Why are GPU comparisons often inaccurate?

[TeslaK20]

When looking at several GPU comparison lists, I came across very bizarre results. Apparently, the GeForce 750 was better than the 780 Ti at 4k gaming, and one GTX Titan was better than a Titan Z (the R9 280 was also "better" than the Titan Z). I dismissed it as nonsense and moved on. But then, I found other GPU comparison lists with nonsense results, not to mention the several times GPUBOSS messed up completely. Can someone explain this to me?

 

[BarbaricSoul]

What GPU comparisons are you looking at? The only way a GTX 750 beats a GTX 780ti is in energy efficiency. The 780ti is the strongest single GPU video card produced. The Titan Z is 2 Titan GPUs on one card. How can a Titan be better than a Titan Z. The R9 280 is just a rebranded HD7950. It is no where near as good a performer as the Titan or Titan Z.

I have no idea about the results you are referring to, but personally, I get all my information about video cards from TechpowerUp's video card reviews. Wizzard does what is considered the most accurate and consistent GPU reviews available. Just get your GPU info here and you will not have this problem.

 

[TeslaK20]

You're right. Such results are impossible. But here are some:
Crysis 3
CAD

There is no way a Geforce 760 is better than a Quadro K5000 at CAD.
And this is from TomsHardware, not NeverHeardOfItInChina.

 

[RCoon]

You're right. Such results are impossible. But here are some:
Crysis 3
CAD

There is no way a Geforce 760 is better than a Quadro K5000 at CAD.
And this is from TomsHardware, not NeverHeardOfItnChina.

Dual GPU's have a higher frame time variance because of SLI/Xfire not being 100% perfect and efficient with latency of frames. Also, bare in mind the 280X (basically a 7970) has had better drivers since it has been out for many many years longer. The Hawaii chip is entirely new, so if you wait for a year or two, those benchmarks will probably be exactly the same, except with the 290X where the 280 is, and the 790 and 8990 where the older dual GPU cards are.

Look up frame pacing.

 

[TeslaK20]

But this doesn't really explain all of it. Again, a Geforce 760 beating a K5000 defies the laws of PhysX (get it?). And drivers shouldn't cause this much variation.

 

[Shyrzos]

You're right. Such results are impossible. But here are some:
Crysis 3
CAD

There is no way a Geforce 760 is better than a Quadro K5000 at CAD.
And this is from TomsHardware, not NeverHeardOfItInChina.

In my experience(not viewing reviews) Autocad 3d is very simple,primarly use the single precision or fp32 compute power of the cards,not the fp64/aka use of cuda power of the quadros. So,according to this,the most fp32 or single precision power(its directly compared to fps in games) of the cards,better points/fps in simple 3d programs,like autocad.
This not applies to profesional 3d programs,like Maya and Max,a difference of day and night of quality and the last ones,use the full gp64 compute capabilities of the quadro,improving framerates/stability(and less graphical bugs too) a lot versus the consumer cards

 

[TeslaK20]

Even so, the 780 or the Titan would be at the top, not the 760. But it is!

 

[Shyrzos]

Even so, the 780 or the Titan would be at the top, not the 760. But it is!

Sorry TeslaK20, i didn´t see the other gtx cards lol. However the differences are very small,but the results are innacurate after all,as you say

This is from pudget systems,another benchmark in autocad 2013 3d,for comparison:

http://images.pugetsystems.com/images/pic_disp.php?id=21684&width=800&height=800

 

[TeslaK20]

I have no idea about the results you are referring to, but personally, I get all my information about video cards from TechpowerUp's video card reviews. Wizzard does what is considered the most accurate and consistent GPU reviews available. Just get your GPU info here and you will not have this problem.

Thank you for suggesting it. While I am still curious what causes it (probably a combination of things), thank you for directing me to a good source. I checked it and it looks way more accurate.

 

[Frag_Maniac]

It's largely because different GPUs play different games with different results, and not everyone benches the same. It can even depend what level of the game is benched or what segment within a given level, and a LOT of reviews won't give that info. You hardly ever see two benches that were conducted exactly the same, and if everyone used the same in-game bench or segment of game, it would still be impossible to tell how each GPU compares overall. Worse yet, fanboy raves, personal preferences, differences in hardware & software configs, etc, etc, can bias end users pro and con, so it can be hard to tell from their input too.

Am I saying the only way to know for sure is try each one, you're damn right I am. Same goes for how well a given GPU OCs. It's hit and miss.

 

[TeslaK20]

It's largely because different GPUs play different games with different results, and not everyone benches the same. It can even depend what level of the game is benched or what segment within a given level, and a LOT of reviews won't give that info. You hardly ever see two benches that were conducted exactly the same, and if everyone used the same in-game bench or segment of game, it would still be impossible to tell how each GPU compares overall. Worse yet, fanboy raves, personal preferences, differences in hardware & software configs, etc, etc, can bias end users pro and con, so it can be hard to tell from their input too.

Am I saying the only way to know for sure is try each one, you're damn right I am. Same goes for how well a given GPU OCs. It's hit and miss.

Well, I'm not about about to start buying every gpu on that list and benchmarking it. I am not the founder of the company that makes my computer's operating system.

 

[Chetkigaming]

autocad just cannot utilize all given resources in best way, thats all.

 

[Frag_Maniac]

Well, I'm not about about to start buying every gpu on that list and benchmarking it. I am not the founder of the company that makes my computer's operating system.

That's not the point. The point is you can't be sure you'll be getting accurate enough info by reading benches, so it's best to buy from a source that is easy to refund with, like Newegg. That way if one card you try is not what you expected, you can try another. This is why it's a good idea to narrow down your choices to one solid option from each chip vendor.