BenetanegiaIt has 33% more shaders than the HD6970.

2048/1536 *100 = 133%

or

2048 - 1536 = 512 // 512/1536*100 = 33%

HD6970 has 25% less shaders.

The joy of percentages. :)

also dont forget that as a card gets faster, it becomes less of a bottleneck, so system-wise the boost will not equal linearly to the speed increase of the card.

will this unlock?

legends84will this unlock?

We won't know until launch.

ShihabyoooI think it will fit between the 570 and the 580. Though I think the pricing would be that of the 580. Of course the price justification would be backed up by hand picked benchmarks of games where AMD GPUs excel at.

Now seriously, anyone else is curious why Nvidia's still lurking in the shadows ? All the news lately have been about the 7k series. Is Kepler going to be an epic ? Or is the gain's so minimal it would outshine Bulldozer's fail ? Someone gimme some news, I'm dying here !

Kepler should be a monster of a GPU, and IMO "MUCH" faster than the HD 7970. This is why AMD won't release the HD 7990 until NVIDIA releases Kepler. The HD 7990 should be an even bigger Monster, just until NV get's there own Dual-GPU card out the door. It's a back and forth game. Though I don't think NV's Dual-GPU will take out the HD 7990 because it seems the HD 7000 series scale very well in Crossfire :)

what ever happened to the XDR2?

imitationSure, why not? A 7970-equipped PC consumes around 350W (max, not peak), so unless you have a no-name china PSU, you'll be fine.

i've had a months old 550 Antec Basiq fail to post my 5870 on a uATX system with a 65 watt AMD CPU. hell, Antec even said in essence that it sucked in my RMA request. "if possible try to return [to vendor] and get Earthwatts."

ShihabyoooI think it will fit between the 570 and the 580. Though I think the pricing would be that of the 580. Of course the price justification would be backed up by hand picked benchmarks of games where AMD GPUs excel at.

Now seriously, anyone else is curious why Nvidia's still lurking in the shadows ? All the news lately have been about the 7k series. Is Kepler going to be an epic ? Or is the gain's so minimal it would outshine Bulldozer's fail ? Someone gimme some news, I'm dying here !

No, Nvidia always releases after AMD. I think the 2 companies know that if they make a card just a little faster than the competitor people will buy it. They don't try to kill the competition, they just try to make the sale.

BrooksyXPerfect. Just in time for my birthday.

Do you guys think a 7950 would run on a 500w psu with an overclocked 2500k? I am currently running it with a 6870 that has a mild OC.

edit: also hopefully by march nvidia will have their next generation out so card prices will reflect competition.

I like the new features of GCN, so If the 7950 could deliver close to the 580's performance for a lower price, while doing it in a lower power budget and delivering all the same features as the 7970, then I predict that it will be a huge success. My guess is a 399$ price point.

imitationSure, why not? A 7970-equipped PC consumes around 350W (max, not peak), so unless you have a no-name china PSU, you'll be fine.

What PSU calculator do you use for that?

Dj-ElectriCDear god, Forget world piece
Forget the hunger
Forget the wars

I want the HD7950 to unlock to HD7970

Thank you, amen.

:slap::slap::nutkick::nutkick:

I look forward to seeing how these are priced and perform.

I assume if there will be 1.5GB models they would mainly be useful for single card and screen setups and the 3GB cards would be better for dual card, triple screen setups.

ZubasaMore like the other way around :ohwell:
Its most likely that the same damned 32 ROPs are holding the card back more than anything.
It is quite evident that even the 5850 is almost as fast as the 5870 clock for clock, and this is also true for the 6950 and 6970.
They keep slapping more shader power on these cards without addressing the most likely bottleneck.

This is something that has been on my mind, why has AMD kept the top end chips at 32 ROPs for 3 generations now?

afwI'm a little confused here :confused: ... the 7970 has 33% more streaming processors and more memory bandwidth than 6970 ... but its only around 30% faster than the 6970 ... y is that ... ??? may be im missing something ... :o

Have you seen other reviews where dx11 games are maxed out the 7970 is sometimes almost 70% faster in really demanding games with tessellation and other features maxed out. 30% is the bottom lowest it ever goes I think average is more around 50%.

Read the overclocking article at guru3d.com and the card almost reaches 6990 levels and in some games even beats out the gtx 590. It is monster overclocker without even increasing volts it maxes out at 1125 on the core all with in Overdrive.

One hell of a card if you ask me.

I am definately interested in performance of these. I bet the prices are going to be higher than as with previous release like the 7970 is. In a way, it makes sense as AMD needs to return to profitability and staying in the lower price segments isn't going to cut it long term. I know most of the sales are there in the low end but bringing up the ASP should be the main goal at this point.

bear jesusThis is something that has been on my mind, why has AMD kept the top end chips at 32 ROPs for 3 generations now?

This seems to be a weird design decision with ATi for a long time.
They have been stuck on 16ROPs all the way from the X800series (the so called pipelines back then) to the HD4800.
ATi seems to favor shader power (particularly pixel shaders since the X1900 series), until when they released the HD2900 with vliw5 which have problem utilizing its full potential.
They simply slap more shaders on it with the 4800s and beyond, and now they fixed the problem by going this 1D shader route which should be easier to code for but still stuck on 32ROPs....

plonk420i've had a months old 550 Antec Basiq fail to post my 5870 on a uATX system with a 65 watt AMD CPU. hell, Antec even said in essence that it sucked in my RMA request. "if possible try to return [to vendor] and get Earthwatts."

Take a look at your PC. Then have a look at mine. Now look back at yours. Now look at my PSU. Look back to yours. Your Antec PSU is now an overpowered beast. :roll:

TylerZamboriWhat PSU calculator do you use for that?

I read that number in some external review, and quickly checking back at the TPU review and using my brain calculator, it makes sense.

ZubasaThis seems to be a weird design decision with ATi for a long time.
They have been stuck on 16ROPs all the way from the X800series (the so called pipelines back then) to the HD4800.
ATi seems to favor shader power (particularly pixel shaders since the X1900 series), until when they released the HD2900 with vliw5 which have problem utilizing its full potential.
They simply slap more shaders on it with the 4800s and beyond, and now they fixed the problem by going this 1D shader route which should be easier to code for but still stuck on 32ROPs....

It is kind of odd but follow me here. The thing that has stuck with me is 3.4 (of 4 or 5) shaders were used in the vliw design per AMD (anandtech cayman review). In the grand scheme that means 54.4 alus per rop for amd since it is a group of 16. 1/4 that number of shaders for fermi since each shader is 2 alus at 2x clock. Likely half that for kepler since it will double shader count while losing the 2x shader clock.

AMD said that is what is efficient for shader usage and independent tests have backed them up. Cayman put specal function in the shaders to up efficiency closer to all 4 alus being used per pass. nvidia uses fixed funtion sf units. 1/8 for gf100 and 110. 1/6 for gf104 and 114. .6/4 = between 1/6 and 1/7.

example. nv = 3.4 x 2 alus x 2x shader clock x 32 rops = 435.2

ie '100%' generalized efficiency with fermi for 32 rops.
gf114 is 384sp with 1/6 sfu rate. Considering shader design and granularity...pretty efficient.


ok. the math works...yay. so...lets try tahiti.


amd = 3.4 x 16 alus x 32 rops = 1740.8
left over for special function in Tahiti (2048-1740.8): about 1/6 ...slightly more.

what is the point of super fun math time?

GCN is very efficient as far as shaders and rops go.

7790 will likely become the new mainstream card for gamers if it is a 256 bus....

TylerZamboriWhat PSU calculator do you use for that?

my i7-920 + 5870 uses 280 watts prime+fur
i7-920 + 560 uses 370 watts prime+fur per my kill-a-watt

RecusEntire AMD Radeon HD 7000 series Specs and Price Revealed

I have a feeling those prices will drop once Nvidia releases kepler.

I would love to get a 1.5gb or even a 3gb 7950 for my birthday but not at those prices. I could probably sell my 6870 for like $100 and even then I couldnt justify spending another $300 out of pocket on a video card.

Sure would be nice though. My 6870 1gb is struggling pushing my 3x 20" Monitor 4800x900 eyefinity setup.

afwIm not well versed in the tech stuff ... but just asking :D ... suppose the 7970 has the same amount of streaming processors, same memory, same core/memory clocks like the 6970... then ... it'll be slower than the 6970... right ??? so only possible +ve factors will be less power and less heat and the new features ... :confused:

does that mean GCN is less efficient/powerful than the old architecture ???

It won't. 7970 is going to be far more complex than 6970 so it will be a lot faster. It will also have more memory. 7870 is going to be a die shrink of 6970 and have the same number of stream processors and ROPs.