开启DLSS 5会损失多少性能。所以今天我们来解决这个问题。没有6倍帧生成，进行适当的开启与关闭测试，并覆盖整个RTX 50系列的结果。甚至包括RTX 5050。
我们只有一天时间测试这个DLSS 5的首个官方实现，使用的是NBA 2K27的早期访问版本和英伟达支持的驱动程序。这就是为什么这篇文章完全专注于性能。
目前，有很多内容值得深入探讨。DLSS 5在本文发布时已在NBA 2K27中可用，所以值得探索其性能，更重要的是，为什么DLSS 5会有这样的表现。因为这里有一些奇怪之处。
DLSS 5有巨大的固定性能成本
DLSS 5的性能似乎主要由两个关键因素决定：首先，DLSS 5极其消耗性能。神经网络处理过程（对最终渲染帧进行后处理）相对于你可能熟悉的其他效果和着色器需要很长时间才能完成。
别急，英伟达……双关语。
关键是，这主要是一个基于输出分辨率的固定成本。降低游戏内设置并不会使DLSS 5部分的渲染时间变快。这在实际中意味着，当你启用DLSS 5时，性能受到运行这个密集效果所需时间的限制，改变游戏设置不会有显著影响。
在RTX 5090上运行NBA 2K27，使用DLAA和最高设置，启用DLSS 5时我们仅获得76 FPS。将预设降至高仅使性能提升2 FPS，不到3%的增益。尽管在禁用DLSS 5时性能有5%的提升。
一路降至低预设，在禁用DLSS 5时直接遇到似乎是240 FPS的引擎上限。但启用DLSS 5后，与最高设置相比性能仅提升9 FPS，仅增加12%。
这里几乎没有可扩展性。基本上，启用DLSS 5后获得的性能就是这样了。
超分辨率几乎无济于事
第二个因素是DLSS 5的性能直接与输出分辨率相关，而不是渲染分辨率。这是因为DLSS 5在超分辨率之后处理最终渲染帧，而不是之前。
有一些修改版实现将DLSS 5处理移到超分辨率之前，但在官方实现中它在之后运行。因此，启用超分辨率对DLSS 5本身的处理成本几乎没有影响。
这里我们使用RTX 5060 Ti 16GB在1440p下运行，选择这个分辨率是为了避免240 FPS上限和潜在的CPU限制。
启用DLAA时，性能从无DLSS 5的129 FPS降至启用DLSS 5的仅46 FPS。这是一个相当大的性能损失，通常，如果我们使用DLAA获得46 FPS，我们会简单地启用DLSS超分辨率，使用质量或平衡预设来大幅提升性能。
但启用DLSS 5后情况并非如此。
使用DLSS质量超分辨率，性能仅提升至50 FPS，仅增加9%。使用更低的超分辨率设置我们没有看到进一步增益。在这个游戏中，50 FPS实际上是RTX 5060 Ti 16GB在1440p下启用DLSS 5的硬性限制。
我们还看到禁用DLSS 5时有一个奇怪的限制，性能从未超过165 FPS，所以低于质量超分辨率也没有扩展。但关键是，在没有启用DLSS 5的情况下，从DLAA切换到DLSS质量确实看到28%的帧率提升。
这种增益在启用DLSS 5时大大减少，因为其固定处理成本主导了总渲染时间。这个渲染时间不会通过超分辨率减少，因为DLSS 5在超分辨率之后处理。所以使用超分辨率时，你只减少了非DLSS 5部分的渲染时间，而一旦启用DLSS 5，这部分在总帧时间中占比变小。
基本上，在保持显示器原生输出分辨率的情况下，你几乎无法改善GPU上的DLSS 5性能。
你可以通过输出较低的非原生分辨率来提升性能，但那样玩起来很糟糕，因为会非常模糊。我相信你们很多人尝试过在1440p显示器上输出1080p游戏，看到那有多糟糕。目前，这类做法是减少DLSS 5渲染时间并显著提升性能的唯一真正方法。
RTX 5090：最佳情况
考虑到所有这些，让我们看看DLSS 5在一系列GPU上的表现，从英伟达旗舰GPU GeForce RTX 5090开始。如果你碰巧拥有这些卡之一用于3D渲染、运行本地AI模型或仅仅是极致游戏，你可以获得相当可玩的DLSS 5体验。
在4K最高设置下，正如我们已经看到的，我们获得了76 FPS，没有启用任何帧生成。在1440p下，性能提升至134 FPS，在1080p下再次提升至182 FPS。当然，这些结果在所有分辨率下都可玩，但DLSS 5与不运行它相比仍有巨大的性能影响。
RTX 5090在4K下运行NBA 2K27慢60%，在1440p下慢40%。在4K下尤其显著，我们说的是192 FPS对比76 FPS，延迟和流畅度有巨大变化。
但在5090美元的GPU上运行DLSS 5并不那么有趣。在那些价格下，这对大多数人来说并不是一个现实的游戏GPU。
它远远超出了绝大多数游戏玩家的承受范围，除非你足够幸运在一年多前以接近建议零售价购买了一块。
RTX 5080和RTX 5070 Ti
所以，让我们看看RTX 5080。在这款GPU上，以我们认为可接受的帧率在4K下使用DLSS 5游戏实际上是不可能的。我们仅获得53 FPS，而禁用该功能时为185 FPS，性能降低了71%。
这是因为DLSS 5在RTX 5080上处理4K帧需要相当长的时间。事实上，它耗时足够长，以至于在任何游戏中达到60 FPS都将很困难。我们稍后将更详细地探讨这些渲染时间。
不过，在1440p下，DLSS 5是可玩的，输出102 FPS，在NBA 2K27这样的游戏中绰绰有余。这仍然是54%的性能降低，所以影响巨大，但从221 FPS的基线开始使其不那么痛苦。
在1080p下，DLSS 5运行在142 FPS，尽管我们不确定有多少人会将RTX 5080与1080p显示器配对。记住，DLSS 5性能与输出分辨率相关。你不能简单地启用超分辨率，以1080p渲染分辨率输出4K，并期望获得接近1080p的结果。
RTX 5070 Ti在4K DLSS 5游戏中更加吃力，仅输出44 FPS。考虑到我们使用较低设置看到的微小增益，并且我们已经使用了DLSS质量超分辨率，在这张卡上几乎不可能达到60 FPS。
RTX 5070 Ti仅在1440p下使用DLSS 5才变得可用，渲染86 FPS。与禁用DLSS 5相比下降了62%。在1080p下运行在126 FPS，比不使用该功能时下降45%。
RTX 5070及以下严重吃力
鉴于RTX 5070 Ti的结果，RTX 5070在4K下启用DLSS 5毫无用处也就不足为奇了。在不使用DLSS 5的情况下，游戏在这个分辨率下实际上相当可玩。我们使用最高设置和DLSS质量超分辨率获得124 FPS，这是一个非常愉快的体验。但启用DLSS 5后33 FPS完全不可行。
1440p也有点勉强，虽然技术上可玩，为67 FPS。这比禁用DLSS 5时的217 FPS下降了近70%。在1080p下，我们勉强超过100 FPS，但仍然看到57%的显著性能下降。
RTX 5060 Ti 16GB在禁用DLSS 5时在4K下也相当可用，但你可以完全排除在该分辨率下使用DLSS 5。这是一个糟糕的体验。
即使是1440p也很勉强。我们从165 FPS降至启用DLSS质量时的50 FPS，下降70%。技术上有可能达到60 FPS，因为DLSS 5本身的渲染时间不超过60 FPS所需的16.7毫秒帧间隔。但正如我们稍后将看到的，实际达到该帧率是一个很大的要求。
RTX 5060 Ti 16GB更适合在1080p下使用DLSS 5，达到78 FPS，尽管与不使用DLSS 5的1080p相比仍降低65%。
然后我们来到8GB型号，从RTX 5060开始。
我们不得不将游戏设置降至低以避免显存警告，但即使如此，DLSS 5在这款GPU上基本不可玩。4K完全不可能，仅21 FPS。在1440p下，我们仅获得46 FPS，考虑到DLSS 5的渲染时间，没有现实路径达到60 FPS游戏。
最多，你可以在1080p下使用DLSS 5，RTX 5060达到73 FPS。但这是与不使用DLSS 5相比70%的性能降低，如果游戏在这种配置下没有已经达到其240 FPS上限，下降幅度会更大。
RTX 5050对于DLSS 5完全无用。这是一款糟糕的GPU，即使在低设置下，启用DLSS 5在1080p下游戏也未能达到60 FPS。
这显然意味着它也无法在1440p或4K下达到可玩性能。事实上，它甚至接近不了，在1440p下仅渲染33 FPS，尽管禁用DLSS 5时输出152 FPS。即使在1080p下，启用DLSS 5性能也降低75%。你只获得原始帧率的四分之一。
DLSS 5实际需要多长时间渲染？
通过启用DLSS 5前后的帧率结果，我们可以估算每款GPU上DLSS 5的渲染时间。这是通过将帧率转换为帧时间，然后计算启用和禁用DLSS 5之间的差异来计算的。这告诉我们由于DLSS 5，每帧额外需要多少时间。
请注意，对于某些GPU，我们提高了设置至DLAA以避免与之前图表相比的瓶颈。RTX 5090在1440p下也受瓶颈限制，因此其在该分辨率下的渲染时间是外推估计。
查看DLSS 5的单独渲染时间揭示了一些有趣的事情：该功能的成本几乎直接与输出像素数和GPU的张量性能相关。
平均而言，DLSS 5在4K下处理时间比1440p长2.2倍。4K包含的像素数是1440p的2.25倍，所以我们在那里看到近乎完美的扩展。
英伟达还声称，例如，RTX 5080的张量核心AI TOPS是RTX 5060 Ti 16GB的2.4倍。在1440p和4K下，根据我们的性能测试，RTX 5080处理DLSS 5的速度大约快2.3倍。
达到60或80 FPS需要什么
有了这些渲染时间数字，我们可以估计在不同情况下预期的性能结果。我们表格中的第一个数字（下方）是游戏在未启用DLSS 5时以120 FPS运行时的估计DLSS 5帧率。
例如，在RTX 5090上4K下，DLSS 5的处理时间约为7.9毫秒。在120 FPS时，游戏其余部分之前渲染需要8.3毫秒。启用DLSS 5并加上那7.9毫秒处理时间，总帧时间变为16.2毫秒，相当于大约62 FPS。
实测DLSS 5渲染时间
n/a = 不可能
RTX 5090
RTX 5080
RTX 5070 Ti
RTX 5070
RTX 50
60 Ti 16GB
RTX 5060
RTX 5050
4K DLSS 5渲染时间
7.9
13.7
17.0
22.1
31.1
37.8
53.1
DLSS 5估计帧率（120 FPS游戏帧率）
62
45
40
33
25
22
16
DLSS 5估计帧率（80 FPS游戏帧率）
49
38
34
29
23
20
15
达到60 FPS所需游戏帧率
114
332
n/a
n/a
n/a
n/a
n/a
达到80 FPS所需游戏帧率
217
n/a
n/a
n/a
n/a
n/a
n/a
1440p DLSS 5渲染时间
3.6
6.1
7.8
10.2
14.1
16.9
24.0
DLSS 5估计帧率（120 FPS游戏帧率）
84
69
62
54
45
40
31
DLSS 5估计帧率（80 FPS游戏帧率）
62
54
49
44
38
34
27
达到60 FPS所需游戏帧率
77
95
113
156
387
n/a
n/a
达到80 FPS所需游戏帧率
112
157
213
444
n/a
n/a
n/a
在所有GPU中，你可以看到DLSS 5即使在之前获得相当不错帧率的情况下也会引入多大的帧率打击。
RTX 5090在4K下表现还算不错，但其他GPU在该分辨率下基本无用。在1440p下，RTX 5070受到超过50%的打击，低端卡的惩罚更大，因为DLSS 5在这些GPU上处理该分辨率需要更长时间。
我们对启用DLSS 5前以80 FPS运行的游戏进行了相同计算。在4K下，对于大多数GPU来说，这不是足够高的起始帧率。在1440p下，你大部分时间也会陷入困境。RTX 5090勉强过线，但RTX 5080在该配置下不会达到60 FPS，尽管性能下降减少到约33%。
此表中的其他数字显示启用DLSS 5后达到60和80 FPS所需的起始帧率。换句话说，这是游戏在添加DLSS 5处理时间之前需要通过设置调整或超分辨率运行的速度。
在4K下，你可以看到大多数这些GPU启用DLSS 5后永远不会达到60 FPS。这是因为DLSS 5处理时间本身超过16.7毫秒。
RTX 5080勉强通过，意味着60 FPS在技术上是可能的，但前提是游戏其余部分在约3毫秒内渲染。这需要大约330 FPS的起始帧率。在1440p下，可玩的帧率在更广泛的显卡上成为可能。RTX 5080需要95 FPS的起始帧率才能在启用DLSS 5后达到60 FPS，需要157 FPS才能达到80 FPS。
RTX 5070 Ti在这里也合理，前提是你满足于60 FPS和约50%的性能打击。在该卡上达到80 FPS也是可能的，尽管有62%的性能打击。
低于这个，我们就进入不可行的领域。RTX 5070可以以巨大的性能成本勉强处理1440p 60 FPS，但除此之外的任何事情都将困难。RTX 5060 Ti在1440p下使用DLSS 5基本不可能。
英伟达表示他们正在继续优化DLSS 5性能，因此这些渲染时间有可能通过进一步优化（模型、软件，也许未来一代GPU的硬件）而减少。我们拭目以待。
但对于发布版本，有些GPU无论你使用什么游戏内设置组合，在显示器的原生分辨率下启用DLSS 5永远无法达到可玩的帧率。
RTX 40系列支持可能即将到来
发布时一个有争议的问题是DLSS 5仅限于RTX 50系列，尽管模组制作者已经让该功能在每一代RTX GPU上工作，一直到RTX 20系列。
RTX 40系列的限制尤其痛苦，因为那些GPU似乎足够强大来运行DLSS 5。例如，RTX 4090在许多游戏工作负载中比RTX 5080更快，而RTX 5080已经可以在某些配置下以可玩帧率运行DLSS 5。
英伟达告诉我们，他们计划在RTX 50系列性能调校得更好后，努力将官方支持扩展到GeForce RTX 40系列，无论这意味着什么。但听起来英伟达确实计划在某个时候将该功能带到至少RTX 40系列。
我不确定英伟达是否从一开始就打算支持RTX 40系列。根据公司通常的运作方式，我怀疑DLSS 5最初是作为RTX 50系列独占设计的，以帮助销售那些GPU（或者甚至可能是为了宣传从未到来的“Super”刷新），但这纯粹是我的猜测。
无论如何，一旦模组制作者让DLSS 5在RTX 40系列硬件上工作并显示没有真正的技术限制，严格的RTX 50独占计划就会变得难以合理化。当然，英伟达尚未确认对RTX 30或RTX 20系列的官方支持。从几个角度来看这是合理的……
首先，DLSS 5设计使用FP8指令，这些指令在Blackwell和Ada架构上硬件加速。RTX 30和RTX 20系列缺乏硬件FP8支持，所以模组制作者不得不将DLSS 5切换到FP16才能以超过几FPS的速度运行。
但即使如此，RTX 3090在许多现代工作负载中仅相当于RTX 5070的性能级别，而RTX 5070使用当前模型和原生FP8加速勉强能以可接受帧率运行DLSS 5。如果该功能在旧卡上无论如何都运行不好，不提供DLSS 5支持是合理的。
这仍然感觉像技术演示领域
我们在这里看到的性能几乎处于技术演示领域。当然，如果你有5000美元的GPU，你可以在4K和1440p下启用DLSS 5获得可玩体验。除此之外，这是一个黯淡的场景。RTX 5080在4K下挣扎，而RTX 5070在1440p下仅勉强可用。
根据我们在GPU上看到的性能，很难想象大多数游戏玩家会使用DLSS 5来实际玩游戏。也许他们会为截图或只是看看而启用它，但对于正常游戏，性能成本巨大。
这不仅仅是因为
低帧率。还因为使用DLSS 5提升性能真的很难。DLSS 5的处理时间主导了整体帧时间，所以降低游戏内设置和使用超分辨率影响相对较小。这是DLSS 5与路径追踪等技术有显著差异的地方，路径追踪也非常苛刻。
超分辨率对提升路径追踪性能有奇效，因为启用超分辨率降低了执行路径追踪的渲染分辨率，带来显著的性能提升。你并不总能获得可玩的帧率，但有一些灵活性，例如，根据你的硬件，可以将40 FPS的体验推到60 FPS以上。对于DLSS 5，当前实现几乎没有灵活性。
如果你在最高设置下启用DLSS 5仅获得40 FPS，无论调整哪些设置都不可能达到60 FPS。
我个人喜欢日常游戏玩家可以用来改善游戏体验的实用功能。DLSS超分辨率就是这样一个例子，它在高端和低端硬件上都能很好地工作。但DLSS 5目前不属于这一类。
我们还没有深入探讨DLSS 5带来的图像质量优势或进行详细的视觉比较，最终这是这里的另一半。还有一个尴尬的可能性，对于某些游戏玩家来说，整个性能讨论可能没有意义。DLSS 5仍然极具争议，因为很多人根本不喜欢它对游戏视觉效果的影响。如果你一开始就不喜欢那些变化，那么DLSS 5是让你损失20%、50%还是70%的性能几乎不重要。
也就是说，过去一周我们看到的大量DLSS 5模组不应被视为该技术所能达到的代表。NBA 2K27是我们第一次看到当这些决定由游戏开发者做出时DLSS 5会是什么样子。
视觉改进是否足以证明巨大的性能成本是合理的，这个问题仍需回答，但到目前为止我们已经确定了DLSS 5在当前一代硬件上运行的成本有多高。
如果你喜欢TechSpot，请考虑订阅。
体验无广告的TechSpot，同时支持我们的工作
我们的承诺：
您的贡献直接资助新内容
结果：
更多技术报道、基准测试和深入分析
查看原文（中文正文为 AI 翻译）
Serving tech enthusiasts for over 25 years. TechSpot means tech analysis and advice you can trust.
When you buy through our links, we may earn a commission.
Reddit
Facebook
RSS
Nvidia's first DLSS 5 benchmarks left out the one number we actually wanted to see: how much performance you lose by enabling the technology. Those "official" NBA 2K27 results leaned heavily on 6X Frame Generation to inflate the output frame rates, a laughable attempt at benchmarking with frame gen enabled in every chart and every configuration.
What we didn't get were proper before-and-after figures showing how much performance is lost simply by turning DLSS 5 on. So we're fixing that today. No 6X Frame Gen, proper on vs. off testing, and results across the full RTX 50 series lineup. Even the RTX 5050.
We've only had a day to test this first official implementation of DLSS 5, using an early access copy of NBA 2K27 and Nvidia's supported driver. That's why this article is focused exclusively on performance.
For now, there's plenty to dig into. DLSS 5 is available in NBA 2K27 as this article goes live, so it's worth exploring its performance and, more importantly, why DLSS 5 performs the way it does. Because there are some quirks here.
DLSS 5 Has a Huge Fixed Performance Cost
DLSS 5 performance appears to be largely dictated by two key factors: the first is that DLSS 5 is extremely performance-intensive. The neural network pass, which post-processes the final rendered frame, takes a very long time to complete relative to other effects and shaders you might be familiar with.
Not so fast, Nvidia... pun intended.
Crucially, this is largely a fixed cost based on the output resolution. Lowering in-game settings does not make the DLSS 5 portion of the render time faster. What this means in practice is that when you enable DLSS 5, performance is capped by the time required to run this intensive effect, and changing game settings doesn't have a significant impact.
On an RTX 5090 running NBA 2K27 using DLAA and max settings, we achieved just 76 FPS with DLSS 5 enabled. Dropping the preset to High increased performance by only 2 FPS, less than a 3% gain. This is despite a 5% increase in performance when DLSS 5 is disabled.
Dropping all the way to the Low preset runs directly into what appears to be a 240 FPS engine cap when DLSS 5 is disabled. But with DLSS 5 enabled, performance improves by only 9 FPS compared to max settings, just a 12% increase.
There's very little scalability here. It's pretty much a case where the performance you get with DLSS 5 enabled is what it is.
Upscaling Barely Helps
The second factor is that DLSS 5's performance is tied directly to the output resolution, not the render resolution. That's because DLSS 5 processes the final rendered frame after upscaling, rather than before it.
There have been some modded implementations that move the DLSS 5 pass before upscaling, but in the official implementation it runs afterward. Therefore, enabling upscaling has little to no impact on the processing cost of DLSS 5 itself.
Here we have the RTX 5060 Ti 16GB running at 1440p, which we chose to help avoid the 240 FPS cap and potential CPU limitations.
With DLAA enabled, performance drops from 129 FPS without DLSS 5 to just 46 FPS with DLSS 5 enabled. That's a pretty huge performance loss, and normally, if we were getting 46 FPS with DLAA, we'd simply enable DLSS Super Resolution using something like the Quality or Balanced preset to drastically improve performance.
But that's not what happens with DLSS 5 enabled.
Using DLSS Quality upscaling, performance improves to just 50 FPS, a measly 9% increase. We saw no further gains using lower upscaling settings. In this title, 50 FPS is effectively the hard limit for the RTX 5060 Ti 16GB at 1440p with DLSS 5.
We also saw a somewhat strange limit with DLSS 5 disabled, where performance never exceeded 165 FPS, so there was no scaling below Quality upscaling there either. But crucially, without DLSS 5 enabled, we do see a 28% frame rate increase going from DLAA to DLSS Quality.
That sort of gain is heavily reduced with DLSS 5 because its fixed processing cost dominates the total render time. This render time is not reduced through upscaling because DLSS 5 is processed after upscaling. So when using Super Resolution, you're only reducing the non-DLSS 5 portion of the render time, which becomes a smaller part of the total frame time once DLSS 5 is enabled.
Essentially, there is very little you can do to improve DLSS 5 performance on your GPU while maintaining your display's native output resolution.
You can improve performance by outputting a lower non-native resolution, but that's a terrible way to play because it will look very blurry. I'm sure many of you have tried outputting a 1080p game on a 1440p monitor and seen how bad that looks. That sort of thing is currently the only real way to reduce DLSS 5's render time and significantly improve performance.
RTX 5090: The Best-Case Scenario
With all of that in mind, let's look at how DLSS 5 performs across a range of GPUs, starting with Nvidia's flagship GPU, the GeForce RTX 5090. If you happen to own one of these cards for 3D rendering, running local AI models, or simply extreme-end gaming, you can get a reasonably playable experience with DLSS 5 enabled.
At 4K max settings, as we've already seen, we achieved 76 FPS without any Frame Generation enabled. At 1440p, performance improved to 134 FPS, and again to 182 FPS at 1080p. Of course, those results are playable at all resolutions, but DLSS 5 still has a huge performance impact compared to running the game without it.
The RTX 5090 ran NBA 2K27 60% slower at 4K and 40% slower at 1440p. That's especially substantial at 4K, where we're talking about 192 FPS versus 76 FPS, a massive change in latency and smoothness.
But running DLSS 5 on a $5,090 GPU isn't that interesting. At those prices, this really isn't a realistic gaming GPU for most people. It's well out of reach of the vast majority of gamers unless you were lucky enough to purchase one near MSRP well over a year ago.
RTX 5080 and RTX 5070 Ti
So instead, let's look at the RTX 5080. On this GPU, gaming at 4K with DLSS 5 isn't really possible at what we'd consider acceptable frame rates. We achieved just 53 FPS, down from 185 FPS with the feature disabled, a 71% reduction in performance.
This is because DLSS 5 takes quite a while to process a 4K frame on an RTX 5080. In fact, it takes long enough that achieving 60 FPS in any game will be difficult. We'll explore those render times in more detail shortly.
At 1440p, though, DLSS 5 is playable, outputting 102 FPS, which is more than sufficient in a game like NBA 2K27. This is still a 54% reduction in performance, so it's a massive hit, but starting from a baseline of 221 FPS makes it less painful.
At 1080p, DLSS 5 ran at 142 FPS, though we're not sure how many people are pairing an RTX 5080 with a 1080p display. And remember, DLSS 5 performance is tied to output resolution. You can't simply enable upscaling with a 1080p render resolution while outputting at 4K and expect to get anything close to the 1080p result.
The RTX 5070 Ti struggles even more with 4K DLSS 5 gaming, outputting just 44 FPS. Given the small gains we saw from using lower settings, and considering we're already using DLSS Quality upscaling here, it's pretty much impossible to hit 60 FPS on this card.
The RTX 5070 Ti only becomes usable with DLSS 5 at 1440p, where it renders 86 FPS. That's a 62% decrease compared to DLSS 5 being disabled. It also ran at 126 FPS at 1080p, down 45% compared to running without the feature.
RTX 5070 and Below Struggle Badly
Given the RTX 5070 Ti results, it should come as no surprise that the RTX 5070 is useless at 4K with DLSS 5 enabled. The game is actually quite playable at this resolution without DLSS 5. We were getting 124 FPS using max settings and DLSS Quality upscaling, which is a really enjoyable experience. But 33 FPS with DLSS 5 enabled is a complete no-go.
1440p is also a bit of a stretch, though technically playable, at 67 FPS. That's down from 217 FPS with DLSS 5 disabled, representing a nearly 70% reduction in performance. At 1080p, we scrape past the 100 FPS barrier, but we're still looking at a significant 57% drop in performance.
The RTX 5060 Ti 16GB is again quite usable at 4K in this game when DLSS 5 is disabled, but you can completely rule out DLSS 5 at that resolution. It's a terrible experience.
Even 1440p is a stretch. We go from 165 FPS down to 50 FPS with DLSS Quality enabled, a 70% drop. It is technically possible to get 60 FPS here because the render time of DLSS 5 itself doesn't exceed the 16.7ms frame interval required for 60 FPS. But as we'll see shortly, actually hitting that frame rate is a big ask.
The RTX 5060 Ti 16GB is much better suited to 1080p with DLSS 5, achieving 78 FPS, though that's still a 65% performance reduction compared to running at 1080p without DLSS 5.
Then we get to the 8GB models, starting with the RTX 5060.
We had to turn the game's settings down to Low to avoid a VRAM warning, but even then DLSS 5 was largely unplayable on this GPU. 4K is completely out of the question at just 21 FPS. At 1440p, we achieved only 46 FPS, with no realistic path to 60 FPS gaming given the render time of DLSS 5.
At best, you can use DLSS 5 at 1080p, where the RTX 5060 achieves 73 FPS. But that's a 70% performance reduction compared to not using DLSS 5, and the drop would have been even larger if the game wasn't already maxing out its 240 FPS cap in this configuration.
The RTX 5050 is completely useless for DLSS 5. This is a terrible GPU, and even on Low settings the game failed to achieve 60 FPS with DLSS 5 enabled at 1080p.
That obviously means it can't achieve playable performance at 1440p or 4K either. In fact, it doesn't even get close, rendering just 33 FPS at 1440p despite outputting 152 FPS with DLSS 5 disabled. Even at 1080p, performance is 75% lower with DLSS 5 enabled. You're getting just one-quarter of the original frame rate.
How Long Does DLSS 5 Actually Take to Render?
With frame rate results from before and after enabling DLSS 5, we can estimate DLSS 5's render time on each GPU. This is calculated by converting frame rates into frame times and then calculating the difference between DLSS 5 being enabled and disabled. This tells us approximately how much additional time it takes to render each frame because of DLSS 5.
Note that for some GPUs we increased settings to DLAA to avoid bottlenecks compared to the previous charts. The RTX 5090 is also bottlenecked at 1440p, so its render time at that resolution is an extrapolated estimate.
Looking at the isolated render times for DLSS 5 reveals something interesting: the feature's cost is almost directly tied to output pixel count and the Tensor performance of the GPU.
On average, DLSS 5 took 2.2 times longer to process at 4K than at 1440p across all GPUs. 4K contains 2.25 times as many pixels as 1440p, so we're seeing near-perfect scaling there.
Nvidia also claims, for example, that the RTX 5080 has 2.4 times the Tensor Core AI TOPS of the RTX 5060 Ti 16GB. At both 1440p and 4K, the RTX 5080 processed DLSS 5 around 2.3 times faster based on our performance testing.
What It Takes to Hit 60 or 80 FPS
With these render-time numbers, we can estimate what sort of performance outcome to expect in different situations. The first number in our table (below) is the estimated DLSS 5 frame rate when a game was previously running at 120 FPS without DLSS 5 enabled.
For example, on an RTX 5090 at 4K, DLSS 5's processing time is about 7.9ms. At 120 FPS, the rest of the game was previously rendering in 8.3ms. Enable DLSS 5 and add that 7.9ms processing time, and the total frame time becomes 16.2ms, which equates to roughly 62 FPS.
Measured DLSS 5 Render Times
n/a = Impossible
RTX 5090
RTX 5080
RTX 5070 Ti
RTX 5070
RTX 5060 Ti 16GB
RTX 5060
RTX 5050
4K DLSS 5 Render Time
7.9
13.7
17.0
22.1
31.1
37.8
53.1
DLSS 5 Estimated Frame Rate (120 FPS Game FPS)
62
45
40
33
25
22
16
DLSS 5 Estimated Frame Rate (80 FPS Game FPS)
49
38
34
29
23
20
15
Required Game FPS to Hit 60 FPS
114
332
n/a
n/a
n/a
n/a
n/a
Required Game FPS to Hit 80 FPS
217
n/a
n/a
n/a
n/a
n/a
n/a
1440p DLSS 5 Render Time
3.6
6.1
7.8
10.2
14.1
16.9
24.0
DLSS 5 Estimated Frame Rate (120 FPS Game FPS)
84
69
62
54
45
40
31
DLSS 5 Estimated Frame Rate (80 FPS Game FPS)
62
54
49
44
38
34
27
Required Game FPS to Hit 60 FPS
77
95
113
156
387
n/a
n/a
Required Game FPS to Hit 80 FPS
112
157
213
444
n/a
n/a
n/a
Across all the GPUs, you can see just how much of a frame rate hit DLSS 5 introduces even in situations where you were previously getting a pretty decent frame rate.
The RTX 5090 holds up reasonably well at 4K, but the rest of the GPUs are pretty useless at that resolution. At 1440p, the RTX 5070 takes a hit of more than 50%, with an even larger penalty on lower-tier cards because DLSS 5 takes longer to process at this resolution on those GPUs.
We've run the same calculations for a game running at 80 FPS before enabling DLSS 5. At 4K, that isn't a high enough starting frame rate for most GPUs. At 1440p, you'll also be on struggle street for the most part. The RTX 5090 just scrapes over the line, but the RTX 5080 won't hit 60 FPS in that configuration, although the performance drop is reduced to around 33%.
The other numbers in this table show the required starting frame rate needed to achieve 60 and 80 FPS with DLSS 5 enabled. In other words, this is how fast the game needs to be running through settings adjustments or upscaling before adding DLSS 5's processing time.
At 4K, you can see that most of these GPUs will simply never hit 60 FPS with DLSS 5 enabled. That's because the DLSS 5 processing time itself exceeds 16.7ms.
The RTX 5080 just scrapes through, meaning 60 FPS is technically possible, but only if the rest of the game renders in around 3ms. That requires a starting frame rate of roughly 330 FPS. At 1440p, playable frame rates become possible on a wider range of graphics cards. The RTX 5080 requires a starting frame rate of 95 FPS to achieve 60 FPS after enabling DLSS 5, and 157 FPS to achieve 80 FPS.
The RTX 5070 Ti is also reasonable here, provided you're okay with 60 FPS and a performance hit of around 50%. Achieving 80 FPS is also possible on that card, although it comes with a 62% performance hit.
Below that, we're getting into the realms of implausibility. The RTX 5070 can sort of handle 1440p 60 FPS at a large performance cost, but anything beyond that will be difficult. The RTX 5060 Ti is pretty much out of the question for DLSS 5 at 1440p.
Nvidia says it is continuing to work on DLSS 5 performance, so it's possible these render times could be reduced through further optimization (the model, on software, and perhaps in hardware with future-gen GPUs). We'll have to wait and see.
With the launch version, though, there are simply some GPUs where no matter what combination of in-game settings you use, you will never achieve a playable frame rate with DLSS 5 at your display's native resolution.
RTX 40 Series Support Could Be Coming
One contentious issue at launch is that DLSS 5 is restricted to the RTX 50 series, even though modders have already gotten the feature working on every generation of RTX GPU, right down to the RTX 20 series.
The RTX 40 series restriction is especially painful because those GPUs appear powerful enough to run DLSS 5. The RTX 4090, for example, is faster than the RTX 5080 in many gaming workloads, and the RTX 5080 can already run DLSS 5 at playable frame rates in some configurations.
Nvidia tells us that it plans to work on expanding official support to the GeForce RTX 40 series once RTX 50 series performance is better tuned, whatever that means. But it does sound like Nvidia plans to bring the feature to at least the RTX 40 series at some point.
I'm not sure whether Nvidia intended to support the RTX 40 series from the beginning. Based on how the company typically operates, I suspect DLSS 5 was originally designed as an RTX 50 series exclusive to help sell those GPUs (or maybe even a feature to headline the "Super" refresh that never arrived), but that's purely speculation on my part.
In any case, plans for strict RTX 50 exclusivity would have become much harder to justify once modders got DLSS 5 working on RTX 40 series hardware and showing there's no true technical limitation. Of course, Nvidia has not confirmed official support for the RTX 30 or RTX 20 series. That makes sense from a couple of perspectives...
First, DLSS 5 is designed to use FP8 instructions, which are hardware-accelerated on Blackwell and Ada architectures. The RTX 30 and RTX 20 series lack hardware FP8 support, so modders had to switch DLSS 5 over to FP16 to get it running at more than just a few FPS.
But even then, an RTX 3090 is only around the performance class of an RTX 5070 in many modern workloads, and the RTX 5070 can barely run DLSS 5 at acceptable frame rates using the current model with native FP8 acceleration. Not providing DLSS 5 support on older cards makes sense if the feature simply won't run well regardless.
This Still Feels Like Tech Demo Territory
The performance we're seeing here is pretty much in tech demo territory. Sure, if you have a $5,000 GPU, you can get a playable experience at 4K and 1440p with DLSS 5 enabled. Outside of that, it's a bleak scenario. The RTX 5080 struggles at 4K, while the RTX 5070 is only borderline usable at 1440p.
Based on the performance we're seeing across GPUs, it's hard to imagine most gamers will use DLSS 5 to actually play games. Maybe they'll enable it for screenshots or just to check it out, but for normal gameplay, the performance cost is enormous.
And that's not just because of the low frame rates. It's also because improving performance with DLSS 5 is really difficult. DLSS 5's processing time dominates overall frame times, so lowering in-game settings and using upscaling has relatively little impact. This is where DLSS 5 differs substantially from a technology like path tracing, which is also extremely demanding.
Upscaling works wonders for improving path tracing performance because enabling Super Resolution reduces the render resolution at which path tracing is performed, delivering a substantial performance boost. You won't always get playable frame rates, but there's some flexibility to take, say, a 40 FPS experience and push it above 60 FPS depending on your hardware. With DLSS 5, there is little to no flexibility with the current implementation.
If you're only getting 40 FPS with DLSS 5 enabled at max settings, it's impossible to achieve 60 FPS regardless of which settings you adjust.
I personally love practical features that everyday gamers can use to improve the gaming experience. DLSS Super Resolution is one such example, and it works well across both high-end and lower-end hardware. But DLSS 5 doesn't fall into that category right now.
We've yet to dive into the image quality benefits DLSS 5 brings or conduct any detailed visual comparisons, and ultimately that's the other half of the equation here. There's also an awkward possibility that, for some gamers, this entire performance discussion is moot. DLSS 5 remains hugely controversial because plenty of people simply don't like what it does to a game's visuals. If you don't like those changes in the first place, it hardly matters whether DLSS 5 costs you 20%, 50%, or 70% of your performance.
That said, the flood of DLSS 5 mods we've seen over the past week shouldn't necessarily be treated as representative of what the technology can achieve. NBA 2K27 is our first opportunity to see what DLSS 5 looks like when those decisions have been made by the game's developers.
Whether the visual improvement is substantial enough to justify the enormous performance cost is the question that still needs an answer, but for now we've established just how expensive DLSS 5 is to run in current-gen hardware.
If you enjoy TechSpot, please consider subscribing.
Experience TechSpot ad-free while supporting our work
Our promise:
Your contribution directly funds new content
The result:
More tech coverage, benchmarks, and in-depth analysis