Which benchmark translates most realistically?

[skaughtz]

What I mean by the title is, between 3DMark/Unigine/PassMark/Etc., is there one benchmark tool that most/more accurately gives you an idea of how your hardware is going to do playing modern games at specific settings?  For instance, if you take two different processors and pair them with the same GPU and then run a benchmark, which would be best to boil down the results to say, "this system is going to get you X more/less FPS in modern games than the other system at these settings"?

 

Perhaps it is a stupid question, but with 3DMark for example, the different tests produce large disparities for the combined test scores, which themselves don't seem to reflect real-world use.  My i5-3570k at 4.2Ghz and RX 580 8GB system can pretty much run most games at max settings at 1080p at 60+FPS.  However, in Fire Strike, which is a DX11 benchmark rendered at 1080p, the 2 graphics scores average 69.52 FPS, while the combined score for that setup is only 23.78 FPS.  Meanwhile, an old A8-7670k at 4.4Ghz I have (think an X4 860K) with an RX 580 averages 67.06 FPS for the graphics scores and 12.02 FPS for the combined score.  Am I misunderstanding the results?  Would the second processor only get me half of the game performance of the first?  I guess that makes sense given those two examples, but where would they diverge?  Low?  Medium?  Max settings?  I don't think you are getting that kind of information based on what 3DMark tells you.

 

TL;DR:  If you wanted to test CPU A vs CPU B, paired with the same GPU, and see which is going to get you better results playing a modern game at 1080p at different settings, what benchmark would be best to use?

[NunoLava1998]

The game itself

[Ankerson]

17 minutes ago, skaughtz said:

What I mean by the title is, between 3DMark/Unigine/PassMark/Etc., is there one benchmark tool that most/more accurately gives you an idea of how your hardware is going to do playing modern games at specific settings?  For instance, if you take two different processors and pair them with the same GPU and then run a benchmark, which would be best to boil down the results to say, "this system is going to get you X more/less FPS in modern games than the other system at these settings"?

 

Perhaps it is a stupid question, but with 3DMark for example, the different tests produce large disparities for the combined test scores, which themselves don't seem to reflect real-world use.  My i5-3570k at 4.2Ghz and RX 580 8GB system can pretty much run most games at max settings at 1080p at 60+FPS.  However, in Fire Strike, which is a DX11 benchmark rendered at 1080p, the 2 graphics scores average 69.52 FPS, while the combined score for that setup is only 23.78 FPS.  Meanwhile, an old A8-7670k at 4.4Ghz I have (think an X4 860K) with an RX 580 averages 67.06 FPS for the graphics scores and 12.02 FPS for the combined score.  Am I misunderstanding the results?  Would the second processor only get me half of the game performance of the first?  I guess that makes sense given those two examples, but where would they diverge?  Low?  Medium?  Max settings?  I don't think you are getting that kind of information based on what 3DMark tells you.

 

TL;DR:  If you wanted to test CPU A vs CPU B, paired with the same GPU, and see which is going to get you better results playing a modern game at 1080p at different settings, what benchmark would be best to use?

 

 

3DMARK is excellent and is stable and repeatable and an excellent way to compare to other systems, actually the best way in reality because it validates the scores. It is not easy on the system at all.

 

Unigine Heaven, Valley and Superposition are also good and more gaming like with more real world like results.

 

Passmark is good for evaluating the whole system and comparing to others. 

[Mira Yurizaki]

20 minutes ago, skaughtz said:

TL;DR:  If you wanted to test CPU A vs CPU B, paired with the same GPU, and see which is going to get you better results playing a modern game at 1080p at different settings, what benchmark would be best to use?

If you're going to limit it to one benchmark suite, 3DMark would be it since you can test with a variety of different levels of graphical quality with mixed workloads.

 

Otherwise it would be better to run games with a benchmarking mode like GTA V, Rise/Shadow of the Tomb Raider, or The Division (1 or 2).

[Mister Woof]

21 minutes ago, NunoLava1998 said:

The game itself

This, some games favor CPU or GPU, some favor single core CPU performance, some threads, some prefer AMD over Nvidia, etc.

 

Each engine is different.