AMD Radeon HD 7900 ''Tahiti'' Pictured, 384-bit Memory Bus Confirmed?

[lrlcr]

TF3 are silent at below about 35% fans, what are you doing to require the fans to go so high?

 

[BS Dave]

That's 11 chips right, surely you'd need 12?

 

[ALXAndy]

Am I the only one who finds it rather odd?

Aren't these cards supposed to be like uber low on nm and thus power and heat?

Why would they need a cooler like that (looks like a foot long, btw) and a 6 and 8 pin then?

Or is it another case of fail by AMD?

 

[Owenb]

Am I the only one who finds it rather odd?

Aren't these cards supposed to be like uber low on nm and thus power and heat?

Why would they need a cooler like that (looks like a foot long, btw) and a 6 and 8 pin then?

Or is it another case of fail by AMD?

28nm will lower the power usage, and heat, at a given performance point. The performance point of these cards will be similar to the 6990 but will use about 50-60% of the power that a 6990 would.

 

[Orcvader]

It's because of the smaller process they can cram more transistors in. More transistors = more heat. That pretty much outweighs the heat reduction from shrinking the process. Not to mention, you can always crank up the clocks as well. Hence why power usuage and heat usually keeps going up and up each gen...

 

[drunkenmaster]

28nm will lower the power usage, and heat, at a given performance point. The performance point of these cards will be similar to the 6990 but will use about 50-60% of the power that a 6990 would.

Yup, not sure whats so hard to understand, a new process can drop leakage and power usage per transistor significantly, but if that was the only thing any chip makers used, ie they dropped power by half with the same number of transistors every gen, we'd have great 3W gpu's, that performed like GF3's.

You drop the top end cards performance to half the power, then double the performance at roughly the same power as the previous top end card.

This card is likely to be around the 4billion transistor mark, depending on how big they went and how good the process was predicted to be, it's likely anywhere from 3.8billion to 4.5billion.

The 4870 was 950billion transistors, the 5870 is 2.15billion, honestly I can't remember what the 6970 is but almost certainly quite a bit more. You can give or take double the transistor count.

Basically a 7970 done on 40nm would use probably 600W, be 650-700mm2, if really really lucky get 1-2 working dies off a wafer, at a cost of $5k per core.

New processes aren't about enabling lower power graphics cards, they are about building something that is financially impossible on the previous process.

 

[Sasahara]

New processes aren't about enabling lower power graphics cards, they are about building something that is financially impossible on the previous process.

They are about enabling lower power graphics cards as well as improving the top end.

 

[varkanoid]

it might be the light or the poor picture but it looks like there are 8 LED's in between the red stripes at the exhaust end of the card ?

 

[iB CoMpL3XiTy]

The Pic shows an 8 pin and a 6 pin per card, however the PCB seems to have traces for 2 x 8 pins, maybe they are 7950's and the 7970's require 2 x 8 pins?

 

[Rroff]

In other words, they probably had finished, boxed, ready to ship 7970's as of potentially up to 6 weeks ago..... just at no where close to launch ready supply.

EDIT: Would seem they are further along than I was thinking I'm probably too used to nVidias development cycle :S just a bit strange we haven't had more leaks from AIBs at this point.

 

[Freddie1980]

EDIT: Would seem they are further along than I was thinking I'm probably too used to nVidias development cycle :S just a bit strange we haven't had more leaks from AIBs at this point.

Not really, for the past 2 years both AMD and Nvidia have been as tight as a Scotsman when it comes to the secrecy of their new products and that includes info from their board partners as well.

 

[Rroff]

I agree nVidia has been a bit tighter on info lately, but I had no problems getting details on the 69xx cards last year.

 

[ALXAndy]

28nm will lower the power usage, and heat, at a given performance point. The performance point of these cards will be similar to the 6990 but will use about 50-60% of the power that a 6990 would.

But it still needs the same power connectors?

Surely if it needed 50% of the power it would use a single 8 pin?

 

[MichaelM]

But it still needs the same power connectors?

Surely if it needed 50% of the power it would use a single 8 pin?

If it performs no faster than the previous generation then yes, that would be the case.

 

[Owenb]

But it still needs the same power connectors?

Surely if it needed 50% of the power it would use a single 8 pin?

The 6990 used two 8 pins, this uses an 8+6. Power in doesn't equal power usage. Cards that use 60watts often have a 6pin even though the PCI slot provides 75Watts. Head room never hurts!

 

[drunkenmaster]

I agree nVidia has been a bit tighter on info lately, but I had no problems getting details on the 69xx cards last year.

There's no need for AMD to ship the cards to AIB's when they are coming in still, probably still only ship them a couple weeks before, and because of new process and lower than expected volume at this point(or more Apple getting in the way than anyone but overpaying idiots want ) AMD could still be tinkering with clock speeds/voltage/shader combinations to get the most possible usage of the wafer..... though that likely would be 7950's that are liable to change unless they have an Nvidia type problem and have to cut shaders off the top part too.

The real question I guess is, will the 7950 have less memory and memory bus than a 7970, even if there was some unlocking method, would it still have fundamentally less performance, even another thought, would it matter. If AMD need more bandwidth than 256bit can provide, there isn't a whole lot of choice that makes sense but 384bit........ but is that way more than they actually need anyway?

Lastly, from the pics people are thinking size somewhere between Cypress and Cayman, it certainly doesn't look from first glance like this is some tiny 1st of two gen 28nm cards that was always a slim possibility.

But it still needs the same power connectors?

Surely if it needed 50% of the power it would use a single 8 pin?

In reality, 2 6 pin's gives you the option of 225W max, 75W x2 cables and 75W slot, 8+6pin gives you max 300W. That realistically only means, overall that they think it will use 226W at some point either stock or when overclocked. If it used 220W or something, and basically couldn't overclock you'd have a lot of very unhappy enthusiasts, over something as trivial as adding another $1 of circuitry to let a second cable be connected?

Likewise the more overall power you can draw from the PSU and the more cables it has, the lower the load per cable as it tends to draw a fairly equal percentage across each source. Another 6pin cable means the mobo and 8pin significantly reduced from their maximum rated current draw.

 

[ALXAndy]

Well just colour me incredibly suspicious. The 6 series launch was laughable. I still remember their roadmap where they showed the 6850 above the 5850 and the 6870 above the 5870.

And what incredible overclocks ! about 60 mhz

Something else that doesn't seem to add up is the cards releasing in January were supposed to be low to mid range cards I thought? with the bigger boys following later in the year?

From what I saw the other day Nvidia have no plans at all to go high end with Kepler untlil this time next year.

 

[Aurhinius]

The top end card i.e dual GPU card will come in march but the other cards are scheduled for CES. This should give ATI time to build up inventory as the TSMC process isn't producing yields.

Nvidia have the same issues but are taking a bit more time to get their products out.

You will probably see something from nvidia towards the end of march where they have to strike back against ATI's dual GPU solution.

Lets also remember nvidia have been about 6 months behind before - GTX 480/580

 

[Bobisuruncle54]

TF3 are silent at below about 35% fans, what are you doing to require the fans to go so high?

I have 2 6950s clocked at 900Mhz and they are noisy as hell. This doesn't happen in every game, but in BF3 they wind up within minutes.

At the moment MSI Afterburner reports the fan speed as 44% for the top card, running idle at 52 degrees, and 40% for the bottom card, running at 37 degrees. The top card is running hotter because I'm using dual screens. The fans kick in to stop the cards going above 80 degrees, and they are very loud at that point.

Thinking of replacing them with a 7000 series card/s rather than replacing the coolers, but I'll have to see how the figures work out.

 

[Marine-RX179]

I have 2 6950s clocked at 900Mhz and they are noisy as hell. This doesn't happen in every game, but in BF3 they wind up within minutes.

At the moment MSI Afterburner reports the fan speed as 44% for the top card, running idle at 52 degrees, and 40% for the bottom card, running at 37 degrees. The top card is running hotter because I'm using dual screens. The fans kick in to stop the cards going above 80 degrees, and they are very loud at that point.

Thinking of replacing them with a 7000 series card/s rather than replacing the coolers, but I'll have to see how the figures work out.

No offense, but how is the airflow of your case, and how much gap are there between the top and lower card?

If the top card is sitting directly on top of the lower card's PCB, then I doubt ANY air cooler can be quiet and keep the temp at at sub 80C.

Also, pair of 6950s on "dump heat into case" type cooler WILL dump lots of heat into the case (especially overclocked), so unless you got some exhaust for the lower half of the case (i.e. fan on lower side-panel as exhaust), lots of the heat will get recycled back into the graphic card coolers.