[AnarchX]

At Chiphell was posted an interesting 3DMark Vantage screen which is supposed to be from a HD 6850. The interesting stuff are the feature test.


Feature Test 1 (texture) & Feature Test 6 (perlin noise ALUs) ~ HD 5750 level
Feature Test 2 (Pixel ROPs) & Feature Test 5 (GPU particles, with high vertex shader load) ~ HD 5850 level

[SimBy]

Originally Posted by Mianca

Viewing those two slides side-by-side, I can just hope that the >150W Max Board Power Barts XT card will actually trump GTX 460 by a fairly huge margin ... anything less than ~ HD 5850 performance (i.e. ~ GTX460 + 20%) would practically correspond to a decrease in Perf/W ...

I really couldn't care less about the controversial branding etc. - but Perf/W is something I (and a lot of reviewers) keep a very close eye on ...

Interesting point. I don't think there's much if any improvement as far as performance/W goes. In contrast to substantial performance/mm2 gain.

Unless Barts XT is faster than Cypress XT but I kinda doubt that.

[Mianca]

Originally Posted by SimBy

I don't think there's much if any improvement as far as performance/W goes.

Yeah, it would seem reasonable that clocking a ~230mm² chip high enough to reach the performance segment they indicated actually came at a considerable Perf/W penalty.

I'm really interested in how they'll pull off a 2xCayman Antilles card with Cayman XT supposedly drawing somewhere between 225W and 300W ...

[Kaotik]

Originally Posted by Mianca

Yeah, it would seem reasonable that clocking a ~230mm² chip high enough to reach the performance segment they indicated actually came at a considerable Perf/W penalty.

I'm really interested in how they'll pull off a 2xCayman Antilles card with Cayman XT supposedly drawing somewhere between 225W and 300W ...

Handpicked Cayman Pros?

[Arnold Beckenbauer]

Originally Posted by Jawed

http://www.microsofttranslator.com/b...a%3dpage%253D1


i.e. 1/2 for POW and 1/2.5 for SIN.

For HD5870, GPU Shader Analyzer shows 129 ALU instructions for the POW shader, i.e. half MUL rate. Here's how 4 POW instructions compile:
...

SIN is a bit of a tangle to compile in GPUSA, but with a bit of fiddling it comes out as 162 instructions, i.e. 1/2.5.

...
As far as I can tell the throughput for XYZT would be the same as XYZWT in all three of these tests.

But I think it rules out XYZW with emulated transcendentals.

HD4850 (700 MHz):

--------------------------------------------------------------------------
                             Instruction Issue
--------------------------------------------------------------------------
512      109.5269       ADD          4         64
512      109.7506       SUB          4         64
512      109.7168       MUL          4         64
512      109.6555       MAD          4         64
512      109.3467       EX2          4         64
512      109.3772       LG2          4         64
512      55.4172       POW          4         64
512      136.9928       FLR          4         64
512      136.7206       FRC          4         64
512      109.6678       RSQ          4         64
512      109.7598       RCP          4         64
512      44.0939       SIN          4         64
512      44.0870       COS          4         64
512      104.8287       SCS          4         64
512      111.4375       DP3          4         64
512      111.4533       DP4          4         64
512      87.8590       XPD          4         64
512      109.5514       CMP          4         64

--------------------------------------------------------------------------
                     Scalar vs Vector Instruction Issue
--------------------------------------------------------------------------
512      110.9219       ADD          1         40
512      108.3673       ADD          4         40
512      110.6266       SUB          1         40
512      108.3721       SUB          4         40
512      110.9219       MUL          1         40
512      108.3673       MUL          4         40
512      110.7065       MAD          1         40
512      108.4488       MAD          4         40

[Mianca]

Originally Posted by SimBy

Unless Barts XT is faster than Cypress XT but I kinda doubt that.

Maybe it's about as fast?

[UniversalTruth]

Originally Posted by Mianca

Maybe it's about as fast?

But then, why is it positioned against so weaker card like GTX460?

[Mianca]

Link.

Awesome achievement in Perf/mm² and (slight) enhancement in Perf/W over Cypress XT if true ...

Moreover, notice the interestingly new shader counts in the chart I linked above?

AMD marketing seriously deserves a "medal of advanced smoke generation" ...

[UniversalTruth]

Uploaded with ImageShack.us


Looks quite promising but we will see.... soon...
Antilles looks like really two times Barts.

[chavvdarrr]

Originally Posted by Mianca

Maybe it's about as fast?

http://gathering.tweakers.net/forum/...34952#34834952
the correct link

[UniversalTruth]

And what exactly will the performance difference between Cayman XT and Antilles be? 5%, 10%?

[AwfuL]

Forget about that chart, guys. It's one member's personal speculations. It's not based off of any leaks.

[Unknown Soldier]

Originally Posted by UniversalTruth

And what exactly will the performance difference between Cayman XT and Antilles be? 5%, 10%?

A better question would be why is there only one Antilles? I thought there was supposed to be a Pro and XT Antilles.

[Mianca]

Originally Posted by AwfuL

Forget about that chart, guys. It's one member's personal speculations. It's not based off of any leaks.

It certainly took some balls to post completely different numbers when everyone is discussing whether Barts XT has 960 or 1280 shaders etc. ...

Makes me wonder.

[Harison]

Originally Posted by Unknown Soldier

A better question would be why is there only one Antilles? I thought there was supposed to be a Pro and XT Antilles.

Exactly, I was earlier thinking about ~500$ for Antiles Pro and ~600$ for XT, add ~50$ if tweakers.net prices are correct.

[rpg.314]

Originally Posted by UniversalTruth

Uploaded with ImageShack.us


Looks quite promising but we will see.... soon...
Antilles looks like really two times Barts.

IMHO, a fake.

[trinibwoy]

Originally Posted by Mianca

It certainly took some balls to post completely different numbers when everyone is discussing whether Barts XT has 960 or 1280 shaders etc. ...

Makes me wonder.

It doesn't take balls. Just a keyboard and an internet connection

[-The_Mask-]

Originally Posted by rpg.314

IMHO, a fake.

They are just my ideas, written in notepad.

[RedVi]

Can't say I find that chart credible at all. Would be nice in most regards if true, the exceptions of course are Antilles being Barts x 2 and Juniper being renamed to 67xx. I don't wish for either of those to become reality.

[Kaotik]

Originally Posted by UniversalTruth

Uploaded with ImageShack.us


Looks quite promising but we will see.... soon...
Antilles looks like really two times Barts.

It indeed says that Antilles = Barts X2, but drivers say Cayman

On other things, what are the chances that Juniper has been updated at least with HDMI 1.4? Since the hardware can do 3D anyway

Also, does 1680 for example split with 4D VLIW?

edit:
scratch that, as it's fake

[-The_Mask-]

Originally Posted by Mianca

It certainly took some balls to post completely different numbers when everyone is discussing whether Barts XT has 960 or 1280 shaders etc. ...

Makes me wonder.

[rpg.314]

http://www.anandtech.com/show/3976/a...ches-next-week

[psolord]

Originally Posted by -The_Mask-

They are just my ideas, written in notepad.

Nice ideas mate, but I almost had a stroke when I saw 450$ for the Cayman and 275$ for Barts. lol

[Alexko]

Official slides say Cayman has 6GT/s memory, and just 1GB of it. So this is almost certainly wrong.

EDIT:

Originally Posted by -The_Mask-

They are just my ideas, written in notepad.

Well that settles it!

[DavidGraham]

Originally Posted by jaredpace

Clusters of SIMDs arranged in 2 or 3 blocks versus a single stack as in the squarer Juniper & Rv770 die.

Juniper had a rectangular look too :


that suggests Cayman could be rectangular too .