[WCCF] AMD Radeon R9 390X Pictured

[Arachnotronic]

Both AMD & GF have publicly presented (official videos, presentations, conferences) their superior node tech. Whether its true or not, we will know soon enough when we compare Grenada vs Hawaii.

Could you please provide me with links? Thanks!

 

[boozzer]

I like how some people are telling others they can only post hopeful, cheery rumors, or else they consider it "thread-crapping".

it is called product crapping. :awe: and based on completely nothing. :biggrin: it isn't thread crapping because they go out of their way to ignore logic/counter arguments and just continuously post the same arguments based on nothing they "stay" on topic most of the time, which is very very clever I am gonna stop now because I don't want to OT the thread.

 

[Glo.]

But remember that a lot of sites tested R9 290X reference with Auto fan or on Quiet mode where it still throttled over time.

An after-market R9 290X with 1030mhz is 15% faster at 1080P and 1440P and than a reference 290X at Computerbase.

Based on Computerbase's testing, they have a 1030mhz 290X 8-10% slower than a reference 980. Once nV releases the 980Ti, 980's price could be lowered $50-100. That will put a lot of pressure on the 390X with 1050mhz clocks. AMD will have the ability to gauge where 980's price is once 980Ti launches next week. The profit margin on a 980 is probably so high that NV could sell them for $399 if it wanted to and still make good $. Things are really tough for AMD this generation since outside of Fiji cards none of their other products get a brand new architecture like Kepler -> Maxwell and they don't get HBM1 either. I have a feeling AMD will be forced to compete on price/performance for R9 300 level of cards until next gen.

Exactly thats why I don't believe that R9 390X will be inefficient, and slower than GTX980.

AMD is not THAT stupid.

 

[jamesgalb]

http://www.hardwareluxx.com/index.p...l-probably-be-released-as-radeon-r9-380x.html

That wasn't from so long ago

further speculation...

Computex = 300 series launch
E3 = Fiji launch

 

[swilli89]

Exactly thats why I don't believe that R9 390X will be inefficient, and slower than GTX980.

AMD is not THAT stupid.

I just can't help but think AMD is taking a gamble here.

Why not knock Hawaii down to $299 380X 4GB and $349 380X 8GB and have Fiji utterly destroy nVidia's value proposition by coming in at $599? They already established that the -80x is the previous gen high end at midrange pricing, why change that now? This alone makes me think its not just a simple rebadge like Tahiti with 7970 -> 280x.

Edit:

http://www.hardwareluxx.com/index.p...l-probably-be-released-as-radeon-r9-380x.html

That wasn't from so long ago

further speculation...

Computex = 300 series launch
E3 = Fiji launch

Ah super interesting. I know that is dated 6 months ago but if true it could be just the ticket they need for level competition with the 980.

 

[cbrunny]

Problem with well priced 290x in the new lineup is that it is not a big enough step up for anyone in the 79xx series. Maybe even the 7xxx series. If there are improvements then it could work well but if not it would probably flop. I know I wouldn't bite anyway.

 

[Alatar]

Where did this "grenada is built on glofo, has been updated to GCN 1.2, maybe even has more shaders etc." nonsense come from? Anand is the only forum where I see it. And worst of all there's no proof what so ever offered.

Guys, that wouldn't be hawaii anymore. And unfortunately what's in codeXL and the leaked drivers is hawaii, no doubts about it.

Show me even an ounce of credible proof of some of these new GPUs and their new device IDs. Where are they?

 

[Despoiler]

Where did this "grenada is built on glofo, has been updated to GCN 1.2, maybe even has more shaders etc." nonsense come from? Anand is the only forum where I see it. And worst of all there's no proof what so ever offered.

Guys, that wouldn't be hawaii anymore. And unfortunately what's in codeXL and the leaked drivers is hawaii, no doubts about it.

Show me even an ounce of credible proof of some of these new GPUs and their new device IDs. Where are they?

Logical fallacy much? The absence of something doesn't disprove something. The burden of proof is on you to prove your point, not anyone else.

On top of that everyone here understands this thread is mostly speculative.

 

[happy medium]

Where did this "grenada is built on glofo, has been updated to GCN 1.2, maybe even has more shaders etc." nonsense come from? Anand is the only forum where I see it. And worst of all there's no proof what so ever offered.

Guys, that wouldn't be hawaii anymore. And unfortunately what's in codeXL and the leaked drivers is hawaii, no doubts about it.

Show me even an ounce of credible proof of some of these new GPUs and their new device IDs. Where are they?

Good point ,I'd like to see it to.:whiste:

 

[Glo.]

Alatar, we know EXACTLY that 2816 GCN core chip made on GloFo 28 nm process with GCN 1.2 architecture is not Hawaii anymore.


That is exactly the whole point.

Were disputing right here, why AMD would rename the codenames for chips if they would have similair core counts. There were some rumors over here and there claiming that new Chips would be Trinidad, Tobago, Bermuda, and Grenada. All from Caribbean Islands.

http://www.guru3d.com/news-story/am...is-380xfiji-is-390x-and-bermuda-is-395x2.html
http://www.techpowerup.com/209719/radeon-r9-380x-based-on-grenada-a-refined-hawaii.html
http://wccftech.com/amd-gpus-fiji-grenada-trinidad-tobago-unveiled-computex-2015/
http://wccftech.com/amd-radeon-300-...ailed-rumors-point-grenada-hawaiis-successor/

 

[happy medium]

Quote:
Originally Posted by Silverforce11
Both AMD & GF have publicly presented (official videos, presentations, conferences) their superior node tech. Whether its true or not, we will know soon enough when we compare Grenada vs Hawaii.

Could you please provide me with links? Thanks!

I'm still waiting for these links too.

 

[happy medium]

Quote from techpowerup on grenada.

http://www.techpowerup.com/209719/radeon-r9-380x-based-on-grenada-a-refined-hawaii.html

" We don't expect the chip to be any more energy efficient at its final clocks, than Hawaii"

Guru 3d on Grenada
quote
" The GPU is different in the sense that it is going to have increased clock frequencies to beat the GTX 980 from Nvidia. We expect a similar TDP"

http://www.guru3d.com/news-story/am...is-380xfiji-is-390x-and-bermuda-is-395x2.html


I 'm not even going to bother reading the other links. I seen nothing about the manufacturer Glo fo or TSMC

 

[RussianSensation]

Exactly thats why I don't believe that R9 390X will be inefficient, and slower than GTX980.

AMD is not THAT stupid.

You are forgetting price and timing context of this generation and soon to be EOL 28nm node. We know that 970/980 cards overclock very well and NV has incredible profit margins approaching Intel levels. Let's say AMD were to spend tens of millions of dollars designing a new mid-range chip on 28nm from scratch only for this stop-gap generation. All NV would need to do is bump the clocks on 970/980 another 150mhz and/or, drop prices $50-100 and it's game over. AMD would have risked all that $ but might not even have won. At least with Fiji chips, AMD can reuse them later with a 14nm/16nm shrink as mid-range next gen parts. HBM1 controller and memory could be easily made into HBM2.

How would 28nm GDDR5 parts look next generation against Pascal? They would get destroyed. With AMD's finances, it's not unreasonable that they made a conscious decision to invest into future tech/SKUs - which means Fiji and just doing a mild refresh on Hawaii using a more mature 28nm node at GloFo / maybe using a High Density stack, adding HDMI 2.0 and some other minor differences. We've seen NV do a rebrand before with GTX680 --> 770 and that 770 card sold well due to 10% higher speed and a lower price than the original 680.

I am personally of the view that AMD will still have to compete on price/performance with everything other than Fiji because NV's 2x perf/watt lead cannot be overcome this generation without a new architecture, lower node or HBM1. I would be shocked if AMD was able to drop 100W of power usage in 1.05Ghz 290X = 390X Hawaii XT GDDR5 parts.

It could make more sense to just leave 390/390X cards at $249-399 levels but instead use Fiji Pro (say 3300-3500 shader HBM1 variants) in the $449-599 space.

I mean let's think about it for a second, what's easier and more cost effective for AMD :

Option 1: Spend tens of millions of dollars to create a GDDR5 only brand new chip with 2816 shaders that competes with a GTX980

Option 2: Make a lower Fiji SKU from non-fully yielding Fiji XT 4096 shader parts.

Since AMD was going to design and manufacture the large die size Fiji XT with HBM1 no matter what, Option 2 makes a lot more sense than Option 1. It guarantees a part that's faster performing than a 290X and it solves AMD's issues of what to do with all those die (surely it's better to sell them as lower tier SKUs than throwing them out!!!).

Another thing is the existinance of Fiji Pro has been almost universally ignored on these forums but it would be a shocker if AMD has a $399 2816 shader GDDR5 GPU and an $850 4096 shader HBM1 part*** and nothing in between.

Everyone is trying to somehow fit 390X Hawaii as a competitor to a 980 while fully ignoring the other possibility here that AMD is going to bring out a cut-down Fiji to stomp all over the 980 in the $500 price range.

***Disclaimer: unless of course Fiji XT is nowhere near a 4096 shader product but is itself just a 3300-3500 shader part.

TL; DR:

I think this forum is making a bold assumption discussing 390X (Hawaii XT with higher clocks and GDDR5) as a $550 980 competitor and $850 Radeon Fiji Fury as a Titan X competitor, but is completely discounting any real possibility of a cut-down Fiji Pro HBM part(s) as 980/980Ti competitors. Historically speaking, AMD/ATI always had a cut-down version of their flagship card. For that reason so many people ignoring the possibility of cut-down Fiji Pro SKU is an eye opener. :hmm:

The branding can be as simple as AMD Radeon Fury Pro and AMD Radeon Fury XT/MAXX. After all, ATI had Rage Fury PRO and the Rage Maxx. There is no reason to suggest that AMD cannot use the Fury branding for more than 1 Fiji product.
http://www.tomshardware.com/reviews/ati-rage-fury-pro-review,133-2.html

 

[krumme]

Could you please provide me with links? Thanks!

Links ?? Beema superior vs jaguar is solid facts that counts in my book??
Wtf is your purpose or question anyway?- a 3 year newer node performs better is hardly surprising even for gf

 

[krumme]

I'm still waiting for these links too.

The numbers for beema is a few pages back.
What links are you looking for. Some official ppt slidehows ?

 

[krumme]

But do you agree that it would have been even more costly to to a complete redesign? Cost/benefit ratio. With their limited resources, AMD must be finishing up a large part of the next gen designs now.

Obviously a new design is more expensive but tonga is designed and as its synthesized in my world when/if they are porting using that design seems most logical.

Its hopeless to predict what is happening and the reason is the wsa agreement that we know nothing about. Mubadala effectively controls amd and is their cash reserve and bank. Therefore strategically the interests of Mubadalas investment in gf is more important than amd. Its bigger investment. Perhaps more shiny from a political perspective.
That means if gf have capacity amd must use it. As amd cpu is half dead and carizo is minor market in the 1b amd must fullfill it looks like they need to port more than just high end gpu.
A 4096 shader gpu is not sold in any numbers and amd cpu line is going down the drain.
To me it looks like this is pointing to more gpu sizes beeing ported.

On the other hand hawai is less than 10% bigger than 980 so a rebrand seems to make sense from that perspective. Its fully depreciated. For desktop it looks fine from a performance point of view vs cost.

So i would say if it wasnt for the wsa and gf it would make more economic sense to just rebrand Hawaii but the wsa can change that. The 1b for me looks like we get a gf process for midrange too.

If we get new process its good because the process gives better perf for beema. But if we dont get new process for gpu then carizo is very good. And if we get new gpu carizo is probably unimpressive. We cant have it all. But lets look at it as if the glass is half full of water and not half empty

 

[Cloudfire777]

AMD needs to do a new dense design of existing chips to get the benefits of reduced power consumption and TDP with GlobalFoundries 28nm process.
Thats what got Kaveri etc more efficient.

Not sure if they will do that with R9 300 rebrands since it will involve resources and more work, but they probably did that when they designed the Fiji XT since it is new and they could do it from scratch.

 

[jpiniero]

Everyone is trying to somehow fit 390X Hawaii as a competitor to a 980 while fully ignoring the other possibility here that AMD is going to bring out a cut-down Fiji to stomp all over the 980 in the $500 price range.

It would make sense if the full Fiji was on water while the cut was on air. Problem is, I don't think either Fury (or the 390X for that matter) is going to sell much unless it's 250 W or less regardless of price or performance, and would be a no-go if it's over 300 W. I'd be a bit worried that the Fury is on water because it needs it to get down to 300 W.

 

[SimianR]

Yeah, it's all complete speculation, but they could have a cut down Fiji as a "R9 395X" or something, air cooled.

 

[tential]

It would make sense if the full Fiji was on water while the cut was on air. Problem is, I don't think either Fury (or the 390X for that matter) is going to sell much unless it's 250 W or less regardless of price or performance, and would be a no-go if it's over 300 W. I'd be a bit worried that the Fury is on water because it needs it to get down to 300 W.

Worried about power consumption on a $700-800 purchase. Like people worried about fuel consumption and serious about buying a sports car....

 

[Zstream]

Option 2: Make a lower Fiji SKU from non-fully yielding Fiji XT 4096 shader parts.

]

Agreed, I never understand why people discounted this considering all gpu/cpu manufacturers do this.

 

[96Firebird]

Worried about power consumption on a $700-800 purchase.

This may sound shocking, but people may have different preferences than you.

What I don't get is, if Fiji is the only new GPU and the rest are either rebrands or ported over to GloFo, what has AMD's GPU division been working on this whole time? I realize they probably put a lot of resources into helping develop HBM and all that, but I really hope they don't just get one GPU out of that. Does it take a lot of work to port over existing hardware designs to another process, ie. TSMC --> GF? I guess I just expected more drastic changes since a year and a half ago when Hawaii debuted, since the only "new" thing we've gotten thus far is Tonga. That is, if the latest rumors are indeed true.

Also, is it possible to cut down portions of a GPU when it involves the interposer and all that? I'm not all that savvy when it comes to the tech, just wondering if there is a difference between the way dies are harvested now and with HBM and the interposer.

 

[swilli89]

This may sound shocking, but people may have different preferences than you.

What I don't get is, if Fiji is the only new GPU and the rest are either rebrands or ported over to GloFo, what has AMD's GPU division been working on this whole time? I realize they probably put a lot of resources into helping develop HBM and all that, but I really hope they don't just get one GPU out of that. Does it take a lot of work to port over existing hardware designs to another process, ie. TSMC --> GF? I guess I just expected more drastic changes since a year and a half ago when Hawaii debuted, since the only "new" thing we've gotten thus far is Tonga. That is, if the latest rumors are indeed true.

Also, is it possible to cut down portions of a GPU when it involves the interposer and all that? I'm not all that savvy when it comes to the tech, just wondering if there is a difference between the way dies are harvested now and with HBM and the interposer.

Excuse me but what has nvidias team been working on then? Still using Kepler for mid range on down from 2013? It's ok for nvidia to rebrand 680 to 770 or to not release a 950 ti or 960 non ti? Also keep in mind 960 ti didn't show up til MONTHS after big maxwell. Why must amd present a full stack when nvidia just released a single new chip in all of 2014?

 

[happy medium]

Excuse me but what has nvidias team been working on then? Still using Kepler for mid range on down from 2013? It's ok for nvidia to rebrand 680 to 770 or to not release a 950 ti or 960 non ti? Also keep in mind 960 ti didn't show up til MONTHS after big maxwell. Why must amd present a full stack when nvidia just released a single new chip in all of 2014?

what was the gtx750 ti ,and the gtx960? Where can I buy a 960ti?

I thought Nvidia's lineup was gtx980,gtx970,gtx960 ,gtx750ti,gtx750? All Maxwell cards. Anything slower than that is not a card to game on anyway.

 

[96Firebird]

I think you're confused, Nvidia released GM107 (750 Ti and 750) in 2014, along with GM204. But why are we ending in 2014? From the beginning of 2014 until now, they have also released GM206 and GM200. Just wondering what has been going on all this time...