• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Radeon HD 7900 Launch set for December 22nd, 2011 - R1000 | Tahiti | GCN

pestul

Member
It's a new architecture.. that will account for an increase in speed as well. 30% this and 50% that means something, but what if some of the processors are 100% more efficient.
 
What you say is certainly not impossible, but it's not a strategy either company has used in recent years.

Regarding the comment that 33% more processors and 50% more bandwidth clearly equals > 50% performance increase, you may want to spend some time investigating Amdahl's Law.

You may want to stop looking at AMOUNTS and realize that this is a NEW architecture completely. In some instances we might see huge amounts of performance increases, especially in future-oriented stuff like tessellation etc.

And didn't the 5870 outperform two 4870's in Crossfire? That was both on the 'new' (at the time) 40 nm process AND within the last 3 years.

I'll be crazy ignorant in my corner over here then.
My 40% comment was in relation to a 580 btw. Dunno if that means I get to leave the corner or not.

You can stay or you can go, but the corner doesn't suit you imo. :p

It's a new architecture.. that will account for an increase in speed as well. 30% this and 50% that means something, but what if some of the processors are 100% more efficient.

Thank you. Somebody that gets it.
 

tokkun

Member
You may want to stop looking at AMOUNTS and realize that this is a NEW architecture completely. In some instances we might see huge amounts of performance increases, especially in future-oriented stuff like tessellation etc.

I was responding to what you actually said.

Will the new architecture increase IPC? Maybe. But we don't have any information on how much, if any. New architectures sometimes even decrease IPC, as with P4 Willamette. Accurately predicating performance based on the information available to the public is pretty much impossible. Therefore, I will make predictions based off of what seems to make business sense.

If they're pricing 7970 at $500 as rumored and the GTX580 is currently at $500, they don't need 50% more performance to attract the type of buyers in that price segment.
 

Antiochus

Member
Reviews may be up by Dec. 22nd, not just paper launch

http://fudzilla.com/graphics/item/25247-amd-radeon-hd-7900-press-event-on-22nd-of-december

According to our sources, the 22nd of December is the date for the Radeon HD 7900 series press event. This means that reviews will go up on 22nd of December and AMD will probably brag about being having the first 28nm desktop cards in 2011, while there will be no cards or partner annoucements until 9th of January 2012.

Some select members of the press have already gotten their hands on cards yesterday and some will get them today

As Fuad is an anti ATI zealot, one would take this with healthy doses of caution:

http://fudzilla.com/graphics/item/25248-radeon-hd-7970-is-around-30-percent-faster

According to our sources, the Radeon HD 7970 is about 30 percent faster in 3DMark when compared to the Radeon HD 6970. These are early performance benchmarks made with the first "usable" driver.

According to the whispers that we are hearing right now, Radeon HD 7970 should end up about 30 percent faster than the HD 6970. These are 3DMark early performance numbers probably done with the 12.1 RC driver that was, at least according to our info, released two days ago. This should put the HD 7970 somewhere around Radeon HD 6990.

We are also hearing that the HD 7970 does even better in game benchmarks and we hope that we'll have more details regarding those pretty soon, as after all, game benchmarks are the ones that really matters when it comes to performance of graphics cards. We still don't know if there will be another driver before first reviews hit the net, but AMD usually holds the final driver as long as it can to control the performance leaks.

Some slides have leaked over at Donanimhaber showing that Tahiti might do a lot better when it comes to tesselation, but once again we are talking about AMD slides so we tend to thread lightly there.

In any case, we'll know for sure on 22nd of December.
 

artist

Banned

artist

Banned
slajdek.png
 
Might be not appropriate for this thread, but I was planning to upgrade from my 4850 to Nvidia, is it true that their high end single cards won't launch until late 2012 ?

If so I think I might pick up 7970/7950
 

Hazaro

relies on auto-aim
Rumors point to May/June/July so it is certainly possible.
Also possible nVidia will counter launch some card or cut prices agressively on their older process.
 

mik83kuu

Banned
I didn't want to get onboard the AMD driver train so I got myself a GTX580 the other day. Here's to hoping nVidia release their new series in the summer :)
 

Wazzim

Banned
We already have a console launch today! In a few hours Japanese stores will open, and Vita will be freed to the masses.

Honestly not really hyped for the launch in like 3 hours, I'm eagerly waiting for the western launch of the system though.
 
It's becoming increasingly likely to me we're looking at 30-50% faster than 6970 here.

As somebody at B3D posted, 50% more memory bandwidth, so that's pretty much your upper performance improvement limit.

I'm sure some will kvetch about this but I wont too much. Reason being is Nvidia and AMD having been marching in performance lockstep for several releases now so there's no reason to think Nvidia's next will be any better until proven otherwise, besides seeming 3+ months away.

Of course I'm not buying these $300+ so I wonder if anything good will filter down to the 200-250 price point anytime soon, or if there will be nice price cuts on the 6000 series? Just bought a 6870 so...
 

DieH@rd

Banned
Of course I'm not buying these $300+ so I wonder if anything good will filter down to the 200-250 price point anytime soon, or if there will be nice price cuts on the 6000 series? Just bought a 6870 so...

Wait for 8xxx. They will come out shortly after gf6xx series. If a tradition continues 88xx will get performance of 79xx, and that will be awesome deal for <300$ card.
 
I've always wondered why AMD kept giving Nividia that advantage, something I believed helped give Nvidia an edge performance wise. 384 bit bus, more POM please.
 

Red

Member
Wait for 8xxx. They will come out shortly after gf6xx series. If a tradition continues 88xx will get performance of 79xx, and that will be awesome deal for <300$ card.

This is what I'm considering. I want to upgrade from my 5870 (mainly for improved tesselation performance + increased video memory), but I'm not sure a hefty investment is worth it right now. The next revision of these cards seems like the smarter deal.
 
I've always wondered why AMD kept giving Nividia that advantage, something I believed helped give Nvidia an edge performance wise. 384 bit bus, more POM please.

What advantage? If you're referring to the bus width, AMD has had the same bandwidth as Nvidia in recent times. When AMD was using a 256 bit bus vs Nvidia's 512, AMD was using twice as fast GDDR5 so it came out about the same. More efficient really since you have to use a gigantic die for a 512 bus, and AMD was using an effective small die strategy.

Just looked it up and GTX 580 is 192 GB/s, HD 6970 175 GB/s, similar. More is better but I doubt it's ever been a major issue.
 
What advantage? If you're referring to the bus width, AMD has had the same bandwidth as Nvidia in recent times. When AMD was using a 256 bit bus vs Nvidia's 512, AMD was using twice as fast GDDR5 so it came out about the same. More efficient really since you have to use a gigantic die for a 512 bus, and AMD was using an effective small die strategy.

Just looked it up and GTX 580 is 192 GB/s, HD 6970 175 GB/s, similar. More is better but I doubt it's ever been a major issue.

It's somewhat of an issue, as current cards are still bandwidth starved. AMD's move to 384-bit was a smart one.
 
What advantage? If you're referring to the bus width, AMD has had the same bandwidth as Nvidia in recent times. When AMD was using a 256 bit bus vs Nvidia's 512, AMD was using twice as fast GDDR5 so it came out about the same. More efficient really since you have to use a gigantic die for a 512 bus, and AMD was using an effective small die strategy.

Just looked it up and GTX 580 is 192 GB/s, HD 6970 175 GB/s, similar. More is better but I doubt it's ever been a major issue.
Its a performance issue I believe, if the article I've read are to be believed, bandwidth makes the difference, even 17GB/s.
 
Going to be expensive!

All signs point to the 7970 being $550.

You could pick up 2x 6970 for that price and if a single 7970 is more powerful than 2x 6970 then I will be a very happy chappy indeed.
 

Corky

Nine out of ten orphans can't tell the difference.
Going to be expensive!

All signs point to the 7970 being $550.

You could pick up 2x 6970 for that price and if a single 7970 is more powerful than 2x 6970 then I will be a very happy chappy indeed.

If a 7970 is more powerful than two 6970 then I'll throw my 560 tis out of the window and go xfire 7970. Something tells me that wont be the case though.
 

pestul

Member
I'm starting to think it might only be 30-50% over the 6970 at release and improve quickly with drivers. Still quite a large gap, but they might actually lose out on the price/performance with this one.
 

Shambles

Member
Looks good but the pricing seems like it's going to be terrible until nVidia release their next lineup. Fine with me though. I'm waiting for Ivy Bridge to upgrade my Q6600/4870. By March there should be some GPU competition, and SSDs should continue to fall in price
 

squidyj

Member
It's a new architecture.. that will account for an increase in speed as well. 30% this and 50% that means something, but what if some of the processors are 100% more efficient.

Will it though? A lot of it is about the shift to GCN and the shift to GCN is about compute, and not shader performance.
 

artist

Banned
Yeah new arch does not always equal more efficient/powerhouse performance

Geforce 6800 series = Awesome
Geforce 8800 series = Awesome
Geforce GTX200 series = Meh
Geforce GTX400 series = Bleh

Radeon 9700 series = Awesome
Radeon X1800 series = Meh
Radeon X1900 series = Awesome
Radeon HD 2000 series = Yuck
Radeon HD 4000 series = Good
Radeon HD 6000 series = Meh

Intentionally skipped generations with minor improvements/shrinks etc.
 

Corky

Nine out of ten orphans can't tell the difference.
Yeah new arch does not always equal more efficient/powerhouse performance

Geforce 6800 series = Awesome
Geforce 8800 series = Awesome
Geforce GTX200 series = Meh
Geforce GTX400 series = Bleh

Radeon 9700 series = Awesome
Radeon X1800 series = Meh
Radeon X1900 series = Awesome
Radeon HD 2000 series = Yuck
Radeon HD 4000 series = Good
Radeon HD 6000 series = Meh

Intentionally skipped generations with minor improvements/shrinks etc.

Out of curiosity, what determines whether or not a generation is good/bad/etc?
 
If AMD can fix their drivers then I might be interested, but for now my 6870 is more annoying than enjoyable. Random BSOD after sleep (video drivers crashing the system) and messing up bad in some games because of dual monitors.
 

SapientWolf

Trucker Sexologist
Yeah new arch does not always equal more efficient/powerhouse performance

Geforce 6800 series = Awesome
Geforce 8800 series = Awesome
Geforce GTX200 series = Meh
Geforce GTX400 series = Bleh

Radeon 9700 series = Awesome
Radeon X1800 series = Meh
Radeon X1900 series = Awesome
Radeon HD 2000 series = Yuck
Radeon HD 4000 series = Good
Radeon HD 6000 series = Meh

Intentionally skipped generations with minor improvements/shrinks etc.
Uh, the 5xxx series wasn't just minor improvements. It was on a level of 100% performance increases in some cases.
 

McHuj

Member
Semiaccurate has a story up that everything above and including 77xx series are all GCN based and 76xx below are rebranded VLIW5 chips.

I figured only 79xx series would be GCN, but that's good news if it's true.
 

tokkun

Member
I went from a 4850 ->5850 and not only were performance gains huge, but the power efficiency and heat output majorly improved. Even if it was only a die shrink, 5xxx was a very strong gpu gen...

No one is saying that the 5xxx was shitty.

The point of that list was that a new architecture doesn't necessarily mean big performance or efficiency gains, so we shouldn't take it for granted that GCN will give us such gains.

The most relevant point of comparison is probably 8800 -> GTX200. That was an architectural upgrade that included a lot of consideration for GPGPU, and that is also supposed to be the case with GCN.

In that case, we know that the GTX200 and Radeon 4800 series were equivalent in performance, but the GTX200 chips were larger and used more power. So there is reason to be guarded about optimism over GCN in terms of improvements in game performance.
 
Top Bottom