New rumor about ATI Southern Islands

[saaya]

So with that said, where does the assumption come from that the HD 5870 architecture is bandwidth limited?

some "wise" reviewers drew that conclusion when comparing rv870 and rv770, and since the bandwidth hadnt increased as much the the sps, they concluded, it must be pretty bandwidth limited
its one of those urban legends that still floats the web...

[Iconyu]

I wouldn't say it was bandwidth limited, but I will say that while the 5870 was a good card there were 'flaws' in its design that meant it rarely got the most out of it's shaders.

The changes needed to fix them did not seem very hard to implement and in fact seemed to be similar to the changes that nvidia have made with the 460.

However we still don't know if these benchmarks are real, so until then let the speculation continue.

[JohnJohn]

I'm going to ROFLMAO if these benchmarks turn out to be fake

They seem too good to be true, and you know what they say about being too good to be true... Fake benchmarks always circulate before a new card is released.. You guys should know that by now, so don't put all of your eggs in one basket eh?

And correct me if I'm mistaken, but memory overclocking never had much of an impact on HD 5870 performance compared to overclocking the core..

So with that said, where does the assumption come from that the HD 5870 architecture is bandwidth limited?

Of course you will be ROLFMAOING if these are fakey, as you must be feeling pretty depressed right now, and your only consolation is the thought of ROLFMAOING at the thought of these being fakey, because you are depressed. Sad panda

[Carfax]

Of course you will be ROLFMAOING if these are fakey, as you must be feeling pretty depressed right now, and your only consolation is the thought of ROLFMAOING at the thought of these being fakey, because you are depressed. Sad panda

Yes, I'm so depressed, that I'm going to put my 480s on Ebay right now and start fasting until the 6870 arrives

Thanks, but no thanks. I tried ATI before......never again....unless their product is superior to Nvidia's on all fronts, and that includes Crossfire.

[Calmatory]

Yes, I'm so depressed, that I'm going to put my 480s on Ebay right now and start fasting until the 6870 arrives

Thanks, but no thanks. I tried ATI before......never again....unless their product is superior to Nvidia's on all fronts.

Welcome to the crowd, (fan)boy!

If these are true, then Your 480 SLI setup will look INFERIOR compared to 6870 on all fronts but not necessarily in performance, no? Then again, it's ok to be different.

Hopefilly Nvidia comes up with something, rather soon...

[Carfax]

some "wise" reviewers drew that conclusion when comparing rv870 and rv770, and since the bandwidth hadnt increased as much the the sps, they concluded, it must be pretty bandwidth limited
its one of those urban legends that still floats the web...

I doubt ATI's engineers would overlook something as obvious as memory bandwidth..

Why is it that so many people assume that the engineers that design these things are stupid?

[Carfax]

If these are true, then Your 480 SLI setup will look INFERIOR, no? Then again, it's ok to be different.

Not really. It may be slower, but inferior overall? I think not.

A few reasons why I, and many others have a strong preference for Nvidia are:

1) Driver support

2) Extra features ie PhysX, 3D vision, CUDA etc

3) Better filtering.

Number 3 in particular has always infuriated me concerning ATI. Since the 4800 series, GPUs have had so much raw power that there's no need to resort to filtering tricks anymore to increase performance.

Yet ATI still has their "brilinear" filtering optimizations to this day

[Manicdan]

I doubt ATI's engineers would overlook something as obvious as memory bandwidth..

Why is it that so many people assume that the engineers that design these things are stupid?

they used faster memory than nvidia, but on a narrower bus
the cost to add a wider bus, and the cost for extra memory chips seems to offer a worse price/perf ratio than to leave it as is.

if they increased the memory bandwidth by 30-40%, then costs would have probably gone up by 20-30%, while perf might have only gone up by ~10%, and power consumption up by 8-15%

AMD must really be hurting if they have the balls to keep their prices the same since launch

[Frontl1ne]

I doubt ATI's engineers would overlook something as obvious as memory bandwidth..

Why is it that so many people assume that the engineers that design these things are stupid?

I know right? I mean, before the 5870 was released everyone was crying about how there wasn't a 2GB version. And LOL when the 2GB version came out, not even high res plus high AA could save the whiners.

[Carfax]

if they increased the memory bandwidth by 30-40%, then costs would have probably gone up by 20-30%, while perf might have only gone up by ~10%, and power consumption up by 8-15%

Exactly, which goes to show that the architecture itself isn't constrained by memory bandwidth. If it was, the performance increase from widening the bus would be much greater than 10%.

The same thing with Fermi. A 512 bit bus would have been useless on the 480 in combination with the greater bandwidth already provided by GDDR5.

The only things that would have increased were complexity and power usage.

So assuming these benchmarks are true, it can't be memory bandwidth that has increased the performance, nor the added shaders. It would have to be something else.

I'm no engineer though so I'm not even going to speculate

[ajaidev]

If the scores are correct this is how it may play out:

1.Nvidia launches a full dual GF104 "384 shaders not 336" based GPU lets call it GTX 495
2.ATi releases the 6870 the card in crossfire makes life for the GTX 495 very hard.
3.Nvidia's partners launch overclocked versions of GTX 495 around max possible TDP.
4.ATi releases the 6970/6950 and become the raining chaps once again for single PCB cards.

[madcho]

Not really. It may be slower, but inferior overall? I think not.

A few reasons why I, and many others have a strong preference for Nvidia are:

1) Driver support

When a new GPU from nvidia is on market, old GPU don't have any update in drivers. Yeah support is better on green way ...

Drivers need to be improved in nvidia, they need improve how it's designed, you can be lost in this pannel ...

2) Extra features ie PhysX, 3D vision, CUDA etc

Nice but What do you do with cuda ? Are you a programer on CUDA yourself ? Or you use the 2-3 programs that runs on it ( and ATIs too ) with some demos ? It's nice Stuff. PhysX is nothing really good for futur graphics. So i don't want it. 3D systems tryed to exist already 10-15 years ago, but was a fail. It's gonna fail again. A lot of people are already wearing glases. They can't wear two in same time. So it's gonna fail again.

3) Better filtering.

Saying something is not a proof. ATi's filtering is far better from old RV770, proof :

http://www.hardware.fr/articles/770-...5870-5850.html

And i've don't found the page on fermi, but i remember to have seen a beter quality from RV870

Number 3 in particular has always infuriated me concerning ATI. Since the 4800 series, GPUs have had so much raw power that there's no need to resort to filtering tricks anymore to increase performance.

Yet ATI still has their "brilinear" filtering optimizations to this day

I don't have any problem with filtering. I can even run HL2 EP2 in AA16x/AF16x, 24x is a bit slow but playable too.

[spursindonesia]

Welcome to the crowd, (fan)boy!

If these are true, then Your 480 SLI setup will look INFERIOR compared to 6870 on all fronts but not necessarily in performance, no? Then again, it's ok to be different.

Hopefilly Nvidia comes up with something, rather soon...

LOL, the card hasn't even arrived, yet the ranting already begun.

I think GTX 480 will still be SUPERIOR .... as a room heater, remember, the cold days are arriving in the northern hemisphere.

It's OK to be satisfied with a product that you like, but to crap in competitor's product thread, well that shows how "mature" that particular person actually is. You CAN & has the right to doubt these numbers, but to bring driver FUD, features FUD, IQ FUD into conversation/discussion, doh ????

[saaya]

I doubt ATI's engineers would overlook something as obvious as memory bandwidth..

Why is it that so many people assume that the engineers that design these things are stupid?

dont think its about people thinking they overlooked it... they thought gddr5 can only clock that high, and its not enough... and they thought that ati made a decision to rather cut costs and go for 256bit only even though it wouldnt be enough bw...

i think theres some truth to that, 320bit would probably have helped rv870, but i dont think it would have done much... and it would have made the cards notably more expensive...
i wonder how much actually... 25$ maybe if you count it all together? pcb, extra memory chips, packaging, additional transistors...
probably less than 25$ i think?

[Mungri]

It's OK to be satisfied with a product that you like, but to crap in competitor's product thread, well that shows how "mature" that particular person actually is.

I were just skim reading, and thought you were carfax and started looking for a huge NO U! Gif.

Then I noticed you are someone else

[RPGWiZaRD]

While Nvidia may have bigger resources for driver support they seem to break stuff often with new drivers too so it's not like Nvidia is like free from bugs either. For example ever since 19x.xxx drivers I've had alt-tab issues with some games, especially UT3, clocks get stuck in 2D mode or otherwise you just get very poor performance after an alt-tab or two (like 250 fps drops to 40-70 fps, not that I play with 250 fps but constant 120 fps rather) and you have to reboot to get back the FPS, highly annoying, currently on 259.32 beta and the issue still persists, might as well go back to 182.47 driver soon cuz that one worked without issues...

Speaking of drivers, does any1 know if ATI supports 120Hz in non-native res yet in latest catalyst? This is such an important thing for me using a 120Hz LCD, I'd gladly pick an ATI card but since I've heard it seems like it's not working with 120Hz in non-native res I've sticked to Nvidia so far cuz 120Hz support is better.

[Carfax]

When a new GPU from nvidia is on market, old GPU don't have any update in drivers. Yeah support is better on green way ...

Drivers need to be improved in nvidia, they need improve how it's designed, you can be lost in this pannel ...

What?

Nice but What do you do with cuda ? Are you a programer on CUDA yourself ? Or you use the 2-3 programs that runs on it ( and ATIs too ) with some demos ? It's nice Stuff. PhysX is nothing really good for futur graphics. So i don't want it. 3D systems tryed to exist already 10-15 years ago, but was a fail. It's gonna fail again. A lot of people are already wearing glases. They can't wear two in same time. So it's gonna fail again.

I don't use CUDA but many other people do. Fermi is not just a gaming GPU you know.

As for PhysX, thats your opinion and you're welcome to it. Personally, I love a good PhysX implementation. It can really change the atmosphere of a game for the better.

Batman AA and Mafia 2 are the best examples of really good PhysX implementations.

As for 3D Vision, again, thats your opinion. But you're wrong on one thing.

3D isn't failing. Why do you think there are so many new HDTVs coming out that support 3D capability?

I recently bought a 58 inch Samsung 3D Plasma in fact, and the 3D material on it was surprisingly good. Nvidia is also coming out with a device that will allow you to play 3D games on 3D capable HDTVs rather than LCD monitors.

Saying something is not a proof. ATi's filtering is far better from old RV770, proof :

http://www.hardware.fr/articles/770-...5870-5850.html

And i've don't found the page on fermi, but i remember to have seen a beter quality from RV870

LOL, you must have missed the big debate we had on the forum concerning ATI cheating in Crysis.

Read this thread.

Also, that D3D AF tester doesn't mean squat when it comes to actual image quality in real games.

Click here, and you'll see that ATI's filtering quality in games is actually inferior to that of Nvidia's.

Guru3d also noticed it as well when they published their Starcraft 2 GPU performance article.

[Manicdan]

What?

the laptops we use at work have one massive problem we see non stop (on different models too)

when docking and undocking, the gpu fails to recognize what happened, and will never see the second monitor, or crash the laptop.

[mindfury]

http://forum.beyond3d.com/showpost.p...postcount=1464

Unfortunatelly, these benchmarks are likely fakes. At least the Unigine "screenshot" is. Notice, that number "2" in "scores" differs from number "2" in "min FPS". Different (bolder) font...

[Calmatory]

Can someone run Unigine Heaven and post a comparison pic? I'd bet a bunch that the Scores number is actually bolded by the benchmark.

[blindbox]

http://forum.beyond3d.com/showpost.p...postcount=1464

Are all the marks from the same source? IINM some marks are from the same person giving Unigine marks and some marks are from someone else.

If they want to fake it at least put some effort into it. It's basically a damn HTML file IINM. Yeah, I'm just dreaming

[mindfury]

Are all the marks from the same source?

Yep,WG_baby@pcinlife

[Xoulz]

Yes, I'm so depressed, that I'm going to put my 480s on Ebay right now and start fasting until the 6870 arrives

Thanks, but no thanks. I tried ATI before......never again....unless their product is superior to Nvidia's on all fronts, and that includes Crossfire.

LOL... it's sad how much you drink from teh nvidia cup.

[blindbox]

Yep,WG_baby@pcinlife

Ah okay. Anyway, what Calmatory said just to confirm said just to confirm.


Oh.. I have unigine. Brb!

[Calmatory]

It seems that the data is retrieved from HTML file, no? If so, then it just proves that the bolded score is part of the Unigine layout and that means nothing.

However... It also means that simply by editing the HTML file one can produce fake screens, no? :| If so, then *** these results!