AMD: Southern Islands (7*** series) Speculation/ Rumour Thread

[MadMaxtrox]

Pure power and computing force dont forcibly transmit to game engine and evnironnement, and you have mention one ... the drivers, or engine developpement .. the choice made during the developpement of the game, and specially the optimisation of it .

the 7970 have someting like 1tflops more in SP and 5x more tflops in DP ( not important for games anyway ) .. The GK20 with a dual Kepler is just at 4.5tflops.... really close of the 7970 .. i have no doubt why Nvidia have some problem with their GK110 against GCN ... they will not be able to match 2x Kepler, and the GCN have a good advance in computing part now . ( hence why maybe Nvidia have not report the GK100 and pass to GK110 directly )

So is the implication here that AMD has fallen significantly behind NV on the SW side? Additionally are you implying that with better SW development / Title developer participation that 7870s would leave 680s in the dust?

Is there a real corollary between synthetic throughput analysis and practical real game results?

 

[Kaotik]

Speaking about old 7970, 250W is the max board power ... not the tdp ..this is the max is designed the card for be used .. tdp was vary between 160 to 190W .. yet it is a bit higher for the 7970ghz, but still it is vary a lot from a game to another ..

Before saying TDP is this and that one has to agree which definition of TDP we're using

TDP (Thermal Design Power) of HD7970 however is nowhere near "160-190W", but either 210W (typical board power) or 250W (max board power), depending on wether the cooling solution is designed to dissipate that 250W or just the typical 210W

 

[gamervivek]

So is the implication here that AMD has fallen significantly behind NV on the SW side? Additionally are you implying that with better SW development / Title developer participation that 7870s would leave 680s in the dust?

Is there a real corollary between synthetic throughput analysis and practical real game results?

http://blogs.amd.com/play/2012/06/20/dirt-showdown-on-gcn/

Though when you consider the original contention, 7970 still comes out looking like a paper tiger.

 

[Rangers]

http://blogs.amd.com/play/2012/06/20/dirt-showdown-on-gcn/

Though when you consider the original contention, 7970 still comes out looking like a paper tiger.

Wow, Nvidia is in for quite the whooping on that game. Think Anand said they're replacing Dirt 3 with it in their benchmark suite going forward too.

 

[boxleitnerb]

Before saying TDP is this and that one has to agree which definition of TDP we're using

TDP (Thermal Design Power) of HD7970 however is nowhere near "160-190W", but either 210W (typical board power) or 250W (max board power), depending on wether the cooling solution is designed to dissipate that 250W or just the typical 210W

There are applications (Furmark, cough) that easily exceed 250W on either 7970 or 7970 GHE. Why isn't powertune throttling there?

 

[Ryan Smith]

Wow, Nvidia is in for quite the whooping on that game. Think Anand said they're replacing Dirt 3 with it in their benchmark suite going forward too.

It's on the list of things to do as part of a general refresh. But we're talking weeks, not days.

 

[Gipsel]

There are applications (Furmark, cough) that easily exceed 250W on either 7970 or 7970 GHE. Why isn't powertune throttling there?

Actually it is throttling in FurMark when Powertune is set to 0. But some sites raise the powertune setting to +20% for the power measurements with FurMark and at +20% at least the normal HD7970 is not throttling, afaik. All the measurements of 300W (only for the card) or even above are done with a raised powertune setting in my opinion (HT4U say so in their test for instance).

 

[Alexko]

Actually it is throttling in FurMark when Powertune is set to 0. But some sites raise the powertune setting to +20% for the power measurements with FurMark and at +20% at least the normal HD7970 is not throttling, afaik. All the measurements of 300W (only for the card) or even above are done with a raised powertune setting in my opinion (HT4U say so in their test for instance).

Honestly, with power-management mechanisms like PowerTune and Boost, I just don't see the point of Furmark anymore.

 

[CarstenS]

Neither do I. It can only serve as a point of reference whether or not the power containment meachnisms are active (on some partner's cards they apparently are not) and working correctly.

But I will question something given as "maximum board power" (or the equivalent of it) when some „real“ applications make the card exceed it.

 

[lanek]

So is the implication here that AMD has fallen significantly behind NV on the SW side? Additionally are you implying that with better SW development / Title developer participation that 7870s would leave 680s in the dust?

Is there a real corollary between synthetic throughput analysis and practical real game results?

Not really, but ofc it will impact in some games ( look the dirt3 performance change between 12.3-4 and 12.7, the 7970 was slower of the 680, and now it is faster in this game ).. Still some games due to their developpement path will still perform better on a card or another.. (architecture or game code etc remember Hawx2 on nvidia hardware or DragonAge2 on AMD cards. )

GCN have all the computing parts enabled, when the 680 have not them. ( specially for DP, just need to see the Tesla10 who have just 0.2 Tflops DP performance, if they was there, they will enable them for this card. .. ).

 

[ECH]

It looks to me that the GHz Edition does really well when using 8xAA. I've only found 2 reviews sites that show this though.
HC
CB
Are there other reviews showing 8xAA results? It wasn't that long ago where 8xAA wasn't much of an option at 1200 res. Now it looks like we can use it by default.

 

[hurleybird]

The GHz edition ASIC is explicitly designed to fit within the same thermal / electrical / physical footprint as the standard 7970 and as such, given that every one of the partners have transistioned to their own designs they will be using their own fansink designs. The "reference" design is not much more than our qualification mule now and is sampled for the purposes of performance testing.

But it looks like the first GHz edition announced is going to use stock cooling

 

[Dave Baumann]

http://vr-zone.com/articles/sapphir...or-x-models-to-hd-7970-ghz-edition/16379.html

 

[RedVi]

6GB I hope they do a re-release of the current Sapphire 7970 OC edition since the current one is very nicely priced. Somehow I doubt this Toxic will be within my price range. Would be a great card for professionals though.

 

[lanek]

It looks to me that the GHz Edition does really well when using 8xAA. I've only found 2 reviews sites that show this though.
HC
CB
Are there other reviews showing 8xAA results? It wasn't that long ago where 8xAA wasn't much of an option at 1200 res. Now it looks like we can use it by default.

There, i dont agree, 8xMSAA have been the rules since a long time, anyway for 1920x1080... sadly there's many game out now who dont support standard AA method and you end with 4xAA + ( FXAA etc ) .. I dont understand why some review even bother to test with medium AA quality setting for this type of card when possible ( if this is not just for compare between 2-4-8xAA ).

Before we get strange AA method in late games, i have allways use 8xMSAA at minimum ( and this since the X1900XTX ( ok paired with a x1950XTX).

But anyway, the Hardwarecanuck review use at same time 2560x1600 + 8xMSAA ... and we know this is exactly where the 7970 perform really well. At this res, the 7970 was allready faster in many games.
( the old 7970 was allready win 4 on their 8 games tested and was equal or at 1 fps under of the 680 in the rest of the games..

@Redvi, well 6gb is not really needed, outside maybe for extreme Eyefinity setup ( 6 screens ) for try save a bit more of graphic quality setting

 

[Alexko]

http://vr-zone.com/articles/sapphir...or-x-models-to-hd-7970-ghz-edition/16379.html

By the way, is it possible for partners to change the way Boost works? For instance its frequency or voltage amplitude?

 

[lanek]

Just for add as i have do some little test this morning ..

DTE look to impact positively the overclocking capability at least with H2o.

As i have mention allready, one of my gpu dont oc as good of the other one.. this card had some problem to pass 1300mhz+ at 1300mv.with 12.7 it pass without problem. (As trixx allow me 1386mv, i will see how much i can go with this one )

 

[boxleitnerb]

Not sure why anyone cares about the reference cooler since it won't be used.

Really?
http://vr-zone.com/articles/club-3d-first-with-radeon-hd-7970-ghz-edition/16400.html

Carrying the model number CGAX-7977SO, the card by Club 3D sticks to AMD's reference design, for both the PCB and cooling assembly, with just company branding stickers to make for the difference.

 

[lanek]

Allready posted 7 post above ..

I believe club3d will use their own cooler only on the "royal " cards.. ( King / queen ), and they wanted to be the first..

At same time, Vr-zone do an article about Sapphire who will use the Ghz edition on the Toxic, and other "customized card " .

 

[MadMaxtrox]

strategy

Before we get strange AA method in late games, i have allways use 8xMSAA at minimum ( and this since the X1900XTX ( ok paired with a x1950XTX).

@Redvi, well 6gb is not really needed, outside maybe for extreme Eyefinity setup ( 6 screens ) for try save a bit more of graphic quality setting

At our lab what we have been interested in for at least the last 4 years is true smooth performance in 3 screen simulation and gameplay. At issue is the resolution configuration. Triple 30" 2560x1600 in portrait (4800x2560). The long term goal has been smooth experience WITHOUT microstuttering. The closest we have been able to achieving this is with 3 card configs, but costing has been problematic as well as some SW not supporting (Scaling has gotten better over time). Monitor cost is already high, so potential single card solutions are of interest.

With that being said, i am not sure that a 6GB frame buffer is as critical as the memory bandwidth for us, but we are back to the issue of SW developers taking full advantage of underling architectural advancements.