AMD FX 8370 vs Intel I7 5960x [GTX 970 SLI 4K benchmarks]

[KarateHottie93]

http://linustechtips.com/main/topic/435411-gtx950-vs-r9-270x/page-3#entry5841445

Somebody wanted a budget GPU the other day and this was my reply...I could have said 270X but I didn't...

 

The perfect card for you, my man

 

http://galaxstore.net/Galax-Nvidia-Geforce-GTX-950-EXOC-WHITE-2GB_p_91.html

[HalGameGuru]

-snip-

 

If this makes you feel better, have at it. No concern of mine. If it was that easy to procure a tank I'd leap at the opportunity, its not illegal to own one.

 

I would say this has been fun, but it hasn't. It stopped being fun when I found out it wasn't variance, even though it was referred to like it was, and was in fact simple averaged framerates. And while I appreciate the acrobatics you put yourself through to attempt to defend your position, which I never disagreed with, that lower variance is better; in the face of none of what we were arguing about being what we thought it was, the fact it went on and in so vociferous a manner when no argument was made against your premise, only against the graphs themselves and their data, strikes of public embarrassment and defensiveness rather than an attempt to find the truth.

 

All I've been interested in is reconciling the data we have been given with the premises we hold for the presumed type of data it was, now that it is all reconciled I have no reason to deal with you any longer. I can go on happy in knowing no mathematical skullduggery was made use of and miscommunication and poor wording led to the whole insane exercise. I hope, in the future, to see actual variance graphs, although the lowest framerates is an interesting metric, couching it in terms of variance was quite unpleasant though.

 

And I am QUITE happy knowing the graphs are as expected, in both what they represent and how they are represented, when considering their ACTUAL data rather than what we assumed they were.

[SpaceNugget]

If this makes you feel better, have at it. No concern of mine. If it was that easy to procure a tank I'd leap at the opportunity, its not illegal to own one.

 

I would say this has been fun, but it hasn't. It stopped being fun when I found out it wasn't variance, even though it was referred to like it was, and was in fact simple averaged framerates. And while I appreciate the acrobatics you put yourself through to attempt to defend your position, which I never disagreed with, that lower variance is better; in the face of none of what we were arguing about being what we thought it was, the fact it went on and in so vociferous a manner when no argument was made against your premise, only against the graphs themselves and their data, strikes of public embarrassment and defensiveness rather than an attempt to find the truth.

 

All I've been interested in is reconciling the data we have been given with the premises we hold for the presumed type of data it was, now that it is all reconciled I have no reason to deal with you any longer. I can go on happy in knowing no mathematical skullduggery was made use of and miscommunication and poor wording led to the whole insane exercise. I hope, in the future, to see actual variance graphs, although the lowest framerates is an interesting metric, couching it in terms of variance was quite unpleasant though.

 

And I am QUITE happy knowing the graphs are as expected, in both what they represent and how they are represented, when considering their ACTUAL data rather than what we assumed they were.

Lol, high horse. admit you are wrong. you seriously think you can get a speed which by definition is a distance over a time without a distance. grow up and get real.

[Guest Strangerbob]

Id like to see benchmarks where the target fps is above 120 on 4k.

 

Like - with powerful single gpu and low game settings. Then i want to see which CPU does better. 

[patrickjp93]

you are not only amazingly ignorant but condescendingly so.

please tell me exactly how many miles per hour 15 minutes is. don't respond with a question. just a number. feel free to ask questions after but i want the first thing in the reply to be exactly how many miles per hour 15 minutes is.

You two having this pissing match is getting on my nerves. Time and FPS are not interchangeable nor reciprocal a of each other. Frame Time = time per frame or time/1frame. This can be instantaneous or average. FPS = frame/1second and can be an instantaneous or average measure. These are reciprocals. These are what may be converted into each other. With these facts on the table, please resolve this, preferably in a private chat.

[SpaceNugget]

You two having this pissing match is getting on my nerves. Time and FPS are not interchangeable nor reciprocal a of each other. Frame Time = time per frame or time/1frame. This can be instantaneous or average. FPS = frame/1second and can be an instantaneous or average measure. These are reciprocals. These are what may be converted into each other. With these facts on the table, please resolve this, preferably in a private chat.

I know. that is literally my entire point. that time is not a reciprocal of fps.

[HalGameGuru]

Lol, high horse. admit you are wrong. you seriously think you can get a speed which by definition is a distance over a time without a distance. grow up and get real.

You can take any time, put it into any other time, and put whatever unit you want in front of it, it doesn't matter, what matters is the TIME. If you cannot see that, that is on you not me. I can take any millisecond time, push it into 1000 to convert it to a full second rate and then put whatever I want to in front of it, because the rate is what mattered. 

 

Neither of us was right, as it wasn't variance, and I am GLAD for that fact because I do not want someone putting out benchmarks needing those kind of massaged numbers to make sense. I can go home happy knowing it was a misunderstanding, and WE mistook their, granted ambiguously worded, context for actual variance.

 

It was presumptive on my part to take an out of context ms reading and pump it into a FPS conversion, it has nothing to do with the maths involved nor what unit. As what I did is exactly what you do when it refers to a frame. I am done.

[HalGameGuru]

Back to more productive things.

 

The overclock.net forum thread shared earlier... either in this thread or another, has continued, at length, and gone into more specifics on some of the possible reasons behind some of the more curious results from benchmarks like these.

 

http://www.overclock.net/t/1569897/various-ashes-of-the-singularity-dx12-benchmarks/

 

DX11 or DX12 there are some interesting incongruities in what we may have expected and some interesting overlap in performance in some instances.

 

As it relates to THIS thread and its focus on GPU limited scenarios at 4k, there appear to be some software and API specific codings that can apparently skew results a bit depending on the system. Asks more questions than it answers, but in this scenario where we are looking at DX11 on nVidia hardware with different CPUs and resolutions, albeit only 2 CPUs at 2 different clocks each, may make us start to question how much of a bottleneck certain things are in these scenarios. And it provides some interesting color commentary on the tangential issue of how nVidia finds avenues of optimization in DX11 that AMD just seems less willing to invest in, and why the current disparity of AMD and nVidia hardware in DX12 seems so high.

 

If I had the money I'd still be leaning towards getting an Intel system built but it makes me feel a little better about having to run an A10 APU for the foreseeable future.

[Nena Trinity]

Awesome I guess.

[Sauron]

That's amazing! although a little expected since cpu bottlenecks are only evident at very high framerates.

[SpaceNugget]

You can take any time, put it into any other time, and put whatever unit you want in front of it, it doesn't matter, what matters is the TIME. If you cannot see that, that is on you not me. I can take any millisecond time, push it into 1000 to convert it to a full second rate and then put whatever I want to in front of it, because the rate is what mattered. 

 

Neither of us was right, as it wasn't variance, and I am GLAD for that fact because I do not want someone putting out benchmarks needing those kind of massaged numbers to make sense. I can go home happy knowing it was a misunderstanding, and WE mistook their, granted ambiguously worded, context for actual variance.

 

It was presumptive on my part to take an out of context ms reading and pump it into a FPS conversion, it has nothing to do with the maths involved nor what unit. As what I did is exactly what you do when it refers to a frame. I am done.

The reason it wasn't a variance was what I was trying to explain to you this entire time. you were just too stubborn to actually read. either that or you are a troll.

[patrickjp93]

I know. that is literally my entire point. that time is not a reciprocal of fps.

I just summarized the mathematics formally so you two would stop. I know I have a tendency to get involved in thread derailments over opinionated things, but the facts are the facts, and you two need to quit.

[HalGameGuru]

@WaxyMaxy, no you were making arguments about variance, which I never disputed, over the premise that what was in the graphs WAS variance, which it wasn't, and how we could reconcile what we know of variance representation and what was in the graph which was not displaying variance in any way the two of us KNEW it should be.

 

@patrickjp93 Thank you, appreciated, I REALLY should have read through the first page of the OP source better the first time around. Could have prevented this whole sophomoric volley from the get go.

[SpaceNugget]

@WaxyMaxy, no you were making arguments about variance, which I never disputed, over the premise that what was in the graphs WAS variance, which it wasn't, and how we could reconcile what we know of variance representation and what was in the graph which was not displaying variance in any way the two of us KNEW it should be.

 

@patrickjp93 Thank you, appreciated, I REALLY should have read through the first page of the OP source better the first time around. Could have prevented this whole sophomoric volley from the get go.

Please don't try to tell me what I am saying. You don't even know what you are saying. but I agree that this is derailing the thread so by all means I am content to let you continue wallowing in your own ignorance.

[HalGameGuru]

Please don't try to tell me what I am saying. You don't even know what you are saying. but I agree that this is derailing the thread so by all means I am content to let you continue wallowing in your own ignorance.

I don't have to tell you what you are saying, I have a copy of every post we've made so far, I can read it verbatim. I never disputed variance, I disputed these graphs.

 

 

Back to what this thread is ACTUALLY about, seeing as how this setup was basically very old architecture versus very recent architecture, would there be any profit in doing this same set of tests with something like an A10-7870K and a more recent i3 or i5 contender? Just to see how much of an effect the old FX arch is having? Or maybe an older i7 analog?

[patrickjp93]

Please don't try to tell me what I am saying. You don't even know what you are saying. but I agree that this is derailing the thread so by all means I am content to let you continue wallowing in your own ignorance.

 

 

I don't have to tell you what you are saying, I have a copy of every post we've made so far, I can read it verbatim. I never disputed variance, I disputed these graphs.

 

 

Back to what this thread is ACTUALLY about, seeing as how this setup was basically very old architecture versus very recent architecture, would there be any profit in doing this same set of tests with something like an A10-7870K and a more recent i3 or i5 contender? Just to see how much of an effect the old FX arch is having? Or maybe an older i7 analog?

If I have to get a mod involved in this pissing match, I will. Both of you, back down. You've gone in a circle. Hal admitted a wrong, and Waxy, you're being dickish. Get back to the thread's original issues.

[SpaceNugget]

I don't have to tell you what you are saying, I have a copy of every post we've made so far, I can read it verbatim. I never disputed variance, I disputed these graphs.

 

 

Back to what this thread is ACTUALLY about, seeing as how this setup was basically very old architecture versus very recent architecture, would there be any profit in doing this same set of tests with something like an A10-7870K and a more recent i3 or i5 contender? Just to see how much of an effect the old FX arch is having? Or maybe an older i7 analog?

 

 

If I have to get a mod involved in this pissing match, I will. Both of you, back down. You've gone in a circle. Hal admitted a wrong, and Waxy, you're being dickish. Get back to the thread's original issues.

 

I read back through all of our posts. I have no problems admitting I have been a dick and should have let this go way earlier. I wanted to prove a point that wasn't really needed. For that I apologise. I disagree with you Hal and that is ok. we can agree to disagree. On the other hand, I do agree with you that the confusion definitely stems from the lack of documentation on the graphs leading to lots of people arguing completely different arguments.

[HalGameGuru]

@patrickjp93 Once again, thank you for the refereeing.

 

But, honestly, that last question. These graphs cause a lot of question marks from the get go, would it be a good idea to redo these tests with more closely related recent chips?

[HalGameGuru]

I read back through all of our posts. I have no problems admitting I have been a dick and should have let this go way earlier. I wanted to prove a point that wasn't really needed. For that I apologise. I disagree with you Hal and that is ok. we can agree to disagree. On the other hand, I do agree with you that the confusion definitely stems from the lack of documentation on the graphs leading to lots of people arguing completely different arguments.

Definitely, to be entirely honest the only reason I thought it was still variance was the subtexts under the graphs in the source referring to it as variance. Once I re-read the first page and saw it was not variance, but an attempt to display a different metric (averaged lowest percentile frame rates) as an analog to the quality of experience variance is meant to illustrate I face palmed and thrashed around for a thesaurus to try to find some other word than variance to use to describe the graphs. It illustrates the point but vastly confuses the issue when an accepted term for a different metric is used.

 

 

In the overclock.net thread they were talking about how one of the main driver level optimizations nVidia uses in DX11 is cutting off the API at the pass and substituting shaders that are less effective for their hardware for more effective ones. Effectively making what an nVidia card and an AMD card are rendering two entirely different things. Is this accurate? And if so does this have a noticeable effect on these and similar benchmarks above and beyond the architectural and non-optimized driver differences?

[SpaceNugget]

Definitely, to be entirely honest the only reason I thought it was still variance was the subtexts under the graphs in the source referring to it as variance. Once I re-read the first page and saw it was not variance, but an attempt to display a different metric (averaged lowest percentile frame rates) as an analog to the quality of experience variance is meant to illustrate I face palmed and thrashed around for a thesaurus to try to find some other word than variance to use to describe the graphs. It illustrates the point but vastly confuses the issue when an accepted term for a different metric is used.

 

 

In the overclock.net thread they were talking about how one of the main driver level optimizations nVidia uses in DX11 is cutting off the API at the pass and substituting shaders that are less effective for their hardware for more effective ones. Effectively making what an nVidia card and an AMD card are rendering two entirely different things. Is this accurate? And if so does this have a noticeable effect on these benchmarks above and beyond the architectural and non-optimized driver differences?

Yea, I didn't read the reply in this thread by the guy who wrote the article until just now. it kind of clears up where you were coming from. Both of us saw that It wasn't variance from the graph and came up with different explanations of how to rationalize the misleading labeling on the graph.

 

and to address your second point, to a degree, every hardware configuration renders slightly different final products and its basically impossible to eliminate that. A common web development technique is to ask a system to render a vaguely described image on an html5 canvas and see what the system responds with. every system responds differently and you can use the image it makes up to fingerprint the system.

[HalGameGuru]

Yea, I didn't read the reply in this thread by the guy who wrote the article until just now. it kind of clears up where you were coming from. Both of us saw that It wasn't variance from the graph and came up with different explanations of how to rationalize the misleading labeling on the graph.

 

and to address your second point, to a degree, every hardware configuration renders slightly different final products and its basically impossible to eliminate that. A common web development technique is to ask a system to render a vaguely described image on an html5 canvas and see what the system responds with. every system responds differently and you can use the image it makes up to fingerprint the system.

 

I get that, its just with all these benchmarks coming out now comparing DX11 to 12, and all these different CPU arch.s being leveraged, if there might be a overarching shift we may miss because of how nVidia optimizes their performance at the driver level.

 

 

I know I got a little hot under the collar but I wasn't trying to argue the conversion I was doing was in any way the right way to do thing, just that if someone were bat-crap crazy enough to try to shoehorn variance into the graphs we were seeing that was the only way I could see the data being massaged to put up those numbers.

[patrickjp93]

@patrickjp93 Once again, thank you for the refereeing.

 

But, honestly, that last question. These graphs cause a lot of question marks from the get go, would it be a good idea to redo these tests with more closely related recent chips?

The tests need to be redone period by a non-biased 3rd party. These results smell to high heaven.

[SpaceNugget]

I get that, its just with all these benchmarks coming out now comparing DX11 to 12, and all these different CPU arch.s being leveraged, if there might be a overarching shift we may miss because of how nVidia optimizes their performance at the driver level.

 

 

I know I got a little hot under the collar but I wasn't trying to argue the conversion I was doing was in any way the right way to do thing, just that if someone were bat-crap crazy enough to try to shoehorn variance into the graphs we were seeing that was the only way I could see the data being massaged to put up those numbers.

 

(as a gamer)

I do see your point, but unless there are ways around the nvidia drivers you might as well just consider it part of the cards instead of a man in the middle. End users are stuck with the drivers they are provided, so for real world situations how the final product is rendered is kind of ir-elephant.

 

(as a fellow nerd)

now if you are a technophile and want to compare the raw horsepower of the cards to find out which is truly the best card just for the sake of it. then you have a bit of a conundrum. I guess you could engineer a benchmark that works around the api to avoid using any compromised api calls like the shader ones you mentioned. I really don't know how different the implementations of the apis are between the two or how many cheats and hacks they have, maybe such a benchmark can't exist, but if it could it would be really interesting to see the results.

 

I remember reading that some drivers detect the presence of a benchmark and temporarily raise thermal limits and voltages past the safety margins to score artificially higher.