NDA for Cayman is lifted 22nd?!

[Silverforce11]

Here you all, enjoy pretty slides: (From Chiphell, http://www.chiphell.com/thread-140791-1-1.html)

http://img694.imageshack.us/img694/1947/697001.jpg

http://img227.imageshack.us/img227/990/697002.jpg

http://img12.imageshack.us/img12/1089/697003.jpg

http://img249.imageshack.us/img249/4616/697004.jpg

http://img840.imageshack.us/img840/530/697005.jpg

http://img196.imageshack.us/img196/96/697006.jpg

http://img26.imageshack.us/img26/7452/697007.jpg

http://img821.imageshack.us/img821/4716/697008.jpg

http://img833.imageshack.us/img833/4254/697009.jpg

http://img19.imageshack.us/img19/135/697010.jpg

http://img815.imageshack.us/img815/8448/697011.jpg

http://img138.imageshack.us/img138/7991/697012.jpg

Their new "power containment" on architecture is very interesting. Sounds like NV throttling "power viruses".

These are from a previous brief. Lots of TBD to be determined.
Funny thing, even with this late briefing they aren't spilling on some specs. Like bandwidth, just a flat >160 and TMUs. heh

Should be some benchmark leaks soon (anticipation is too much to handle, im going to get a cold beer), AIBs getting briefed now (22nd afternoon in china).

Edit: There's more here, http://forums.vr-zone.com/news-around-the-web/939375-69xx-slides.html

 

[Hauk]

I smell another well balanced GPU coming..

 

[chewietobbacca]

Don't think it's actually lifted.. the slides are from the October presentation. They probably gave more updates this past week... but a good sneak nontheless!

 

[Rifter]

wake me up when anandtech has one to benchmark. all this marketing BS and fanboy fighting is getting out of control.

 

[Ares1214]

Wait, they are making ANOTHER form of AA?!

 

[RaistlinZ]

Hm, will the power containment limit overclockability? Is it going to be something the user can turn on/off? I really couldn't care less about power usage on a $400 video card, I just want performance for my money spent.

 

[BFG10K]

Wait, they are making ANOTHER form of AA?!

It's coverage sampling.

 

[Skurge]

It's coverage sampling.

What is that?

 

[BFG10K]

What is that?

Coverage sampling decouples color/z/stencil information from the sample and leaves only a binary value as to whether it falls inside a polygon or not.

nVidia has had it since the G80's launch in 2006.

 

[Skurge]

Coverage sampling decouples color/z/stencil information from the sample and leaves only a binary value as to whether it falls inside a polygon or not.

nVidia has had it since the G80's launch in 2006.

What advantages or disadvantages does it have versus normal AA?

 

[BFG10K]

What advantages or disadvantages does it have versus normal AA?

I wrote a piece that describes it in detail, but I can't link to my own articles.

Google GT200 anti-aliasing investigation - it's the first hit.

 

[Douglar]

Here are the things that I saw in the slides that stood out:

Dual 8th Gen tessellation units
* off chip buffering
* up to 3x tessellation of the 5870
Dual rasterizers
* Fed from the same dispatcher
* Each works with one tessellation unit
256 Bit memory
* clocked faster than 1.25Ghz (5Ghz)
Integrated power control processor
* dynamically sets clocks for each functional block
* looks like a real performance maximizer in the current power constrained environment

 

[HurleyBird]

Things are starting to shape up, although the big unknown is still the "off chip buffering". It almost sounds like higher end SKUs will have some kind of tesselation co-processor.

 

[Martimus]

I wrote a piece that describes it in detail, but I can't link to my own articles.

Google GT200 anti-aliasing investigation - it's the first hit.

Here is the link: http://alienbabeltech.com/main/?p=10638

BFG, can you give a quick 30,000ft overview of the differences?

 

[SirPauly]

What advantages or disadvantages does it have versus normal AA?

What they were to me is a way to garner more image quality with a lower performance hit but more importantly, smaller memory footprint.

 

[blastingcap]

I wrote a piece that describes it in detail, but I can't link to my own articles.

Google GT200 anti-aliasing investigation - it's the first hit.

Thanks; last week I asked apoppin what his opinion was of NV's CSAA vs MSAA because some people say CSAA is like, half of MSAA, but I don't know if he saw my question.

P.S. Do I detect a bit of pride where you wrote: "Google GT200 anti-aliasing investigation - it's the first hit"? Nice article.

 

[SirPauly]

Thanks; last week I asked apoppin what his opinion was of NV's CSAA vs MSAA because some people say CSAA is like, half of MSAA, but I don't know if he saw my question.

P.S. Do I detect a bit of pride where you wrote: "Google GT200 anti-aliasing investigation - it's the first hit"? Nice article.

It's not really CSAA vs MSAA based on MSAA is the foundation for Coverage sample AA. CSAA is really an enhancement to improve the quality of MSAA. In the past, all one had was adding multi-sampling and when using higher resolutions, the hit, specifically memory footprints of going to x4 to x8 multi-sampling was high, if one didn't have enough memory. What CSAA did was help address this and allows more flexibility to improve IQ for many.

 

[tviceman]

Well back to the topic at hand...

I have to admit, if Cayman isn't memory bandwidth starved, a 1920 shader card may put the gtx580 to shame. I still question whether or not that will happen because if the card is as fast as, or slightly faster, than gtx580 with whatever current beta drivers it's using, then delaying was a bad decision especially if the chip/card will be priced lower than the gtx580.

1920 shaders with several tessellation units sounds amazing on paper - I am really looking forward to seeing the reviews and benchmarks.

 

[blastingcap]

BSN* has historically not always been accurate, especially during the Fermi delay, but Theo was right about the Barts timeframe. He even disclosed the location of the deep dive meeting and such.

Anyway, FWIW, he's saying that Cayman debuts Dec. 15, not Dec. 7 like some other people are saying. Perhaps Dec. 7 is some other Cayman-related date.

AMD itself said Dec. 13 or after, though it's possible they could change their minds and move it up.

If I were a betting man, I'd bet on Dec. 15 as the launch date.

http://www.brightsideofnews.com/new...tures-1920-cores2c-hd-6990-antilles-3840.aspx "Our sources informed us that HD 6950 and HD 6970 are scheduled to debut on December 15, at 12:01AM EST. We're working hard on securing a sample for launch review, but it is not easy given that we were taken off the press list with no particular reason."

 

[happy medium]

So release is set for mid December for the 6950/6970?
Mabe they are waiting for the gtx 570 release on the 7th?

 

[Arkadrel]

Whats funny is they dont give 100% specific values!

they just say:
more than >160 GB/s bandwidth
more than >20 SIMD engines
more than >80 Texture units.


EQAA looks good, something nvidia had but ati didnt.
Power Contrainment unit thingy, looks good.

the Z/Stencil ROP units, and Colour ROP units are 2x-4x faster... That seems like one of the major changes over the 5870's.

2GB ram... for people that run 2-3 monitors these cards will be golden, in that they wont be ram limited by the high res. Compaired to the 1GB cards of the 5xxx series.

one of the slides just says up to 3x tessellation performance of the 5870. That doesnt sound like that much of a improvement though.


Im really looking forwards to benchmarks.

 

[Grooveriding]

Whats funny is they dont give 100% specific values!

they just say:
more than >160 GB/s bandwidth
more than >20 SIMD engines
more than >80 Texture units.


EQAA looks good, something nvidia had but ati didnt.
Power Contrainment unit thingy, looks good.

the Z/Stencil ROP units, and Colour ROP units are 2x-4x faster... That seems like one of the major changes over the 5870's.

2GB ram... for people that run 2-3 monitors these cards will be golden, in that they wont be ram limited by the high res. Compaired to the 1GB cards of the 5xxx series.

one of the slides just says up to 3x tessellation performance of the 5870. That doesnt sound like that much of a improvement though.


Im really looking forwards to benchmarks.

There is a lot of talk on Chiphell about very sparse concrete data on performance and it's rumored AMD is dead-set on nvidia not getting any specifics about 6970/6950 performance.

Chiphell also has some rumors that nvidia wants to release an overclocked 580 at 6970's release. As I said before, 6970 is likely going to overall be faster than a 580.

It's worth noting when debating 580's performance, the 580 is much like what the GTX260 216core was to the GTX260, with the addition of improved thermals. Clock for clock the 580 is 6% faster than a GTX 480, which is not surprising as it is a 480 with the full 512 shaders available which gives it that 6% lead.

Whereas from rumors and some of these leaked specs the 6970 has some major changes, new additions and is a bigger die compared to the 5870. I think there is more room for a bigger performance leap than what nvidia could do with what is basically a full shader count GTX 480 in the 580.

AMD also has to contend with NV managing to get out a dual-gpu card, they could pull that move around 6970's launch and prior to 6990's launch. NV is doing a big push to try and get themselves back the halo crown and it looks like AMD is trying to counter their efforts by keeping solid performance data on their new parts out of nvidia's hands.

 

[T2k]

There is a lot of talk on Chiphell about very sparse concrete data on performance and it's rumored AMD is dead-set on nvidia not getting any specifics about 6970/6950 performance.

Chiphell also has some rumors that nvidia wants to release an overclocked 580 at 6970's release. As I said before, 6970 is likely going to overall be faster than a 580.

It's worth noting when debating 580's performance, the 580 is much like what the GTX260 216core was to the GTX260, with the addition of improved thermals. Clock for clock the 580 is 6% faster than a GTX 480, which is not surprising as it is a 480 with the full 512 shaders available which gives it that 6% lead.

Whereas from rumors and some of these leaked specs the 6970 has some major changes, new additions and is a bigger die compared to the 5870. I think there is more room for a bigger performance leap than what nvidia could do with what is basically a full shader count GTX 480 in the 580.

AMD also has to contend with NV managing to get out a dual-gpu card, they could pull that move around 6970's launch and prior to 6990's launch. NV is doing a big push to try and get themselves back the halo crown and it looks like AMD is trying to counter their efforts by keeping solid performance data on their new parts out of nvidia's hands.

Exactly. Anyone paid attention last year knows Fermi had very little headroom while ATI had pretty much free ticket to do whatever they want to do.
OC'd 580 or not I think AMD will pretty much own Nvidia for the next 6-8 months, just as I expect it...

 

[tyl998]

It's not really CSAA vs MSAA based on MSAA is the foundation for Coverage sample AA. CSAA is really an enhancement to improve the quality of MSAA. In the past, all one had was adding multi-sampling and when using higher resolutions, the hit, specifically memory footprints of going to x4 to x8 multi-sampling was high, if one didn't have enough memory. What CSAA did was help address this and allows more flexibility to improve IQ for many.

This is referring to RAM on your GPU right? Your system RAM does not help at all right?

 

[happy medium]

So is the NDA lifted on the 22nd or is this another mabe ,could be, rumor thread?