AMD: R7xx Speculation

[satein]

I think no-x posted some 3dmark fillrate numbers a while back that pointed to 32.

Given that the test is for HD 4850 with GDDR3 memory... Is it possitlbe that it got hit by bandwidth limit?

I might be wrong anyway, it is only my assumption.

 

[Wirmish]

Ummm... Isn't Perlin Noise a shader intensive bench?

Oh... sorry.

Originally Posted by Hornet331
Simple mathematics

several fillrate tests from 4850 in the wild show ~19500MT/s and clock of the Card (TMUs) is 625Mhz

with 32TMUs and 625mhz you get a theoretical fillrate of 20.000MT/s
with 40TMUs at 625mhz you get 25.000MT/s.

i dont think RV770 is that inefficent.

Since there are 800 SP, 40 TMUs makes more sense.

 

[Wirmish]

EE Times - 06/16/2008

The decision to use a two-chip strategy for the high end was made more than two years ago, based on an analysis of yields and scalability. It was not related to AMD's recent financial woes, said Rick Bergman, general manager of AMD's graphics division.

AMD says its 4850 device at about 110 W and $199 will deliver about 75 percent of the performance of Nvidia's high-end GTX280 which costs $649 and dissipates 236W. Two of the AMD parts on a board will hit graphics benchmarks about 30 percent higher than the Nvidia device, the AMD spokesman added.

A 4870 version of the product sporting slightly higher performance will cost $299. Both AMD chips are made in a 55nm process, compared to 65nm for the Nvidia chips, and measure about 16x16mm compared to about 24x24mm for the Nvidia part.

Bergman said the AMD focus on a more mainstream design will enable it to roll out this fall a version for notebook computers that consumes less than 70W. "There's no way this new Nvidia core will be in notebooks this fall," Bergman said.

 

[itaru]

http://translate.google.co.jp/trans.../2008/0617/amd.htm&sl=ja&tl=en&hl=ja&ie=UTF-8

AMD, 200 dollars to achieve a 1 TFLOPS Vietnam Radeon HD 4850 to June 25 shipment Movies and Games to the fusion of "Cinema 2.0" to promote

blend of movie and game plan to achieve Cinema 2.0
1 TFLOPS more than Radeon HD 4800 series, said it shipped next week
depart from the traditional design of the new GPU architecture

new ruby demo
wow!!

 

[Nuker]

Has AMD aquired VALVe recently? Nonetheless that demo looks really nice, a substantial improvement over the previous ruby demos. I wonder if there is some raytracing involved with that demo, considering those accurate reflections would be very hard to represent with animated textures or other raster techniques, I presume.

 

[Mat3]

From the same EETimes article mentioned above, this quote is the one that stood out for me:

"I predict our competitor will go down the same path for its next GPU once they see this," Bergman said. "They have made their last monolithic GPU."

 

[Anarchist4000]

Instead it will launch a part aimed at the graphics mainstream with a proprietary interconnect that links two chips on a board to deliver top performance.

Both AMD chips are made in a 55nm process, compared to 65nm for the Nvidia chips, and measure about 16x16mm compared to about 24x24mm for the Nvidia part.

Found those two quotes in the article somewhat interesting. 256mm sounds about right and a confirmation the interconnect is there. They could be talking about the crossfire link but I'd assume it's something more.

 

[Blacklash]

Wonder what the odds are that the HD 4870 will be able to beat the GTX 280 in most games. I am looking at 1920x with high filtering. Ideally, I'd like to use 8-16xAA. I want the fastest single GPU I can buy.

 

[nAo]

Has AMD aquired VALVe recently? Nonetheless that demo looks really nice, a substantial improvement over the previous ruby demos. I wonder if there is some raytracing involved with that demo, considering those accurate reflections would be very hard to represent with animated textures or other raster techniques, I presume.

Was the demo developed by Valve?
IIRC AMD and Valve always had a good relationship.

 

[nAo]

I'm sure those numbers are skewed towards RV770 but they look damn impressive!

Someone at NVIDIA is crying right now..

 

[Wirmish]

Read this : AnandTech

It's useful to point out that, in spite of the fact that NVIDIA doesn't support DX10.1 and DX10 offers no caps bits, NVIDIA does enable developers to query their driver on support for a feature. This is how they can support multisample readback and any other DX10.1 feature that they chose to expose in this manner.

"We support Multisample readback, which is about the only dx10.1 feature (some) developers are interested in. If we say what we can't do, ATI will try to have developers do it, which can only harm pc gaming and frustrate gamers."

So nVidia can use "multisample readback" (the same way ATi use it in Assassin's Creed to accelerate the AA stuff) without the need for DX10.1, for all TWIMTBP games !

But at the same time nVidia is not ashamed to "ask" UbiSoft to remove his DX10.1 pathway for the Radeons 3000 with a fucking patch.

We now know why 4xAA si so fast on nVidia's cards...


"The Way It's Meant To Play Dirty"

 

[Andrew Lauritzen]

So nVidia can use the "multisample readback"

Clearly I'm just being dense here, but what about MSAA readback is a DX10.1 feature? You can read back MSAA samples in DX10. You can't read back a MSAAed depth buffer, but I don't believe you can do that on NVIDIA in DX10 either. What do they mean here when they say they already support DX10.1 "multisample readback"? Are they just mincing words and saying that "hey, we support a feature that was optional in DX10 and required in DX10.1" similar to saying "we already support fp32 filtering" (also option in 10, required in 10.1)?

 

[CarstenS]

Found those two quotes in the article somewhat interesting. 256mm sounds about right and a confirmation the interconnect is there. They could be talking about the crossfire link but I'd assume it's something more.

I don't know if there's marketing speak at work, but...
"Instead it will launch a part aimed at the graphics mainstream with a proprietary interconnect that links two chips on a board to deliver top performance."

Some "proprietary interconnect" has also been present on current and last-gen cards from AMD - namely their crossfire-connection.

So, I'm not altogether doubting this theory, but I'd rather see it in practice, having been burnt by too many a marketing person's speak in the past by both vendors.

 

[AnarchX]

I'm sure those numbers are skewed towards RV770 but they look damn impressive!

http://pc.watch.impress.co.jp/docs/2008/0617/amd_14.jpg

Someone at NVIDIA is crying right now..

More interessting on this slide, are the "16 mm", so RV770 is 256mm², like several times said.

 

[Wirmish]

This is incredible -> Short video

This is real-time rendered !

 

[Entropy]

More interessting on this slide, are the "16 mm", so RV770 is 256mm², like several times said.

And this is a marketing slide, so you can be absolutely sure that the AMD number is rounded down, whereas the nVidia one is rounded up. Really, until someone goes for the chip with a micrometer, we just won't know the decimals on the 16.xx.
I wonder how they measure "performance" in that slide. FLOPS probably, since they state that the RV770 is 2.8/((24/16)^2)=2.8/1.5^2=1.244 or 25% faster than the GT200. Somehow I doubt that is true, although it would stir things up a bit at the highest end.

 

[slapnutz]

This is incredible -> Short video

This is real-time rendered !

Can open it. Is there another way to get to it?

 

[AlphaWolf]

Can open it. Is there another way to get to it?

Here's the page its linked from

You probably just need the latest flash drivers.

 

[Raqia]

That was really short. AMD's got its presentations up here:

http://www.amd.com/us-en/Corporate/AboutAMD/0,,51_52_15438_15106,00.html?redir=cin01

with some nice photos of its other demo which stars a Scorpion it seems:

http://www.flickr.com/photos/amd_unprocessed/

So are we more likely to see a set of miracle drivers from nVidia or ATI in the coming weeks for their latest cards?

This is incredible -> Short video

This is real-time rendered !

 

[Ninjaprime]

I'm sure those numbers are skewed towards RV770 but they look damn impressive!

Someone at NVIDIA is crying right now..

Interesting... look closely at the RV770 die shot. I see 6 rows of 16 processors. 480 SP confirmed? Or am I just seeing things?