Geforce GTX 1060 Thread: faster than RX 480, 120W, $249

[hsjj3]

3dMark of 1060 @ 1911mhz vs RX 480 @ 1350mhz:

http://www.3dmark.com/compare/fs/9202637/fs/9188962#

It looks like the 1060 may catch up at over 2ghz.

Come on. That's a fully stock GTX 1060 vs a decently OCed RX 480 (6.6% core and 10% memory).

The GTX 1060 would also be able to do the same over clocks, if not more, and achieve at least 10% higher score.

 

[hsjj3]

1911MHz is the stock result for 1060?

Yes. I've seen FE 1070/80 boosting that high at stock settings. GPU Boost 3.0 does take the clock speeds much higher over the rated boost, though sadly most of it is a meaningless initial peak.

 

[littleg]

With GPU boost I think reviewers are going to have to change their benchmarking methods a bit. It's no longer good enough to just fire up a game, run the benchmark and get the result. For a meaningful analysis the card will need to be heated up in a gaming scenario for a good 20-30 minutes before benching and it needs to be done in a closed case like most of us use.

 

[brandonmatic]

Personally, I'd say that a harmonic mean difference of 5% across a wide and representative sample of games in favor of the 1060 would make it legitimately "5% faster." Otherwise it's probably not a particularly meaningful statement.

 

[boozzer]

Personally, I'd say that a harmonic mean difference of 5% across a wide and representative sample of games in favor of the 1060 would make it legitimately "5% faster."

that still hinges on if they only include outliners like project cars + tomb raider and not hitman. and even if all 3 are included, the results would be skewed towards [+] because it is 2 outliners vs 1.

sponsored games needs to be dropped from any and all benchmarks, unless it can be proven to be fair. :thumbsup:

 

[Arachnotronic]

that still hinges on if they only include outliners like project cars + tomb raider and not hitman. and even if all 3 are included, the results would be skewed towards [+] because it is 2 outliners vs 1.

sponsored games needs to be dropped from any and all benchmarks, unless it can be proven to be fair. :thumbsup:

People play sponsored games, so they are relevant to GPU buyers.

 

[Silverforce11]

Yes. I've seen FE 1070/80 boosting that high at stock settings. GPU Boost 3.0 does take the clock speeds much higher over the rated boost, though sadly most of it is a meaningless initial peak.

It's not a meaningless peak if it's for a short benchmark, which is how many review sites do theirs actually.

 

[Arachnotronic]

It's not a meaningless peak if it's for a short benchmark, which is how many review sites do theirs actually.

My 1080 FEs sustain >1733 over very long game play sessions at stock. I don't run them at stock anymore, though.

 

[boozzer]

People play sponsored games, so they are relevant to GPU buyers.

then do so individually, don't include them in the overall score like the ones from TPU. voila. :thumbsup:

 

[Arachnotronic]

then do so individually, don't include them in the overall score like the ones from TPU. voila. :thumbsup:

Why not, though?

 

[boozzer]

Why not, though?

skewed results that is why. it is how avgs work, the first thing you do is to take out outliners. one of the first thing I learned in my lap days. especially since they are not naturally occurring outliners but sponsored ones. that is why they need to be taken out of any avgs that wants to be taken seriously.

 

[brandonmatic]

that still hinges on if they only include outliners like project cars + tomb raider and not hitman. and even if all 3 are included, the results would be skewed towards [+] because it is 2 outliners vs 1.

sponsored games needs to be dropped from any and all benchmarks, unless it can be proven to be fair. :thumbsup:

I think the harmonic mean is meant to deal with outliers. They're included, but they have a much lower effect on the average. Seems like a reasonable way of dealing with games like Project Cars without excluding them altogether.

 

[hsjj3]

So is there any way to achieve sustained high boost clocks? If no, is it possible to at least disable GPU Boost 3.0 partially? It's kind of irritating to the OCD people cause the clock speeds are dropping. Would rather see one constant line. GPU Boost 2.0 wasn't as bad.

And yeah, as mentioned above, this gives rise to higher than deserved benchmarking scores.

 

[JDG1980]

People play sponsored games, so they are relevant to GPU buyers.

I agree that big-name titles like the new Tomb Raider game should be included even if they favor one GPU vendor. But the use of Project CARS as a benchmark was much harder to justify. It wasn't a very popular game and it was grossly biased in favor of Nvidia, perhaps more so than any other title. (And if you want to criticize Ashes of the Singularity, at least AotS is the harbinger of new technologies like Async Compute and Explicit Multi-GPU. What real innovation did Project CARS bring to the table?)

 

[hsjj3]

What I fail to understand is, why does Hitman even on DX11 absolutely suck on Nvidia cards compared to AMD? The gulf between the two is the widest I've ever seen in DX11 for an AAA game.

 

[littleg]

I agree that big-name titles like the new Tomb Raider game should be included even if they favor one GPU vendor. But the use of Project CARS as a benchmark was much harder to justify. It wasn't a very popular game and it was grossly biased in favor of Nvidia, perhaps more so than any other title. (And if you want to criticize Ashes of the Singularity, at least AotS is the harbinger of new technologies like Async Compute and Explicit Multi-GPU. What real innovation did Project CARS bring to the table?)

As an individual benchmark or data point it's fine but when you're compiling averages like TPU does then you need to remove outliers.

 

[SolMiester]

While they should know better that doesn't exonerate the IHV for lying or over hyping.

who what when?

 

[truckerCLOCK]

Why not, though?

Defeats the purpose of an A/B test. Why even bother testing the cards against each other. Playing field has to be level.

 

[boozzer]

I think the harmonic mean is meant to deal with outliers. They're included, but they have a much lower effect on the average. Seems like a reasonable way of dealing with games like Project Cars without excluding them altogether.

http://mathworld.wolfram.com/HarmonicMean.html
https://www.youtube.com/watch?v=F8_5GgSrxE8 video guide
still doesn't solved the problem of sponsored/wrong data being used to calculate the avg. end result becomes useless.

how could the end result be correct when the data use to calculate it is wrong to begin with? :thumbsdown:

if you could explain how harmonic mean can deal with false data inputs, I am all ears.

 

[antihelten]

how could the end result be correct when the data use to calculate it is wrong to begin with? :thumbsdown:

A game being biased towards on vendor or the other doesn't make it wrong. Besides all games are biased to some degree.

You could argue that they are not representative of a given person's game patterns, but then again they might also be perfectly representative of another person's game patterns. For instance I wouldn't be surprised if there are people out there who has basically played nothing but Project CARS since it came out, given that it is one of the premier car sim games out there.

So as a tech site constructing a suite of games for reviews, the best you can really do is look at what games people are playing today (using steamspy for instance), but also try and incorporate games that you think may be representative of future game tech patterns to insure that your readers can also have somewhat of an idea of future performance (Ashes could for instance be argued to be one such game). Of course trying to predict what will be popular in the future is an inherently error prone exercise. If a game like Project CARS fits into either of the above two criteria (popular today or representative of tomorrow's tech), then it is perfectly reasonable to include it.

 

[IllogicalGlory]

http://mathworld.wolfram.com/HarmonicMean.html
https://www.youtube.com/watch?v=F8_5GgSrxE8 video guide
still doesn't solved the problem of sponsored/wrong data being used to calculate the avg. end result becomes useless.

how could the end result be correct when the data use to calculate it is wrong to begin with? :thumbsdown:

if you could explain how harmonic mean can deal with false data inputs, I am all ears.

It's effective for dealing with outliers on the large end of scale, but enhances the effect of outliers on the low end.

Since the harmonic mean of a list of numbers tends strongly toward the least elements of the list, it tends (compared to the arithmetic mean) to mitigate the impact of large outliers and aggravate the impact of small ones.

 

[MarkizSchnitzel]

skewed results that is why. it is how avgs work, the first thing you do is to take out outliners. one of the first thing I learned in my lap days. especially since they are not naturally occurring outliners but sponsored ones. that is why they need to be taken out of any avgs that wants to be taken seriously.

But the outliner here would be a game that is obviously favouring one side, and it's NOT popular.

For example, if 5 top selling games were to favour one or the other company, would you take outliners as well? Those results would not be relevant for the people they are actually meant for, gamers. Just saying.

 

[boozzer]

It's effective for dealing with outliers on the large end of scale, but enhances the effect of outliers on the low end.

so essentially useless and actually the worst possible case for the use of harmonic means then. score 1 = me.

But the outliner here would be a game that is obviously favouring one side, and it's NOT popular.

For example, if 5 top selling games were to favour one or the other company, would you take outliners as well? Those results would not be relevant for the people they are actually meant for, gamers. Just saying.

doesn't matter if the games are popular or not, as long as it is sponsored data, it become useless when you want to use it to calculate the avg.

the data is representative of each sponsored game but not the avg. that is why I stated that each of those 5 games should have their own benchmarks but 100% not be included when you calculate the avg. so gamers of those popular games would still have their benchmarks. :thumbsup:

the only point where your scenario even remotely makes sense is when sponsored games is 50+% of all the games use in the benchmarks. but by that point, there should just be 2 averages, 1 for the sponsored games and one for the ones without. voila. :thumbsup:

why hide the fact they are sponsored games? give gamers all the info and let them decide. when you hide, that means you got something to hide. :thumbsdown:

 

[Mikeduffy]

All this arguing about which games to use - just use a random pick of games released within the past year. This makes perfect sense to me. I would also attempt to disable vendor specific options like HBAO+ and TressFX.

I was so puzzled as to why Anandtech used Crysis 3 for the 480 review - who cares how well it performs on a old game.

 

[boozzer]

All this arguing about which games to use - just use a random pick of games released within the past year. This makes perfect sense to me. I would also attempt to disable vendor specific options like HBAO+ and TressFX.

I was so puzzled as to why Anandtech used Crysis 3 for the 480 review - who cares how well it performs on a old game.

this would 100% be fine if there is no avg included in the given review. but you know the marketing people love their avgs.