Let me give you the red pill about the graphics cards

Let me give you the red pill about the graphics cards.
Benchmarks are bullshit.
You should look at Gtexel/second (number of texture pixels whichare processed per second) of a graphics card to compare them.

and also the frame pacing

How is this any better than comparing gflops?

Graphics dev here. I can confirm that this is entirely accurate for my programs - performance is linearly correlated to texture fill rate (GT/s) in all GPUs I've tested.

>specs matter more than actual performance in a real life situation
lol ok so a car with 500 horsepower going at 100 mph is actually faster than a car with 300 horsepower going at 200 mph?

No, but a car with 500 horsepower can output more power than one with 300.

What difference does that make when it doesn't actually utilize the extra power?

in benchmarks, the benchmarker has the power to bottleneck a unit of a card he dislikes. for example, amd may choose to reserve less size on a chip for raster units, and then you apply 32x antialiasing in your benchmark and all other components in the chip would be underutilized. which is never the case, since you dont need 32x antialiasing in a normal scenario. it could be vice versa.

this. I always get pissed off by people benchmarking at 4K with 8x AA turned on..

what matters for the average consumer in descending order:

can it run my game? (60 fps in fullHD on max settings)
how much does it cost?
phew, fuck this, how much does this one cost?
can it run my game? (anywhere between 25 and 60 fps in fullHD on medium-max settings)
how loud is it?
does it look cool?
how much energy does it use?

let me explain. in the charts: core config consists of 3 elements: Unified Shaders : Texture Mapping Units : Render Output Units
if you set a high antialiasing you bottleneck render output unit. if you have high number of light sources and ambient occlusion you bottleneck your shaders. if you want unusually high frame rates, you can lower the settings but you'll end up having underutilized shaders but your bottleneck will be texture mapping units, you can still avoid render putput unit bottleneck by lowering antialiasing, but buying a card with high number of texture mapping units, which means higher gtexel fillrate, is better in long term.

I personally think that ROPs are the most valuable thing, AA is just so valuable.

so your favorite cards should be the ones which have 64 render output units or high pixel fillrate, see the core config: Unified Shaders : Texture Mapping Units : Render Output Units

A card needs to output everything it calculates, pixel throughput is definitely an important aspect of an ASIC.
Gaming benches aren't useful for exploring the technical aspects of a given card, but they are indicative of whats going on on die.

we can still avoid ROP bottleneck by not using 32x or 16x antialiasing you know. guys in amd probably think "anyone whos not a retard won't use that much of antialiasing while using hd resolutions. who needs retarded amount of AA in full hd?" they can make an antialiasing oriented card if thats what you want. if amd makes a card which is good at 64x msaa, fantards would just pick some other unit in gpus and compare cards based on it

If Polaris 10 was a 64 ROP design it would outperform the R9 390X across the board. Whether or not AA is heavily used, P10 still suffers from its lower pixel throughput compared to Hawaii.

It was just designed to be cheap. Though it does look much more impressive if you compare it to Hawaii, which its much more similar to by the numbers.

Fairly accurate.

> Though it does look much more impressive if you compare it to Hawaii,

Fuck, I meant Tahiti.

>max settings
we are talking about bottlenecks withing the gpu, and some retard drops the "max settings" phrase. just ban this faggot.
cost? just calculate gtexel/price

Basic question here...

Can I judge which card is the best for me, solely by looking at the specs? What am I looking for?

My goal is basically to run any game on the highest possible settings, and the card being as cheap as possible (most bang for buck).

Let me give you the red pill about the CPUs.
Benchmarks are bullshit.
You should look at clock frequency of a CPU to compare them.

depends on how much antialiasing will you use. if you are fine with 4x-8x you can go for texel fillrate/price or gflops/price
false gpus are parallel processors. their computing power is core count X frequency for each pipleline unit.

Is there any reason why I wouldn't want... say... 9000x AA? I understand that it's something which affects to look of lines and edges, smoothing them out or something. Is there a point in which it's pointless to continue?

google "diminishing returns". adding more processing doesnt improve quality at some point

Tessellation is the prime example for graphics right now.
64X being barely different from 16X, and entirely indistinguishable from 32X,

even if I had a high end gpu I'd still use lower antialiasing and tesellation just to keep the heat / fan noise low for better experience

pumping and hoping it helps us to put an end to fanboy shitposting

When was it mentioned that the car cannot utilize its 500 horsepower?

that was a terrible analogy anyway

Sup Forums kiddies are too dumb for that shit