X-Bit Labs review 68xx vs. 460/470

[Arkadrel]

From Xbitlabs:
1600x900 res is the battlefield they performance most equal, higher res (1900x1080, 2560x1600 favors the ATI cards abit more than the nvidia ones)

http://www.xbitlabs.com/articles/video/display/radeon-hd6870-hd6850_19.html


Easier to read than that picture above thats impossible to make out what things mean on it:

-------------6870--vs--460-1gb: (1600x900)
AvP_________31.8__vs__32.2
BF:bc_______55,3__ vs__50.1
CoD MF2_____96.5__vs__86.6
Crysis:WH____32.8__vs__28.9
FC2_________72.1__vs__80.4
Metro________35.1__vs__33.9
stalker_______39.2__vs__41.9
Justcause2____43.8__vs__40.6
Lostplanet2____28.5__vs__31.7
Masseffect2____95.8__vs__95.5
Dirt2__________68.4__vs__77.7
Hawx__________79___vs___ 83
battleforge______48.2_vs__43.6
Sc2_____________60__vs__63.2
------------------7----------7



-------------6870--vs--460-1gb: (1920x1080)
AvP__________27.9_vs_25.6
BF:BC________44.6_vs_39.2
CoD MF2______78,7_vs_73.1
Crysis:WH_____26.1_vs_21.9
FC2__________62.9_vs_68.7
Metro_________29.8_vs_25.8
stalker________30.4_vs_31.9
Justcause2_____36.5_vs_32
Lostplanet2_____23.1_vs_26.4
Masseffect2_____78.8_vs_68.8
Dirt2___________65.5_vs_62
Hawx___________73_vs_74
battleforge______38.8_vs_35
Sc2____________53_vs_51.4
-----------------10------4


About topic above:
Yeah His 460 768mb version numbers seem higher than most review sites....
Anandtech shows the 6850 beating the 460 768 in like every test... but in this benchmark from Xbitlabs the 460 768 it doesnt look that way.

I honestly doubt some of their numbers in many reviews. a while back that would get horrible performance with the gtx260 that they had and I contacted them about it and on more than one occasion they literally just changed the numbers in the benchmark after I complained.

possible explaintion for 460 768 higher than most review sites scores?

 

[GaiaHunter]

These cards are just too close in performance, so specific games, specific IQ settings different scenes in the same game and drivers optimizations will make a difference.

I know that the 5970 is the fastest card, the GTX480 is the fastest single GPU card, the 5870 seems to be better at 2560x1600 than the GTX470 but after that it is too close.

 

[Arkadrel]

If you game in 1600x res, then stock 460 vs stock 6870 = equally well.
If you game in 1900x res, then stock 460 vs stock 6870 = 6870 better overall.

but yeah... the 6870 and 460 1gb are very close in performance.

5970 > 480 > 5870 > 470 > 6870/460 1gb > 6850/460 768 >

 

[GodisanAtheist]

Where is the 5830, the turd these cards are supposed to replace?

 

[Dribble]

The most interesting thing in the xbitlabs article was the comment in the comments section about how IQ has been sacrificed by Ati in their latest drivers, included several links e.g. ugly texture flickering:
http://www.tweakpc.de/hardware/tests/grafikkarten/amd_radeon_hd_6870_hd_6850/s09.php

 

[Arkadrel]

Where is the 5830, the turd these cards are supposed to replace?

lol I had forgotten all about the 5830s ^-^


5970 > 480 > 5870 > 470 > 5850 > 6870/460 1gb > 5830 > 6850/460 768 >

fixed

 

[Skurge]

ATI Catalyst:

* Anti-Aliasing: Use application settings/Standard Filter
* Morphological filtering: Off
* Texture Filtering Quality: High Quality
* Surface Format Optimization: Off
* Wait for vertical refresh: Always Off
* Anti-Aliasing Mode: Quality
* Other settings: default

I don't know, but when you choose Quality for AA in CCC it uses SSAA. Just something some one mentioned to me.

 

[nyker96]

How does the gtx460 768 beat a 6850 in half the test @ 1900x1080? That would mean a gtx460 1gb is faster.
I don't get these results.

I also find the gtx460 768 vs 6850 results suspicious. Almost every other site like Anand etc comparing these two shows 6850 winning most of the benches at these resolutions. This I believe the good people at xbit might have screw something up.

 

[biostud]

Here in Denmark the GTX 460 1gb is a little cheaper than the 6850, so to me the GTX 460 is the best option here.

 

[Vdubchaos]

One of the reasons HardwareCanucks is one of my favourite review websites right now is because they aren't cool with "canned", "rolling demos" benches. They even did a short article on that. Based on the fact that each review website will differ in their methodologies, it makes it extremely hard to draw conclusions on which review is the most accurate. Personally, I read many reviews and just try to get an idea of a general trend (I am not saying this is the right method, just what I like to do). I mean if we start comparing exact numbers across reviews, it's just going to leave us frustrated (45 fps on Website #1, 32 fps on Website #2, 38 fps on Website #3, 2 of them are using Fraps, 1 is using a pre-built bench, etc. too many variables).

It's way too easy to fall back on "Wreckage tactics" - find at least 1 website with contradictory results and present it as evidence of a general trend. Oh look, I just found a GTX470 beating an HD5870 in Crysis Warhead. Therefore, the 470 is faster than HD5870 in Crysis....However, if we stack up 10 reviews, we'll probably find 9 of them showing the HD5870 to be faster than the 470 in Crysis. Therefore, it's fair to say that HD5870 is 'generally' faster in Crysis.

Well, 5870 BETTER be "Generally" faster in Crysis with it's $50-100 price difference

IMO Nvidia has an edge latelty. If I was to buy a card today I would get 460 1gb > 6850 and 470 > 6870

Not because "Nvidia or ATI" is better blah blah blah

Simply because new 6800 cards are rewarked versions of 5770 and 5830, 2 cards in which I have 0 interest in. And differences that 6800 has from 5770/5830 are all worthless to me.

Also Nvidia cards are generally cheaper than AMD counterparts.

But the differences are VERY minimal IMO. It's really more of a personal preference as we all play different games and let's face it, in some games Nvidia is faster and in others AMD is.

What I really want to see is new 6950 come in at around $250 or so. Sorry but I will not spend $300 for a "really nice card"

 

[The Sauce]

I think that it's hilarious that my cheaper, 1-year-old 5850 beats the pants off of these newer cards, and even despite higher power and TDP, overclocks significantly better, beating them even more soundly. The 5850 is looking like a better investment every day.

 

[Hauk]

I think that it's hilarious that my cheaper, 1-year-old 5850 beats the pants off of these newer cards, and even despite higher power and TDP, overclocks significantly better, beating them even more soundly. The 5850 is looking like a better investment every day.

You have good reason to be proud of a 5850 purchase. I'd not be troubled by it's power consumption, at all. Great purchase..

 

[RussianSensation]

The most interesting thing in the xbitlabs article was the comment in the comments section about how IQ has been sacrificed by Ati in their latest drivers, included several links e.g. ugly texture flickering:
http://www.tweakpc.de/hardware/tests/grafikkarten/amd_radeon_hd_6870_hd_6850/s09.php

True. ComputerBase also investigated this. I posted a thread specifically about this. Basically, AMD seems to have reduced default image quality to gain another 5-6% performance difference starting with Catalyst 10.10.

I don't know, but when you choose Quality for AA in CCC it uses SSAA.

I believe AMD cards can only use SSAA in DX9 mode or below. They can't run SSAA in DX10/11 (Maybe BFG can confirm this). Quality = Standard settings in CCC. Starting with Catalyst 10.10, "Quality" setting is no longer comparable to "Quality" in Catalyst 10.9 (or older versions). With current drivers, High Quality for AMD = Standard Quality for NV.

 

[Teizo]

Some of the problems with the initial 6850 reviews could be that those reviewers got a 6850 with 1120 shaders and it benched higher than it was supposed to and the reviewers didn't catch it. Those that did catch it had the 6850 dead even with the vanilla 460 irrc (for the most part), so these results are not all that surprising.

The 6850 is a helluva chip though even though it runs a tad hotter and louder (though I imagine the non-reference ones should alleviate that like the review concluded), and besides we have AMD and the 6850 to thank for the price of the 460 dropping. Bottom line is buying either is a good deal and neither decision is a bad one. No need for the fanboyish stupidry on either side of the equation which seems viral these days, even more so than in the past I remember.

 

[RussianSensation]

What I really want to see is new 6950 come in at around $250 or so. Sorry but I will not spend $300 for a "really nice card"

Maybe in 6-8 months it will fall to $250. With HD6870 occupying the $239 bracket, HD6950 is probably going to be >$300-$350.

 

[Arkadrel]

I think the 6870 vs 460 stock clocks, the 6870 is the winner. But it costs more too,
if you get a 460 evga FTW things look pretty even, with maybe slight egde to the 460 evga ftw.

 

[biostud]

I think that it's hilarious that my cheaper, 1-year-old 5850 beats the pants off of these newer cards, and even despite higher power and TDP, overclocks significantly better, beating them even more soundly. The 5850 is looking like a better investment every day.

Did you get your 5850 with a special offer? otherwise the 5850 is not cheaper than the 6870.

 

[The Sauce]

I believe it was around $220-$240 when I got it.

 

[Rezist]

So whats with all the different reviews? I mostly only read HardOCP and anad both which put the 6800 series performing far better then Xbit did.

 

[Arkadrel]

@Rezist

toyota (page1 post #9) seems to imply that alot of their reviews have questionable numbers... like they pulled some of them out of thin air, or when you asked them why their numbers where differnt next day theyd just magically be altered.

If you dont think their review is credible just look at a few others instead.

 

[RussianSensation]

So whats with all the different reviews? I mostly only read HardOCP and anad both which put the 6800 series performing far better then Xbit did.

They do perform better in their reviews.

@ Xbitlabs
HD6870 beat the GTX460 1GB in almost every test at 1920x1080 4AA. The tests where it lost are the exact same tests where GTX460 beats the HD6870 in other reviews:

Hawx 2, Lost Planet 2 (heavy tessellation), Stalker (Sun Shafts) (AMD's weak spot). The only outlier is Far Cry 2 from what I can see since HD6870 seems to perform better in that game (Guru3d).

Their conclusions seem to be not much different than anywhere else.

Xbitlabs Review Conclusions:

1) The HD6870 easily cleaned up the GTX460 1GB in their review. HD6870 was faster in DX10/DX11 games at 1920x1080 4AA including: (a) AvP, (b) BF:BC2, (c) Crysis, (d) Metro 2033, (e) Just Cause 2, (f) Mass Effect 2, (g) Dirt 2, (h) BattleForge (i) Starcraft 2 (All these wins for HD6870 are All TRUE in every other review on the web I can think of).
2) Moreover, the Radeon HD 6870 was generally faster than the Radeon HD 5850 (TRUE).
3) HD6870's tessellation was better than HD58xx series (TRUE).
4) HD6850 was slower than GTX460 768mb/GTX460 1GB in some games at lower resolutions (TRUE).
5) HD6850 catches up to GTX460 768mb at higher resolutions but still loses to GTX460 1GB (TRUE).
6) GTX460 > HD68xx cards in Hawx 2, Lost Planet 2, STALKER: Sun Shafts (TRUE).
7) HD6870 > GTX470 in Crysis, Mass Effect 2, < GTX470 in Hawx 2, Lost Planet 2, STALKER: Cop, CD4:MW2, Battleforge (TRUE), and their performance in BF:BC2, Just Cause 2 and Dirt 2 is more or less equal (TRUE).

Plus, they tested with Catalyst AI: Disabled, and with image Quality set to High Quality. So their numbers for HD68xx series will be 5-6&#37; slower on average than in most other launch reviews in case you are comparing HD68xx vs. HD5870 vs. GTX470 series. Therefore, HD5870 will actually be about 10%-12%+ faster than an HD6870 (this is also TRUE).

Hardware Canucks almost mimmicks these findings.

 

[Hauk]

@ Russian Sensation, thanks for taking time to break it down.

Some seemed quick to criticize thier review, while presenting little or no evidence as to why. Some also seemed to overlook varying testing methodologies, or that reviewers may test different parts of games with no standard benchmark.

 

[RussianSensation]

Thanks for taking time to break it down.

Np! I just wanted to see myself what the fuss was all about. When but when I actually looked at the review, surely HD6870 > GTX460 1GB and HD6850 lost some compared to the GTX460 1GB. I think some people start comparing actual #s from 1 review to another as opposed to looking at the overall conclusion/trend.

For example, AnandTech enabled everything under the sun in Metro 2033 showing the top end cards producing something ridiculous like 18-23 fps. However, in the real world you'll probably either run tessellation OR AA, but not both since none of the high end cards can do that. Now under the Anandtech scenario, GTX470=HD5870=GTX460=HD6850/70 = all unplayable. Another review will relax some of the settings and get playable results in the 30-40 fps. Both are right.

 

[IndyColtsFan]

Thanks for the breakdown, RS. This is making my upgrade decision easier.