Let the 6970 Rumors Begin

Status
Not open for further replies.
255W TDP for a single card.... excellent that it beats the 480 quite considerably.

I'll be in for 2 if they have Eyefinity+Crossfire fixed.
Posted via [H] Mobile Device
 
That's double the performance of my current 5870 so if that proves to be true I will be getting a new vid card soon so long as the price is fair.
 
Its going to be interesting to see what the Gen 8 "scalability and off-chip buffering" brings. :)

tess_small.png
 
That's double the performance of my current 5870 so if that proves to be true I will be getting a new vid card soon so long as the price is fair.

6970: 3dmarks: 23,499
5870: 3dmarks 19,337

That isn't even remotely double. That is 20%.

More like 20% overall and Double on heavy Tessellation (IOW fixed tessellation).
 
hybrid design is doing quite well. Cant wait till 28nm and souther island, my gtx 480 should be enough until then.
 
Off chip buffering? What is that exactly?
A buffer that's not on the chip. :p

But what you're really asking is, how does it improve performance. We'll just have to wait another month to find out.
 
I'll believe it when I see it. Real games will be more modest than that unigine rubbish me thinks.
 
If the 6970 only has 1GB, I rather have the 1.5GB+ Nvidia GTX 580.
More of a reason why I'm going with the 6990.
 
What will 28nm bring? Thanks.

28nm will bring double the transistor count in the same physical size. Double the transistors can mean double the execution units.

Every big leap in performance has come from a big leap in the transistor count.
 
Off chip buffering? What is that exactly?

Just guessing, but it's probably that the triangle data from the Tesselator needs to be buffered before it can be used for the rasterisers & pixel shaders
- at the moment this buffering is on-chip, which means the buffers are small
- this means that the chip is having to task-swap between Tesselation tasks and Rasteriser/pixel shader tasks continually
- whereas, in the HD6900 this data can use larger off-chip buffers, and so the amount of task swapping is reduced....
(because, in DX11, the Tesselation shaders (HS & DS) are just like any other shader, and run in the main shader core)
 
255W TDP for a single card.... excellent that it beats the 480 quite considerably.

I'll be in for 2 if they have Eyefinity+Crossfire fixed.
Posted via [H] Mobile Device

That's alot of power
- Fermi-class power....
:eek:
 
That's alot of power
- Fermi-class power....
:eek:

It'll be interesting if they can come up with a way of taming that kind of power, or if we're in for a "surprise, it's really hot and loud" event like the GTX 480 debut.
 
28nm will bring double the transistor count in the same physical size. Double the transistors can mean double the execution units.

Every big leap in performance has come from a big leap in the transistor count.

Hm okay, thanks for that. I thought something along those lines... But what I was also asking is, what products will the 28nm process be used for? The next gen, 7XXX series? Someone proposed in another thread it might be used in low-end/budget 6000 series cards to replace the 57xx, 56xx, etc. Is that true?
 
Hm okay, thanks for that. I thought something along those lines... But what I was also asking is, what products will the 28nm process be used for? The next gen, 7XXX series? Someone proposed in another thread it might be used in low-end/budget 6000 series cards to replace the 57xx, 56xx, etc. Is that true?

It could be, The 7000 series is definitely going to be 28nm based. They could use lower end 6xxx cards as test samples before they jump head first with their power hungry cards, they did that with the 4770, and it was quite a benefit for them.
 
6970: 3dmarks: 23,499
5870: 3dmarks 19,337

That isn't even remotely double. That is 20%.

More like 20% overall and Double on heavy Tessellation (IOW fixed tessellation).

I think those are the over all scores with the GPU + CPU. The GPU score maybe more revealing as to the actual performance vs 5870. The 1120 SP 6870 can out do a 5850 that has 320 more SPs. Just think what a 6970 will do with even more optimizations. It will be far more then 20%.
 
Finally some news on the card I've been waiting for. Regardless of the power issue, I plan on sticking mine under water right away.
 
If the 6970 only has 1GB, I rather have the 1.5GB+ Nvidia GTX 580.
More of a reason why I'm going with the 6990.

You do realize, if the 6970 has only 1GB then chances are the 6990 effectively will have only 1GB as well? The 5970, for example, has 2GB total on board but only 1GB of effective memory.
 
I really hope they offer a 2Gb version on release and we don't have to wait several months for a special version or something one of the manufacturers have done.
 
I really hope they offer a 2Gb version on release and we don't have to wait several months for a special version or something one of the manufacturers have done.

More likely than not, 2GB model will provide only minimal increase in performance at 1900x1200 and below. Of course it's always nice to have more.
 
More likely than not, 2GB model will provide only minimal increase in performance at 1900x1200 and below. Of course it's always nice to have more.

Part of that depends on the game engine and type of game. RTSs & MMOs often benefit from more memory for smoother game play.
 
More likely than not, 2GB model will provide only minimal increase in performance at 1900x1200 and below. Of course it's always nice to have more.

True.. but not many people buy a top end card now for 1920x1200. It's all about Eyefinity and 2560x1440+, where that extra framebuffer can come into play.
 
True.. but not many people buy a top end card now for 1920x1200. It's all about Eyefinity and 2560x1440+, where that extra framebuffer can come into play.

quite few games used over 1GB VRAM even in 1680*1050-1920-1200... Crysis is one of it.
 
I'm waiting for the HD 6970 myself. It should beat out the GTX 480 by a pretty good margin performance wise. IF, major keyword, if the GTX 580 rumors up to be true or at the least, near true, then the HD 6970 will not be able to beat the Fermi refresh. The HD6990 should be the top all around performance part though unless Nvidia has something up there sleeves. Note that this is all based on my opinion & what has been gathered through the rumor mill.
 
6970: 3dmarks: 23,499
5870: 3dmarks 19,337

That isn't even remotely double. That is 20%.

More like 20% overall and Double on heavy Tessellation (IOW fixed tessellation).

6970 = In the Unigine Heaven benchmark, the card managed 36.6fps

5870 = 17.3 fps in Unigine Heaven at the same settings.

That's over double the performance.

GTX 480 = 29.5fps in Unigine Heaven.
 
Just guessing, but it's probably that the triangle data from the Tesselator needs to be buffered before it can be used for the rasterisers & pixel shaders
- at the moment this buffering is on-chip, which means the buffers are small
- this means that the chip is having to task-swap between Tesselation tasks and Rasteriser/pixel shader tasks continually
- whereas, in the HD6900 this data can use larger off-chip buffers, and so the amount of task swapping is reduced....
(because, in DX11, the Tesselation shaders (HS & DS) are just like any other shader, and run in the main shader core)

Sounds good to me, bring it on.
 
6970 = In the Unigine Heaven benchmark, the card managed 36.6fps

5870 = 17.3 fps in Unigine Heaven at the same settings.

That's over double the performance.

I did say it was double on Tessellation. Something that many consider very weak/broken in AMD cards.

But it won't be double overall. It will be more like 20%-30% overall + fixed Tessellation.

Still a good result, but not double.
 
I did say it was double on Tessellation. Something that many consider very weak/broken in AMD cards.

But it won't be double overall. It will be more like 20%-30% overall + fixed Tessellation.

Still a good result, but not double.

Ironically, the 6800's are actually doing in better in game than in lame benchmarks, but that Unigine leak was actually fake so dont bother with the #'s
 
I want the extra memory for playing on Eyefinity across three Dell 27" monitors. I mainly play WoW at 5760x1200res, and with the new Dx11 options to the WoW engine in Cataclysm, the game can get pretty demanding actually, with Ultra settings, and 4x AA, etc...

How does Crossfire work in Eyefinity ? And does the WoW engine take advantage of dual VideoCards ?
 
Now to replace my 5870 with 5950's or 5970's to see eyefinity in all it's glory! :D I'm excited for this release.
 
Anyone know if the HD69xx is still using the 5D shaders (like the HD68xx), or if it's using the 4D shader architecture.

The rumor was that NI was using the new 4D shader architecture, but the HD68xx being 5D seems to hint otherwise for the HD69xx also.
 
Anyone know if the HD69xx is still using the 5D shaders (like the HD68xx), or if it's using the 4D shader architecture.

The rumor was that NI was using the new 4D shader architecture, but the HD68xx being 5D seems to hint otherwise for the HD69xx also.

The hints only say that Cayman is a superset of Barts, meaning it will have more features/changes than Barts has, and that it's geared for high end GPU domination

No specifics though
 
I did say it was double on Tessellation. Something that many consider very weak/broken in AMD cards.

But it won't be double overall. It will be more like 20%-30% overall + fixed Tessellation.

Still a good result, but not double.

Those numbers do represent overall performance, no? Will have to wait and see when the card is released and HardOCP and other sites do a review but if it is only a 20% improvement over my 5870 then it will be a hard sell for me.
 
I did who the fuck knows how I butchered that one, lol.

It's ok, between tryign to type 6850 and 6870 when talking about 5850's and 5870's while comparing them to 4850s and 4870s and referecing vidias GTX4600's and oh wait :confused::confused::confused:
 
Well, the numbers are very impressive if true. Unless Nvidia pulls a rabbit out I'm in for one, and i really don't care what it costs.
 
Status
Not open for further replies.
Back
Top