Fermi's lead over Cypress shrinks w/ new drivers

[Paratus]

I find it very interesting that the two 'real world' reviews that I've read, this one and [ H ], find the competing cards tob much closer to each other than canned benchmarks suggest.

 

[luv2increase]

I find it very interesting that the two 'real world' reviews that I've read, this one and [ H ], find the competing cards tob much closer to each other than canned benchmarks suggest.

Last I checked, I don't find the select few games they chose as evidence of the true performance between the cards.

I think it is even more interesting how different review sites show different results all the time. I see it happen all the time for awhile now. One review site A with driver set X will have worse results than review site B with the same driver set X and so on. To be blunt, I don't think any of us can take any GPU review "completely" seriously because there are always biases and money to be made.

I don't doubt one bit that reviews are swayed to either the reviewer's preference or determined by who paid him/her/them the most money. What better investment for AMD and Nvidia then to get to the heart of those who are going to sway the minds of the consumers.

 

[tincart]

Why are you harping on with this patent nonsense?

The 5870 and GTX480 are "very" expensive graphics cards.

False equivalence. The GTX 480 is in a different price bracket from the 5870 and yet you seem very interested in presenting them as equal. This is not the case. The 5870 is expensive. The GTX 480 is "very" expensive. The 5970 is very very expensive.

To buy an expensive graphics card you have to:

1. Have the money
2. Have an unsatiable appetite for increases in gaming performance

Re #1: Trivial. You need money to buy anything
Re #2: Insatiable. Are you spouting out some canned marketing slogans for us? I don't buy a new graphics card because I have an appetite for increases in gaming performance, I buy one so I can play the games I want at a good frame rate. The vast majority of buyers probably don't care about performance beyond having a fully playable (e.g. 60fps) experience.

What we loosely term "enthusiasts" do care about such things and they are a minute portion of the overall market.

Both these points are trivial.

Due to facts 1 & 2, MANY people buying the 5870 and GTX480 will buy more than just one for multi-GPU gaming goodness. Due to this, seeing benchmarks of single-GPU performance differences between the 5870 and GTX480 simply don't matter if you have 2 x 480s or 2 x 5870s or you want or plan to buy 2 x 480s or 2 x 5870s. Due to this, multi-GPU performances differences between the 5870 and GTX480 are more important than their single-GPU config results.

Show me the evidence.

Since you are so confident about this assertion that you had to bold your words, I want to see your sources. Exactly how many people buy multiple high-end cards? Which industry sales information are you using to come to that conclusion? What is the ratio of people who buy one vs. two cards? What % of total sales do current CF or SLI purchasers represent?

Either answer me these specific questions and provide me with reliable sources for that information or admit you have no idea what you're talking about.

 

[luv2increase]

Why are you harping on with this patent nonsense?



False equivalence. The GTX 480 is in a different price bracket from the 5870 and yet you seem very interested in presenting them as equal. This is not the case. The 5870 is expensive. The GTX 480 is "very" expensive. The 5970 is very very expensive.



Re #1: Trivial. You need money to buy anything
Re #2: Insatiable. Are you spouting out some canned marketing slogans for us? I don't buy a new graphics card because I have an appetite for increases in gaming performance, I buy one so I can play the games I want at a good frame rate. The vast majority of buyers probably don't care about performance beyond having a fully playable (e.g. 60fps) experience.

What we loosely term "enthusiasts" do care about such things and they are a minute portion of the overall market.

Both these points are trivial.



Show me the evidence.

Since you are so confident about this assertion that you had to bold your words, I want to see your sources. Exactly how many people buy multiple high-end cards? Which industry sales information are you using to come to that conclusion? What is the ratio of people who buy one vs. two cards? What % of total sales do current CF or SLI purchasers represent?

Either answer me these specific questions and provide me with reliable sources for that information or admit you have no idea what you're talking about.

Are you serious? If we did a poll right now out of all the members of the Anandtech forums who have the 5870 and 480, I bet you'd be quite surprised to see just how many have more than one. I don't think you've cruised around forums long enough to realize what "enthusiasts" do. I don't think you know what the word "enthusiast" means when it comes to PC hardware.

When I said "have the money", I obviously meant an adequate money supply that won't break the bank and take from the money needed to pay necessities like living expenses. How you could of inferred anything else is beyond me and quite troubling. The nitpicking is odd which you've displayed here.

You are trying to lump "average PC gamers" and merge them with the "enthusiast" crowd. Maybe you haven't been around long enough to realize, but they are different sets of people. lol

Lastly, I am really dumbfounded about how you took the word insatiable and turned into some sort of twilight zoned marketing blah blah --> I am at a lost for words. Oh my goodness ROFL

Pheww... :\

 

[konakona]

Maybe you went a bit too far when you stated them as "facts". I do agree with you personally though, if I had enough $$$ to get a 480 at all, I would probably get a couple AND most definitely watercool it. In the real world, I would get a 5850 instead which is what I exactly did.

 

[HurleyBird]

I have the money, I have the drive for performance.... and I have a single 5870.

 

[Nox51]

I have the money, I have the drive for performance.... and I have a single 5870.

For some reason that made giggle.

Right, I'll take my hat and leave, don't mind me.

 

[3DVagabond]

Strange how everyone is now avoiding the elephant in the room. This thread is (nominally) about the Techspot findings regarding real-world performance of 4 cards. I can't fathom how CF vs. SLI scaling is taking priority in discussion.

It's called a straw man argument. Gets the discussion centered on a strength of Fermi, SLI scaling, and away from the one on one real world gaming comparison that the review talks about.

 

[Will Robinson]

Those tests results look pretty reasonable to me.
So current DX11 video card standings:
1.The Overdog HD5970
2.Fermi GTX480
3.Cypress HD5870
4.AMD HD5850
5.Fermi GTX470
6.ATi HD5830
7.NVidia GTX465
8.AMD 5770

How's that look?^_^

 

[Tsavo]

Strange how everyone is now avoiding the elephant in the room. This thread is (nominally) about the Techspot findings regarding real-world performance of 4 cards. I can't fathom how CF vs. SLI scaling is taking priority in discussion.

These benches don't prove anything because the bench runs themselves will vary from run to run on the same card. The results can't be compared to each other on the same card much less on a different card.

 

[Lonyo]

These benches don't prove anything because the bench runs themselves will vary from run to run on the same card. The results can't be compared to each other on the same card much less on a different card.

Depends on the testing methodology.
If they do say 5 runs and get close results (within normal benchmark variance), then it's reasonable.

It doesn't seem to indicate anywhere they have done that though, so it is a good idea to be slightly wary of the results, but there is no reason to discredit them entirely because simply if these results can be obtained, then it warrants further testing in real world situations to see how reliable they are, and to reconsider the various cards from a value proposition and be more critical of canned benchmarks.

 

[Tsavo]

Depends on the testing methodology.
If they do say 5 runs and get close results (within normal benchmark variance), then it's reasonable.

It doesn't seem to indicate anywhere they have done that though, so it is a good idea to be slightly wary of the results, but there is no reason to discredit them entirely because simply if these results can be obtained, then it warrants further testing in real world situations to see how reliable they are, and to reconsider the various cards from a value proposition and be more critical of canned benchmarks.

There's no way for anyone to reproduce their benchmarks. No reproducibility = no validity.

What I'd like to see is all the popular games offering the ability to record a custom timedemo that others could use on their own hardware.

 

[Nox51]

There's no way for anyone to reproduce their benchmarks. No reproducibility = no validity.

What I'd like to see is all the popular games offering the ability to record a custom timedemo that others could use on their own hardware.

I'd dosagree, if you follow their exact same method and get results within a few percentage points it should be valid.

 

[Lonyo]

There's no way for anyone to reproduce their benchmarks. No reproducibility = no validity.

What I'd like to see is all the popular games offering the ability to record a custom timedemo that others could use on their own hardware.

Even canned demos don't always show the exact same performance. Most sites talk about "margin of error". To say the results aren't valid because they don't use a timedemo means that anyone's real world experiences of playing games isn't valid. Which is ludicrous, because most people buy graphics cards to play games and not to run benchmarks.
If anything these results are more valid for a consumer, because they reflect (according to the site) real world gameplay experience.
http://www.anandtech.com/show/2233/4

During actual game play the perception was the CrossFire setups were more fluid and "seemed" faster than the single card setups although the net timedemos said otherwise. We also ran our custom timedemo and the scores were just the opposite with the 975X scoring up to 12% better.
At present, we would have to say that our Q4 test is almost completely CPU limited, and this is often not the case during actual gaming.

Just one example of where built in demos vs custom runs show differences, and gameplay shows a difference to canned benchmarks.

http://www.anandtech.com/show/2673/3
Here's built in benchmarks vs a custom benchmark in Far Cry 2 (like you want). It goes from the GTX280 beating the HD48701GB and GTX260 beating HD4870 512MB to the complete reverse when the custom timedemo is run.

Also in some manner they seem to have reproduced the gameplay throughout their runs (maybe they did record a run through and get the computer to replay through it), since the graphs show similar peaks and troughs. Even if you don't have access to the benchmark they used, they do seem to have been able to maintain some consistency which indicates there results are reasonably accurate and consistent. (I.E the FRAPS framerate graphs show similar performance peaks and troughs).
You would also think that playing the games would reproduce their benchmarks (in terms of general trends, not specific results) and give a congruent picture.

I'd say this testing, even if the end user can't reproduce it (while the site seems to have been able to by somehow recording custom runs or by playing through the same bit of the game), is a lot more valid than the majority of canned benchmarks which often don't reflect real world performance.

 

[Tsavo]

Even canned demos don't always show the exact same performance. Most sites talk about "margin of error". To say the results aren't valid because they don't use a timedemo means that anyone's real world experiences of playing games isn't valid.

That wasn't the most idiotic thing I've ever read.

It was close though!

 

[Keysplayr]

I would have to say, that if TR is going to release benchmarks contradictory to most that are out there, then they most definitely provide captured videos showing exactly what they did, what courses thay had run, exact settings for said capture and showing fraps fps counter in each. This would allow for reproducability by us and other web sites allowing us to confirm their findings. I feel I'm not alone in thinking that providing a means of reproducability would either remove doubt, or confirm it.

 

[zebrax2]

Does it have to be the same. Can't one just do a control run themselves then check if their result follow the same pattern as techspot?

 

[Lonyo]

I would have to say, that if TR is going to release benchmarks contradictory to most that are out there, then they most definitely provide captured videos showing exactly what they did, what courses thay had run, exact settings for said capture and showing fraps fps counter in each. This would allow for reproducability by us and other web sites allowing us to confirm their findings. I feel I'm not alone in thinking that providing a means of reproducability would either remove doubt, or confirm it.

Contradictory?

BFBC2
http://www.xbitlabs.com/articles/video/display/geforce-gtx-465_7.html#sect2
Contradictory.

Crysis Warhead
http://www.techspot.com/review/283-geforce-gtx-400-vs-radeon-hd-5800/page4.html
Not contradictory

Just Cause 2
http://alienbabeltech.com/main/?p=17869&page=4
http://www.guru3d.com/article/geforce-gtx-470-480-review/25
Contradictory (compare 8xAA with 8xAA). Yes older drivers. Anyone know somewhere that tested Just Cause 2 with 8xAA?

Metro 2033
http://www.xbitlabs.com/articles/video/display/geforce-gtx-465_8.html#sect1
No AAA, but not contradictory.

Resident Evil 5
http://alienbabeltech.com/main/?p=17869&page=14 (older drivers)
http://www.xbitlabs.com/articles/video/display/geforce-gtx-465_9.html#sect2 (lower AA)
Probably not contradictory, but no exact comparison available.

STALKER
http://www.anandtech.com/show/3745/nvidias-geforce-gtx-465/10
Not contradictory
http://www.xbitlabs.com/articles/video/display/geforce-gtx-465_8.html#sect2 <- contradictory with both the OPs link AND Anandtech's results.

Wolfenstein
http://www.anandtech.com/show/3745/nvidias-geforce-gtx-465/13
Not contradictory
http://www.xbitlabs.com/articles/video/display/geforce-gtx-465_8.html#sect3 <- Xbitlabs is again contradictory to both AT and the OPs link.

So basically out of the 7 games tested, 3 results seem to agree with other sites, 2 results likely agree with other sites (but without using an exact comparison) and only 2 of the 7 test results were actually contradictory.

Now, I am sure you will claim I cherry picked the benchmarks I compared to, which I did. I cherry picked for ones which are using similar settings, and ones which show the same results where possible (although i don't hunt and only checked 4 sites for results).
The point is that the results are NOT particularly contradictory, outside 2 of the tests. Yet people still try and scream that they are.

And why aren't the contradictory? Because of SETTINGS.
Again, like I have already said in this thread, it's all about the settings, and blindly using benchmarks to say X is faster than Y is pointless. Typically the GTX4xx cards do have a lead over HD58xx cards, but in fact, they don't always, and you can find various situations where the results "contradict" the general apparent trends, but that's because those general trends don't always hold true.

There is nothing wrong with this review, there is nothing massively contradictory on the whole, part of the "problem" is that the benchmarks and settings used favoured ATI, hence the apparently strong showing.
The anomalous results do indeed warrant further consideration, but that's 2 of the 7 benchmarks.

And you will notice that I used Xbitlabs quite a lot, and in 2 of the Xbitlabs benchmarks, the results were... CONTRADICTORY to those of this fair site, Anandtech.
Does that mean Xbitlabs needs to be considered "fail" or wrong? Or maybe Anandtech does, because in 2 games its results agree with the site in the OP and disagree with the Xbitlabs results.
No, it just means they used different settings and different benchmarks on different computers and got different results. That doesn't mean they fail, or that there is contradiction, it just means that these graphics cards do not flat out follow a strict set of performance criteria that mean X is always faster than Y, even in a single given game.

 

[HurleyBird]

In the perfect world all sites would do as you suggest, keys. Unfortunately, sites seem to not provide as much data as they could. A prime example of this is min/max frame rate values which by themselves mean almost nothing (dropping down to 5fps for the very first frame of the benchmark means something entirely different then dropping down to 5fps repeatably during heavy action). Most sites nowadays are guilty of this. Even more meaningless is max frame rate... in fact the more spikes in frame rate the less fluid the game becomes, stable frame rate is the key.

In terms of "real world" vs. canned benchmarks, I think the old saying is that there is no such thing as a free lunch -- obviously there are a lot of timedemos out there that do not simulate real gameplay loads. On the other hand, it is also obvious that testing "real world" gameplay is inheritely less accurate (although repeated runs can decrease margin of error), and even running fraps in the background changes the testing environment slightly.

One has to wonder if there's another alternative, like assembling two otherwise identical system and some way controlling them with the same mouse/keyboard as to make the runs more repeatable.

Of course, providing a graph of framerate over time can help determine if the playthroughs are generally the same (by comparing the paths that both cards leave) as well as show when and how often frame rates dip (which unlike a single min frame rate value, is actually useful)

 

[Keysplayr]

The whole purpose of my post was for confirmation of TechSpots results. How would one do that without mimmicking their procedure?
I mean, sure, you can do it your way, but it won't be comparable to TS. And that was my intended point.

 

[Idontcare]

I would have to say, that if TR is going to release benchmarks contradictory to most that are out there, then they most definitely provide captured videos showing exactly what they did, what courses they had run, exact settings for said capture and showing fraps fps counter in each. This would allow for reproducibility by us and other web sites allowing us to confirm their findings. I feel I'm not alone in thinking that providing a means of reproducibility would either remove doubt, or confirm it.

It is definitely in their best interest to do as you suggest.

Consider when Anand made his big splash over jmicron controllers in SSDs...he went to great lengths to meticulously document a way that would enable others to easily verify his observations (and thus arrive as his same conclusions as well).

Your suggests here are not only completely reasonable and prudent but they are also the hallmark of a good reviewer.

 

[GaiaHunter]

[H] already did a canned vs real game play article once (8800GTX vs 3870 and 3870x2) and their findings were pretty much that there could be substantial differences between canned benches and real game play.

I think it would be good if reviews did both but alas they can claim time constrains to do so.

 

[Lonyo]

[H] already did a canned vs real game play article once (8800GTX vs 3870 and 3870x2) and their findings were pretty much that there could be substantial differences between canned benches and real game play.

I think it would be good if reviews did both but alas they can claim time constrains to do so.

Yup. This is a non-issue.
There is nothing wrong with what TR has done, and nothing wrong with its findings.
They are not particularly contradictory, and they just agree with existing sentiments that timedemos and real gameplay don't always show the same results.

To claim TR needs to do more is to ignore the long and glorious past of the canned timedemo and its failings. TR isn't the one which needs to prove itself, built in non-representative game benchmarks/demos need to, and so far they never have.

 

[HurleyBird]

Agreed, there should be some reproducability both ways. As much as a description (and/or video) of what is done in the playthrough along with a graph of FPS/time would be superb. At the same time, if a reviewer wants to use a timedemo it stands to reason that he should compare it to an averaged set of FRAPS runs in order to tell how accurate of a representation it is of the game. Obviously, some games might leave things out like AI in timedemos, and gfx card companies might also try and optimize for the timedemos themselves as opposed to the actual game.

 

[ArchAngel777]

I have the money, I have the drive for performance.... and I have a single 5870.

I have the money and the desire for performance, but I am sitting with a GTX 280. No impulse to waste money on something that really isn't going to improve my current gaming experiences. Nothing I play struggles with the settings that I am content to run.