New rumor about ATI Southern Islands

[Manicdan]

how well do you trust your sources?

[LightSpeed]

how well do you trust your sources?

Me? Full confidence in my source, except for maybe the performance approximate, well, because its an approximate

[iMacmatician]

900 MHz core would be 2.76 TFLOPS on 1536 SPUs or 3.46 TFLOPS on 1920 SPUs. The latter would be amazing.

[Carfax]

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

If I drink from "teh Nvidia cup," you drink from teh ATI fountain.

Cayman will be the 6810/6830/6850/6870 (yes, four based on Cayman) to be released in November. The 6870 will be around 10 - 15 % faster than the GTX480, have clocks of 900/1500 and have 2GB memory.

10-15% isn't all that much. Nvidia can make up that difference with driver optimization, as they still have much milking to do with the Fermi architecture.

They've done it before, many times.

[spursindonesia]

They are fakes or at the very least modified based on the evidence in this thread and other threads.

Keep prices down? AMD is selling whatever quantities it has of 5xxx. When the 6xxx gets released. There not going to be a lasting price drop on 5xxx products because they are going to be to or close to out of stock already.

If these scores are true, AMD will jack up the price of their card another 100 dollar up the line at the very least to reflect their performance increase. They are already selling out at their current prices. And there's nothings wrong with this, but AMD is not the saint of a company people make it out to be, nor is nvidia the devil.

But its going to be pretty scary for the consumer. 6870, 499 MSRP 600+street price. 6970 750MSRP 850+ dollar street price.

This is almost a certainty to happen in regards to pricing because these card are so underproduced and cards supply are so constrained that the consumer will pay way more than MSRP and along with the increased pricing, its going to be the most expensive generation of cards ever.

If this generation from AMD has taught us anything, supply can screw with the consumer just as much as any company.

The scary thing is if Nvidia doesn't have anything to respond with, which they won't unless they have something underwraps which is unlikely, the consumer is going to be overpaying till Nvidia get a new generation going. The worst part is a solution might not be in sight till perhaps even longer than 28nm considering how unscalable fermi seems at this point considering the size and performance for a new generation at this point. Considering Nvidia may be selling gtx 470-480 at cost, this is really bad for the consumer when a company has to sell a product below cost. It does horrible damage to a company as seen with AMD after the Core 2 duo generation.

You have the right to doubt these rumours/leaks, take a negative perspective, and be a pessimist about it, no problem mate.

But IMHO, all these bad pricing situation in the past 4-5 Q has more factors affecting it rather than just ATi being greedy or wanting to maximize their profit in the short term. This situation is IMHO quite similar with A64 X2 s939 situation of yesteryears, where ATi is limited in supply, that even if they want to slash prices according to cost/profit calculation & expanding the omnipotent marketshare against Intel the behemoth, they still can't. Especially in the graphic card bussiness where market share is very important considering game developers inclination of optimising more to the market leader's mArch design, i think ATi will certainly jump into more sales to grab more market share if situation permitted.

Well, perhaps not like RV 670 & RV 770 experiences where they were quite desperate to maintain market share, not to mention expanding it, but the current situation is certainly not in the best interest for them especially in the long term, if they're not actually supply constrained. And then, you have to take account on materials price inflation, the higher price + low yielding TSMC 40 nm process, & the state of influx of US$ weakening. IMHO, if TSMC promise of significantly increasing their capacity in the 3rd Q 2010 comes true & can supply ATi's demand better, we'll see the situation of inflated ATi's card price abating in the next few Qs.

My .02 centz on it, regards.

[iMacmatician]

10-15% isn't all that much. Nvidia can make up that difference with driver optimization, as they still have much milking to do with the Fermi architecture.

They could also make up close to half of that by a 512 core "GTX 485" providing they don't run into any problems.

[tajoh111]

fermi is fine. almost every chip in the last 10 years has been designed for process scaling. i wouldnt be surprised if they could get a 1024sp fermi on 28nm. they may do an Si spin and get a considerable improvement on 40nm too. i think GTC will reveal what they are planning.

AMD's losses were not only from inferior products but buying ATi and the TLB bug fiasco. nvidia doesnt have those issues.

Whats to stop AMD from doubling up with SI or NI. If they can get more performance for less space, they will keep on getting better and better each generation.

Nvidia needs to get more performance out of the transistors they have and keep the size the same or lower. The simplest way to do this would be to up that shader clock if they can. It was originally rumors thought that this card was going to have a shader clock between 1600-2000mhz. If they can keep the core clock down while increasing the shader clock, this architecture will start to have legs.

The problem with this generation compared to the prior is that they removed the MUL operation which supposedly would not have a drop in performance(it was found in the gtx 280), however it actually did(or drivers still havent reached maturity). If the gtx 295 was clock like a fermi card, it would almost certainly be faster. Also the gtx 480 loses pretty soundly to SLI gtx 285. Per transister, fermi is worse than the gtx 280, which is pretty bad considering it is a new architecture.

The only thing I can think of to turn fermi around at this point is get the power down, up the shader clock and get those original TMU reenabled. Fermi needs 25% more performance at least to be considered a success and to justify its power consumption.

[blindbox]

Oh you guys, I have Unigine so... see this. If you want to remove bold, you'd have to purposely remove obviously.

<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
<html><head>
<title>Unigine benchmark result</title>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8"/>
<style type="text/css">
body { background: #1a242a; color: #b4c8d2; margin-right: 20px; margin-left: 20px; font-size: 14px; font-family: Arial, sans-serif, sans; }
a { text-decoration: none; }
a:link { color: #b4c8d2; }
a:active { color: #ff9900; }
a:visited { color: #b4c8d2; }
a:hover { color: #ff9900; }
h1 { text-align: center; }
h2 { color: #ffffff; text-align: center; }
.right { text-align: right; }
div.orange { color: #ff9900; }
div.highlight { color: #ffffff; }
div.copyright { margin: 20px; text-align: center; }
table.result { border: 0px; margin-left: auto; margin-right: auto; }
table.result td { border: 0px; padding: 3px; font-size: 200%; }
table.detail { border: 1px solid #b4c8d2; border-collapse: collapse; margin-left: auto; margin-right: auto; }
table.detail td { border: 1px solid #b4c8d2; padding: 3px; }
</style></head><body>
<h1><a href="http://unigine.com/products/unigine/">Unigine</a></h1>
<h2>Heaven Benchmark v2.1</h2>
<table class="result">
<tr><td class="right">FPS:</td><td><div class="orange"><strong>1234567890.</strong></div></td></tr>
<tr><td class="right">Scores:</td><td><div class="orange"><strong>1234567890.</strong></div></td></tr>
<tr><td class="right">Min FPS:</td><td><div class="orange"><strong>1234567890.</strong></div></td></tr>
<tr><td class="right">Max FPS:</td><td><div class="orange"><strong>1234567890.</strong></div></td></tr>
</table>
<h2>Hardware</h2>
<table class="detail">
<tr><td class="right">Binary:</td><td><div class="highlight">Windows 32bit Visual C++ 1500 Release May 21 2010</div></td></tr>
<tr><td class="right">Operating system:</td><td><div class="highlight">Windows 7 (build 7600) 64bit</div></td></tr>
<tr><td class="right">CPU model:</td><td><div class="highlight">AMD Phenom(tm) II X3 720 Processor</div></td></tr>
<tr><td class="right">CPU flags:</td><td><div class="highlight">3400MHz MMX+ 3DNow!+ SSE SSE2 SSE3 SSE4A HTT</div></td></tr>
<tr><td class="right">GPU model:</td><td><div class="highlight">ATI Radeon HD 4800 Series            8.762.0.0 1024Mb</div></td></tr>
</table>
<h2>Settings</h2>
<table class="detail">
<tr><td class="right">Render:</td><td><div class="highlight">direct3d10</div></td></tr>
<tr><td class="right">Mode:</td><td><div class="highlight">1920x1080 4xAA fullscreen</div></td></tr>
<tr><td class="right">Shaders:</td><td><div class="highlight">high</div></td></tr>
<tr><td class="right">Textures:</td><td><div class="highlight">high</div></td></tr>
<tr><td class="right">Filter:</td><td><div class="highlight">trilinear</div></td></tr>
<tr><td class="right">Anisotropy:</td><td><div class="highlight">16x</div></td></tr>
<tr><td class="right">Occlusion:</td><td><div class="highlight">enabled</div></td></tr>
<tr><td class="right">Refraction:</td><td><div class="highlight">enabled</div></td></tr>
<tr><td class="right">Volumetric:</td><td><div class="highlight">enabled</div></td></tr>
<tr><td class="right">Replication:</td><td>disabled</td></tr>
<tr><td class="right">Tessellation:</td><td>disabled</td></tr>
</table>
<div class="copyright"><a href="http://unigine.com/">Unigine Corp.</a> &copy; 2005-2010</div>
</body></html>

[Manicdan]

If the gtx 295 was clock like a fermi card, it would almost certainly be faster. Also the gtx 480 loses pretty soundly to SLI gtx 285. Per transister, fermi is worse than the gtx 280, which is pretty bad considering it is a new architecture.

this
nvidia is way behind in perf per mm2, and amd is about to get even better. if they continue to lead in perf per $, then gaming market share will continue to head toward them.

the 28nm race could change everything though

[Carfax]

this
nvidia is way behind in perf per mm2, and amd is about to get even better. if they continue to lead in perf per $, then gaming market share will continue to head toward them.

the 28nm race could change everything though

This is only relevant if you take Fermi as a gaming GPU.....which it isn't.

Fermi was designed for both HPC and gaming.. The HPC market is much more profitable than the high end gamer market (and from the benchmarks I've seen, Fermi decimates ATI in that area), so ATI leading in perf per mm2 isn't as important as you'd believe.

[Iconyu]

This is only relevant if you take Fermi as a gaming GPU.....which it isn't.

Fermi was designed for both HPC and gaming.. The HPC market is much more profitable than the high end gamer market (and from the benchmarks I've seen, Fermi decimates ATI in that area), so ATI leading in perf per mm2 isn't as important as you'd believe.

Does that same defence stand for the 460? It's bigger than the 5870... and designed with lobotomised HPC capabilities.

[trinibwoy]

Cayman will be the 6810/6830/6850/6870 (yes, four based on Cayman) to be released in November. The 6870 will be around 10 - 15 % faster than the GTX480, have clocks of 900/1500 and have 2GB memory.

Barts will be very similar to Cypress but on a new PCB and different clocks, maybe some SP's disabled and have 1GB memory.

Only 15% faster than the GTX 480 but with four different SKUs that are faster than Cypress? It would mean there would be tiny 5-10% differences between each of them, not likely.

[Manicdan]

This is only relevant if you take Fermi as a gaming GPU.....which it isn't.

Fermi was designed for both HPC and gaming.. The HPC market is much more profitable than the high end gamer market (and from the benchmarks I've seen, Fermi decimates ATI in that area), so ATI leading in perf per mm2 isn't as important as you'd believe.

perf per mm2 translates to price and profit.
if amd felt threatened, they would have lowered the price.

funny you bought 2 gpus that arnt even for gaming, lol

[Iconyu]

Only 15% faster than the GTX 480 but with four different SKUs that are faster than Cypress? It would mean there would be tiny 5-10% differences between each of them, not likely.

Nope, more like yields are flaky, but three levels of recycled cores does seem unlikely. I guess it depends on how many shaders and how many banks.

[mindfury]

[rolleyes>記得看清標題“是真的還是假的”
WG_Baby 发表于 2010-8-30 09:59

WG_Baby claimed that he never said those screenshots were legitimate.

[LightSpeed]

10-15% isn't all that much. Nvidia can make up that difference with driver optimization, as they still have much milking to do with the Fermi architecture.

They've done it before, many times.

Remember, the Cayman is a different architecture as well. If Fermi can be tweaked for more performance (which already recieved a perf boost), Cayman can be as well.

Only 15% faster than the GTX 480 but with four different SKUs that are faster than Cypress? It would mean there would be tiny 5-10% differences between each of them, not likely.

Well, 15% is still an approximate from them (Im thinking conservative). Might as well be 20%, they are still trying to set the clockspeed straight. Initial target as far as I know was 950Mhz but its likely that it will end up at 900. Also, I frankly dont remember if I was told 512SP Fermi or 480SP, gotta check..

Also, the four different SKU's are faster than a reduced Cypress (Barts). There should be atleast be a 40% difference between that and the top end Cayman. That leaves 3 to fit in the middle somewhere I guess.

One thing is for sure though, the reduced Cayman versions look to be able to clock like a madman. They are likely to have some SP's disabled though, but they certainly look to be great bang for the buck cards.

Finally, as many places have already noted, it will be AMD Radeon

[saaya]

10-15% isn't all that much. Nvidia can make up that difference with driver optimization, as they still have much milking to do with the Fermi architecture.
They've done it before, many times.

This is only relevant if you take Fermi as a gaming GPU.....which it isn't.

yet thats what they are selling 90%+ of fermi chips as... but nOoOoOo its not a gaming chip... pff... that would be silly...

sounds the standard bad looser excuse we hear so much these days "if you lose, claim you didnt really try to win to begin with"

[Mechanical Man]

yet thats what they are selling 90%+ of fermi chips as... but nOoOoOo its not a gaming chip... pff... that would be silly...

sounds the standard bad looser excuse we hear so much these days "if you lose, claim you didnt really try to win to begin with"

Yeah.

In the end, Fermi is crappy from gaming perspective. It is more powerful yes, but with what a cost..

[mibo]

WG_Baby claimed that he never said those screenshots were legitimate.

So he admits that all sceenshots are fake?
Did he want to test the response time of news sites and forums?

[mindfury]

So he admits that all sceenshots are fake?
Did he want to test the response time of news sites and forums?

According to nApoleon@chiphell,the Vantage P and Unigine scores are legitimate.

http://www.chiphell.com/thread-119587-1-1.html

某卡(平台未知)
3DMark Vantage P24499
Unigine Heaven (1920 1200 4AA+16AF) 36.6
同平台下GTX 480的Unigine Heaven (1920 1600 4AA+16AF)为29.5
nApoleon 发表于 2010-8-30 11:38

[Carfax]

Does that same defence stand for the 460? It's bigger than the 5870... and designed with lobotomised HPC capabilities.

Yes and no. The 460 is much more geared towards gaming than the 465/470/480, but it still has many HPC qualities that increase the size of the die.

Basically, none of the Fermi derived GPUs are purely gaming or HPC GPUs. Instead, they are hybrid processors.

[Carfax]

perf per mm2 translates to price and profit.
if amd felt threatened, they would have lowered the price.

But as I said, the HPC market is far more profitable than the gamer market, so increasing the HPC capability of your GPU leads to more profit than just focusing on gaming only.

Just think, how much does a high end Quadro GPU cost? Thousands of dollars....for ONE!

Increasing the HPC and Scientific capability of their GPUs while retaining their ability to be used as primary gaming GPUs, is one of the best moves Nvidia has ever come up with.

Because of Fermi, Nvidia has strengthened it's hold on the HPC market, which like I said, is inherently much more profitable than the gaming market.....even though less GPUs are sold.

funny you bought 2 gpus that arnt even for gaming, lol

They are for gaming. I should have been more specific, but I thought I clarified in my following sentence:

Fermi was designed for both HPC and gaming

Fermi was designed for BOTH gaming and HPC, and as such, it's pretty amazing since they are the top performers in both areas.....at this time.

[flopper]

There is a increasing need for HPC which amd also will adress most likely in the upcoming 6000 series, but also down the line with new generations as the market expand.

the oem with fusion makes sense a lot, no extra videocard or such features on the motherboard allows the cheaper things to be build all the things a oem like.

I see amd in a better position than Nvidia after they got things togeheter.

AMD still lacks a good PR and communicating department but they havent hired me yet.

[Carfax]

Any long term Nvidia user will tell you that it's nothing for Nvidia to squeeze about 15% extra performance (average) or more out of their GPUs between the introductory drivers, and the fully optimized drivers; which may take about a year or more to realize..

In Fermi's case though, it may be sooner (or later) because the GPU is radically different compared to previous generations.

The 200xx drivers have already given significant increases in performance after only a few months.

yet thats what they are selling 90%+ of fermi chips as... but nOoOoOo its not a gaming chip... pff... that would be silly...

I never said Fermi wasn't a gaming chip

You just selectively read one sentence of mine, instead of looking at the entire post. I said that fermi was designed for BOTH gaming and HPC..

sounds the standard bad looser excuse we hear so much these days "if you lose, claim you didnt really try to win to begin with"

Do you honestly believe that Fermi is a failure? If so, then you are truly deluded..

[Carfax]

There is a increasing need for HPC which amd also will adress most likely in the upcoming 6000 series, but also down the line with new generations as the market expand.

If AMD genuinely desires to address the HPC market, they will have to invest more transistors into their designs for dedicated HPC function....which will mirror what Nvidia has done with Fermi.

Of course, the AMD fan boys will herald this as the greatest thing ever even if the power envelope is increased dramatically, much as what happened with Fermi