[WCCF] AMD Radeon R9 390X Pictured

[MrTeal]

What was the pic with the curved copper areas?

It was a visual artifact of the interposer and thermal compound.

 

[maddie]

Maddie



Are you saying Fiji will be faster than the 295x2? What?

The worst thing we can do is set the bar high. Actually, the rumors of a slow Fiji can help it have a huge buzz if it launches faster than the 980ti and TitanX.

I think the 980ti is in short supply and surely the people who were gonna wait, they will wait.

Possible

Hawaii = 438mm2 [fact]
Shader portion = 292 mm2 [2/3 area]
Non shader = 146mm2

Fiji = 600mm2
Lets assume a 10% increase in density.
Non shader < 146mm2 [HDL, HBM memory controller]
Say 10% for HDL and and say further 10% for memory controller = 146 x 0.9 x 0.9 =118mm
Shader = 482mm2
With 10% density increase, we get 482 x1.1 = 530mm2
Compare with Hawaii = 530/292 = 5120 shaders [80 CU]

Note:
This is a 10% density increase.
15% is many more shaders. 2 X Hawaii

 

[LTC8K6]

It was a visual artifact of the interposer and thermal compound.

Could be, but I really don't think so.

Looks different to me.

http://forums.anandtech.com/showpost.php?p=37445180&postcount=1751

 

[crashtech]

@maddie, where does the 10% density increase come from?

 

[Azix]

I'm getting under 600mm^2 (599mm^2) measuring on screen, but yeah its around full maxwell size. It should beat titan-x handily but what really matters to most is what the cards below fiji xt can do and are priced at.

 

[sze5003]

So from what I've gathered here, I should be buying the 390x?

 

[maddie]

@maddie, where does the 10% density increase come from?

My previous post said If 10% density increase you get 5120 shaders. 15% gets many more. As to density increase, read below.

Carrizo analysis on Anandtech:

Same foundry and process as new GPUs.
The GPU portion should have less that 33% increase in density as it was already more dense than the CPU

http://www.anandtech.com/print/9337/amd-shows-freesyncoverhdmi-concept-at-computex-2015
A big part of the reduction in die area comes from the set of high density libraries being used by AMD. Above were three examples provided where >33% gains were made in silicon area. Typically using a high density library design is a double edged sword &#8211; it reduces die area and potentially leaves more area for other things, but the caveat is that it may be more prone to defects in construction, require additional latency or have a different frequency/voltage profile. AMD assures us that these changes are at least like-for-like but most of them contain other improvements as well.

It&#8217;s worth noting here that AMD has described the high density library project internally as the equivalent of a moonshot, essentially the developers were part of a &#8216;skunkworks&#8217; division attempting to make drastic changes in order to improve performance. The high density library is one such successful project from that.




I suggest everyone, especially the naysayers, look at the movie "SPARE PARTS"

 

[MrTeal]

Could be, but I really don't think so.

Looks different to me.

http://forums.anandtech.com/showpost.php?p=37445180&postcount=1751

It's not.

Look at the bottom left corner of the interposer, you can see that it extends to a square shape and the circle where the thermal paste touches is shinier. Also, you can see that small rectangle that's part of the interposer, half in the shiny part, half out. In the good die shot, you can see that square in the bottom right.

 

[EightySix Four]

@maddie, where does the 10% density increase come from?

AMD is using newly developed high density libraries that are supposed to increase the design density ~10%.

 

[LTC8K6]

It's not.

Look at the bottom left corner of the interposer, you can see that it extends to a square shape and the circle where the thermal paste touches is shinier. Also, you can see that small rectangle that's part of the interposer, half in the shiny part, half out. In the good die shot, you can see that square in the bottom right.

When I rotate the bottom pic to the same orientation, it still looks different to me.

Oh well, it's not really important.

Let's hear about the cards.

 

[crashtech]

@maddie (and EightySix Four), the size increase alone might not make Fiji a Titan beater in my very rough estimations, but a 10% density increase sure could, and that isn't even taking into account the new bandwidth available. This is looking very interesting!

OK, back on the red team bandwagon, at least until about June 16th...

 

[Azix]

AMD is using newly developed high density libraries that are supposed to increase the design density ~10%.

and glofo 28nm SHP supposedly is more dense. So is this double improvement in density or do I fail?

 

[maddie]

AMD is using newly developed high density libraries that are supposed to increase the design density ~10%.

read this:
http://forums.anandtech.com/showpost.php?p=37451377&postcount=2107

 

[maddie]

and glofo 28nm SHP supposedly is more dense. So is this double improvement in density or do I fail?

You are correct in saying two separate density improvement factors in play and they are also cumulative.



Not double but maybe 10-15% improvement in density which can give double shaders on a 600mm2 die.

 

[Azix]

So better than 295x2 is possible. 14nm will be ridiculous.

 

[MrTeal]

When I rotate the bottom pic to the same orientation, it still looks different to me.

Oh well, it's not really important.

Let's hear about the cards.

Sorry, I highlighted the areas I'm talking about. You can see the surface feature of the interposer through the copper looking area.

 

[jackstar7]

[H] has TT's pics of a Powercolor hybrid... of some kind. It's stated to be a 390x.

http://www.hardocp.com/news/2015/06/03/powercolors_nextgen_radeon_r9_390x_pictured#.VW85_kYYG_s

 

[MrTeal]

So better than 295x2 is possible. 14nm will be ridiculous.

I'm not sure how much better performance we'll get at the top end with 14nm, at least at the start. It will depend on scaling.
It's taken years for AMD to attempt a 600+mm^2 die. Given the troubles that 14nm has had, it might be awhile before we see a die anywhere near size. I think we might see something closer to the 300-350mm^2 die that Tahiti introduced at. It should be faster and a lot more efficient, but maybe the massive increase we'd hope for.

 

[dacostafilipe]

Shader portion = 292 mm2 [2/3 area]

There's a picture of a Tonga die in the new Carrizo docs that you could use to calculate the % of shader area, PHY area, etc ...

 

[krumme]

I doubt we will get 10% density improvement for gpu logic using hdl libraries.
As i remember the hdl competences is comming from the gpu side so its knowledge implemented today. Secondly i would asume gpu logic is more repeated so to speak so the benefits is less. It might be wrong whats your take on it?

 

[krumme]

It is such a pleasure to see a big die. About time. Thank you - we have waited for that in 10 years or so. Now we get serious blasting perf and competition.
980 is soon a 250usd card when all the gf capacity starts flooding the market. Now nv gets competition from oil. Lol and we get some progress.

 

[Cloudfire777]

Here is a dead giveaway that R9 390X is R9 290X rebrand (as if we didnt have enough).

R9 390X board is marked as "LF R29FA"
Picture

They use this on PowerColour R9 290X cards
http://www.coolingconfigurator.com/step1_complist?gpu_gpus=1477

 

[.vodka]

Fair enough. But the chip itself, this tweaked Hawaii that's being thrown around as Grenada, could be for example pin compatible and there wouldn't be a need to change the PCB. It's a possibility.

You still can't assume it's a direct rebrand.

 

[maddie]

I doubt we will get 10% density improvement for gpu logic using hdl libraries.
As i remember the hdl competences is comming from the gpu side so its knowledge implemented today. Secondly i would asume gpu logic is more repeated so to speak so the benefits is less. It might be wrong whats your take on it?

AMD is getting 33%+ on the CPU size so I took the 10% as a reasonable # to aim for in estimating the possible increase in shaders.

Kaveri has 2.41 billion
Carrizo has 3.1 billion

Overall CPU + GPU 28% increase in density.

I think 10-15% for GPU is a fair assumption.

 

[krumme]

I'm not sure how much better performance we'll get at the top end with 14nm, at least at the start. It will depend on scaling.
It's taken years for AMD to attempt a 600+mm^2 die. Given the troubles that 14nm has had, it might be awhile before we see a die anywhere near size. I think we might see something closer to the 300-350mm^2 die that Tahiti introduced at. It should be faster and a lot more efficient, but maybe the massive increase we'd hope for.

I agree but how much is big die maxwell selling? I mean if there is a market for those crazy high cost cards we might see more than 350mm2 14nm from day one. Not that i expect something like 600mm2 from day one. Certainly not. Perhaps if mobiles is less attractive in 2 years