[Digital Foundry] AMD Radeon RX 9060 XT 16GB review vs RTX 5060 Ti vs... PlayStation 5 Pro?

[-]

ResolutionMotor4909@reddit

5070 turned out to be the best deal xD who'd ever know

[-]

mybrainisoutoforderr@reddit

even 5060 ti fucks 9070 xt in the ass

[-]

mybrainisoutoforderr@reddit

i would be honored to be penetrated by nvidia

[-]

I'd like to use this review as an example of how modern GPU testing should be. Everything tested: raster, latency, RT, path tracing, real world testing of GPU features like vendor exclusive upscalers that the consumer is going to use (FSR4 for AMD, DLSS for Nvidia). He touched on MFG and the Nvidia suite allowing the 5060ti to run path traced Cyberpunk 150+ fps with great image quality thanks to DLSS, Ray reconstruction while having latency that's normal on a controller and more.

Every base relevant to a consumer who's thinking of buying a 9060 or 5060 was touched giving a complete view of what gaming on these cards is actually like. Hopefully we'll see other reviews evolve more towards the real world testing (what is the real world gaming experience the consumer will have with each card) direction like DF has.

[-]

aminorityofone@reddit

i respectfully disagree. DLSS, FSR, XeSS, PSSR Framegen all cause graphically anomalies and DLSS is the best. These software solutions to fix a hardware issue is no different than software rendering vs hardware rendering in the 90s. Hard was always better visually, and even today that holds true. There needs to be baseline performance without any software upscaling going on and then a review with upscaling on so the consumer can make their own judgement. This is partially hampered by the fact that youtube compression hides many artifacts. Lastly, reviewers need to review the card objectively, 'real world' is not objective and will be full of bias and incorrect testing. For example, in the real world, a person will have 3rd party programs installed on their computer or music playing in the background, discord or other voice chat, maybe facebook messenger etc. All of these things do impact a cpu/gpu/ssd/ram performance but non of these can be accurately measured as each person will have different amounts of programs running. AKA, the scientific method. Remove as many variables as possible and make all testing on the same platform (same mobo, ram, cpu ssd etc). That way consumers get a best possible performance baseline to base their purchase choice on.

[-]

nukleabomb@reddit

But people use these cards in real world conditions. They're not gonna drop a 9060xt or 5060ti on a 9800x3d system. It'll be a 5600 or 7600 system, which brings a cpu bottleneck to the equation.

Then comes the fact that people will play around with the settings to achieve their desired fps. Some find it fine to play at 30 with max eye candy, some do locked 60. So aim for a high refresh rate at 120 or 144 depending on the monitor and some play at unlocked with vrr.

Some want hdr, and for those RTX hdr maybe the decider. Some play on a 4k tv in the living room, where an upscaler will have high value. Frame gen can help players who already play on a controller and don't mind the latency in single player games. Competitive gamers might find reflex and anti lag to be the deciding factor.

There's a lot more to how graphics cards are used than pure scientific method. This is similar to cars being tested for a week on track or being tested as a daily for a month. The month long review will hold more water because it covers the everyday use of the car better, which a buyer is very likely going to do.

[-]

SEI_JAKU@reddit

"Real world conditions" is me going to use a 9060 XT with a 11800X3D.

[-]

Strazdas1@reddit

youre going to be playing games that still bottleneck on CPU though. So the GPU performance isnt that important.

[-]

SEI_JAKU@reddit

Well yeah, that's why this is a pretty good combination. CPUs are cheaper, so you spend more on a great CPU and spend less on a still-good GPU. A lot of what I do is CPU-bound anyway, so this combo works for me.

Really looking forward to Zen 6 at this point, if it's anywhere near as good as Zen 5 was. Crazy how people ever hated Zen 5 at all, that HUB video trashing the 9600X oughta be treated as the misinformation that it is.

[-]

aminorityofone@reddit

you completely missed the point. Please define a real world condition that can be repeatedly tested accounting for variables.

[-]

ResponsibleJudge3172@reddit

A repeatable but unusable data point is not scientifically sound to use in practice.

[-]

_I_AM_A_STRANGE_LOOP@reddit

Trying to chase pure objectivity here is a fool's errand, as things like image quality post-upscale, percieved impact of latency/frame pacing, and even what looks 'good' in the first place are entirely subjective. To do a review without a lot subjective impressions provides barely more value than looking at a collection of 3dmark results. This is the case in almost every critical field, as any product that can be assessed purely objectively loses the need for actual human reviewers. Providing context for the numbers is what provides the value, and what that looks like, is, well, this video and its choices to focus on these more nebulous aspects again from a deliberately subjective lens.

[-]

ResponsibleJudge3172@reddit

They are not even really subjective either. There are artifacts and other things that you show improved or worsened over native as DF usually does

[-]

timorous1234567890@reddit

But people use these cards in real world conditions. They're not gonna drop a 9060xt or 5060ti on a 9800x3d system.

Why not? If they play more games like Stellaris or Civ or Cities Skylines or PoE2 then CPU performance can matter a lot more than GPU performance so a 9060XT / 5060 Ti may provide plenty good enough GPU grunt for their use case.

That is the thing with the real world, not everybody exclusively plays the latest AAA games at launch...

[-]

zacker150@reddit

All real-time rendering techniques, including the Luddites' beloved rasterization, cause graphical anomalies.

The only reason we're having this argument is that bar charts can only show one dimension at a time.

[-]

aminorityofone@reddit

The nvidia bots are in full force here, even when i say dlss is the best. Even when i say do raster review and do an upscaling review (which literally 'guesses' at what the pixel should be) I suppose you hate the fact that GN only does FSR for their reviews in testing.

[-]

Strazdas1@reddit

A better question is why dont you hate that GN only does FSR? Thats objectively stupid.

[-]

zacker150@reddit

I suppose you hate the fact that GN only does FSR for their reviews in testing.

As a matter of fact, I do.

GN claims they do it to ensure an "apples-to-apples" comparison, when the real reason is that their chosen method of visualization - bar charts - only lets them compare one dimension at a time.

Instead, I propose an alternative method of visualization that I call the GPU possibilities frontier.

[-]

XYHopGuy@reddit

that's a lot of words to incorrectly describe both rendering and the scientific method.

DLSS, FSR, XeSS, PSSR Framegen all cause graphically anomalies

So does rasterization.

These software solutions to fix a hardware issue is no different than software rendering vs hardware rendering in the 90s

uh if you're referring to fixed function pipelines the solutions are completely equivalent. There is no "hard is better." The unified shader model was a gigantic step forward for graphics and not due to any approximations as you've suggested here.

AKA, the scientific method. Remove as many variables as possible and make all testing on the same platform (same mobo, ram, cpu ssd etc)

You've described an aspect experimental design, which is a component of the scientific method but not the method itself. Kinda would need a hypothesis first, but even then, the experiment would need to be designed such that it could disprove said hypothesis.

[-]

_I_AM_A_STRANGE_LOOP@reddit

It's hilarious to me that people have moved onto the idea that raster is some sort of "ground truth" representation of the scene in some nebulous way compared to RT and neural features, rather than an EXTREMELY janky collection of almost-good-enough hacks that result in an OKish output

[-]

wankthisway@reddit

here needs to be baseline performance without any software upscaling going on and then a review with upscaling on so the consumer can make their own judgement.

is that not what testing raw raster, and then testing with all that stuff on, is?

Lastly, reviewers need to review the card objectively, 'real world' is not objective and will be full of bias and incorrect testing...AKA, the scientific method.

Dude, what do you think these reviewers are doing? I feel like you're just rambling.

[-]

aminorityofone@reddit

did you read the original comment or just read mine? I was saying reviewers need to do what they are doing and not do 'real world' testing as OP said they would prefer

[-]

NGGKroze@reddit

As it should be. Some other techtubers brag how much effort and time they put (and they really do), but still don't respect the features the GPU's has.

Upscalers are big part of gaming in the last 3 years or so as well as FG. Most big games now ship with RT and some even baked in RT, so by not testing any of that, those reviewers are doing disservice to both the GPU's tech and the devs optimization.

[-]

QuixotesGhost96@reddit

VR performance?

[-]

_I_AM_A_STRANGE_LOOP@reddit

Amen, this is a league above in terms of actual consumer utility. Bar charts alone cannot tell the story anymore

[-]

gokarrt@reddit

DF are tech-forward. when the only innovation in the market is the features, they test the features.

[-]

Sevastous-of-Caria@reddit

I dont know if US or Reddit users differ a bit different. But I never see people getting excited to play mfg path traced cyberpunk on low end cards like these. I saw all of 4060ti sales happening on oem machines and pc cafes. Where Gta5, CS2 and valorant was the rage. And sadly I dont think they even know what MFG even is. Because I introduced features to multiple people buying these.

[-]