Radeon HD 7000 Revealed: AMD to Mix GCN with VLIW4 & VLIW5 Architectures

[boxleitnerb]

If 7950 is below or on par with GTX580, then a 7970 will only be 20-25% faster - otherwise they would put an additional model there. So IF that arrangement is true, Tahiti XT won't exceed GTX580+25%. And that would surely be dissapointing.

[Lanek]

Lol, Nvidia fanboys are going out of the forest ?

[Soultaker52]

I for one, am just hoping it leave my 5870 in the dust. I want a new card, and if they impress, the 7970 will be it. I had another 5870 in the system for crossfire, but man, it was nothing but headaches and crashes. I guess crossfire just doesn't agree with me? Lol.

[boxleitnerb]

Fanboy is an insult. Please refrain from that in the future. People are just speculating here.

[SKYMTL]

One thing to remember is that NVIDIA did their "big move" towards a true ground up DX11 architecture with Fermi and paid the price. If AMD does the same thing, then they will pay the same price and learn the same (very hard) lessons.

AMD has essentially been using the same basic architecture since the DX10 days and has gradually added modules for DX11 compatibility. So, the move to a ground up DX11 architecture may be a tough one.

[3lfk1ng]

@Soultaker52
Could have also been your meager 750watt PSU. Two 5870's in Crossfire consume 670watts using this setup.

[kuroikenshi]

One thing to remember is that NVIDIA did their "big move" towards a true ground up DX11 architecture with Fermi and paid the price. If AMD does the same thing, then they will pay the same price and learn the same (very hard) lessons.

AMD has essentially been using the same basic architecture since the DX10 days and has gradually added modules for DX11 compatibility. So, the move to a ground up DX11 architecture may be a tough one.

I concur with you Mr Rodent.

[PrometheusCon]

7950? Oh that brings back memories...

GX2.

[Soultaker52]

@3lfk1ng
Hmm.. I was thinking that could have been related. Though I've seen people running such setups fine with lesser PSU's. The crashing happened when clock speeds changed though. When I forced it to stay at my OC'd speeds constantly, it wouldn't crash. The performance just didn't feel amazing though. Maybe I'm expecting too much?

[tajoh111]

One thing to remember is that NVIDIA did their "big move" towards a true ground up DX11 architecture with Fermi and paid the price. If AMD does the same thing, then they will pay the same price and learn the same (very hard) lessons.

AMD has essentially been using the same basic architecture since the DX10 days and has gradually added modules for DX11 compatibility. So, the move to a ground up DX11 architecture may be a tough one.

One more way AMD is going to also pay a monster price is driver performance. If GCN is much more different, much of the driver optimizations for their old architecture will not come about. Considering how long it took for the 3870 to get performance up to snuff with the 8800gt, it might be a year before we see good driver performance from AMD.

A 40 percent jump or say 25 percent faster than a gtx 580 wouldn't be that bad, with immature drivers it could go to 40% in half a year. Expecting 80-100 percent is too much considering the driver team from AMD and the new architecture. No one gets it right the first time anymore and I think AMD is taking risks, expect to see some eratic performance jumps.

[dasa]

@Soultaker52
Could have also been your meager 750watt PSU. Two 5870's in Crossfire consume 670watts using this setup.

i doubt thats the problem
two 5870 in cf only use ~300w while gaming or up to ~400w with fur + other components
i was running two gtx480 @875mhz 1138mv stable 940mhz 1150mv benchmarking with a 2600k at 4.4 1.3v 24\7 stable 5ghz 1.45v benchmarking also on a tx750w psu for several months power draw at gaming clocks was ~500-600w at the wall and peaked at ~740w benching although being on 240v may have helped
the psu was working a bit when overclcoked its fan would become rather loud but at stock clocks the psu fan stayed quiet
two gtx480 at stock use ~470w gaming ~610w max fur

my mate had two 5850 in cf and had nothing but problems with them while at the same time my 4870x2 was fine (oh also tried two 4870x2 on this psu for a quick bench...)
from what i have seen around the net it looks like the 5000 cards just had a lot more driver problems in cf than most cards from ati


it will be nice to finally see some real data would love to buy two or three hd7000 cards myself to replace these 480

[hurleybird]

One thing to remember is that NVIDIA did their "big move" towards a true ground up COMPUTE architecture with Fermi and paid the price. If AMD does the same thing, then they MIGHT pay the same price and learn the same (very hard) lessons.

Fixed that for you. It's likely that AMD's big push into compute will come up with some of the issues Fermi had, but as a reviewer who I'm assuming hasn't gotten his hands on hardware yet, I wouldn't be talking in such absolutes about unreleased parts. Kind of looks bad, and preconceived notions can be a dangerous thing.

[LordEC911]

If 7950 is below or on par with GTX580, then a 7970 will only be 20-25% faster - otherwise they would put an additional model there. So IF that arrangement is true, Tahiti XT won't exceed GTX580+25%. And that would surely be dissapointing.

There are multiple sets of performance numbers out... I wouldn't read too much into that one.

One more way AMD is going to also pay a monster price is driver performance. If GCN is much more different, much of the driver optimizations for their old architecture will not come about. Considering how long it took for the 3870 to get performance up to snuff with the 8800gt, it might be a year before we see good driver performance from AMD.

A 40 percent jump or say 25 percent faster than a gtx 580 wouldn't be that bad, with immature drivers it could go to 40% in half a year. Expecting 80-100 percent is too much considering the driver team from AMD and the new architecture. No one gets it right the first time anymore and I think AMD is taking risks, expect to see some eratic performance jumps.

Or maybe it has the opposite effect.

[BababooeyHTJ]

should I sell the 6950 now? how soon is this coming?

Without any pressure from Nvidia I would expect AMD to release the 7xxx series at a price in line with their current offerings just like they did with the 5xxx series. Thats the best way to win marketshare.

One more way AMD is going to also pay a monster price is driver performance. If GCN is much more different, much of the driver optimizations for their old architecture will not come about. Considering how long it took for the 3870 to get performance up to snuff with the 8800gt, it might be a year before we see good driver performance from AMD.

A 40 percent jump or say 25 percent faster than a gtx 580 wouldn't be that bad, with immature drivers it could go to 40% in half a year. Expecting 80-100 percent is too much considering the driver team from AMD and the new architecture. No one gets it right the first time anymore and I think AMD is taking risks, expect to see some eratic performance jumps.

I've never seen any huge performance boosts across the board with drivers. With 6950 I can't think of any games that saw some serious gains with just driver updates and thats not a bad thing. That card did really well out of the gate and that includes crossfire performance. New games is one thing but not in older titles.

[SKYMTL]

Fixed that for you. It's likely that AMD's big push into compute will come up with some of the issues Fermi had, but as a reviewer who I'm assuming hasn't gotten his hands on hardware yet, I wouldn't be talking in such absolutes about unreleased parts. Kind of looks bad, and preconceived notions can be a dangerous thing.

No absolutes were discussed in my post. I was just mentioning that from an architectural standpoint, there are many hurdles for AMD to overcome in order for their theoretical DX11 to reach NVIDIA's current levels, particularly with AA enabled. I never said they couldn't do it. Quite the contrary actually since the amount of engineering intelligence housed in AMD's GPU division is mind boggling.

[prava]

Fixed that for you. It's likely that AMD's big push into compute will come up with some of the issues Fermi had, but as a reviewer who I'm assuming hasn't gotten his hands on hardware yet, I wouldn't be talking in such absolutes about unreleased parts. Kind of looks bad, and preconceived notions can be a dangerous thing.

SKYMTL is probably the most honest reviewer I have yet to see. With that said, though, he has his own opinion...as we all do, and as long he isn't under NDA he is as free as we are to discuss the rumours we keep getting

[tbone8ty]

One thing to remember is that NVIDIA did their "big move" towards a true ground up DX11 architecture with Fermi and paid the price. If AMD does the same thing, then they will pay the same price and learn the same (very hard) lessons.

AMD has essentially been using the same basic architecture since the DX10 days and has gradually added modules for DX11 compatibility. So, the move to a ground up DX11 architecture may be a tough one.

I agree but I think amd will have less wafer issues to deal with than nvidia did with the big original fermi chip.

look at the driver progression of the gtx 480 it definetly got better with age like a good wine. Drivers updates helped it a lot more so than a lot of other architectures.

we will see ;-)

[AliG]

I agree but I think amd will have less wafer issues to deal with than nvidia did with the big original fermi chip.

look at the driver progression of the gtx 480 it definetly got better with age like a good wine. Drivers updates helped it a lot more so than a lot of other architectures.

we will see ;-)

I personally am thinking about getting an NV card next round purely because of drivers. CCC crashes on me way too much for my liking

[SKYMTL]

I agree but I think amd will have less wafer issues to deal with than nvidia did with the big original fermi chip.

we will see ;-)

Actually, that is what I was referring to. A true DX11 rendering pipeline + additional compute resources takes up a ton of die space. That's what NV had to find out the hard way.

[cold2010]

AMD Radeon HD 7000 fiasco as Bulldozer - likely Yes!

AMD sent to several manufacturers info-slides about the new Radeon HD Graphics 7000. According to first estimates results in the slides, AMD-fans will be disappointed. Game performance against the current ty-slow cards with Cayman not significantly increased, but consumption is relatively high for an advanced 28nm process. Stay tuned .... some slides will be here soon!

[tajoh111]

There are multiple sets of performance numbers out... I wouldn't read too much into that one.



Or maybe it has the opposite effect.

I just don't see it though. The more powerful these things cards get, the more driver work needs to come into play. Most games, can't close to come to utilize all the power of a card, especially because most of them are console ports, and driver's must be specifically written to take advantage of these things. All games that are out of the moment for PC, are written to take advantage of current technology, not future technology like 7xxx. As a result, the driver team is really responsible for getting performance up, because they are doing all the work for scraping out performance, rather than the software companies making the games who mostly cater to the lowest common denominator and barely care for new hardware. Unless a game is sponsored by a hardware company, there is little purposes for them to code for new hardware.

AMD current driver set is so focused on vliw architecture that they are going to have to start from the ground up. In addition, it just seems hard for AMD to put out a good product as of late, Bulldozer and cayman were both let downs. Considering how much markets they are trying to get into, I have a feeling is AMD is spreading itself too thin. I have a feeling, GCN, is going to have a lot of growing pains, it won't get it completely right this generation and it will get alot better with generation 2. It is amazing that AMD has this much of a headstart over NV. I have a feeling things were rushed to get there and AMD just has to be 15-25% performance jump over a gtx 580 to justify an above $450-550 price tag. The 5870 atleast according to techpowerup, was 25% faster than a gtx 280. And look how well the sales of that card were. This is the optimal pricepoint for the highend because we are in a recessions and people don't want to pay 600+ for cards anymore.

AMD doesn't need crazy performance to sell and it doesn't need a completely polished product to sell. What is more important is getting a headstart on the competition, as AMD probably learned from the 58xx generation or on the opposite end during the 88xx NV generation. In addition, Nvidia is no intel when it comes to branding strength, people won't wait for NV, as they would for an Intel product. It in there best interest to leave room for performance if they face no competition for 5-8 months before NV drops there product, that way, when they release 8xxx, they have a product(a quick refresh) that looks great as well without sinking a huge amount of money into R and D and time into it. If southern islands is too good off the bat, unless it is fundamentally broken, it will be hard to have 8xxx look good if they are stuck at 28nm. Also considering it is second generation, it would likely beat nvidia's first generation of keplar. Intel has products that could at the moment grind AMD into dust, but they don't release them to get as much return on investment on their current research and their own products would become competition(as well as anti-trust issues).

[Tim]

AMD Radeon HD 7000 fiasco as Bulldozer - likely Yes!

OBR, lol.

[boxleitnerb]

HD7400-7600M just rebrands?
http://techreport.com/discussions.x/22120

I thought these were supposed to be 28nm chips.

[Kuroimaho]

OBR, lol.

I remember how people were laughing at his posts about bulldozer until bulldozer launch day...

[generics_user]

I remember how people were laughing at his posts about bulldozer until bulldozer launch day...

he's adding too much hate into his articles to be taken seriously; it doesn't matter what or where his sources are