AMD Radeon HD 7970 Review: 28nm And Graphics Core Next, Together As One

[SKYMTL]

Since the 2 last titleses, Batman have been developped in collaboration with Nvidia.... the graphic engine have been developped ( in collaboration ) with Nvidia ( and it is a success ) the bench inside the game is really the worst thing you want for a review

Thoses games are clearly cherry picked, but shogun2, who is certainly the most demanding game today, gcn show his potential.

Wait, you're saying that Batman was developed with the help of NVIDIA. And exactly who helped develop Shogun 2? Who do you think helped develop the built in benchmark in Shogun? Check the second loading screen....last I checked, it said Gaming Evolved. Deus Ex? Same thing. So of course AMD pulls FAR ahead.

[LiquidReactor]

Quite dissapointing.

Yes, very, for a brand new card thats priced at $550, i expected more. If this was priced at $399 or even $449 it would be more appealing. Im sure once nVidia comes out with the new stuff, prices will be lowrered to around 450 for the 7970.

With all due respect, you guys must have mental issues. HD 7970 is competing with 6990 and 590 and its a single die card. AMD gave us a new single gpu powerhouse and you guys are complaining?

[kuroikenshi]

^^ this!

[Tao~]

GJ, AMD. I was surprised to read the first comments on review-sites and on this page labeling this as disappointing. I think people fail to realize the fact that this is a working card with a new arch , highest single card perf. , better-than-class thermals/power on a new manufacturing process that is leaky as hell. I mean - just considering the fact that they got 4 billion transistors working there perfectly in Dec 2011 is amazing. Specially when I remember Fermi.

Those whining over the price tag should realize that GTX 580 3GB costs almost the same and is inferior in every possible way - unless you are using it to heat your room up.

[Klarko]

Thing that impressed me the most was the power saving options personally.

[HelixPC]

With all due respect, you guys must have mental issues. HD 7970 is competing with 6990 and 590 and its a single die card. AMD gave us a new single gpu powerhouse and you guys are complaining?

It is a powerful card, BUT judging from the MIN fps, not AVG fps, its only 5% faster than the GTX 580, which is on Fermi tech, finalized in 2009. I believe the price should be much lower, and the only reason AMD is price gauging is because there is no competition at this moment in time. Things will change around march/april time.

[Lanek]

Wait, you're saying that Batman was developed with the help of NVIDIA. And exactly who helped develop Shogun 2? Who do you think helped develop the built in benchmark in Shogun? Check the second loading screen....last I checked, it said Gaming Evolved. Deus Ex? Same thing. So of course AMD pulls FAR ahead.

He this was not a critic .... dont take it like that... But we know ( as for Shogun ) Nvidia clearly work with drivers at the same time the game is developped with Batman .. This said it is too the same for F12011 and Shogun2 on AMD side.

Its why today, it is really hard to find a good panel of games for review a gpu...

It is a powerful card, BUT judging from the MIN fps, not AVG fps, its only 5% faster than the GTX 580, which is on Fermi tech, finalized in 2009. I believe the price should be much lower, and the only reason AMD is price gauging is because there is no competition at this moment in time. Things will change around march/april time.

Have you seen the minimum fps in Metro, F1 2011 and Shogun2 ? ( ofc they are AMD evolved games for the 2 last, so crazy optimised ), but the minimum is faster of 580SLI and this is at 5000+ resolution, not 1920x1080... I can understand some result are a bit disappointing ( BF3 and Skyrim are good example and widely used ), but if you look too on the other side, this thing have some crazy horsepower, it just need sometimes for deliver where it dont right now.

Dont shoot the messenger, but wait to see the performance applied to other games, the codes who is running by GCN is completely different of VLIW4 or 5... the tweaks of VLIW4 dont apply with it... im sure we will see a huge increase in performance in the next months... If not in all games, in some.

I dont know where you have seen the GTX580 have been finalised in 2009, the 480 was allready released in march 2010 .. ( not important anyway )

Now we can compare with Nvidia one, but it is too really important to compare with 5870-6970 ... And only the computing side.. this is an incredible step forward.

(damn christmas, im allready drunk )

[kuroikenshi]

It is a powerful card, BUT judging from the MIN fps, not AVG fps, its only 5% faster than the GTX 580, which is on Fermi tech, finalized in 2009. I believe the price should be much lower, and the only reason AMD is price gauging is because there is no competition at this moment in time. Things will change around march/april time.

Nope, it should be sooner, HD7990 comes out earlier than that...

[zalbard]

Dont shoot the messenger, but wait to see the performance applied to other games, the codes who is running by GCN is completely different of VLIW4 or 5... the tweaks of VLIW4 dont apply with it... im sure we will see a huge increase in performance in the next months... If not in all games, in some.

People were saying the same about Cayman, that going from VLIW5 to VLIW4 requires driver optimisations and that there would be performance improvements down the road. I am still waiting for these.
Don't make the same mistake twice. Remember, competition works on its drivers as well.

[Lanek]

People were saying the same about Cayman, that going from VLIW5 to VLIW4 requires driver optimisations and that there would be performance improvements down the road. I am still waiting for these.
Don't make the same mistake twice. Remember, competition works on its drivers as well.

Open the June Anandtech article about GCN.... and watch the comparaison between old VLIW architecutre and the new Vertex + scalar architecture code ( code resulting was allready shown) .... you will see what i mean by a big difference... VLIW4 and 5 was not so different.
basically, in VLIW4, the last fixed function who was only used in some cases, have been removed, and this function added in the 4 Vertex ( ALU ), it was something like a compromise with VLIW4.... but the main architecture was finally extremely identic in the code .... ( sorry for my english, i hope it is readable )

In my work, a different language need a different approach. I believe this is the same here ..

maybe im wrong, i hope not... but maybe im wrong.

[zalbard]

Open the June Anandtech article about GCN.... and watch the comparaison between old VLIW architecutre and the new Vertex + scalar architecture code ( code resulting was allready shown) .... you will see what i mean by a big difference... VLIW4 and 5 was not so different.
basically, in VLIW4, the last fixed function who was only used in some cases, have been removed, and this function added in the 4 Vertex ( ALU ), it was something like a compromise with VLIW4.... but the main architecture was finally extremely identic in the code .... ( sorry for my english, i hope it is readable )

I read it. Perhaps you should do it again, too.
The article is about compute capabilities of CGN. There is no doubt that there are many changes to computing between Cayman and CGN. For gaming, however, the effect of optimisations is uncertain. While there is a good chance that drivers may improve gaming performance, I wouldn't preach big gains from future driver releases. They may as well be no different from those gained by Nvidia (read: just optimisations for poorly coded future titles).

[madcho]

I read it. Perhaps you should do it again, too.
The article is about compute capabilities of CGN. There is no doubt that there are many changes to computing between Cayman and CGN. For gaming, however, the effect of optimisations is uncertain. While there is a good chance that drivers may improve gaming performance, I wouldn't preach big gains from future driver releases. They may as well be no different from those gained by Nvidia (read: just optimisations for poorly coded future titles).

The drivers will improve shader to GCN, for old games too, when a game run, it run using VLW4 optimisations shaders. It could be more faster with nvidia's one. And the guy from hardware.fr tridam think the same, so saying there is room from improvement from drivers, and games, is not a mistake.

[XSAlliN]

It's kinda weak for that price - at that performance 2 years from now won't worth more than 150 $.

[spursindonesia]

I think people has the right to be yapping about AMD's pricing policy if NVidia makes a swift counter move with their own price cut, say GTX 580 down to US$ 399. They have not so far, so where's the rationality of the complaint ???

I actually don't really mind with people bashing AMD for their perceived bad driver -it has some merit, though mostly all related to NVidia's "better" relation with game developer. But bashing their pricing policy while it has a lot of merit in it ? I'm not sure we all live in the same universe.

[Kallenator]

Great iteration by AMD, nothing to complain about!

[SKYMTL]

There are a few things that need to be set straight with the ravings of both sides.

- Southern Islands IS a huge step in the right direction for AMD. The architecture is efficient, performs well and is massively impressive regardless of what NVIDIA fans say while burying their heads in the sand.

- Here's a shocker for you: AMD is behind the technology curve. Sure Tahiti is able to pull ~20-25% (on average) ahead of the GTX 580 but we have to remember that we're comparing it to a PREVIOUS generation Fermi architecture, not a comparable next generation Kepler part. When you look at it in that context, AMD may still be fighting one hell of an uphill battle in 2012 and need to get product in the channel....NOW.

- I know the above statement sounds like absolute insanity to some around here but it's important to remember that Southern Islands was originally supposed to be introduced on 32nm before the process was cancelled. Think of it as a Tick (New Architecture) to NVIDIA's Tock (Fermi v2) and that would have put Tahiti in a great position. Instead, AMD had to release "Northern Islands" at a low price AFTER NVIDIA's refresh and push their new architecture back to a time when it would be competing in a less dominant position against Kepler.

- Additional performance through "optimizations" for GCN? Quite possibly but I really don't think so since the goal sticks for driver dev are always in movement. Plus, while there may be increases in today's games, will you be playing Crysis 2 6 months from now? How about MW3? Arkham City? No, probably not. You'll be on to the latest and greatest game and while AMD may be upping performance on the games seen in yesterday's reviews, time has proven that it is their support for NEW games that continues to lag behind. So "performance improvements through drivers" be damned (IMO) until they can show proper and immediate support for new releases. Remember, NVIDIA's driver dev. is hardly standing still either so once everything evens out, you won't see any additional separation between the GTX 580 and Tahiti-based cards in upcoming games. And you can quote me on that.

- We can't only think of Tahiti from a purely gaming performance standpoint. AMD has built in enough appealing features like ZeroCore Power that there are more than enough reasons to consider a HD 7970 over any other card currently on the market. From a personal perspective, the HD 7970's power consumption alone has caused me to immediately replace a GTX 580 3GB in my main gaming system. It may be winter up here in Canada but even a few bucks every month saved on my power bill will add up.

- The HD 7970 is priced lower than most GTX 580 3GB cards and slightly above reference GTX 580s. It consistently outperforms both. The fact that people can continue to complain about pricing simply boggles my mind.

[FischOderAal]

Thank you SKYMTL, your posts are much appreciated. You always have a fair amount of criticism for both parties.

[cold2010]

^ this

[ajaidev]

There are a few things that need to be set straight with the ravings of both sides.

- I know the above statement sounds like absolute insanity to some around here but it's important to remember that Southern Islands was originally supposed to be introduced on 32nm before the process was cancelled. Think of it as a Tick (New Architecture) to NVIDIA's Tock (Fermi v2) and that would have put Tahiti in a great position. Instead, AMD had to release "Northern Islands" at a low price AFTER NVIDIA's refresh and push their new architecture back to a time when it would be competing in a less dominant position against Kepler.

- Additional performance through "optimizations" for GCN? Quite possibly but I really don't think so since the goal sticks for driver dev are always in movement. Plus, while there may be increases in today's games, will you be playing Crysis 2 6 months from now? How about MW3? Arkham City? No, probably not. You'll be on to the latest and greatest game and while AMD may be upping performance on the games seen in yesterday's reviews, time has proven that it is their support for NEW games that continues to lag behind. So "performance improvements through drivers" be damned (IMO) until they can show proper and immediate support for new releases. Remember, NVIDIA's driver dev. is hardly standing still either so once everything evens out, you won't see any additional separation between the GTX 580 and Tahiti-based cards in upcoming games. And you can quote me on that.

heheh if you mean to use the TICK-TOCK from Intel POV then Southern Islands would be a TOCK and Fermi a TICK.

The reason why Southern Islands was rescheduled is know to all this did not only impact AMD but also Nvidia thus future plans would have suffered. Game developers are not keen to work with AMD it maybe the fact that AMD just does not care as much as Nvidia does "It took AMD 6 days to respond to some queries from a developer and Nvidia did the same in a few hrs". The future games will no doubt take into account GCN's design and be made accordingly its not like the developer does not do anything for GPU optimizations. AMD needs to develop better relationships with developers and increase their S/W engg percentage.

Intel has very good relationship with developers not as good as Nvidia but we are very supportive. People may not know this but as of now there are very few people in the S/W side for IGP's. We were suppose to get transferred to a new graphics wing but then things got messy with LRB dont know when things will get on track. Speaking off LRB it has a very good future things are working in its favor with the whole GPU market "This is a hint" things are still on the down low but there are mad men at work on something very special.

[SKYMTL]

heheh if you mean to use the TICK-TOCK from Intel POV then Southern Islands would be a TOCK and Fermi a TICK.

I'm not going by manufacturing processes but rather architectures / DirectX product generations and the revisions thereof. Fermi v2 (GTX 580) would be considered a "Tock" to the GTX 480's "Tick"....in the GPU world that is. Kepler = Tick

HD 3800 = Tick, HD 4800 = Tock

HD 5800 = Tick, HD 6900 = Tock

Southern Islands = Tick even though the "Tock" after Cypress wasn't supposed to happen.

The reason why Southern Islands was rescheduled is know to all this did not only impact AMD but also Nvidia thus future plans would have suffered. Game developers are not keen to work with AMD it maybe the fact that AMD just does not care as much as Nvidia does "It took AMD 6 days to respond to some queries from a developer and Nvidia did the same in a few hrs". The future games will no doubt take into account GCN's design and be made accordingly its not like the developer does not do anything for GPU optimizations. AMD needs to develop better relationships with developers and increase their S/W engg percentage.

Intel has very good relationship with developers not as good as Nvidia but we are very supportive. People may not know this but as of now there are very few people in the S/W side for IGP's. We were suppose to get transferred to a new graphics wing but then things got messy with LRB dont know when things will get on track. Speaking off LRB it has a very good future things are working in its favor with the whole GPU market "This is a hint" things are still on the down low but there are mad men at work on something very special.

IMO, that's all hearsay. I'd think AMD has perfectly fine relationships with developers but their Gaming Evolved program (a parallel to NVIDIA's TWIMTBP) lacked resources up until 2011.

[solofly]

I'm just glad NV is no longer on top, that way I'll get my Kepler sooner...

[Tao~]

@ SKYMTL - You have put some nice points, and while I agree with most of them I also have something else to say -

AMD is behind the technology curve. Sure Tahiti is able to pull ~20-25% (on average) ahead of the GTX 580 but we have to remember that we're comparing it to a PREVIOUS generation Fermi architecture, not a comparable next generation Kepler part. When you look at it in that context, AMD may still be fighting one hell of an uphill battle in 2012 and need to get product in the channel....NOW.

IMHO, that doesn't matter. What matters is Tahiti is shipping and Kepler doesn't even have a launch date. And even when Kepler launches can it be more efficient - that is, will it be able to deliver a better performance-to-cost of manufacturing ratio , because that is what matters to the company. For AMD, the time before high-end Kepler becomes available is the time when they dictate the prices. What I'm trying to say is, AMD build a smaller ship, earn higher profit for six months before Nvidia launch next-gen and then price drop to the actual target price of the chip, while Nvidia earn lower profit in the initial period because of lagging performance with previous-gen to target profit in the next 6 months because they have a bigger chip. AMD effectively is undercutting Nvidia by 6 months.

- Additional performance through "optimizations" for GCN? Quite possibly but I really don't think so since the goal sticks for driver dev are always in movement. Plus, while there may be increases in today's games, will you be playing Crysis 2 6 months from now? How about MW3? Arkham City? No, probably not. You'll be on to the latest and greatest game and while AMD may be upping performance on the games seen in yesterday's reviews, time has proven that it is their support for NEW games that continues to lag behind. So "performance improvements through drivers" be damned (IMO) until they can show proper and immediate support for new releases. Remember, NVIDIA's driver dev. is hardly standing still either so once everything evens out, you won't see any additional separation between the GTX 580 and Tahiti-based cards in upcoming games. And you can quote me on that.

You are right in that performance of newer games is what catches the eye - but I feel that newer drivers will be more focused on improvements to GCN than older architectures - so while VLIW4 chips gain 5% in the next year on some "new" title from Catalyst 12.1 to 12.6, GCN will gain 10%. How it compares to Nvidia is depends on how similar Kepler will be to Fermi architecturally. Older tech falls out of favor very fast is something I have learnt from experience.

[iddqd]

There's no evidence to suggest that Kepler isn't simply a bugfixed and pimped out version of Fermi, anyway - gpu vendors love riding the exact same architecture for several generations.

[Iconyu]

- Additional performance through "optimizations" for GCN? Quite possibly but I really don't think so since the goal sticks for driver dev are always in movement. Plus, while there may be increases in today's games, will you be playing Crysis 2 6 months from now? How about MW3? Arkham City? No, probably not. You'll be on to the latest and greatest game and while AMD may be upping performance on the games seen in yesterday's reviews, time has proven that it is their support for NEW games that continues to lag behind. So "performance improvements through drivers" be damned (IMO) until they can show proper and immediate support for new releases. Remember, NVIDIA's driver dev. is hardly standing still either so once everything evens out, you won't see any additional separation between the GTX 580 and Tahiti-based cards in upcoming games. And you can quote me on that.

One thing I might have misread is that AMD have made their architecture less reliant on compiler tweaks to get great shader performance. With VLIW4/5 AMD pretty much had to see the game and then tweak the drivers for each title, which was one reason their support for new games was flaky. So if I did read correctly the card can handle unoptimised shaders better, would that leave AMD more time to focus on bugs?

[AliG]

One thing I might have misread is that AMD have made their architecture less reliant on compiler tweaks to get great shader performance. With VLIW4/5 AMD pretty much had to see the game and then tweak the drivers for each title, which was one reason their support for new games was flaky. So if I did read correctly the card can handle unoptimised shaders better, would that leave AMD more time to focus on bugs?

If that's the case, then that's really good news for those of us that are concerned about driver stability. However I think that also means we won't be seeing any big performance improvements over time either