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

And right there you proved Dave's point.

The 48xx cards greatly eroded the Radeon's brand positioning and image for the worse.
snip




Oh come on.. AMD is sitting the performance/price disadvantage on their brand right now based on what?

- Brand placement? nVidia has it all with most AAA PC titles being TWIMTBP and having their logo appear in the beginning of each game. Is AMD changing this?

- Custom game enhancements? None that I've heard of. nVidia has all of those: PhysX, unlocked AA modes, GPGPU assisted effects like Just Cause 2. Is AMD changing this?

- GPGPU? nVidia is light-years ahead with almost every GPGPU-enabled software having a "main" CUDA path and then sometimes they go out of their way to enable an OpenCL path. Now that the GCN is a much more compute-friendly architecture, I have no doubts that this will change, but the current paradigm is: nVidia is better for computing. No brand advantages there, yet.

- Multi-monitor gaming? Yeah, Eyefinity was here first, but how many people exactly have such a setup? Now nVidia has it too for single-card usage, so expect all TWIMTBP games to feature this. Plus, 2.5 years after the release of the HD5000 series we're still waiting for custom resolutions. We'll see who comes up with that first, as it clearly seems to have been forgotten by AMD.

- 3D displays? nVidia is also miles ahead here. Sure, AMD stated they would follow the "open" way, which is nice and all, but when a game supports stereo 3D, which system does it use? 3D Vision.

- Driver quality? Hardly.. nVidia is a lot faster in releasing updated drivers when a new AAA game comes out. Probably related to the fact that their engineers partially develop the games themselves. I have no problems whatsoever with AMD drivers, but clearly they're not on top and clearly there's no brand advantage here.



So what exactly is AMD sitting on here? For 99% of the people who purchase graphics cards in the $500 range, where exactly is the advantage in purchasing a HD7970 instead of a GTX680?

They honestly think their brand will be empowered by having a top-end graphics card that fails in price/performance ratio compared to the nVidia counterpart, while they lose in almost everything else that's brand-related?

To me, that's not brand enhancement, that's brand destruction.


I think Rangers is right, the only valid reason for the price not to come down is most likely the fact that the GTX680 is sold out everywhere so it hasn't affected the sales of the 7900 series, yet.
 
I just wish that AMD would release a card for the 5800 series users to upgrade from that's actually worthwhile, releasing the 7800 series at the same price/performance as the 6900 series doesn't do it since if we didn't upgrade to the 6900 series the 7800 series doesn't really offer any more performance and it's no cheaper.

So what exactly are 5800 series users like myself supposed to do? All I can say is if Nvidia release the 660/670 series cards and they are both faster and cheaper than the 7800 series then I can see AMD losing a lot of it's older fan base because they simply aren't competitive atm.
 
Oh come on.. AMD is sitting the performance/price disadvantage on their brand right now based on what?

- Brand placement? nVidia has it all with most AAA PC titles being TWIMTBP and having their logo appear in the beginning of each game. Is AMD changing this?

- Custom game enhancements? None that I've heard of. nVidia has all of those: PhysX, unlocked AA modes, GPGPU assisted effects like Just Cause 2. Is AMD changing this?

- GPGPU? nVidia is light-years ahead with almost every GPGPU-enabled software having a "main" CUDA path and then sometimes they go out of their way to enable an OpenCL path. Now that the GCN is a much more compute-friendly architecture, I have no doubts that this will change, but the current paradigm is: nVidia is better for computing. No brand advantages there, yet.

- Multi-monitor gaming? Yeah, Eyefinity was here first, but how many people exactly have such a setup? Now nVidia has it too for single-card usage, so expect all TWIMTBP games to feature this. Plus, 2.5 years after the release of the HD5000 series we're still waiting for custom resolutions. We'll see who comes up with that first, as it clearly seems to have been forgotten by AMD.

- 3D displays? nVidia is also miles ahead here. Sure, AMD stated they would follow the "open" way, which is nice and all, but when a game supports stereo 3D, which system does it use? 3D Vision.

- Driver quality? Hardly.. nVidia is a lot faster in releasing updated drivers when a new AAA game comes out. Probably related to the fact that their engineers partially develop the games themselves. I have no problems whatsoever with AMD drivers, but clearly they're not on top and clearly there's no brand advantage here.



So what exactly is AMD sitting on here? For 99% of the people who purchase graphics cards in the $500 range, where exactly is the advantage in purchasing a HD7970 instead of a GTX680?

They honestly think their brand will be empowered by having a top-end graphics card that fails in price/performance ratio compared to the nVidia counterpart, while they lose in almost everything else that's brand-related?

To me, that's not brand enhancement, that's brand destruction.


I think Rangers is right, the only valid reason for the price not to come down is most likely the fact that the GTX680 is sold out everywhere so it hasn't affected the sales of the 7900 series, yet.


Dont come with driver, i pass my time to fix problem on Nvidia card for my brother and friend, and go see on Nvidia driver threads (guru3Dfor example ) for see how it is in reality. they have wait 4months a WHQL, and nobody use it, as it buggy as hell. Go see how much are complaining about updated profile for their games who is still not there. Or problem who have still not be fixed yet. You think it is pinkland on Nvidia side ?

- For 3D gaming, im right with you. Even if yet there's only few monitors ( Asus and one Acer ), who support 3D vision 2.0. you better have to own them. And personally if i want buy a monitor, this will not thoses one.

- Multi- monitor, AMD have still a good advance, Nvidia is just trying to catch on them. And outside the custom resolution ( well you can set custom resolution since 12.2 Eyefnity 2.0 ), Nvidia is not even yet on same level with feature ( 5:1 system on a single card, 6x 2560x1600 monitors on single cards, 16kx16K resolution on a single card etc ) .., yes Nvidia have rework their copy, Bezel correction etc ...

- and you will use what for GPGPU ? a 680 who is slower in computing of a GTX480, and 5 to 10x slower of a HD7970 in computing ? (without saying, the shape is now to move from CUDA to OpenCL, computing professional dont like at all the homogenous CUDA computing, and now the shape is to go on Heteregenous computing hardware. http://www.hpcwire.com/hpcwire/2012-02-28/opencl_gains_ground_on_cuda.html
 
Last edited by a moderator:
Sorry, meant mixed resolutions and PLP. Mainly PLP.
IIRC we haven't talked about enabling PLP, mainly because it is not possible to do without specific hardware support for it under the mechanism that Eyefinity works (which is not the same as SoftTH).
 
So what exactly are 5800 series users like myself supposed to do? All I can say is if Nvidia release the 660/670 series cards and they are both faster and cheaper than the 7800 series then I can see AMD losing a lot of it's older fan base because they simply aren't competitive atm.

I don't think 5850/5870 owners have much worry about upgrading IMO, unless you're dealing with 1080P+ resolutions. Two weeks ago, I was using my sorely overclocked 5850 to run my Dell 2007WFP (1680x1050) and had almost NO issues in most games. However, with my impending upgrade to a Dell U2711 (2560x1440) I knew I was going to need some additional fillrate.

To be very honest, I was basically armed and ready to buy the GTX 680, but the only thing that really turns me off about it is that GPU boost clock stuff. It may be absolutely perfectly fine, but this is NVIDIA's first incarnation of that tech AFAIK, and I don't trust the first incarnation of something that can affect my overclocks that drastically.

So, 7970 it was. But only because I knew the 5850 wasn't gonna crank out the necessary oomph for my games at this rez. Otherwise? Screw it, the 5850 was doing just fine. It's being willed to my brother (along with my 2007WFP) to run in his Intel Q9400 + 6GB ram + Win7 Home Premium box. He'll replace the aging 4850 I willed him almost three years ago ;)
 
- and you will use what for GPGPU ? a 680 who is slower in computing of a GTX480, and 5 to 10x slower of a HD7970 in computing ?

Afaik, it's that much slower in DP only, which may not be all that important for consumer software.
Then again, you're right, the 680 is slower for computing but people will still buy a GeForce for computing.
That's the power of the brand. A luxury that, in computing, AMD can't afford yet.


IIRC we haven't talked about enabling PLP, mainly because it is not possible to do without specific hardware support for it under the mechanism that Eyefinity works (which is not the same as SoftTH).

I never said you promised it, I said we were waiting for it ;)

But knowing that it can't be done with current hardware is a bit of a let-down.
Isn't it possible to at least do a "trick" with a windowed mode, as an official feature?
 
Afaik, it's that much slower in DP only, which may not be all that important for consumer software.
Then again, you're right, the 680 is slower for computing but people will still buy a GeForce for computing.
That's the power of the brand. A luxury that, in computing, AMD can't afford yet.

People involved in commercial or hobbyist GPGPU work would surely be the least likely to be influenced by simple brand power.
They still may choose the product which from a purely hardware basis has less theoretical performance/$, but that would be due to factors such as the software stack, community and professional support, and a proven track record of not abandoning previous generation customers (these are just examples, not snide comments towards AMD or Nvidia).
 
- Brand placement? nVidia has it all with most AAA PC titles being TWIMTBP and having their logo appear in the beginning of each game. Is AMD changing this?

Perfect example of what will never change unless AMD can re-position the Radeon brand as a premium brand once again. That doesn't happen if they have to continually offer their cards as the "discount" alternative to Nvidia.

- Custom game enhancements? None that I've heard of. nVidia has all of those: PhysX, unlocked AA modes, GPGPU assisted effects like Just Cause 2. Is AMD changing this?

- GPGPU? nVidia is light-years ahead with almost every GPGPU-enabled software having a "main" CUDA path and then sometimes they go out of their way to enable an OpenCL path. Now that the GCN is a much more compute-friendly architecture, I have no doubts that this will change, but the current paradigm is: nVidia is better for computing. No brand advantages there, yet.

PhysX is debatable. I find it worthless. At best it's a cheap gimmick in games. At worst it's a cheap gimmick that drags down performance when enabled (on supported video cards) for minimal gains. I realize that other people feel differently, for them it probably is an advantage.

As for GPGPU, GTX 680 has taken a step both forwards and backwards at the same time. Sometimes faster than GTX 580 sometimes radically slower.

And when it comes to consumer gaming, CUDA is going to matter less and less as DXcompute matures. For prosumer markets, the GTX 680 has taken a step backwards. OpenCL may become relevant but I don't have high hopes for that. I expect DXcompute to mature far faster and be far more prevalent in the consumer gaming sector.

And there, the 79xx cards will likely have an advantage in some cases while the GTX 680 may have some advantages in some cases. I'd call it a potential wash at this point.

- Multi-monitor gaming? Yeah, Eyefinity was here first, but how many people exactly have such a setup? Now nVidia has it too for single-card usage, so expect all TWIMTBP games to feature this.

Nvidia have certainly caught up but they are still some distance behind in multi-monitor gaming. But certainly a niche, but a very important niche for cards costing 500 USD or more. As is the 2560x1600 gaming market. Both area's BTW where the 7970 closes the gap in performance and surpasses the GTX 680 in more cases.

- 3D displays? nVidia is also miles ahead here. Sure, AMD stated they would follow the "open" way, which is nice and all, but when a game supports stereo 3D, which system does it use? 3D Vision.

Again a wash here. For Nvidia you have to buy their 3D vision whatever mabobber thing. For AMD you have to buy software to force it in games that don't support it naturally.

For Nvidia users that want to use, say a 3D HDTV, they have to buy the same software that AMD users have to buy.

Speaking of 3D HDTV's. If a game natively supports AMD's HD3D (like Deus Ex: HR), then you don't even have to buy anything extra if you already have a 3D capable display.

Speaking of native support. All of the above will become less and less relevant as Microsoft starts to incorporate stereoscopic support into DirectX.

As to the quality of the experience. The last comprehensive test I saw that pitted Nvidia's 3D vision, versus AMD with HD3D and the 2 available software enablers shows a general wash. In some games Nvidia did better with 3D vision. In some games AMD was better with those software enablers.

- Driver quality? Hardly.. nVidia is a lot faster in releasing updated drivers when a new AAA game comes out. Probably related to the fact that their engineers partially develop the games themselves. I have no problems whatsoever with AMD drivers, but clearly they're not on top and clearly there's no brand advantage here.

Again, until AMD can once again price their cards as premium graphics cards, they won't be able to make enough profit to expand their driver team. If they even need to do that.

They've started being like Nvidia with releasing frequent BETA drivers to address problems in games. Nvidia are a bit faster at it, whether that's due to experience of doing it for such a long time or that their driver team is better funded, who knows. But AMD are certainly catching up fast in that area.

AND, they still maintain their WHQL'd monthly driver releases. And Nvidia still can't touch this. Like PhysX above, it's not a bonus for some people, but it is a big bonus for others.

So what exactly is AMD sitting on here? For 99% of the people who purchase graphics cards in the $500 range, where exactly is the advantage in purchasing a HD7970 instead of a GTX680?

A card that was in retail 2.5 months before the competition. A card that at low resolutions is only 10-15% slower than the competition at stock clocks. 5-8% slower at higher resolutions at stock on average.

And when both cards are overclocked (which is highly likely with people spending 500+ USD on a video card) it is roughly the same speed at low resolutions. But takes the lead at higher resolutions. And suddenly depending on your gaming environment the 7970 becomes the better value (well if they were priced the same).

So, yes, I can certainly see eventually dropping it to 499 USD to match the GTX 680. But until supply and demand for the GTX 680 start to even out, there is very little pressure on AMD to adjust the price of the 7970. AMD can certainly thank Nvidia for that gift.

Hence, premium price to rebuild mindshare as a premium brand.

BTW - just so we're clear, I don't like it personally. But then I'd love it if both companies could sell enthusiast class video cards for 300 USD or less or possibly as high as 350 USD. The reality is that neither company is going to make money if they do that.

And if AMD are forced out of the video card market due to having to price their cards differently than Nvidia despite having similar performance and swapping performance leads with them... Then things will rapidly get a whole lot worse for the consumer.

Regards,
SB
 
seriously wtf are you smoking, first your arguments dont actually make sense second you start changing the past. if you mean 2900/3800 its not anywhere near as back or white as you paint it ( shock horror there:devilish:).

Oh, what about your arguments?!? :LOL: Don't make it personal, because that one leads to only one thing- a ban. :D
No, I'm perfectly aware that 4870 for example has inferior performance even if compared to GTX 260.

See it yourself:

http://www.techpowerup.com/reviews/Palit/GeForce_GTX_260_Sonic_216_SP/29.html

Market demand will set pricing

Market demand will likely set something but until then it is the terribly fat obstinate ego of some people at the marketing department that is doing that. And it is true for a free, proper market, not for a duopoly market in which some guys (maybe loyal to AMD) are asking* the competition for a 50$ price increase of their product so not to have a direct competition. lol

so obviously the market is happy to pay the current prices for a 7970.

lol and that's why it's everywhere full of such topics like the current one.

Read the comments here:

http://www.techpowerup.com/163261/Radeon-HD-7970-Price-Cuts-Not-Any-Time-Soon-Report.html

It's not me who call them "money grabbing bastards".
Sorry.
 

Apples and oranges, 512 MB 4870 vs 896 MB 260.

It's not me who call them "money grabbing bastards".
Sorry.

No, it is all the other people who don't understand supply and demand.

There is supply of 7970s, every e-tailer has stock. AMD not cutting price means there is demand for the card. The supply can be the result of one of two things:
1. Supply constrained Nvidia. People shopping for a 680 give up and buy the next best thing
2. The 7970 perceived as being good value for money.

Either way, as long as AMD is shifting 7970s there is no reason to lower price. Note that all those bitching about AMD not slashing prices actually wants the card.

Cheers.
 
Oh, what about your arguments?!? :LOL: Don't make it personal, because that one leads to only one thing- a ban. :D
No, I'm perfectly aware that 4870 for example has inferior performance even if compared to GTX 260.

See it yourself:

http://www.techpowerup.com/reviews/Palit/GeForce_GTX_260_Sonic_216_SP/29.html
hardly personal, there is just this thing called reality and your way from it.

reality,
1. nvidia had to give a big hair cut and money back on the 280 once 4870 was released.
2. 260/280 were 570mm sq RV770 was 270mm sq
3. GT200 448/512 bit bus vs RV770 256
4. memory 898/1024 vs 512
5. everyone was guessing around 640 SP's not 880 ( RV770 even outdid itself there)

then lets look at review from the day.
http://www.guru3d.com/article/radeon-hd-4870-review--asus/9 onwards
http://www.anandtech.com/show/2556/13 onwards
http://www.tomshardware.com/reviews/radeon-hd-4870,1964.html ( note the review title "Radeon HD 4870: Better Than GTX 260! "

the reality is RV770 beat the 260 in raw performance both with and without AA until it became memory limited which was above 1080P res. In quite a few tests it equaled a 280.

RV770 was a pocket rocket that smashed GT200 in all metric's except absolute raw performance but it came close enough that Nvidia felt the need for big price reductions on the 280.



Market demand will likely set something but until then it is the terribly fat obstinate ego of some people at the marketing department that is doing that. And it is true for a free, proper market, not for a duopoly market in which some guys (maybe loyal to AMD) are asking* the competition for a 50$ price increase of their product so not to have a direct competition. lol

that makes no sense and again contradicts your own argument ( own goal to win the battle but to loose the war). The only way people will buy a product for more then what they think it is worth is if the brand has a value.

lol and that's why it's everywhere full of such topics like the current one.

Read the comments here:

http://www.techpowerup.com/163261/Radeon-HD-7970-Price-Cuts-Not-Any-Time-Soon-Report.html

It's not me who call them "money grabbing bastards".
Sorry.
http://xkcd.com/386/

wow someone on the internet having a whinge, who would have thunk it. The reality that you cant escape is that if AMD wasn't maximizing its profits at the current price they would adjust it to do so.
 
Last edited by a moderator:
I think AMD could talk to their partners, have them liquidate stock of the current OC models and then make a reference model with 1050~ (remember that the process will get better with time), and the new OC models start over with the 1100. In theory, anyway
That's every fanboy's wet dream (i still even count myself in that category). But in reality this kind of reshuffling inside DAMN marketing brainy-vision is exclusively reserved for their new series HD8000, which will be more or less rename of this old one, with few new chips to fulfill gap between current Pitcairn and Tahiti chips. And maybe, but highly unlikely, with some grander dies than Tahiti chips. TSMC 28nm is still young process which will stay here for years as they say and pretty economically unviable in short terms (for tsmc ofc). And HKMG @28nm offers a huge gains over last 40nm process which we saw as huge OC headroom nowadays.

I think you are right. I mean, it would be a double-edged sword for AMD to raise the performance of their HD7970 because it would noticeably lessen the HD8000's performance increase, and with the current OC models AMD could do it right.
Someone already explained to you very well. If they raise performance of HD7000 series, which is deliberately cutdown, then they'd need to reinvent the wheel in 12-15 month timeframe since Dec2011, when HD7950 was released.
GCN is young architecture which requires to be paid out.And if we look over to sorrow TSMC 28nm process node that AMD is well aware, then DAMN don't have anything to pull out except to overclock current architecture as much as possible in next generation HD8000. And possibly to squeeze 256-384 more SPs in Pitcairns successor and offer it as "budget deal" against top end TahitiXT (something akin RV570 vs. R580+) which probably could've been made on TSMC 25/24nm offering dumb shrink from current 28nm node (I believe this is possible and 28nm transistor cost wont raise at least not noticeably enough to make it a news just as in 90->80nm and 65->55nm shrinks)
So all we could look for in HD8000 is probably some nice tweaks (improvements) in gpu filtering and memory controller, and that huge overclocking potential to be heavily abused in renewed products.

In actuality, I think the price is right on the 7970. The 7970 RTSO of the 680 in GPGPU, and in this market it counts. The difference in performance isn't so high.
Non-OC HD7970 is below GTX680 in GPGPU performance, except in DP because GK104 is not meant to be a GPGPU part.

If you're a tech enthusiast who buy for gadget appeal, you'll question why the HD7870, using similar PCB and other components as the HD6870 costs twice as much. That would imply that AMD at least charges four times as much for the actual, same size GPU. If that is due to disastrous yields, then time will fix that. If it is due to wanting to increase their margins however, those who count themselves as technological cogniscenti will balk at paying a huge marginal increase in money for a small marginal increase in capabilities, mostly achieved by switching lithographic process. They're getting fleeced, and know it.

Well pointed out. DAMN's just become cocky as they won a huge deals with their DX11 tech in consoles segment. And they just wanna milk out their PC customers.
And beside all that no way 28nm is so much more disastrous than 40nm, TSMC 40nm was pretty much a hack when they start HD5800 production, and 28nm was much longer in development. And they can always switch to GloFo now if TSMC are charging them too much.

And above all Barts (HD6800) was ~270mm² chip iirc and Pitcairn (HD7800) is ~220mm². And PitcairnXT requires less power components, and could be accompanied with cheaper cooling, because it's less TDP chip than BartsXT, even when HD7800 is max OCed (w/o chip overvoltage)

And for DAMN it's TRADITION to have fleeces. They done that already sitting on first x86-64 uP (K8) and pretty much doing zero-research from 2002-2007 and then we have to wait just foour years to get poorely performing brat Bulldozer whwn they finally polished up K8 which become K10.5 (or whatever they wanna call it) and ofer nice performance ebven on some dumb shrik to 32nm. But they simply canceled it because these days nobody needs some older product to outcompete their spoiled bratty brain-child of theirs.
 
I think Rangers is right, the only valid reason for the price not to come down is most likely the fact that the GTX680 is sold out everywhere so it hasn't affected the sales of the 7900 series, yet.

I know a guy in retail, he says he can't keep either of the cards in stock. The ball is totally in TSMCs court on this one. We have no clue how long the supply situation is going to stay like this -- 40nm was bad, and both manufacturers went for it in volume much *later* than they did to 28nm. 4770 was out in April, 5 months before Evergreen. And the supply was tight through to end of January. This time there were no such pathfinder chips -- if the supply situation is similar this time around, AMD will probably feel no pricing pressure until late August.

Of course, 40nm was abnormally and ahistorically bad. Then again, everyone is saying how every node from here on out will be harder than the previous one.

I think 680 is, in this market, underpriced.
 
Non-OC HD7970 is below GTX680 in GPGPU performance, except in DP because GK104 is not meant to be a GPGPU part

Sorry? Not sure that DPFP is actually exposed in anything yet, at least it currently isn't for Tahiti, so there haven't been performance tests. The Compute performance comments are coming from the fact that the performance of GTX 680 not just lower than Tahiti, but also lower than GTX 580 in a number of cases. More has been removed / reduced on GK104 that reduces its compute performance than just the DPFP rates.
 
Sorry? Not sure that DPFP is actually exposed in anything yet, at least it currently isn't for Tahiti, so there haven't been performance tests. The Compute performance comments are coming from the fact that the performance of GTX 680 not just lower than Tahiti, but also lower than GTX 580 in a number of cases. More has been removed / reduced on GK104 that reduces its compute performance than just the DPFP rates.
OpenCL exposes DPFP for HD7970. See SiSoft Sandra results, for example.
 
OpenCL exposes DPFP for HD7970. See SiSoft Sandra results, for example.

Ehh, I was fine using doubles on Tahiti through DirectCompute/C++ AMP too so...what exactly do we mean by DPFP not being exposed? Or does the acronym stand for something other than I think?
 
Back
Top