[BitsAndChips]390X ready for launch - AMD ironing out drivers - Computex launch

[happy medium]

Its not enough for a 4K high end gaming card

4GB is NOT future proof

ok gotcha, thanks

 

[Samwell]

Isn't AMD switching from TSMC to GloFo for next gen? That's what I thought but perhaps I am wrong. I think GloFo's/Samsung's joint node is 14nm not 16nm.

I think in parts, i'm pretty sure they'll use both processes. Actually that's the best they can do because they split the risc. If one company has some problems, then they'll get at least chips from the other company.

 

[2is]

you realize you just brush aside my 290x cf and contradicted yourself in the same post to boot?

To answer your question, I want to move away from multi-gpu setups. It was great when I first got it but a little less so now that the cards aren't as well supported. I find myself either waiting longer for patches or making manual adjustments in nVIdia control panel to get it working optimally. Not always the case, but it happens more often than it used to, which was practically never. Though I may still keep one of my 680's as a dedicated PhysX card.

The better question is, do YOU realize this is about the 5th time in this thread your reading comprehension has failed you?

Do YOU realize this is about the 3rd time you have attacked boozzer? Probably not, and you can have a day off for failing to follow the forum rules..

-Rvenger

 

[Red Hawk]

I'll simply it. Existing console ports are already getting to around 3GB of vram usage at 1080p running at max settings, and that's with low to no amounts of AA and before we even take mods into account if you're into such things. Even a game as old as BF4 will flirt with a 4GB ceiling at 200% resolution scale with settings cranked. These are all scenarios that exist now, at 1080p so yeah... 4GB without an 8GB option is a failure.

Resolution scaling...is not the same as playing at 1080p. You are using a 1080p monitor, but you are not actually playing at 1080p. So no, I don't think that scenario is a useful measure of a game's performance at 1080p.

 

[2is]

You are really trying aren't you? I may be wrong but this is what I get from his logic.



I don't think he denies it bothers him. Its a problem but you can turn down settings (as he has been doing) to get the game into 2 GB. Its workable but not ideal. That said its not something he wants to go through again.

Because he learns from his mistakes? Once bitten twice shy.

He bought 680 SLI a while back and is trying to make it stretch. Many people try out CF and SLI and decide they don't like it. He wants a single GPU card. However, any reasonably priced single GPU card isn't terribly more powerful than 680 SLI when not vram limited.

Also I don't believe he is even in the group of people who bought $700 cards.

I don't doubt that you are right. But this is a costlier approach than sitting through mid/high texture settings and waiting for a large substantial increase. Again, no CF/SLI.

And you know what?

You are comparing an overclocked 7970 setup to an underclocked 680 SLI setup. If he has 680 SLI he gains ~+5% perf over the 690. If he is using aftermarket cards and has them overclocked then he can gain another 10-15% over the 690. If he reduces settings appropriately then the 690 tanking due to vram can be reduced further.

On launch the Ares II did almost 20% better than the 690 (removing games with no CF support) at 1440p/1600p.

In 2012 no one could pull up instances where 3 GB on 7970 ghz CF was advantageous over 680 SLI 2 GB either.

390X is frankly in a different performance class. Would 4 GB be a problem with ~titan X performance?

390X only having 4 GB would be like AMD releasing the 290X series with only 3 GB.

Sure its good to have 4 GB and 8 GB versions. Having only a 4 Gb version is a little iffy though. That said, any card releasing > than the 980 level of performance should have more than 4 GB.

Again with the inability to price anything. Price in canadian no tax with rebates.

Cheapest 980 $630 - http://www.vuugo.com/computer-hardw...ards-ZT-90204-10P.html?tracking=5108053939762

Windforce - $640 http://www.ncix.com/detail/gigabyte-geforce-gtx-980-wf-a6-103002-1334.htm?affiliateid=7474144

Strix - $ 685 http://www.ncix.com/detail/asus-geforce-gtx-980-strix-cc-102671-1334.htm?affiliateid=7474144

superclocked - $690 http://www.newegg.ca/Product/Produc..._-na&utm_medium=affiliates&utm_source=afc-%zn

Windforce looks the best

Titan X - $1212 http://www.vuugo.com/computer-hardw...d-GTXTITANX-12GD5.html?tracking=5108053939762

(Titan X is all reference so just crank up the speeds yourself - you will be cooler limited)

This is why I have to take anything you say about perf/price with a huge grain of salt... the prices are always wrong.

Why compare to the 980? The 2nd cut down chip (970) is half the price for 85% the performance?

Nail on the head. Thanks for having the patience to not only read his entire post, but respond to it in kind. Maybe it was my fault and I wasn't clear, but you seemed to get what I was saying just fine.

Resolution scaling...is not the same as playing at 1080p. You are using a 1080p monitor, but you are not actually playing at 1080p. So no, I don't think that scenario is a useful measure of a game's performance at 1080p.

The resolution is rendered at 200% and scaled down to 1080p, (assuming your monitor is 1080p) so yes, I am playing at 1080p. It's a feature that allows you to get much better detail on an existing 1080p setup and it requires quite a lot of VRAM

 

[RampantAndroid]

Resolution scaling...is not the same as playing at 1080p. You are using a 1080p monitor, but you are not actually playing at 1080p. So no, I don't think that scenario is a useful measure of a game's performance at 1080p.

By scaling, you mean super/multi sampling? That's really moving the goalposts...

 

[Red Hawk]

By scaling, you mean super/multi sampling? That's really moving the goalposts...

Supersampling, yes, but not multisampling. Supersampling is taking a higher resolution image and fitting it on a lower resolution screen. I don't think it's moving goalposts to disregard performance when doing that as representative of performance when actually rendering a game at that lower resolution.

 

[cusideabelincoln]

Spot on +1.

AMD releasing the 390X with 4GB would be like them releasing the 6970 with 1GB or the 7970 with 2GB. When is the last time AMD's flagship didn't ship with more memory then the previous gen? AMD has always been ahead of nvidia as far as vram is concerned and for good reason as most of Nvidia cards would vram bottleneck much quicker. Really hope the 390x is 8 GB.

The released the 4870 as a 512MB card, only matching the previous 3870, and then a few months later released the 1GB version. They released the 5870 as a 1GB card, only just matching the previous 4890.

 

[LTC8K6]

??????
The link is for the paper in the post. It works for me.

A New Perspective on Processing-in-memory Architecture Design

The link does work, but go back and look at the actual link.

It's contaminated by viglink. The viglink garbage is added automatically if you just copy the link. It makes the link about a mile long.

http://en.wikipedia.org/wiki/VigLink

 

[RussianSensation]

Again with the inability to price anything. Price in canadian no tax with rebates.

Cheapest 980 $630 - http://www.vuugo.com/computer-hardw...ards-ZT-90204-10P.html?tracking=5108053939762

Windforce - $640 http://www.ncix.com/detail/gigabyte-geforce-gtx-980-wf-a6-103002-1334.htm?affiliateid=7474144

Strix - $ 685 http://www.ncix.com/detail/asus-geforce-gtx-980-strix-cc-102671-1334.htm?affiliateid=7474144

superclocked - $690 http://www.newegg.ca/Product/Produc..._-na&utm_medium=affiliates&utm_source=afc-%zn

Windforce looks the best

Titan X - $1212 http://www.vuugo.com/computer-hardw...d-GTXTITANX-12GD5.html?tracking=5108053939762

(Titan X is all reference so just crank up the speeds yourself - you will be cooler limited)

This is why I have to take anything you say about perf/price with a huge grain of salt... the prices are always wrong.

Why compare to the 980? The 2nd cut down chip (970) is half the price for 85% the performance?

Really my prices are always wrong? Ok, whatever dude. You just linked several cards with $680-760 prices and the Titan X cards you linked is not in stock.

The reason I used 980's price is to show how expensive NV cards currently are in Canada, which leaves a lot of room for R9 390 nonX (that everyone in this thread wants to ignore) and R9 390X to change the price/performance points. I'll be sure to remind you of this on the date of R9 300 series release that yes, R9 390 non-X does exist. There isn't any point in talking about a 970 if we are discussing the potential of 390X to be 40-50% faster than a 290X since these are not 2 cards that compete with each other. 980 is squarely positioned in the $650+ CDN space (before taxes) which means it's all we have to go by right now when discussing R9 390 series of cards' pricing. Not even sure how you decided to bring up 970 into the mix unless you think R9 390 will be 10% faster than a 290X?

The Windforce 3X card you linked is $760 CDN, $740 after rebate, not $640. Other prices you list are after rebate but in Canada you are taxed on a pre-rebate price. Conveniently you forgot to add the taxes to your prices too unless you are an Aboriginal. Try again. $660-700 CDN before tax is an accurate price for most 980 cards and that's what I listed more or less. Sure, if I spent more time I could find a card for $630-640 CDN but you completely missed my point -- these cards are priced very high, which means AMD has plenty of room to disrupt the price/performance of current cards with both the R9 390X and the 2nd tier 390.

Now, you are saying he will buy just a single $700 card for 1080P, instead of a pair? OK, that makes his case even worse. Strapping 6-8GB of VRAM to play games at 1080P on a single such card is a waste of VRAM. In that case, R9 390 4GB CF would walk all over a single GM200 6GB or an R9 390X for 1080P. If he wants to use 4K VSR/DSR on a 1080P monitor in the latest AAA games today, a single such card isn't fast enough to handle it. Why? Because a Titan X can't either. I can't wait as you and him start collecting benchmarks over the next 2 years at 1080P and 1440P with VSR/DSR to prove how a single GM200 6GB/R9 390X can hit 60 fps on everything maxed out with high MSAA/super-sampling. Also, make sure to save benchmarks where 4GB of VRAM becomes a big bottleneck for 1080P. Maybe you'll find something after 2 years but by then many will already have moved on to 14nm/16nm HBM2 cards.

The resolution is rendered at 200% and scaled down to 1080p, (assuming your monitor is 1080p) so yes, I am playing at 1080p. It's a feature that allows you to get much better detail on an existing 1080p setup and it requires quite a lot of VRAM

Which is the equivalent of a 2-2.5X the performance hit on the GPU itself to start with or like taking a Titan X and trying to power through modern AAA games at 4K without AA. How does that work out? Badly.

Also, if you are going to promote how you are so interested in using super-sampling (VSR/DSR) and how that will expose VRAM bottlenecks, why aren't you considering that AMD's cards have superior IQ in VSR vs. DSR? NV's cards exhibit more flickering and blur with DSR. If you are going to be objective in the discussion, then you can't just look only at performance then when discussing super-sampling methods.

"In Assassin's Creed 4: Black Flag, the result falls surprisingly clearly in favor of AMD's VSR. For VSR manages the feat visibly better to address not only the flicker in 2,560 × 1,440. In the same breath, the game will also be displayed more sharply than the Nvidia solution. Thus helping to DSR and the smoothness switch is not on, as this could make the image more than just blurrier or more flickering."

"Native downsampling of AMD and Nvidia compared"
327MB video download is provided directly from the website:
http://www.computerbase.de/2014-12/amd-vsr-nvidia-dsr-downsampling-catalyst-omega-vergleich/4/

NV appears to be using a Gaussian filter for its DSR implementation, which has been known to be a blur-fest. Computerbase has long investigated and many times confirmed that NV's native super-sampling IQ is inferior to AMD's and this has been the case even during GCN vs. Kepler generation.

Users from OCN also noticed:

"There's a slider for the intensity of the one filter type DSR is limited to (Gaussian), no options to change the filtering type. The old-fashioned Nvidia downsampling method of creating a custom resolution produced a sharper image as does GeDoSaTo (which also has many more options including filter type options) than DSR. Evidently, so does VSR. Simply turning the DSR slider to 0 means only a 2x2 downsampling resolution (which is a massive 5120x2880 for me and my 2560x1440 monitor) is actual effective as anything else results in disgusting shimmering/aliasing-like artfacts and generally poor IQ. DSR is still young, I hope Nvidia expand on it. But at the moment, its functionality is extremely limited in comparison to other methods because of its limitation to that one filter type. Blurring a downsampled image to any degree is counter-intuitive and should be avoided, it's the first thing I noticed when I used DSR (that it wasn't as clear as other downsampling methods I've tried; disappointing)." ~ OCN

But since you already said 4GB of HBM1 is a fail no matter what, down-sampled/super-sampled IQ doesn't matter then, just performance and VRAM amounts, right?

 

[2is]

Also, if you are going to promote how you are so interested in using super-sampling (VSR/DSR) and how that will expose VRAM bottlenecks, why aren't you considering that AMD's cards have superior IQ in VSR vs. DSR? NV's cards exhibit more flickering and blur with DSR. If you are going to be objective in the discussion, then you can't just look only at performance then when discussing super-sampling methods.

"In Assassin's Creed 4: Black Flag, the result falls surprisingly clearly in favor of AMD's VSR. For VSR manages the feat visibly better to address not only the flicker in 2,560 × 1,440. In the same breath, the game will also be displayed more sharply than the Nvidia solution. Thus helping to DSR and the smoothness switch is not on, as this could make the image more than just blurrier or more flickering."

But since you already said 4GB of HBM1 is a fail no matter what, IQ doesn't matter either, right?
Video - https://www.youtube.com/watch?v=BP47lIjr9oU&list=UUv3PUA2RWLQI8k7tGb2oc8g

http://translate.google.com/transla...-omega-vergleich/5/#diagramm-rome-2-1920-1080

Super sampling was one of several examples I provided where vram could be an issue, not the only example. What good is better IQ if your performance is going to tank in new games because you need more VRAM?

I haven't seen the specifics between VSR and DSR but considering it's a driver implantation on both cards, then it's reasonable to assume drivers can also be optimized for a better implementation. What are you goanna do about a 4GB hardware limitation? Download more VRAM?

I'm sorry my opinion on a 4GB 390x bothers you so much, but it is, and will remain my opinion. If it's available in 8GB flavors in addition to 4GB then kudos to AMD for getting it right. If it's limited to 4GB, then, well... You know how I feel. It is what it is. Focus on what card will make you happy and go with that. Your efforts to try and get me to adopt your opinions and reasoning will not succeed. I want my next card to have ≥ 6GB of VRAM. I've illustrated my reasons. You don't have to agree with them, but they are my reasons all the same and ≥ 6GB is what I want.

Now... If you want to discuss specifics between VSR and DSR, a new thread is probably the best place. In that thread we can also talk about how AMD's last non-beta driver is 6 months old.

 

[Serandur]

Users from OCN also noticed:

"There's a slider for the intensity of the one filter type DSR is limited to (Gaussian), no options to change the filtering type. The old-fashioned Nvidia downsampling method of creating a custom resolution produced a sharper image as does GeDoSaTo (which also has many more options including filter type options) than DSR. Evidently, so does VSR. Simply turning the DSR slider to 0 means only a 2x2 downsampling resolution (which is a massive 5120x2880 for me and my 2560x1440 monitor) is actual effective as anything else results in disgusting shimmering/aliasing-like artfacts and generally poor IQ. DSR is still young, I hope Nvidia expand on it. But at the moment, its functionality is extremely limited in comparison to other methods because of its limitation to that one filter type. Blurring a downsampled image to any degree is counter-intuitive and should be avoided, it's the first thing I noticed when I used DSR (that it wasn't as clear as other downsampling methods I've tried; disappointing)." ~ OCN

That post looks familiar...

I've never been cross-forum quoted before.

I just want to add, having now owned some 290Xs, that VSR does indeed produce a superior image quality, though the 290X is unfortunately limited to 3200x1800 downsampling (via VSR, anyway). Not a lot of flexibility in internal rendering resolutions with VSR, but a significantly superior filter. I hope Fiji corrects that limited range.

 

[Erenhardt]

Its not enough for a 4K high end gaming card.
People have mentioned that GTX 680 did fine against 7970 despite less VRAM but neither cards had the pure power to drive very high res with maxed out settings skyrocket the VRAM usage like 390X/980Ti do.

I strongly disagree.

40FPS is quite playable for me. Contrary to 17 fps.

50% more VRAM is a big deal. I know some tried hard to downplay that, but it really gives a second wind for 7970.

VRAM requirements will go up, not down.

 

[Cloudfire777]

Uhm, I just noticed something.
R9 360, 370 and 380 are all manufactured on 28nm from TSMC. Not Global Foundries.
The hell???

Why would they split up GPUs and manufacture one part of 300 series from TSMC and another one at GloFo?

Are we not getting 28nm SHP for GPUs after all? Will they continue with TSMC exclusively for 300 series? After the dissappointment of not getting 20nm cards, this is not good. :'(

 

[3DVagabond]

They've announced them as OEM models and appear to be pure rebrands. Probably offered them at a song to clear out the rest of the old inventory in one large chunk.

 

[R0H1T]

Uhm, I just noticed something.
R9 360, 370 and 380 are all manufactured on 28nm from TSMC. Not Global Foundries.
The hell???

Why would they split up GPUs and manufacture one part of 300 series from TSMC and another one at GloFo?

Are we not getting 28nm SHP for GPUs after all? Will they continue with TSMC exclusively for 300 series? After the dissappointment of not getting 20nm cards, this is not good. :'(

Yes, since GF wafers is what they want & can have'em for cheap (WSA?) IIRC. All things up to GCN 1.2 probably gonna be made on TSMC & anything else (newer) on GF, at least that's what I'm hoping for ^_^

 

[Erenhardt]

They've announced them as OEM models and appear to be pure rebrands. Probably offered them at a song to clear out the rest of the old inventory in one large chunk.

Yep. The only way those will sell those old chips is through OEM.

DYI market knows better and will not touch last gen product with a foot long stick when something new is announced. Would have to offer an amazing perf/$ to be worth a second thought.

OEMs on the other hand sell system to uninformed buyers. If they see R9 300 on the box, they will think they get the latest... Cruel world.

 

[Cloudfire777]

So you guys think there isnt any production going on at TSMC for AMD, so they are just selling out already manufactured inventory from the 200 series?

I guess it makes sense, and just do the manufacturing nescessary at GloFo for the Fiji/Grenada/390 series.
I was thinking they actually produced the 360/370 and 380 right now at TSMC, which is why I wondered why split up the production when 28nm from GloFo is better.

 

[Erenhardt]

So you guys think there isnt any production going on at TSMC for AMD, so they are just selling out already manufactured inventory from the 200 series?

I guess it makes sense, and just do the manufacturing nescessary at GloFo for the Fiji/Grenada/390 series.
I was thinking they actually produced the 360/370 and 380 right now at TSMC, which is why I wondered why split up the production when 28nm from GloFo is better.

Probably...

From what I gather it is not that simple as taking your hammer and nails to the next foundry, and start producing there. Each chip is designed for a target production process. Porting it to different fab requires a make-over of the IC. Some parts are easier to port, some require a lot of engineer work.

For example amd designed console chips with portability in mind, so those can be easy moved to other fab/process.

I wouldn't expect amd to port their old gen gpus to new process only for very shortlived R9 300 series rebrand cards.

 

[Enigmoid]

Really my prices are always wrong? Ok, whatever dude. You just linked several cards with $680-760 prices and the Titan X cards you linked is not in stock.

The reason I used 980's price is to show how expensive NV cards currently are in Canada, which leaves a lot of room for R9 390 nonX (that everyone in this thread wants to ignore) and R9 390X to change the price/performance points. I'll be sure to remind you of this on the date of R9 300 series release that yes, R9 390 non-X does exist. There isn't any point in talking about a 970 if we are discussing the potential of 390X to be 40-50% faster than a 290X since these are not 2 cards that compete with each other. 980 is squarely positioned in the $650+ CDN space (before taxes) which means it's all we have to go by right now when discussing R9 390 series of cards' pricing. Not even sure how you decided to bring up 970 into the mix unless you think R9 390 will be 10% faster than a 290X?

You said yourself its a waste to buy to top end SKU. The second from the top always has much better value. Therefore, and I quote you "the 980 is the worst buy possible".

Everything is expensive in canada. The cheapest 290 I can find is $310 vs. $240 in the states. Same with intel processors or AMD CPUs.

The Windforce 3X card you linked is $760 CDN, $740 after rebate, not $640. Other prices you list are after rebate but in Canada you are taxed on a pre-rebate price. Conveniently you forgot to add the taxes to your prices too unless you are an Aboriginal. Try again. $660-700 CDN before tax is an accurate price for most 980 cards and that's what I listed more or less. Sure, if I spent more time I could find a card for $630-640 CDN but you completely missed my point -- these cards are priced very high, which means AMD has plenty of room to disrupt the price/performance of current cards with both the R9 390X and the 2nd tier 390.

Your prices include rebates as well.

In Canada:

MSI Gaming 980 = $690-700 CDN
Asus STrix 980 = $700 CDN
EVGA Superclocked 980 = $700+ CDN
Gigabyte Windforce 980 = $700+ CDN

Asus Titan X = $1369 CDN

The asus strix was (when I looked and still is) $720 with a $15 rebate. So I included rebates as well.

Thats NCIX being sneaky. If you go through other sites (pcpartpicker) you get this.

Now, you are saying he will buy just a single $700 card for 1080P, instead of a pair? OK, that makes his case even worse. Strapping 6-8GB of VRAM to play games at 1080P on a single such card is a waste of VRAM. In that case, R9 390 4GB CF would walk all over a single GM200 6GB or an R9 390X for 1080P. If he wants to use 4K VSR/DSR on a 1080P monitor in the latest AAA games today, a single such card isn't fast enough to handle it. Why? Because a Titan X can't either. I can't wait as you and him start collecting benchmarks over the next 2 years at 1080P and 1440P with VSR/DSR to prove how a single GM200 6GB/R9 390X can hit 60 fps on everything maxed out with high MSAA/super-sampling. Also, make sure to save benchmarks where 4GB of VRAM becomes a big bottleneck for 1080P. Maybe you'll find something after 2 years but by then many will already have moved on to 14nm/16nm HBM2 cards

He is saying he doesn't want CF/SLI because of the problems. This 2is has to respond to as he is the one buying.

Which is the equivalent of a 2-2.5X the performance hit on the GPU itself to start with or like taking a Titan X and trying to power through modern AAA games at 4K without AA. How does that work out? Badly.

Judicious choice of settings.

Also, if you are going to promote how you are so interested in using super-sampling (VSR/DSR) and how that will expose VRAM bottlenecks, why aren't you considering that AMD's cards have superior IQ in VSR vs. DSR? NV's cards exhibit more flickering and blur with DSR. If you are going to be objective in the discussion, then you can't just look only at performance then when discussing super-sampling methods.

I didn't comment on this.

But since you already said 4GB of HBM1 is a fail no matter what, down-sampled/super-sampled IQ doesn't matter then, just performance and VRAM amounts, right?

Would the titan X at $700 with 4 GB have been a fail? With the same amount of vram as the 980?

 

[desprado]

R9 390X is based on R9 285(GCN 1.2) architecture.
R9 285 has 1792 shaders unit and it is beaten in some games at 1080p by GTX 960 which has 1024 shaders units
If u multiply by 2 than u make
GTX 980 with 2048 shaders unit
just for an example R9 3xx has 3584 shader units
Now Coming to titan X it has 3072 shaders unit and if it has to be beaten in all terms than AMD required 5376 shader units of GCN 1.2 but right now R9 390X has 4096 GCN 1.2 shaders.
Yes of course AMD R9 390X will beat Titan X in 4K but not in 1080p or 1440p resolution and it is just an assumption and i could be wrong

 

[Cloudfire777]

R9 390X is based on R9 285(GCN 1.2) architecture.
R9 285 has 1792 shaders unit and it is beaten in some games at 1080p by GTX 960 which has 1024 shaders units
If u multiply by 2 than u make
GTX 980 with 2048 shaders unit
just for an example R9 3xx has 3584 shader units
Now Coming to titan X it has 3072 shaders unit and if it has to be beaten in all terms than AMD required 5376 shader units of GCN 1.2 but right now R9 390X has 4096 GCN 1.2 shaders.
Yes of course AMD R9 390X will beat Titan X in 4K but not in 1080p or 1440p resolution and it is just an assumption and i could be wrong

It fits VR-Zone`s +20% projections of 390X over GTX 980.
GTX 980: 2048 cores
R9 390X: 4096 shaders

You said AMD needs 1.75 more shaders than an Nvidia equivalent to match it. (3584/2048 = 1.75).

R9 390X have 2.00 more cores than GTX 980, so thats pretty close to the +20% from VR-Zone. But that is if HBM doesnt boost it much over GDDR5 (The extra bandwidth of HBM is more than what is required) and 390X is GCN1.2

God knows what will happen

 

[desprado]

It fits VR-Zone`s +20% projections of 390X over GTX 980.
GTX 980: 2048 cores
R9 390X: 4096 shaders

You said AMD needs 1.75 more shaders than an Nvidia equivalent to match it. (3584/2048 = 1.75).

R9 390X have 2.00 more cores than GTX 980, so thats pretty close to the +20% from VR-Zone. But that is if HBM doesnt boost it much over GDDR5 (The extra bandwidth of HBM is more than what is required) and 390X is GCN1.2

God knows what will happen

I am taking about horse power and of course HBM wont matter on 1080p and 1440p but 4k is totally a different story.

GTX 960 and R9 285 have some what equal benefits bit the real factor is that R9 285 has 1792 sharder were as GTX 960 has 1024 shader.

 

[Glo.]

If R9 290X with old technology is right now 10-20% slower than GTX980, then bigger die, with higher core count and newer technology should be faster than GTX980.

3584 GCN cores with 4 GB of HBM should be 20% faster than GTX980.
4096 GCN core with 4 GB of RAM and 1050 MHz core clock, air cooled should be 10% slower than Titan X
4096 GCN core, with 8 GB of HBM, 1200 MHz of core clock, and water cooled should be faster than Titan X.


That is my prediction.

 

[desprado]

If R9 290X with old technology is right now 10-20% slower than GTX980, then bigger die, with higher core count and newer technology should be faster than GTX980.

3584 GCN cores with 4 GB of HBM should be 20% faster than GTX980.
4096 GCN core with 4 GB of RAM and 1050 MHz core clock, air cooled should be 10% slower than Titan X
4096 GCN core, with 8 GB of HBM, 1200 MHz of core clock, and water cooled should be faster than Titan X.


That is my prediction.

Yes but that is not merit.

R9 285 vs GTX 960 is very interesting because GTX 980 is 90%+ faster than GtX 960 and 3584 GCN 1.2 cores will only be 5% to 10% faster not more than that is what i think base on facts.

1024 Maxwell Core vs 1792 GCN 1.2 Core has some what equal performance.