[WCCF] AMD Radeon R9 390X Pictured

[soccerballtux]

Again it seems they think 390X is Fiji.

The highly expected card is Fiji / Fury.

The highly expected AMD card is very short and water cooled.

No one who is paying any attention should have thought that powercolor card was the "highly expected" card from AMD.

Strange.

honestly anything from AMD at all is highly expected IMO

 

[soccerballtux]

wasn't aware that AMD was going to be the only seller for the WCE version

so optimistic that the rumors that Fiji being slower than 980Ti is the cut down version and not the WCE version

I'm no where near the market for these cards but AMD needs a win after getting beat up so badly by Maxwell

what rumors? with 4096 shaders, that doesn't make even the least bit of sense

 

[crashtech]

I've seen that 4096 shader figure before, is it official?

 

[dzoni2k2]

I'm quite sure its not official.

Wow my 1st post.

 

[AnandThenMan]

wasn't aware that AMD was going to be the only seller for the WCE version

Is AMD doing this to prevent leaks? So they can control every aspect of the design? Both? None? So many questions surrounding this whole launch.

 

[crashtech]

I'm quite sure its not official.

Wow my 1st post.

Well I was wondering because the die shots, coupled with assertions that there has been a density increase, would lead one to imagine there might be even more than 4096. Anybody have a link to a credible source of that number?

 

[IEC]

Well I was wondering because the die shots, coupled with assertions that there has been a density increase, would lead one to imagine there might be even more than 4096. Anybody have a link to a credible source of that number?

It's equally likely that it has 4096 as it is that it has even more than that. No authoritative information either way until the launch

 

[RampantAndroid]

Is AMD doing this to prevent leaks?

Yeah, the last thing you want is water all over the inside of your case.

Could be a supplier agreement that AMD has with whoever made the cooling setup.

 

[Cloudfire777]

wasn't aware that AMD was going to be the only seller for the WCE version

so optimistic that the rumors that Fiji being slower than 980Ti is the cut down version and not the WCE version

I'm no where near the market for these cards but AMD needs a win after getting beat up so badly by Maxwell

If I dont remember wrong the sites said the last few days that Fury X (Fiji XT) will be water cooled/reference only (Like Titan X was) while Fury (Fiji Pro) is open for AIBs different designs.

But like Titan, AMD may open up for other designs with Fury X later on.

 

[AnandThenMan]

Yeah, the last thing you want is water all over the inside of your case.

Your day job is safe. On the WC is this the first time a board is specifically designed for it? I mean the size of the card could not exist without a water cooled system right?

 

[Creig]

wasn't aware that AMD was going to be the only seller for the WCE version

I guess it's possible, but AMD has generally let their AIBs actually produce the cards while AMD focuses on just building the GPUs. I think the last actual AMD branded card I've heard about was a 270X. But maybe they're going to try and generate brand recognition by producing their own flagship cards with a large AMD logo on the box.

 

[RampantAndroid]

Your day job is safe. On the WC is this the first time a board is specifically designed for it? I mean the size of the card could not exist without a water cooled system right?

I'm guessing the size of the card shrinking is an artifact of needing less power regulation on board, since HBM is supposed to be way more efficient (and it is also smaller.) I wouldn't be surprised if we see NV cards also shrink a bit with HBM2.

 

[dzoni2k2]

Well I was wondering because the die shots, coupled with assertions that there has been a density increase, would lead one to imagine there might be even more than 4096. Anybody have a link to a credible source of that number?

Agree. If HD4000 series launch taught us anything is that AMD is quite capable of surprising everyone Would 5632 shock me? No.

 

[krumme]

It's equally likely that it has 4096 as it is that it has even more than that. No authoritative information either way until the launch

Naa but at =>600m2 and probably even denser and less area for mem controller what is the chance of 4096 shaders?
I would say its less than 5%
Performance we dont know because of primarily freq target and perf/w target but we are as close as confirmation we can be before facts hits us this is more than 4096. I will eat my hat if its less than 5k

 

[RussianSensation]

Well I was wondering because the die shots, coupled with assertions that there has been a density increase, would lead one to imagine there might be even more than 4096. Anybody have a link to a credible source of that number?

There is no credible source for 4096 shaders. It's basically speculation based on a collection of information.

The 1st point of reference are November 11, 2014 SiSoftware Sandra leaked benchmark scores which show:

Result ID AMD Radeon Graphics Processor (4096SP 64C 1GHz, 4GB 1.25GHz 4096-bit) (OpenCL)

The 2nd point of reference are some slides that may or may not have been made for AIBs by AMD. These showed up a long time ago as well.

The 3rd point of reference for 4096 shaders, 256TMUs is the large die size Lisu showed this week. Recall that AMD managed to increase shaders and textures 37.5% and double the ROPs with R9 290X over Tahiti, but die size only grew 24.4% (438mm2 vs. 352mm2).

Therefore if the die size > 550mm2, it becomes more plausible how a 4096 shader card, with 256 TMUs and 64-96 ROPs could actually exist.

More than 1 year ago, a 4224 shader Bermuda XT was being thrown around but the specs had GDDR5 memory. Therefore, these are for sure made up.

 

[maddie]

Well I was wondering because the die shots, coupled with assertions that there has been a density increase, would lead one to imagine there might be even more than 4096. Anybody have a link to a credible source of that number?

Posting this again to help those who might not have seen it.

Take as a possibility.

Points to consider:
1) GloFlo 28nm has 10-20% better density than TSMC
http://www.eejournal.com/archives/articles/20111114-gate/
The result is an overall density penalty of 10-20%
This is TSMC penalty vs GloFlo
2) AMD has developed new high density libraries [known internally as a Moonshot program]
3) Fiji is estimated by some to be 600mm2
4) HBM allows lowering of memory controller space requirements
5) Synapse was advertizing 80+ blocks for graphics processor tapeout [80+ x 64 = 5120+ shaders]

If you do the calcs than 2 X shaders is possible
Possibility of Fiji = or > 295x2

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

Fiji = 600mm2 [Assumed]
Lets assume a 15% increase in density [middle range for GloFlo improvement]
Non shader < 146mm2 [HDL, HBM memory controller]
Say 15% for Gloflo and and say further 10% for memory controller = 146 x 0.85 x 0.9 =111mm
Shader = 600-111 = 489mm2
With 15% density increase, we get 482 x1.15 = 554mm2 [TSMC equivalent]
Compare with Hawaii = 554/292 = 1.9x2816 = 5350 shaders [80+ CU]

Note* This ignores any HDL improvements which would give even more.

 

[.vodka]

The 350mm2 chip mentioned in that slide is believed to be Tonga. 50*64=3200sp while we know that Tonga is a 32CU design (2048sp). I don't think the blocks mentioned there is related to the number of CUs.

But the rest of your speculation has basis for more than the rumored 4096sp.




Ah, june 16, it can't come soon enough. Anyone wanna build a time machine? It could be of use...

 

[soccerballtux]

It's equally likely that it has 4096 as it is that it has even more than that. No authoritative information either way until the launch

I agree with this, which is why I go with the 4096 figure. can't be wrong

 

[soccerballtux]

There is no credible source for 4096 shaders. It's basically speculation based on a collection of information.

The 1st point of reference are November 11, 2014 SiSoftware Sandra leaked benchmark scores which show:

Result ID AMD Radeon Graphics Processor (4096SP 64C 1GHz, 4GB 1.25GHz 4096-bit) (OpenCL)

The 2nd point of reference are some slides that may or may not have been made for AIBs by AMD. These showed up a long time ago as well.

The 3rd point of reference for 4096 shaders, 256TMUs is the large die size Lisu showed this week. Recall that AMD managed to increase shaders and textures 37.5% and double the ROPs with R9 290X over Tahiti, but die size only grew 24.4% (438mm2 vs. 352mm2).

Therefore if the die size > 550mm2, it becomes more plausible how a 4096 shader card, with 256 TMUs and 64-96 ROPs could actually exist.

More than 1 year ago, a 4224 shader Bermuda XT was being thrown around but the specs had GDDR5 memory. Therefore, these are for sure made up.

far be it from me to call you a troll or a fanboy, you know your stuff!!!

so did you emmigrate to CA from RU and now you just post on ATF all day?

 

[maddie]

The 350mm2 chip mentioned in that slide is believed to be Tonga. 50*64=3200sp while we know that Tonga is a 32CU design (2048sp). I don't think the blocks mentioned there is related to the number of CUs.

But the rest of your speculation has basis for more than the rumored 4096sp.




Ah, june 16, it can't come soon enough. Anyone wanna build a time machine? It could be of use...

I thought about Tonga but decided to pursue the possibility of another option.

The Wccftech article claiming Tonga as the 350mm2 die is 9 months old. Tonga would have taped out long before that.

Anyhow, its all speculation, hopefully enlightened speculation.

 

[crashtech]

There is no credible source for 4096 shaders. It's basically speculation based on a collection of information.

The 1st point of reference are November 11, 2014 SiSoftware Sandra leaked benchmark scores which show:

Result ID AMD Radeon Graphics Processor (4096SP 64C 1GHz, 4GB 1.25GHz 4096-bit) (OpenCL)

The 2nd point of reference are some slides that may or may not have been made for AIBs by AMD. These showed up a long time ago as well.

The 3rd point of reference for 4096 shaders, 256TMUs is the large die size Lisu showed this week. Recall that AMD managed to increase shaders and textures 37.5% and double the ROPs with R9 290X over Tahiti, but die size only grew 24.4% (438mm2 vs. 352mm2).

Therefore if the die size > 550mm2, it becomes more plausible how a 4096 shader card, with 256 TMUs and 64-96 ROPs could actually exist.

More than 1 year ago, a 4224 shader Bermuda XT was being thrown around but the specs had GDDR5 memory. Therefore, these are for sure made up.

About the leaked slide showing 4096, could it be real, but not representative of the flagship card?

 

[Kenmitch]

About the leaked slide showing 4096, could it be real, but not representative of the flagship card?

It's possible. Savage die?

 

[alcoholbob]

4096 cores would be a double full Tonga. AMD has designed previous flagships this way, such as the 6970 being effectively double the 6870.

 

[Kenmitch]

4096 cores would be a double full Tonga. AMD has designed previous flagships this way, such as the 6970 being effectively double the 6870.

If it's there new trend then wouldn't they double down Hawaii?

 

[crashtech]

Maybe it is 4086, but with some 600mm^2, a possible density increase, and less area required for the memory controller, it seems like there is some room there for more.