AMD Radeon HD 7970 GHz Edition 3 GB

[pantherx12]

Well i have yet to see a full comparison including the latest GTX 6xx / 7xxx series in terms of compute power but last time i checked it myself the GTX 570/580/590 wiped the floor with the HD 6xxx series so i think they would also do the same with the 7xxx series, unless AMD managed to increase their performance by 200-400%.

They increased performance by 400%

They are monstrous when it comes to compute

my 7850 beats all the 6xxx series at math including dual gpu cards.

OpenCL performance is especially epic and 7xxxx cards smash All of nvidias cards right now.

Can download clbenchmark to test

I think vs my card 680 wins 50% of the benchmarks so the cards definitely have different strengths.

My scores are attached.
GPU is @ stock speeds for the test.

 

[raghu78]

that's true.. but please remember, in gaming compute it only utilizes fp32 or hardly uses fp64,, which is nvidias fp32 compute performance is still not way too behind of radeons. so i guess, there is nothing too lose of kepplers or any nvidias compute gaming. it is just how nvidia has big advantage in tesselation, and radeon (now) has big advantage in compute... it is draw.

You do realize that the "comparison" you are showing here is flawed in many ways right ?
1st I doubt they used the latest drivers for the GTX5xx series, naturally that is since there's no way they still have the samples and even if they did it would take too much time to do benchmarks all over again.
2nd In some charts we see the GTX570/580 but in None do we see the GTX590.
3rd In other charts we don't even see the GTX5xx series.

Aside these flaws however it seems to me that even the GTX580 is doing extremely well even against the 7970GHz Edition and we are talking about a card with already almost 2 years in the market.
Regardless i really doubt there's a faster compute card than the GTX590 right now so i can't understand your last statement. Unless of course you speak about a single chip card.

Please read carefully

"Our first compute benchmark comes from Civilization V, which uses DirectCompute to decompress textures on the fly. Civ V includes a sub-benchmark that exclusively tests the speed of their texture decompression algorithm by repeatedly decompressing the textures required for one of the game’s leader scenes. Note that this is a DX11 DirectCompute benchmark"

Thats just a standard DirectCompute benchmark using compute shaders. no fancy fp64 stuff. GTX 680 is slower than GTX 580. Thats not so good looking . If you can't beat your last gen flagship product with the current gen flagship that really is pathetic.


Even in SmallLuxGPU the GTX 680 loses to GTX 580. its not really inspiring stuff. The DX11 Compute shader based fluid simulation benchmark is respectable for GTX 680 as it beats GTX 580 easily by 50% margin. But still its on par with HD 7970 Ghz edition.

As far as tesselation performance is concerned Nvidia's lead is not very big. Also with extreme tesselation Unigine Heaven 3.0 is only slightly faster at 1080p on GTX 680 (1006 core/ 1058 - 1110 Mhz boost) wrt Sapphire HD 7970 (1 Ghz) while slightly slower at 1600p.

http://www.legitreviews.com/article/1962/12/

Simply put GTX 680's major advantage is perf/watt. So for 1080p if you want best perf/watt GTX 680 is the better option. But for ultra high res single monitor (1600p) and multi monitor gaming and compute HD 7970 is the card.

 

[HellasVagabond]

Did you even read what i said ? Really ? I don't see you saying something else. The GTX 6xx is worse in compute power than the GTX 5xx series so what ?
Up until today NV has been better in Tessellation from AMD, noone can deny that, sure one day AMD May take the lead, again so what ? You can't always be best. Same goes for computing power since NV was better as well from AMD up until the GTX 5xx series. Again right now whoever wants the best computing power should get an GTX 590, are you against that ?

 

[atikkur]

Please read carefully

"Our first compute benchmark comes from Civilization V, which uses DirectCompute to decompress textures on the fly. Civ V includes a sub-benchmark that exclusively tests the speed of their texture decompression algorithm by repeatedly decompressing the textures required for one of the game’s leader scenes. Note that this is a DX11 DirectCompute benchmark"

Thats just a standard DirectCompute benchmark using compute shaders. no fancy fp64 stuff. GTX 680 is slower than GTX 580. Thats not so good looking . If you can't beat your last gen flagship product with the current gen flagship that really is pathetic.


Even in SmallLuxGPU the GTX 680 loses to GTX 580. its not really inspiring stuff. The DX11 Compute shader based fluid simulation benchmark is respectable for GTX 680 as it beats GTX 580 easily by 50% margin. But still its on par with HD 7970 Ghz edition.

As far as tesselation performance is concerned Nvidia's lead is not very big. Also with extreme tesselation Unigine Heaven 3.0 is only slightly faster at 1080p on GTX 680 (1006 core/ 1058 - 1110 Mhz boost) wrt Sapphire HD 7970 (1 Ghz) while slightly slower at 1600p.

http://www.legitreviews.com/article/1962/12/

Simply put GTX 680's major advantage is perf/watt. So for 1080p if you want best perf/watt GTX 680 is the better option. But for ultra high res single monitor (1600p) and multi monitor gaming and compute HD 7970 is the card.

that was really what i meant.. the major representative benchmark of compute for gaming here is CIV V and Fluid Sim, no fancy fp64 will be used for gaming, only fp32. so nothing to worry about of nvidia 6xx compute in gaming, still can on par with radeon.

what do you expect of ocl smallux implementation in gaming? realtime raytracing quality gaming? not a chance with todays technology.

for tesseleation, lets have a look at this data too:
http://www.geeks3d.com/20120702/asu...ew-gk104-overclocking-opengl-direct3d-test/4/

scroll down to tessmark section, this is pure tesselation benchmark. nvidias tesselation starting to fly off radeons at their extreme setting:


moderate__evga 680 : msi 7970 - 1097 : 1032 fps___6,29%
normal____evga 680 : msi 7970 - _776 : _643 fps___20,68%
extreme___evga 680 : msi 7970 - _411 : _247 fps___66,39%
insane____evga 680 : msi 7970 - _174 : __73 fps___138,35%

 

[raghu78]

scroll down to tessmark section, this is pure tesselation benchmark. nvidias tesselation starting to fly off radeons at their extreme setting:

You can't just be doing tesselate alone in game. In a game you have lots of work to do in rendering. Tesselation in a more generic benchmark like unigine heaven stresses all aspects of the card - shader power, tesselation, ROP, bandwidth.

http://www.legitreviews.com/article/1962/12/

Sapphire HD 7970 (1Ghz ) with the latest drivers is slightly behind at 1080p and slightly ahead at 1600p wrt GTX 680 in Unigine heaven with extreme tesselation, 8xAA, 16xAF. So you are not going to see huge differences in games.

As for compute performance here is Dirt Showdown which uses Direct Compute shaders for advanced lighting and global illumination.

http://www.rwlabs.com/article.php?cat=&id=636&pagenumber=2

you can read the explanation for advanced lighting and global illumination

http://www.rwlabs.com/article.php?cat=&id=636&pagenumber=3

the performance at ultra settings is close to 50% faster on HD 7970 wrt GTX 680.

few other sites which have tested Dirt Showdown

http://techreport.com/articles.x/23150/6
http://www.hardware.fr/articles/869-14/benchmark-dirt-showdown.html

 

[onthemour]

With all 7970s reference flashed to ghz bios amd is on throne for sure

I can't believe the massive oc boost i get with this bios on my sapphire 7970 reference. i wa at 1235/1700 before now i hit 1325/1785!!!!!!!!!!!!!!!!!! I CAN PLAY BF3 5760*1080 WITH ULTRA

So I guess I can max all games out on 3 screens with one 7970. I think its time for developers to make more demanding games

 

[radrok]

I think its time for developers to make more demanding games

Or it is time for Display manufacturers to increase the Pixel count?

 

[Steevo]

So purchasing a 7970 and overclocking it for less money is just as good of an idea. Amazing.


AMD, when you gonna take the next step, you know, and actually work with software devs and push along the standards you have been crying for?

 

[Aquinus]

So purchasing a 7970 and overclocking it for less money is just as good of an idea. Amazing.

So is waiting for Sea Islands. I'm sure AMD will turn up something interesting.

AMD, when you gonna take the next step, you know, and actually work with software devs and push along the standards you have been crying for?

I think they do this already, but they should expand upon it. It's not easy to design GPUs to be as complex as they are these days. We have more capability on what we can do as far as what kinds of circuity we make and how it is made, but every GPU manufacturer will still have to figure out the best way to do it and as those tools become more advanced, what they make changes with how GPUs are designed to scale with that.

 

[radrok]

So is waiting for Sea Islands. I'm sure AMD will turn up something interesting.

Do you think that the Sea Islands launch is near?

 

[Aquinus]

Do you think that the Sea Islands launch is near?

No, but I'm also satisfied with my 6870s, so why upgrade. I'm not expecting to see Sea Islands until Q4 2012 or Q1 2013 and even then it really depends on how much improvement AMD has made from GCN to GCN2 that will determine if I want to upgrade or not, money aside.

 

[Nordic]

I think all this sea island stuff coming very very soon is all fud. People were saying in august initially... and... was fud. Sea islands will come just not as soon as some people say. Aquinus's assessment seems sound.

I got a 7970 recently. At stock it is much more powerful than I need. My hope is for it last a few years at my resolution of 1080p. Maybe if I ever do get a 1440p or higher monitor I will upgrade.

 

[radrok]

No, but I'm also satisfied with my 6870s, so why upgrade. I'm not expecting to see Sea Islands until Q4 2012 or Q1 2013 and even then it really depends on how much improvement AMD has made from GCN to GCN2 that will determine if I want to upgrade or not, money aside.

Agreed, I think we are fine with graphics power offered by new graphics cards, we need faster CPUs in the IPC front, and an eight-ten core LGA 2011 :|

 

[Steevo]

Fewer cores more gigahertz is my need. I dont need 8 cores or threads right now, but 4or 6 that run 5 or 6.

 

[radrok]

That would be ideal for gaming but I could use more SB/IB cores honestly, 20 threads would be awesome

 

[Nordic]

That would be ideal for gaming but I could use more SB/IB cores honestly, 20 threads would be awesome

For my needs right now I want less cores and higher clocks. I am thinking about disabling 2 cores of my 2500k and seeing if I can overclock it a good bit higher. I can do 4.5ghz now and 1.37 volts.

 

[radrok]

For my needs right now I want less cores and higher clocks. I am thinking about disabling 2 cores of my 2500k and seeing if I can overclock it a good bit higher. I can do 4.5ghz now and 1.37 volts.

I don't know if it is worth it, you won't get much higher clocks at same voltage with 2 disabled cores, but you'll get lower temps at same voltage for sure.

 

[Nordic]

I don't know if it is worth it, you won't get much higher clocks at same voltage with 2 disabled cores, but you'll get lower temps at same voltage for sure.

I'll start another thread if I do it. Show my results

 

[ValenOne]

They increased performance by 400%

They are monstrous when it comes to compute

my 7850 beats all the 6xxx series at math including dual gpu cards.

OpenCL performance is especially epic and 7xxxx cards smash All of nvidias cards right now.

Can download clbenchmark to test

I think vs my card 680 wins 50% of the benchmarks so the cards definitely have different strengths.

My scores are attached.
GPU is @ stock speeds for the test.

With NVIDIA Kelper's SMX unit, 192 core shares 64 KB cache/buffer and 256 KB register storage.

With AMD's GCN's compute unit(CU), 64 cores shares 64 KB local data store (LDS) and 256 KB register storage. That's not including 16 KB L1 cache, 8KB branch cache and 16KB read only/32 KB shared cache (for quad CUs).

In relation to local storage and registers, AMD GCN's cores has less contention compared to the NVIDIA Kelper i.e. AMD GCN has higher stamina before it spills over the external VRAM. When it hits external VRAM, 79x0 has a higher memory bandwidth.

Fermi's SM unit has 32 cores sharing 64 KB cache/buffer and 128 KB register storage. Fermi has less contention compared to Kelper.

 

[ValenOne]

No, but I'm also satisfied with my 6870s, so why upgrade. I'm not expecting to see Sea Islands until Q4 2012 or Q1 2013 and even then it really depends on how much improvement AMD has made from GCN to GCN2 that will determine if I want to upgrade or not, money aside.

7970 doesn't follow 7770-to-7878 scaling. 8970 could scale from 7770-to-7870 i.e. 48 ROPs powered GPU with matching 384bit wide external bus.

There's a Tahiti GPU with 256 bit VRAM i.e. FirePro W8000. http://www.amd.com/us/products/workstation/graphics/ati-firepro-3d/w8000/Pages/w8000.aspx#3

 

[onthemour]

I bet we will see 8000 series in a couple of months. Single screen users won't need then but high resolutions will. Games will not push the graphics until the consoles come out so single screeners have no point to upgrade anymore. Sea islands will be expensive because they know only a small part y of the e m arket will bbuying them

 

[amc77]

Facts

Again you miss the point that even the 7970GHz edition with overvoltage and OC loses to the GTX 680 in several games and where it comes on top we are talking about a 3% increase at most. So i really don't know what you are trying to say. That Some people have OCed the 7970 more ? So what ? Some will have also OCed and GTX 680 more, again so what ? We are not looking for the exceptions, we are looking for the rule and the rule is that the 7970 at stock is no match for the GTX 680 at stock and the 7970GHz edition still falls behind the stock GTX 680 in several games so if one was to slightly OC the GTX 680 it would certainly beat the 7970GHz.
I am not saying that the 7970Ghz is a bad card but it's just an OCed 7970 and nothing more, hardly something for people to get excited for.

'We've known since May, the existence of a new high-end single-GPU graphics card SKU in the works, at AMD. Called the Radeon HD 7970 GHz Edition, the SKU is being designed to regain AMD's competitiveness against NVIDIA's GeForce GTX 680. We're hearing a few additional details about the SKU. To begin with, AMD has worked with TSMC to refine the chip design. The Tahiti XT2 will be able to facilitate significantly higher clock speeds, at significantly lower voltages, than the current breed of Tahiti XT chips.

Tahiti XT2, or Radeon HD 7970 GHz Edition, will ship with a core clock speed of 1100 MHz, 175 MHz faster than the HD 7970. The GPU core voltage of Tahiti XT2 will be lower, at 1.020V, compared to 1.175V of the Tahiti XT.'

Just sayin...

 

[Nordic]

'We've known since May, the existence of a new high-end single-GPU graphics card SKU in the works, at AMD. Called the Radeon HD 7970 GHz Edition, the SKU is being designed to regain AMD's competitiveness against NVIDIA's GeForce GTX 680. We're hearing a few additional details about the SKU. To begin with, AMD has worked with TSMC to refine the chip design. The Tahiti XT2 will be able to facilitate significantly higher clock speeds, at significantly lower voltages, than the current breed of Tahiti XT chips.

Tahiti XT2, or Radeon HD 7970 GHz Edition, will ship with a core clock speed of 1100 MHz, 175 MHz faster than the HD 7970. The GPU core voltage of Tahiti XT2 will be lower, at 1.020V, compared to 1.175V of the Tahiti XT.'

Just sayin...

Source?