Possible ATI 4870 vs 9800GX2@Legit Reviews

[wingless]

http://forums.legitreviews.com/about15370.html

Check out that post. The original source is posted as well (in some European language I don't recognize).

If this is true, and I hope it is, a single 4870 is WAYYYY THE HELL faster than a 9800GX2 in a game that ATI traditionally sucked at. That would simply be f'ing amazing. Remember how the C2D benchies came out months before their release and everybody said "Hell no, those can't be true" our of disbelief? Well I hope ATI can pull one of those off (for AMD's sake). On the flip side, we've heard rumors that the damn thing will only be TWICE as fast as a 2900Pro, which seems a bit more realistic.

Either way, I'd just be happy if its a lot faster than my 2900XT at Folding@Home though. I could care less if its 5fps faster or slower than an Nvidia card.
.

 

[moocow]

If thats true. =O

 

[vixro]

lol 2GB of ram in crysis. who does these stupid tests?

Schmitzer
Legit Little One


Posts: 1

 

[Mr.Guvernment]

i hope AMD can pull a c2d like intel did to AMD, the cost is what will really interest me after ATI said they were not going to make high end cards, but mid range cards more so and multi-gpu solutions instead of massive high end flagships...

 

[Neuromancer]

3800mhz memory speed?

I would love to see a new ATI card kick some butt in areas they normally do not. But I reserve judgement until I see them on the shelf and then reviewed.

 

[Kuroimaho]

@vixro That guy just posted some links.

The results seem too good to be true, unless the card sports 2 RV770 chips, especially if we consider the driver which is most likely still under tweaking.

 

[dudleycpa]

Interesting - Crysis has no AA in these tests

 

[Cheator]

Hmm... well we can hope but who knows.

 

[Mr.Guvernment]

3800mhz memory speed?

I would love to see a new ATI card kick some butt in areas they normally do not. But I reserve judgement until I see them on the shelf and then reviewed.

+ infinity

like any other claims, i will wait until real people have real proof.

 

[Farinorco]

I don't believe those benches. If the rumours are more or less in the right track and ATi new chip has +50% stream processors, +100% texture mappers, near double memory bandwidth thanks to the use of much faster GDDR5, core and shaders clocks divided, and some corrections and optimizations to address several issues, achieve this kind of results would be impossible.

I'd expect a chip without the problems of RV670 with texture intensive games, very good in shader intensive games (probably much better than G9x chips in certain specific games), much better suited to high end segment (that's it, high resolutions -1920x1200 and above- with filters) than RV670 and G92, and little else. So, I think they'll can beat G92 and take the performance crown in a bunch of games, maybe not in texture intensive games played at low resolutions and with no (or few) filters, and maybe not in games that don't allow to take the most of its stream processors because of its vectorial architecture. But a performance like that shown in those charts? No way!

That's more or less around what I'd bet it would be the performance of a HD4870X2 or a HD4870 crossfire. Then it would be absolutely believable, and if those charts are showing any kind of real benchmarks, then probably they're a benchmark of that, and no a single GPU RV770 testing. But I suppose those charts are simply fictional.

Only way it could be in this way is that AMD has decided to implement multicore architecture in next chips (there are rumours about them working on this for a time now), and its single GPU would be a pair of cores, each with the specifications previously given (like an upgraded RV770 each one). That's highly doubtful, because the resulting chip would be probably too big, hot and power hungry to the known AMD standards... and certainly wouldn't match with the supposedly known 100-150W TDP's.

I would consider those charts like an indication of HD4870X2/HD4870 CF performance, if anything. It would be an excellent product for the cost, and it would take the performance crown for a while, probably until nVidia next high end launch (which surely will be a hot&hungry&expensive watt-eater part with no single GPU rival from AMD). And it would be much more useful to the high end user (1920x1200 with filters and up) than the current offerings, being the first true upgrade since 8800GTX/Ultras to this users...

Sorry about this long and boring piece of text...

 

[Anubis_386]

Loooks very promising .. I was thinking of trading in my HD3870X2 (mostly cause the fan is buggered and rattles) for a 9800GX2 but I might hold off now ..

 

[ratbuddy]

Interesting - Crysis has no AA in these tests

I noticed that too, and that's where ATI cards traditionally fall apart

OTOH, if those results are true, the 4870X2 will be my next card

 

[NettieZoom]

While it would be awesome if true I will wait for some real reviews.

They probably ran the 9800 with full AA/AF and then the 4870 ran neither LOL

I want to know if it drops to its knee's at high resolution with lots of AA like ATI's current lineup

 

[Farinorco]

I don't think that ATi's current cards suffer more performance loss than nVidia G92's with antialiasing. What I've seen in the bunch of reviews around the Internet is that there are games in which the performance loss is higher with ATi, others with nVidia. But having a general inferior performance compared to G92 makes ATis run really slow when taking more impact from AA.

I'd say that this may be because of not having specific AA hardware logic and solving this through the more general use stream processors: depending on how intensive use of stream processors makes the game this could be limiting or giving more room to AA compared to a specific hw. And both RV670 and G92 have serious troubles at high resolutions and filters because of the memory bandwidth bottleneck.

I supose that an increase of 50% of stream processors, the divided and increased clock speed of them, and the nearly doubled bandwidth (not to mention possible tweaks that ATi engineers could do) should give much more room to that. But I supose that it'll be the same vs future nVidia offerings and using future games until nVidia has to adopt that kind of more generic, flexible and programmable architecture that now AMD/ATi are using. I'd expect the same kind of behaviour with HD4870 than HD3870.

 

[Neuromancer]

I noticed that too, and that's where ATI cards traditionally fall apart

OTOH, if those results are true, the 4870X2 will be my next card

They do not "fall apart" They take more of a hit then the nVidia cards becasue ATi stuck with Multisampling Anti Aliasing, and nVidia adopted Coverage samling Anti aliasing.


Taken from nVidias website..

CSAA Guidelines
Performance

CSAA performance is generally very similar to that of typical MSAA performance, given the same color/z/stencil sample count. For example, both the 8x and 16x CSAA modes (which use 4 color/z/stencil samples) perform similarly or identical to 4x MSAA.

Since reviews never specify whether they are using 8xQ (or 16xQ) mode, which they most liekly are not, then it does not compare directly to MSAA. Apples to Oranges

 

[Shiggity]

These new cores look pretty exciting, I think the jump really may be this big.

Not to mention the memory on the next series of cards is going to be absolutely rediculous. If these benchmarks are close to true, we're getting like double the memory speed. Nvidia moving to the 512bit bus and ATI clocking their GDDR5 outrageously high.

 

[TTP]

I noticed that too, and that's where ATI cards traditionally fall apart

OTOH, if those results are true, the 4870X2 will be my next card

The 2900XT had aa issues. The 3000's do not. And if you check out Assassin's Creed, the ATI 3000's running DX10.1 gives noticeably better performance with AA.

 

[satandole666]

If these cards are this good and the drivers aren't terrible I expect I'll have at least one of them soon after launch.

I've been waiting for the next huge leap forward since the G80. Let's hope this next generation (4xxx and G200) will fulfill that for me.

 

[jokers_greg]

crysis and synthetic benchmarks... surprise? I'd like to see further tests before I get excited

 

[PhysX]

http://forums.legitreviews.com/about15370.html

Check out that post. The original source is posted as well (in some European language I don't recognize).

Turkish it looks like.