Do GPU Benchmarks Really Matter?

[MEC-777]

So throughout the many hours I've spent researching the various GTX 970 cards on the market in trying to decide which one to buy, I've started to notice an interesting trend...

 

Generally speaking, when I read the customer reviews, there tends to be two different types of reviews:

 

1) Those who overclock, push the card to it's limits, run benchmarks (firestrike, furmark, valley, heaven, etc.) and report on how the card performs in regard to those aspects. 

 

2) Those who simply buy the card just to play games, don't care much about overclocking and don't usually run or report on benchmarking scores.

 

 

As you know, there are "entry-level" cards (EVGA, PNY blower cards and pretty much any stock-clocked cards), mid-range (MSI tiger OC, EVGA superclocked/SSC, Gigabyte Windforce, etc.),  and the tweaked, enthusiast cards (MSI Gaming, Gigabyte G1, Asus STRIX, EVGA FTW, Zotac AMP! edition, etc.).

 

Now, what I've been noticing is that the types of people from group 1 will tend to give the entry-level and mid-range cards lower marks, simply because they either don't OC as well, ran hotter when pushed, throttled when pushed, and or didn't score as high in benchmarks as a result of being pushed. But those from group 2 were mostly very happy with the performance of these same entry-level and mid-range cards because they performed extremely well in gaming.

 

So I guess what I'm asking is; why do so many of us put so much emphasis on overclocking and benchmarking? When you really boil it down, we buy these cards to play games. Sure, I understand a decent OC on one of the higher-end cards can enable it to match the performance of a 980, which is really awesome. But when you're playing games at well over 60fps on any of the cards, entry-level or high-end, does it really matter if it can hit 1500mhz or not, or achieve a certain score in firestrike?

 

You have one person saying "don't buy this EVGA superclocked blower card because it runs hot/loud and throttles when overclocked to 1400+..." and then you have another person saying "This card is great, not too loud, not too hot, amazing performance in any game I throw at it with max ultra settings..."

 

So for someone like me, who may do a little bit of overclocking (nothing extreme, maybe +100-150 core) and cares more about actual in-game performance, it just makes it that much more difficult to decide. 

 

What do you guys think? Do benchmarks really matter or do you just worry more about actual gaming performance? 

[AlwaysFSX]

You're comparing a trend to usefulness. Of course benchmarks matter. They work, don't they? It's a quantifiable value to determine how a card ranks. Who and how they're used doesn't matter.

[CowsGoRoar]

Benchmarks are something I refer to, but I always look for real game performance, noise, cooling and other things like that.

[Enderman]

Thats because the people who don't run proper benchmarks are inexperienced and don't understand that "my GPU can play games" is different from "my GPU can play games at ultra settings"

[OriginalPromise]

To answer your question, cards that score higher on benchmarks should get better performance in gaming.

There are many different factors that determine "real-life gaming performance", so it really depends.

But generally speaking, higher is better.

[Shaqo_Wyn]

Depends on what you're looking for in a GPU. I agree with you, as long as it can play most or all games at 60+ fps on the resolution I want I'm happy. However, I also want it to do this while being reasonably quiet and maxed out at 80 degrees. Good benchmarks cover this aspect too.

 

When looking at a GPU for say BF4 it's nice to see benchmarks confirming that the GPU you're looking at will achieve the desired result. So in this way benchmarks do matter. 

 

As for overclocking, I don't really care about having the best overclocking GPU out there. Those are silly e-peen problems one shouldn't be concerned with IMO.

[TheRetiredSlave]

I don't look at benchmarks as being rock-solid numbers; I view them as being general guides to a GPU's performance..

 

If I read a review and see that [card 1] beats [card 2] in most tests, I know that [card 1] will be a stronger performer in any game that I'm going to play. If I look at a chart that shows [card 1] running games at an average of ~50 FPS on the game's Ultra preset, I know that I'm going to need to tweak a few settings down below ultra in order to achieve an average of 60 FPS.. If that same chart shows [card 2] running the same games at ~45 FPS at the Ultra preset, I know that I'd have to tweak a few more settings lower in order to get a solid 60.

 

As for comparing the same GPU from two different manufacturers, I don't think it really matters that much. The performance of any given GPU will be mostly bound by the GPU itself, and an extra couple Mhz won't change *that* much.. If you're only going for ligher overclocks (+100 Mhz or so), then the difference between two models matters even less. Buy whichever you think looks cooler or matches your color scheme better..

[MEC-777]

I understand what benchmarking is for and that it is a good tool for pitting various cards against one another to see how they all stack up, but benchmarking software does not necessarily reflect gaming performance and loads in the same way. Furmark, for example, is brutal on GPUs and may show higher than normal temps and throttling, but playing an actual game - even a highly GPU demanding game, may not show the same higher than normal temps and throttling. My point here is that some may report throttling or abnormal performance/behavior that would otherwise almost never be observed during actual game play, skewing the judgment of the overall performance of a given card. 

 

I too play around with benchmarks, just for fun and to see how my system performs compared to others, but I don't consider those results as meaningful beyond that. If the card runs my games maxed out, butter smooth, with reasonable temps and never throttles then I'm more than happy and as far as I'm concerned, this makes the abnormal performance during benchmarking (if observed) mostly irrelevant.  

 

@Enderman - Just because someone doesn't run benchmarks doesn't mean they are inexperienced and doesn't know the difference between running a game and running a game at ultra settings. That's just BS. A good friend of mine who is very PC/tech savvy and a hardcore PC gamer, never runs any benchmarks, but knows very well the difference in game settings. He just simply doesn't care about benchmark scores. He cares whether his games look good, run smooth and fast at max settings and whether his GPU is reliable in delivering said performance. 

[s4spencer]

A lot of the reviews for OCing and aftermarket coolers (MSI/gigabyte/EVGA/etc.) come from people who put a considerable amount of time drooling over specs and performance. Lets say a 970 stock vs a 970 windforce. When it comes down to it, does the cooling and OC'ing mean anything to someone simply wanting to play a pretty game at a decent frame rate? No, not at the moment at least. These are higher end cards that can handle any game just fine. The best bang-for-your-buck is the $100 cheaper stock options. But then there's the issue of "this game in 2016 would be playable if I could just snag another 10 FPS". If you plan on upgrading frequently, and high settings are the same as ultra to you, then your wasting your $100 on branded cards, especially if tech doesn't impress you

[Enderman]

@Enderman - Just because someone doesn't run benchmarks doesn't mean they are inexperienced and doesn't know the difference between running a game and running a game at ultra settings. That's just BS. A good friend of mine who is very PC/tech savvy and a hardcore PC gamer, never runs any benchmarks, but knows very well the difference in game settings. He just simply doesn't care about benchmark scores. He cares whether his games run smooth and fast at max settings and whether his GPU is reliable in delivering said performance. 

Well making sure your games run smooth and fast at max settings is basically called a benchmark...

[MEC-777]

Well making sure your games run smooth and fast at max settings is basically called a benchmark...

By benchmarking, I'm referring to benchmarking software like firestrike etc. My friend never runs any of that. What he does do is runs fraps in-game to see how it's running. In-game benchmarking is the most useful benchmarks. IMO, as they show actual game performance, not just some "score". 

 

@s4spencer - That's a good point. By stressing the card now, it gives you an idea of how it might handle even more demanding games in the years to come. 

 

I don't plan on upgrading frequently and high settings are not the same as ultra settings to me. So then perhaps I should opt for one of the more "higher-end" variants...

[Enderman]

By benchmarking, I'm referring to benchmarking software like firestrike etc. My friend never runs any of that. What he does do is runs fraps in-game to see how it's running. In-game benchmarking is the most useful benchmarks. IMO, as they show actual game performance, not just some "score". 

 

@s4spencer - That's a good point. By stressing the card now, it gives you an idea of how it might handle even more demanding games in the years to come. 

 

I don't plan on upgrading frequently and high settings are not the same as ultra settings to me. So then perhaps I should opt for one of the more "higher-end" variants...

That's what I meant...

In-game benchmarks are what most people do, and they are still considered benchmarks.

 

If someone buys a graphics card and doesn't even look at the fps they are getting then they are 'inexperienced' and don't know how to compare performance.

 

So your title is a bit misleading because in game fps is a benchmark too, not just a program like firestrike. You should have specified whether you were talking about 'artificial benchmarks' instead of 'all benchmarks'

[MEC-777]

That's what I meant...

In-game benchmarks are what most people do, and they are still considered benchmarks.

 

If someone buys a graphics card and doesn't even look at the fps they are getting then they are 'inexperienced' and don't know how to compare performance.

 

So your title is a bit misleading because in game fps is a benchmark too, not just a program like firestrike. You should have specified whether you were talking about 'artificial benchmarks' instead of 'all benchmarks'

Yeah, perhaps I should have titled it 'Do Synthetic Benchmarks Matter?" instead. Too late to edit. Oh well, good discussion anyways.

[Briggsy]

Synthetic benchmarks can be optimized more than games. In the specific case of Maxwell, 3Dmark Firestrike is an excellent choice because there is not a broad range of colours being rendered, and the background sometimes stays still for periods of time, and Maxwell's new colour compression absolutely loves it. load up Valley where the colours and scenery are changing frame by frame, and the colour compression is useless.

 

do synthetic benchmarks matter? yes, if only to create an equal playing field to compare apples to apples. once you start comparing different brands or architectures, the equal playing field turns to mud.