AMD: R9xx Speculation

Cool. Do you know how many guys love Downsampling? And do you know that only nVidia's gpu capable of it under Vista and Win7 right out of the box?
I don't care about one of the few specific IHV Feature. nVidia gave me SGSSAA in DX10 and DX11. Something AMD is not doing until today.

 

On the topic of HAWX 2, Nvidia are currently giving away a copy of the game with every single GTS 450 and GTX 460 sold in the UK, despite the fact it isn't even launched yet, so there's definitely some sort of "co-marketing agreement" going on there.

 

The question was in regards to tessellation. Ignoring that your non-answer is a politicians answer. "I don't know what can be done, but something must be done."

It was an interesting presentation on the tessellation method, but I don't see how that means AMD is not interested in tessellation. It's well known Nvidia has a larger developer relations staff relative to AMD. This isn't tessellation specific. I've got no argument if someone thinks AMD should invest more in dev rel, but think there's a leap to jump to the conclusion that AMD isn't interested in promoting "reasonable" tessellation. The issue is what's reasonable.

It's obvious from the presentation Nvidia thinks generating > 6000 triangles from a single triangle patch is still limiting while AMD disagrees.

 

In what DX10 and DX11 games is it feasible to use supersampling though? AMD supports SGSSAA in all DX9 titles and they even automatically adjust the LOD bias (unlike Nvidia) so I'd hardly say they're far behind there, not by a huge margin anyway. Offering basically "free" AA with fantastic quality to damn near every title under the sun is a much bigger deal in my book and the quality is fantastic.

If you want to see some examples, I linked them in a post in the alternative AA thread:

 

That makes much more sense. I was quite puzzled for a while at the notion of "controller" doubling in size for a 15% increase in clock rate.

 

Holy crap, he's gone! YAY!

 

Get some sweet looking demos out using adaptive tessellation that run well on their hardware? What better way to prove their point than to produce something that runs faster and looks better than the uber-tessellated stuff nVidia is pushing? Then everybody will see for themselves that nVidia's approach is wasteful and pointless.

Well I think there are some things better on paper but AFAIK the only discernable difference was the filterinq quality that was fixed with NI.

Ok, I thought the change in default quality only came to light after the NI launch.

 

Or is it maybe that HD5850 should have that wattage if all 1600SPs were enabled on same clock As many expected HD4800 type relaunch, but poor 40nm yields borked that.

It's a hype thing cause when you look to buy "that card" usually dont go around snooping for other card reviews, and if youre not constantly follow graphics market you'll be quite happy with your buy ... (or maybe really different settings)

Edit: Hey but it has same 62fps @19x12

What makes this chips so large?! It doesn't even support FP64, nor even have 1280SPs, but accordingly it sports somewhat smaller Redwood MC (x2 to become 256-bit, and not supporting differential gddr5 signaling makes him 50% smaller than one in Cypress/Juniper).

So what's the legend behind it? It's not like only one more tessellation dispatcher aka. RPE (what it stands for?) needs to consume enormous die space.

In fact it's not all so shiny ruby as you can see from above. But they tweaked performance as much as they could on 1120SPs that wont go much further 960MHz almost reaching HD5870 consumption with 60MHz increase. But it's great enough as gaming solution in ring vs. GTX460 1GB.

All they improved is "broken" AF filtering that shouldn't be broken in first place and MLAA (application) looks like technique we saw 12+ yrs ago with jpeg popularization. So we could at least ask why it aint be there before? It's definetely cheap enough trick that done almost same job as EdgeAA with "huge improvements" on reducing shader utilization to get job done "with similar results". At least i hope we'll saw thus MLAA enabled in drivers for HD5000 series and only then we might get better comparison on HD6800 improvements over HD5700/5800, beyond already mentioned second tessellation dispatching coprocessor/unit - RPE.

Well smaller die space is great for (now) AMD, but that doesnt meant too much for me as consumer because HD5850 on launch had Dirt2 coupon and was selling at 260USD promo price while a year later we get small tesselation+filtering improvements that should be there in first place a year ago. And new DirectCompute application gimmick - MLAA. So all we can hope is that AMD can better reduce prices on Barts than it could on x1.33 larger Cypress.

 

Hi Dave,

I think the $240 price is more welcoming and i still think this should be called the 6770 1GB or 6790 1.5GB.....

I think the most disappointing thing is the very very ....very poor overclocking headroom, what happened? Did the Redwood mc destory every thing? The next most disappointing thing...there is virtually no hardware arch improvements....why is it even designated a 6 series...i cannot tell...

Dave...could you do us a favor and drop a hint about the 69xx cards...will those be worth waiting for...if we have a budget of $350? going from a 4870 1GB...considering now a 5870 or 6850....i know 'em will be more affordable next month....'em will be a good step up from 4870....

 

Many are complaining of the pink tint problem with Cat 10.10. It happens when you move the slider for the Hue/Color, etc. under the Color Tab. If you hit the default button the tint will disappear. However, you can't use the Hue/Color, etc. This needs to be fixed.

 

Nope and nope. There are changes between RV770 and Cypress.

The engine configuration....?

Large in relation to what?

Personally I have no concept of what an "RPE" is. I also don't know where you got the notion that there is another tessellation unit.

Research papers on the concept of MLAA are fairly recent. The hardware changes in the DX11 pipeline make it more efficient to implement. We’ve gone from concept to high performance implementation pretty quick here.

As for other differences, we have higher tessellation performance, UVD3, DP1.2 readiness, HDMI 1.4a with windowed hardware overlays in Stereo 3D modes, native hardware wide gamut color correction, to name a few things.

 

Are you certain i heard it's only 8% on IT equipement. So noting from shopping-holic tourism after all. But great prices thou, only 10€ more than in A and they have 20% VAT.

 

Large considering only 70% instead 80% active engines

So it's still one coprocessor just with improved ipc?

Comprehensive one .... So would that meant that MLAA went enabled for all DC capable ex.ATi cards?

All that are nice improvements but does DP1.2 bandwidth doubling require 2x die space, and after all didnt UVD in R600 consume less than 1-0,5% die space?

 

I tried it but don't see an option for MLAA.

Surface Format is there though. Or is it under Adaptive Multi-Sampling?

 

Should be under "smoothvision hd: Anti-Aliasing. There should be new checkbox next to "use application settings."

 

Anyone got a fix for the pink tint issue for Cat 10.10? People want to use the hue and color but if they do everything turns pink.

 

yea , I'm sorry , got a flu shot and now feeling the effects sooo my head isn't all there tod y !

 

Oh Mr.Dave , the things you go through reading our harsh comments ! , I just want you to know that we have nothing but the outmost respect for you and your company , and I hope you don't get upset with our constant interrogation .

(and yes, I am assuming the role of the good cop now )

 

It was driving me nuts too, thanks Thatdude90210.

 

He is looking at the aspect which happens to be important: what is fraking annoying to his eyeballs. I mean hell why not render in black and white... after all color is only "a single aspect."