[VC][TT] - [Rumor] Radeon Rx 300: Bermuda, Fiji, Grenada, Tonga and Trinidad

Page 13 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

RussianSensation

Elite Member
Sep 5, 2003
19,458
765
126
This is problably just another demo to showcase DX12. I want to see games.

That's a very good point. If a developer purposely makes a 5-10 min demo with an overwhelming number of draw calls, it would be akin to making a 5-10 min demo with billions of tessellation triangles to expose geometry bottlenecks. Unless we see a real world game based on a real world DX12 game engine showing a 4-10X increase in performance over DX11, it is just a theoretical showcase of what could happen should a developer overuse draw calls in a scene. Yet, if DX12 really provides a 4-10X increase in performance for AMD and NV in real world games, every non-DX12 card will be basically worthless for next gen gaming. In that case how would developers be able to make a next gen game that would run at 60 fps on a DX12 card (< 5% of the entire market) and just 6-15 FPS on DX11 cards (95% of the market)? I can't reconcile how that would work in practice.

If the differences were to be that dramatic in games, the developer would have to wait until DX12 becomes mainstream or he would be risking alienating 95% of the GPU market.

AMD claims full DX12 support for all GCN parts based on collaboration with MS on the new API starting spring 2014:
http://www.amd.com/en-us/press-releases/Pages/amd-demonstrates-2014mar20.aspx

Whether true or not we can't test yet until Windows 10 and a 1st DX12 game. Star swarm demo shows that GCN parts respond to DX12 but we don't know if there is any hidden benefits for DX12 optimizatoon between GCN 1.0-1.2, and 390X's presumed 1.3/2.0.
 
Last edited:

nvgpu

Senior member
Sep 12, 2014
629
202
81

tential

Diamond Member
May 13, 2008
7,348
642
121
There are some rumours that AMD may unveil more info on R9 300 series around GDC or Computex. However, based on comments from Gibbo and various sources, it seems the card would only launch June-July 2015, with wider availability expected for Q3. None of this information has been proven or disproven. AMD has been very secretive it seems. As far as 28nm vs. 20nm goes, the data doesn't align. For starters, it's difficult to believe AMD built a 550mm2 20nm chip but was limited to just 438mm2 on less expensive and higher yielding 28nm. Secondly, it seems 20nm is for low power devices. If AMD went with a 20nm power efficient design, I doubt it would have a 300W TDP, or the performance would be off the charts. 20nm and HBM -> the performance should be much more than 40-45% faster than 290X. Therefore, at least on the surface, the current leaks don't align with 20nm+HBM. This is just my opinion and I could be wrong.

780Ti was more or less ~ 2X faster than a 580 after a new architecture+node shrink. I do not think either GM200 or a 390X will be ~ 2x faster than 290X/780Ti. This is a compromise year for GPUs where NV and AMD are forced to use other techniques than a node shrink to improve absolute performance and perf/watt. NV has done a great job so AMD should also show some improvements over the 290X. Chances are the jump to 14nm flagship in 2017 will be much greater than GM200/390x will be over 290X/780Ti but it would also mean waiting ~ 2 years I would say. However, both cards should bring a much more substantial boost for 290 series and 780 series owners who decided to skip the 980.

So 2016 would be the first year we'd see a node shrink in god knows how long right?

Wouldn't it stand to reason that if we've waited all this time for a node shrink, mos tlikely AMD and NVidia both have GPU designs planned for this that were impossible without that node shrink that have been pushed back as they haven't been able to actually get it?
 

S.H.O.D.A.N.

Senior member
Mar 22, 2014
205
0
41
I'm rather more concerned with the release date of R3XX shifting further into the future with each iteration of the rumor. I mean, a June release would make it 19 months since R2XX came out.

It's nice that the 280/290 turned out to be such high value, well supported products, but come on, how long can you string those designs along?
 
Feb 19, 2009
10,457
10
76
R290X these days is only ~10% behind a 980, the gap is much closer since Maxwell's launch with GCN getting gains, particularly against Kepler, where its now >780Ti.

Ofc if its all GameWorks title, that gap is still ~20-25% (or worse).

Unless AMD manage to reduce power use on Hawaii considerably, they won't be able to run it at 1.2ghz and be competitive because the power use will skyrocket.

It looks like the full Tonga will replace the low end 280X class, while the new >550mm2 take the upper segment. That definitely leaves a gap for Hawaii to fill.

BUT, given that R390X will be such a massive chip, they may well end up harvesting it 2 extra steps down, getting R390 and perhaps R385X (which will compete against 980). So we may not get Hawaii recycling UNLESS their inventory is ridiculous. This could be the case as they ramped up to meet the mining demand but mining went bust so they have heaps of Hawaii dies lying around to offload for cheap.

So, a few scenarios could be realistic.
 

loccothan

Senior member
Mar 8, 2013
268
2
81
loccothan.blogspot.com
Only R390X and R390 will be New GCN 1.3, rest of the GPU's will be rebrands with better SOI technology ;-) So it will be better than the Old ones Perf/W Ratio as always.
Like the new FX 8370.
Now im waiting for my new (RMA, one of the fans broke) R9 280X XFX DD Black Edition (1080/6200) 225 heat.W. Im very pleased how XFX manage my RMA :D They just after week give me Brand new GPU !!! My old (almost 2 years) XFX DD GHzEd 1050/6000 (278 heat.W) have Fan issue.
XFX is the Best for Radeon...
My Next GPU will be XFX R390X
 
Last edited:

ocre

Golden Member
Dec 26, 2008
1,594
7
81
I was always sure that Hawaii would be rehashed into the new 300 series. But after reading so many people say that there was no way, I really started doubted myself.

The reason I was sure is because I thought Hawaii was originally planned for 20nm and was pulled in, just like nvidia did with maxwell.
My personal feeling was both Hawaii and tonga was pulled in. Fiji was the big daddy planned for 20nm and pulling it in takes some special engineering skill.

If there are any out there quick to denounce, just think for a minute. Gpus take years to develop and no one knew way back then that there would be a catastrophe at tsmc. AMD had been first with top to bottom node shifts for several generations and had not the 20nm node flopped, they would have done that same thing once again. I also believe AMD had a lot riding on the node shift and was counting on it but because they saw the shape of 20nm hp, they went ahead and pulled Hawaii in while working towards improvements for tonga and Fiji.

Its not really so far fetched when you consider that usually the past has shown that AMD launches entire generations on each node. If you can understand that Fiji was meant for 20nm and now AMD is going to produce it on 28, when you see Hawaii and tonga filling the rest of the series, it might make more sense.

I am not saying that AMD isn't working to improve these chips when they finally become official 300series chips. I think AMD will do as much as they can actually.

The one thing I worry about is this rumor that Fiji will be glofo 28nm. Has there been any glofo 28hp chips shipping? Is the glofo process mature? That could be a huge risk.

Does anyone know for sure this will be glofo instead of tsmc?

Being from a different fab entirely, that could make it really interesting.
Not saying that glofo 28nm is a bad thing, they might be able to produce even better chips. It's just that tsmc has a very very mature process now. That is something you could really use when your chips are over 500mm^2
 

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
However, based on comments from Gibbo and various sources, it seems the card would only launch June-July 2015, with wider availability expected for Q3.

Gibbo has a business to run and cards to sell right now. It would be to his benefit for his customers to believe there's nothing new imminent so they will buy what he has today.
 

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
I haven't heard that said about XFX for a long time. They really messed up with their overcheapened 7xxx coolers, but apparently enough time has elapsed that people have forgiven them for it.

He's talking about service and I've heard it mentioned often that they have the best service.
 

Headfoot

Diamond Member
Feb 28, 2008
4,444
641
126
There's 2 kinds of DX12 support, DX12 compatible and DX12 COMPLIANT. Big difference between the 2.

GPUs released 3 years ago certainly do no support the new hardware rendering features.

http://www.anandtech.com/show/8544/microsoft-details-direct3d-113-12-new-features

Maxwell GM20x does support all those new features, Nvidia even provides OpenGL extension equivalents to expose those features.

http://blog.icare3d.org/2014/09/maxwell-gm204-opengl-extensions.html

I assume you are insinuating the GCN does not have DX12 compliant features. Unless you can provide a source for this, this has not been confirmed anywhere. In fact, AMD has been making noises that they DO support some/all of these features. Especially considering that there are now 3 revisions of GCN floating around. My speculation is that GCN 1 covers a few of the features, GCN 1.1 (Hawaii, Bonaire) covers more/all of them, and GCN 1.2 Tonga covers all.

It's certain that both nVidia and AMD had some input on the features required to be "Full DX12" and that AMD most likely would have pushed to get GCN under that banner if at all possible
 

tviceman

Diamond Member
Mar 25, 2008
6,734
514
126
www.facebook.com
R290X these days is only ~10% behind a 980, the gap is much closer since Maxwell's launch with GCN getting gains, particularly against Kepler, where its now >780Ti.

Ofc if its all GameWorks title, that gap is still ~20-25% (or worse)

And, of course, GTX 980 overclocks 20% on air while 290x gets sweats magma getting 10% on air.

Anyways, I am sure AMD can respin / retool Hawaii to come with higher clocks without suffering higher power draw.
 
Last edited:

nvgpu

Senior member
Sep 12, 2014
629
202
81
GCN 1.0-1.2 certainly does not support those features, otherwise they wouldn't have been called new.

Maxwell was designed for DX12 and top to bottom support, GM200, GM204, GM206 supports them in hardware.
 

Paul98

Diamond Member
Jan 31, 2010
3,732
199
106
GCN 1.0-1.2 certainly does not support those features, otherwise they wouldn't have been called new.

Maxwell was designed for DX12 and top to bottom support, GM200, GM204, GM206 supports them in hardware.

Until you have evidence to support that please stop posting it.
 

waffleironhead

Diamond Member
Aug 10, 2005
7,049
554
136
I already gave the evidence here, I can't help it if some people are just blinded by their AMD fanboyism and can't accept the truth.

http://www.anandtech.com/show/8544/microsoft-details-direct3d-113-12-new-features

Nvidia exposes all those features in OpenGL extensions for GM20x GPUs.

http://blog.icare3d.org/2014/09/maxwell-gm204-opengl-extensions.html

Where is the evidence of "certainly does not support new features"? You made the claim, now provide the evidence.