AMD Radeon 7000-Series 28nm (Southern Islands) | 7990 7970 7870 7770 | Discussion

[boxleitnerb]

i'll have a preview up tonight

i was referring to the next AMD GPU after 7970 - the one that will probably take on the GTX 680 in Q1
^_^

Interesting. So 7970 with a handbrake now and a (single chip) 79"80" later when needed (probably higher clocks)?

 

[RussianSensation]

20%to 30% faster than the gtx 580 is way out of the 6970's league.

In the context of a $500 GTX580, it looks good, but in the context of a net generation, it's disappointing imho. How can you have a card manufactured on a full node down (ie.,28nm) that is only 25-30% faster than a 1-year-old GTX580, considering GTX580 can easily overclock 20% on 40nm....

AMD said new gen every ~12mos.

Do you have as source for this? AMD is already more than 2 years late with introducing a new generation.

 

[Lonbjerg]

That's not short.

Starcraft 2, download a replay with lots of action. Load replay, win.

You can probably remove older games like Far Cry 2 and a few others off your list to make room for newer games. Id hesitate to remove Valve games since so many people play their games on the same engine. But stuff like LP2 and Hawx2, do ppl even play that these days in any meaningful numbers?

that would not be real world performance...and thus useless?

 

[Silverforce11]

that would not be real world performance...and thus useless?

Replays hammer GPUs fine when 200 vs 200 armies fight. It's perfectly indicative of gameplay performance when you play SC2 online, especially so in 4v4 matches.

 

[apoppin]

That's not short.

Starcraft 2, download a replay with lots of action. Load replay, win.

You can probably remove older games like Far Cry 2 and a few others off your list to make room for newer games. Id hesitate to remove Valve games since so many people play their games on the same engine. But stuff like LP2 and Hawx2, do ppl even play that these days in any meaningful numbers?

How many people still play the original Crysis? Or Crysis 2 for that matter. Metro 2033 hasn't sold many copies this year yet it offers awesome visuals and there will be a sequel. ANY game months after it launches wanes quickly in popularity. Gamers always move onto the next game. Yet everyone also plays niche games and games they like and some prefer bargain bin and older games with maxed out IQ.

If i wanted to do "popular", i'd bench MW3. But it isn't demanding like BF3 is.

The idea of benching many games is to give an idea of a card's overall ability to handle many kind of games and game engines. If i only bench the very latest games, we will only know how a certain card handles them - and there are tech sites that specialize in just new games - maybe 5 or 6 total. Do you think that is representative of a card's overall abilities?

There will be a Far Cry 3 probably based on that same Dunia engine. FC2 has very nice visuals and one of the very best created benchmarks anywhere.

My *goal* is to have more and better benches. Unfortunately, they take a lot of time to create and time is of the essence; especially when i am backed up with *committed* reviews until CES. Then it starts all over again.

apoppin, quit posting and get that preview of yours up.

OK, break's over here - i'm back to work. Looks like i'm writing until early tomorrow AM, now
:whiste:

 

[badb0y]

How many people still play the original Crysis? Or Crysis 2 for that matter. Metro 2033 hasn't sold many copies this year yet it offers awesome visuals and there will be a sequel. ANY game months after it launches wanes quickly in popularity. Gamers always move onto the next game. Yet everyone also plays niche games and games they like and some prefer bargain bin and older games with maxed out IQ.

If i wanted to do "popular", i'd bench MW3. But it isn't demanding like BF3 is.

The idea of benching many games is to give an idea of a card's overall ability to handle many kind of games and game engines. If i only bench the very latest games, we will only know how a certain card handles them - and there are tech sites that specialize in just new games - maybe 5 or 6 total. Do you think that is representative of a card's overall abilities?

There will be a Far Cry 3 probably based on that same Dunia engine. FC2 has very nice visuals and one of the very best created benchmarks anywhere.

My *goal* is to have more and better benches. Unfortunately, they take a lot of time to create and time is of the essence; especially when i am backed up with *committed* reviews until CES. Then it starts all over again.


OK, break's over here - i'm back to work. Looks like i'm writing until early tomorrow AM, now
:whiste:

apoppin!!!!! Eagerly awaiting the preview.

 

[Silverforce11]

Btw, here's that BF3 leak:


I like the 3x1080p res with MSAA result.
Would probably be the same at 1600p with 4xMSAA. Vram wall.

Edit: Below, BF3 64MP is very demanding on hardware. Don't bother benching the SP.

 

[RussianSensation]

Here is my shortened current benching list; in bold is brand-new:

• Crysis
• Metro 2033
• Civilization V
• Crysis 2
• Total War: Shogun II
• BF3
• Dragon Age 2
• Witcher 2
• Anno 2070

If you are strapped on time and want to do a quick preview using the most demanding games, and then add the rest of the games to get an overall picture of performance . The games above are the only games that actually stress modern GPUs. The other games you listed are a walk in the park for the likes of HD6970/GTX580. I understand the logic of testing a game like FC2, but you are just going to be getting some high frames like 85-100...


Dragon Age 2, for example, is WAY more demanding than STALKER: COP. There is even less merit to test COP since STALKER 2 is cancelled for now. The reason I left Civ5 in there is to see if AMD incorporated multi-threading in their DX11 driver for HD7970. If you are going to be testing Serious Sam for example, might as well test the latest Serious Sam 3.

 

[3DVagabond]

In the context of a $500 GTX580, it looks good, but in the context of a net generation, it's disappointing imho. How can you have a card manufactured on a full node down (ie.,28nm) that is only 25-30% faster than a 1-year-old GTX580, considering GTX580 can easily overclock 20% on 40nm....



Do you have as source for this? AMD is already more than 2 years late with introducing a new generation.

They released the 6900 last year, That was their next gen part after evergreen. It was stuck @ 40nm and was scaled back.

Sorry, I don't have the quote at my fingertips. I'm surely not going back to over a year ago looking through AMD press releases. It was out there publicly, not some leak. It was stated by the then CEO.

 

[badb0y]

Btw, here's that BF3 leak:

I like the 3x1080p res with MSAA result.

Is this reality?

 

[Silverforce11]

You can have an educated guestimate performance based on the architecture and press slides. 2048 1D shader is efficient, no wastage like in VLIW. Thus, 2048 1D shaders perform onpar with ~2560 VLIW shaders.

Or to put it in perspective vs Cayman, its 66% more effective with just SP alone, minus clock speed or any other improvements.

The only negative is the 32 ROPs. If they didn't buff it, its going to be a major bottleneck. So performance could range from utter crap (think BD-esque) to awesome sauce, depending on how good their drivers are and if their ROPs are vastly improved or not. With such a big variance, your guess is as good as anyone else who doesn't currently have a sample card in hand.

 

[RussianSensation]

Btw, here's that BF3 leak:

Internally created slides from AMD or NV showing their own products should be taken with a huge grain of salt. Remember how HD6970 was 10-20% faster than GTX480? vs. reality.

They released the 6900 last year, That was their next gen part after evergreen. It was stuck @ 40nm and was scaled back.

HD6xxx series is not a new generation. For all intents and purposes it's a refresh with a slightly revised internals like X1900 series was to X1800 series or 7800GTX 256mb --> 7900GTX. HD6970 and GTX580 were each about 15% faster than HD5870 and GTX480. Historically speaking, HD6970 should have been called HD5890 and GTX580 should have been GTX485. That's all there is to it. That makes them refreshes, no matter how much NV or AMD want to spin it. Also, HD6970 only matched the speed of a GTX480, which automatically means it's from the same generation. HD4890 was also 15% faster than HD4870, does that make HD4890 a new generation?

You can have an educated guestimate performance based on the architecture and press slides. 2048 1D shader is efficient, no wastage like in VLIW. Thus, 2048 1D shaders perform onpar with ~2560 VLIW shaders.

Or to put it in perspective vs Cayman, its 66% more effective with just SP alone, minus clock speed or any other improvements.

If your observation is corrent, then this slide showing HD7970 to be 40-60% faster than GTX580 would be more accurate if 2048 1D ~ 2560 VLIW. This is in contrast to the 20-30% rumors we've been hearing. Obviously, again provided by AMD...so take it with a grain of salt.

Source

 

[Silverforce11]

reality

Depends on which reality. It's ~10% faster than Gtx480.

Edit: The 20-30% rumor is from Fud, who clearly stated 30% faster than 6970 and near 6990 levels. FUD.

 

[AtenRa]

380mm2 is not correct.

Performance scaling vs 6970/580 will be higher at 2560x1600 and Eyefinity

DX-11 games will see a bigger performance scaling against 6970.

I dont think we will see reviews on the 22 Dec.

In VLIW optimized games, 2048 ALUs almost equal to 2048 VLIW SPs

 

[Idontcare]

I dont think we will see reviews on the 22 Dec.

Honestly, for the love of humanity I hope to hell we don't see reviews anywhere that close to Christmas or any other major holiday.

Reviewers are people too, they have family and friends that they need to have the time to be enjoying this close to Christmas.

Would be a depressing thing to do to a reviewer to put them on a review timeline that hogs up all their family time while they try and crunch out a review right before xmas

 

[RussianSensation]

Honestly, for the love of humanity I hope to hell we don't see reviews anywhere that close to Christmas or any other major holiday.

Reviewers are people too, they have family and friends that they need to have the time to be enjoying this close to Christmas.

Would be a depressing thing to do to a reviewer to put them on a review timeline that hogs up all their family time while they try and crunch out a review right before xmas

Even worse if AMD won't actually have cards for sale until January 9th. Why rush reviewers for a paper launch (just to please investors I suppose)? That means all those prayers to Santa will go unnoticed for a lot people. :biggrin:

 

[RobertR1]

Honestly, for the love of humanity I hope to hell we don't see reviews anywhere that close to Christmas or any other major holiday.

Reviewers are people too, they have family and friends that they need to have the time to be enjoying this close to Christmas.

Would be a depressing thing to do to a reviewer to put them on a review timeline that hogs up all their family time while they try and crunch out a review right before xmas

I agree with all that!

but...

....I still want to see the numbers on the 22nd

 

[Nintendesert]

It really isn't. He's a respectable hardware reviewer and probably has the largest benchmark suite amongst all GPU reviewers. He is not at all biased in any of his reviews and, if anything, always puts a positive spin in review summaries even if products end up being disappointing. I have never once seem him post fud about anything and claim it to be straight up fact.

So you say. That thread however is no different than what we've been doing here. So why quote it as a source of anything concrete? I want someone that actually has a card to give us some real numbers. I saw how he defended Bulldozer with some horrible logic, forgive me if I don't take his speculation as gospel. I simply want real info to decide on a 7970 or keep waiting for Kepler.

 

[VulgarDisplay]

I know it's not real info, but I myself don't see kepler seeing the light of day until December 2012. I'd be perfectly happy buying a 7970 now and having a card that is going to dominate 1920x1080 for the next 2 years rather than waiting for a card 11 months away. I'm also going to just assume that used 7970's will still be selling for decent amounts once kepler does come out so I see no harm in jumping on the GCN train early.

 

[Will Robinson]

"Waiting for Kepler" seems to be the new mantra for some here but what if its a dud?
AMD showed NVDA the way with the HD4000,HD5000 series cards while HD6990 still holds the open class crown.
NVDA almost screwed the pooch with Fermi 480.....perhaps we will see another big,hot chip.

 

[3DVagabond]

HD6xxx series is not a new generation. For all intents and purposes it's a refresh with a slightly revised internals like X1900 series was to X1800 series or 7800GTX 256mb --> 7900GTX. HD6970 and GTX580 were each about 15% faster than HD5870 and GTX480. Historically speaking, HD6970 should have been called HD5890 and GTX580 should have been GTX485. That's all there is to it. That makes them refreshes, no matter how much NV or AMD want to spin it. Also, HD6970 only matched the speed of a GTX480, which automatically means it's from the same generation. HD4890 was also 15% faster than HD4870, does that make HD4890 a new generation?

Sometimes I wonder if 2 people use your acct.?

You can't seriously be comparing 5800->6900 being the same as 4870->4890. As far as 480->580 goes, I posted that nVidia was the one on a 2yr time table for new arch's.

 

[Mopetar]

Would be a depressing thing to do to a reviewer to put them on a review timeline that hogs up all their family time while they try and crunch out a review right before xmas

If the launch is going to come shortly after the announcement, there's no reason why AMD couldn't have put review hardware out weeks ago. Just because the date is the 22nd doesn't mean that the cards are just arriving today and that everyone has to be in a mad rush. Ideally the review has been done for a week now and everyone's just waiting for the publication deadline to lift.

 

[RussianSensation]

Sometimes I wonder if 2 people use your acct.?

You can't seriously be comparing 5800->6900 being the same as 4870->4890. As far as 480->580 goes, I posted that nVidia was the one on a 2yr time table for new arch's.

Of course HD4890 was just a bump in clocks, but you surely don't believe HD6900 is a "new generation" for just a 15% performance increase? I suppose it depends how you define a new generation. Tweaking VLIW-5 into VLIW-4 is just a minor adjustment imho. To me, it's about performance, not about architecture. For example, when NV increased the specs on G80 architecture from 8800GTX to GTX280, that's still pretty much the same architecture with minor tweaks, but constitutes a new generation. Under similar terms, AMD's architecture was larged unchanged from 9700Pro --> X1950XTX, and similarly from 2900XT --> HD6970 (for the most part). In each of those periods, the performance increase was dramatic and yet the architecture was much unchanged. So I tend to view "real" generations based on relative performance, not their underlying architecture to be honest.

Based on performance, I don't agree that GTX580 and HD6970 warranted a change in the first number, which normally indicates a generational leap. I don't consider a 15% performance increase a "generational leap".

Let's hope HD7970 comes in closer to 40-50% faster than GTX580, rather than the 25-30% rumor.

 

[Grooveriding]

Of course HD4890 was just a bump in clocks, but you surely don't believe HD6900 is a "new generation" for just a 15% performance increase? I suppose it depends how you define a new generation. Tweaking VLIW-5 into VLIW-4 is just a minor adjustment imho. To me, it's about performance, not about architecture. For example, when NV increased the specs on G80 architecture from 8800GTX to GTX280, that's still pretty much the same architecture with minor tweaks, but constitutes a new generation. So I tend to view "real" generational jumps based on relative performance to be honest.

Based on performance, I don't agree that GTX580 and HD6970 warranted a change in the first number.

Let's hope HD7970 comes in closer to 40-50% faster than GTX580, rather than the 25-30% rumor.

Performance wise they didn't at all. I think his argument is that the 580 and 480 are almost exactly the same under the hood except for some tweaks. Whereas the 5870 and 6970 have some real differences.

Unimportant distinction though. I also only care about performance from one card to the next and could give a toss what is happening under the cooling shroud to give it to me. :awe:

 

[Grooveriding]

If the launch is going to come shortly after the announcement, there's no reason why AMD couldn't have put review hardware out weeks ago. Just because the date is the 22nd doesn't mean that the cards are just arriving today and that everyone has to be in a mad rush. Ideally the review has been done for a week now and everyone's just waiting for the publication deadline to lift.

This is my hope, that sites like anand, hwc and techpowerup have had the cards for a week or so and we get benches this Thursday.

Otherwise we're just going to get the slides we've all seen officially released and a press statement that they are launching the cards for sale on Jan 9th.