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

[SimianR]

I don't think that they would rebrand the 290x into the 390 (non-x), I'm guessing the 390 will be a cut down chip with lower clock speeds and an air cooler and if the 290x and 290 are rebranded they'll be the 380 if anything. But the 290x and 290 really can't be cheap cards to make so I can't imagine AMD wants to keep selling them at the mid range price points.

With all these rebrands this onion article seems even more relevant:

http://onion.com/1EcX4Wd

 

[Alatar]

Said the same thing I said on OCN, I know the "enhance" memes etc. are hilarious but a picture really does the best job at showing some things:

Compared to the XFX 290X DD PCB:

exact same component placement as the hawaii card. (at least in the places that aren't covered by the cooler or VRAM cooling plate

 

[Shehriazad]

First of all I really doubt that the 390s are going to be rebrands...and 2nd...people throw around the word rebrand rebrand just like how some tech sites do nowadays.

If you are thinking that they will literally take some 290X...change the GPU clock and name it a 380X and take a 280X and rename it into a 370X...then you're dead wrong I say.

All future cards will support Freesync. The enhanced color compression is also very likely to be aboard on all of the GDDR5 based cards. (Not really sure they'd bother with that on a 8GB HBM card, but why not xD)



Optimization will happen...and yes...it is very likely that we will see a rebranded 290X...but it won't be anything higher than a 380X...and even then it will receive some minor facelifting.


And what the fuss is with the rebranding on the AMD side is beyond me, anyway.....rebranding is SO very popular.

GT 710, 720, 730, 740 and the GTX 770...those were all rebrands...some even rebrands of rebrands. Yet the GTX 770 was VERY popular. Some of the 700 series cards are even FERMI...yes...Fermi.

But now look at that... GTX 780/ti, 760 and 750/Ti were the non rebrands of that series(50% rebrands!). Rebranding happens ALL the time...don't act like AMD is some kind of villain for doing it...we have been supporting it for years.

 

[digitaldurandal]

If you are thinking that they will literally take some 290X...change the GPU clock and name it a 380X and take a 280X and rename it into a 370X...then you're dead wrong I say.

Welcome to the party. There are 19 pages prior to this one and it has been all but confirmed that the 370/x and 380/x are rebrands. Though you are right that it is not a 290x rebrand, they are looking to be Tonga based cards with one of them being the 285 and one of them being the full Tonga that has been seen in mobile.

 

[RussianSensation]

Welcome to the party. There are 19 pages prior to this one and it has been all but confirmed that the 370/x and 380/x are rebrands. Though you are right that it is not a 290x rebrand, they are looking to be Tonga based cards with one of them being the 285 and one of them being the full Tonga that has been seen in mobile.

I think this forum needs to have a clear definition of what the term re-branding means because this is partly what's creating the confusion.

For example, do we mean re-branding as in "re-badging" such as when HD5750/5770 became HD6750/6770 with absolutely 0 changes? Or are we using the term re-branding to mean taking an existing architecture/SKU and improving upon the predecessor as in HD4870 to HD4890?

I would not use the term re-branding for HD4890 even though underneath it's an identical architecture to the HD4870. It sounds to me like some people in this thread are using the term so loosely that they would call a Hawaii-based R9 380X with 10% faster clocks and 10% lower power usage a "re-brand."

But take a look at this -- at 1440P - a popular stop-gap resolution between 1080P and 4K against the current competition - XXX only leads a 290X by 13% according to Computerbase.

Now let's imagine a 10% faster R9 380X based on 290X at $399. Would you say that's a bad videocard? When a certain $330 card was hailed as revolutionary, it only beat an after-market R9 290 by 6-7% and undercut it by only $50-70 vs. market prices of the 290 on launch date. Yet, it was hailed universally as a price/performance "revolution." So, what would reviewers say about a 10% faster R9 290X "aka" R9 380X priced at $399? :hmm:

 

[as9hW]

exact same component placement as the hawaii card. (at least in the places that aren't covered by the cooler or VRAM cooling plate

Note the different holes and the lack of VRAM on the 'newer' card. It should be no surprise that a company who designs boards will re-use designs/layouts.

There's not enough information to conclude either way.

 

[Alatar]

Note the different holes and the lack of VRAM on the 'newer' card. It should be no surprise that a company who designs boards will re-use designs/layouts.

There's not enough information to conclude either way.

The "390" has a VRAM cooling plate like all 290X DD cards that covers VRAM. You can see at least two screw holes being used for keeping it (and the whole cooler) in place. The only screw hole that might not be there is the one being covered by a small mosfet heatsink (or similar) in the "390" pic so we can't say about that one, but other than that all screwhole placements in the pic are confirmed identical.

The same VRAM plate can be seen from the other pic of the "390" as well. It extends to the top of the PCB where the GDDR5 is.

 

[digitaldurandal]

I think this forum needs to have a clear definition of what the term re-branding means because this is partly what's creating the confusion.

For example, do we mean re-branding as in "re-badging" such as when HD5750/5770 became HD6750/6770 with absolutely 0 changes? Or are we using the term re-branding to mean taking an existing architecture/SKU and improving upon the predecessor as in HD4870 to HD4890?

I would not use the term re-branding for HD4890 even though underneath it's an identical architecture to the HD4870. It sounds to me like some people in this thread are using the term so loosely that they would call a Hawaii-based R9 380X with 10% faster clocks and 10% lower power usage a "re-brand."

But take a look at this -- at 1440P - a popular stop-gap resolution between 1080P and 4K against the current competition - XXX only leads a 290X by 13% according to Computerbase.

Now let's imagine a 10% faster R9 380X based on 290X at $399. Would you say that's a bad videocard? When a certain $330 card was hailed as revolutionary, it only beat an after-market R9 290 by 6-7% and undercut it by only $50-70 vs. market prices of the 290 on launch date. Yet, it was hailed universally as a price/performance "revolution." So, what would reviewers say about a 10% faster R9 290X "aka" R9 380X priced at $399? :hmm:

Fair enough.

I don't know what, if any, changes they are making to the card. Only that evidence has suggested it is based on Tonga. I don't think for the cards under Tonga that there will be any changes though TBH, because it is hardly worth their time on this stop gap until the new nodes are producing cards.

I don't know enough about Freesync to know if they can shoehorn that feature onto new reused cards or not.

We will have to see what comes out really.

 

[boozzer]

First of all I really doubt that the 390s are going to be rebrands...and 2nd...people throw around the word rebrand rebrand just like how some tech sites do nowadays.

If you are thinking that they will literally take some 290X...change the GPU clock and name it a 380X and take a 280X and rename it into a 370X...then you're dead wrong I say.

All future cards will support Freesync. The enhanced color compression is also very likely to be aboard on all of the GDDR5 based cards. (Not really sure they'd bother with that on a 8GB HBM card, but why not xD)



Optimization will happen...and yes...it is very likely that we will see a rebranded 290X...but it won't be anything higher than a 380X...and even then it will receive some minor facelifting.


And what the fuss is with the rebranding on the AMD side is beyond me, anyway.....rebranding is SO very popular.

GT 710, 720, 730, 740 and the GTX 770...those were all rebrands...some even rebrands of rebrands. Yet the GTX 770 was VERY popular. Some of the 700 series cards are even FERMI...yes...Fermi.

But now look at that... GTX 780/ti, 760 and 750/Ti were the non rebrands of that series(50% rebrands!). Rebranding happens ALL the time...don't act like AMD is some kind of villain for doing it...we have been supporting it for years.

hey, the perf/watt is now dead, they need something else to hold onto so they can sling mud at amd

 

[RussianSensation]

I don't think for the cards under Tonga that there will be any changes though TBH, because it is hardly worth their time on this stop gap until the new nodes are producing cards.

I think for a lot of people on this forum, Tonga is a write-off. Even in November and December of 2014 we have already seen after-market R9 290 cards selling for $200-250. Today, in the US at least, one can readily find good after-market 290s at those prices (XFX R9 290 with lifetime warranty for $240 or Asus DCUII 290 for $240). As a result, someone looking to spend only $200-250 has less incentive to be waiting for an R9 370X/380 card imo. What are the chances either of those cards will offer a substantial improvement in price/performance over a $240 after-market R9 290?

We almost have to look back at some history of AMD's pricing to understand this. When HD7970Ghz launched, regular 7970 fell in price enough to make it more attractive overall. In fact later cards like 1Ghz HD7970 Gigabyte Windforce sold for $80-100 less than the officially labelled HD7970Ghz edition cards.

When HD6850/6870 launched, 5850 and 5870 were actually more attractive and many gamers weren't that happy with the price/performance of the 6850/6870 cards vs. clearance prices on 5850/5870. Certainly the price differential between a 5870 and a 6970 was massive relative to their small difference in performance in non-tessellated scenarios. When HD5770 launched, HD4870 and HD4890 offered better price/performance and actually the latter was about 15% faster on 5770's launch but its market price was similar.

I think a lot of people in this thread are just setting themselves up for disappointment. If people expect AMD to launch a $250 card that's 30-40% faster than an R9 290 and a $400 card that's 50% faster than an R9 290, they are going to be disappointed for sure.

I mean the fastest single GPU in the world today costs $1000 but it's "only" 45-48% faster than the R9 290X. AMD isn't exactly in a pressured position to repeat HD4870 at $299 or HD5870 at $369. Relative to what's out there right now a card 40% faster than an R9 290X priced at $649 would sell well.

---

R9 290X just has a tarnished reputation but its price/performance is phenomenal. The competitor's card is only 6-8% faster at 4K for an 85% higher price. The reality is it is AMD's game to lose. They know the price and performance of every single competing card. How hard can it be to offer superior price/performance at every level with R9 300 series? It shouldn't be hard at all -- AMD's biggest problem is marketing/branding.

 

[raghu78]

I think a lot of people in this thread are just setting themselves up for disappointment. If people expect AMD to launch a $250 card that's 30-40% faster than an R9 290 and a $400 card that's 50% faster than an R9 290, they are going to be disappointed for sure.

I am confident AMD can release a R9 380 with 2816 sp (GCN 2.0) which matches or beats the GTX 980 by 5% at USD 300 - 350 and a R9 380X with 3072 sp (GCN 2.0) which is on avg 10 - 15% faster than GTX 980 at USD 400 - 450. btw I am sure that R9 380X is HBM based. Cost (die size), power and performance optimized with 1/16 fp64 rate and a die size around 380 - 400 sq mm. Remember HBM reduces the size of the memory controller on die. GF 28SHP has higher transistor density than TSMC 28 HP. GF 28SHP is also less leaky than TSMC 28HP so that helps with efficiency.

I mean the fastest single GPU in the world today costs $1000 but it's "only" 45-48% faster than the R9 290X. AMD isn't exactly in a pressured position to repeat HD4870 at $299 or HD5870 at $369. Relative to what's out there right now a card 40% faster than an R9 290X priced at $649 would sell well.

R9 290X just has a tarnished reputation but its price/performance is phenomenal. The competitor's card is only 6-8% faster at 4K for an 85% higher price. The reality is it is AMD's game to lose. They know the price and performance of every single competing card. How hard can it be to offer superior price/performance at every level with R9 300 series? It shouldn't be hard at all -- AMD's biggest problem is marketing/branding.

AMD's problem is they have fallen behind Nvidia on efficiency which has definitely affected the image of their products. With HBM and GCN 2.0 AMD have a golden opportunity to correct that. I think AMD will be aggressive with their pricing on Fiji XT (R9 380X) while they will relax a bit with R9 390X (Bermuda XT) as it has to account for yield challenges due to a > 500 sq mm die and a large interposer. R9 390X should come in at USD 800 (worst case) or USD 700 (best case).

 

[Head1985]

No way 380x have HBM and i think you are too much confident...
AMD dont have money for multiple SKU like Nvidia.Thats why so many people think only FiJI is new SKU and rest are rebrands.

 

[xixou]

So, no architectural improvements since 2011? Not likely. They aren't going to spend the money on the latest most advanced memory a year ahrad of nVidia and then slap it on a 4 year old design.



Pretty obvious the "enhancement" of that image is fooked.

Well, I just took the color inversion mask from irfanview
to make it more readable.

I did not add any pixel or so.

 

[xixou]

Oh my god AMD better release something soon or VC&G is going to devolve into madness.

Me too ^^

 

[dacostafilipe]

Said the same thing I said on OCN, I know the "enhance" memes etc. are hilarious but a picture really does the best job at showing some things:

Compared to the XFX 290X DD PCB:

exact same component placement as the hawaii card. (at least in the places that aren't covered by the cooler or VRAM cooling plate

It's also a similar PCB used on the 280X, even if it's not 512bit.

Could be that they want to save costs by reusing the same PCB as the 380X?

The VRAM plate could be there because of the needed PCB strength?

Could also be fake, I know

 

[Alatar]

It's also a similar PCB used on the 280X, even if it's not 512bit.

Could be that they want to save costs by reusing the same PCB as the 380X?

The VRAM plate could be there because of the needed PCB strength?

Could also be fake, I know

XFX 280X PCB is somewhat similar: http://images.bit-tech.net/content_images/2014/05/amd-radeon-r9-280-review-feat-xfx/xfx280-12b.jpg

however there are clear differences compared to the "390" we see here. Component placement just isn't the same at all, while with the 290X DD and this "390" it's identical.

 

[AtenRa]

No way 380x have HBM and i think you are too much confident...
AMD dont have money for multiple SKU like Nvidia.Thats why so many people think only FiJI is new SKU and rest are rebrands.

If you have the money to make a 500-600mm2 GPU, you have the money to just scale down the same chip to make more SKUs.

 

[raghu78]

No way 380x have HBM and i think you are too much confident...
AMD dont have money for multiple SKU like Nvidia.Thats why so many people think only FiJI is new SKU and rest are rebrands.

we will see who is right . My call - Bermuda (R9 390X / R9 390) 520 - 550 sq mm , Fiji (R9 380X / R9 380) 380 - 400 sq mm will have HBM. Trinidad (R9 370X / R9 370) 250 - 270 sq mm will be GDDR5.

If you have the money to make a 500-600mm2 GPU, you have the money to just scale down the same chip to make more SKUs.

well said :thumbsup:

 

[Shehriazad]

If you have the money to make a 500-600mm2 GPU, you have the money to just scale down the same chip to make more SKUs.

This.


And people need to get over their hate of rebrands...Nvidia and AMD both love using rebrands like mad for 2 out of 3 rounds of GPUs...so what's the deal?

Not to mention rebrands can potentially be awesome (See GTX 770)


Here some ideas on AMD rebranding done right for RX 300:

-A Tonga with +10% to 15% performance and 4GB Vram instead of 2? That would be a ROCK SOLID product. 370X Boom!

-A Tonga XT as PC GPU and not just inside an iMac? Possibly as 3 and 6GB version? Fuuuuuuudge yea! Say hello to the new 380!

-A Hawaii XT with +5% to +10% base clock, possibly the same or slightly lower TDP and additional color compression optimization? Again...SOLID product. 380X Fudge yea! (Not to mention 3rd party vendors now know which cooling solutions work well on that card, less terribly cooled versions)




And again...those are just examples and not facts...but as you can see, rebrands have potential to be awesome, too.
Also I wonder what will happen when people find out that Nvidias 1000 series (or whatever they will call it) will probably be at least 50% rebrands as well....700 series did it as well and went as far back as 2 generations(Seriously, there is Kepler, Fermi AND Maxwell in the 700 series, people who don't know any better are likely to receive the bad end of the stick).
Nvidia as well loves to release 2 "lines" (400/500, 600/700) of the same architecture and then rebrand the crap out of it.

 

[HurleyBird]

If you have the money to make a 500-600mm2 GPU, you have the money to just scale down the same chip to make more SKUs.

Making a whole new line makes sense when you're:

A) On a new node and want to extract performance from that node, or
B) Have made major architectural advances (eg. G71 -> G80, Kepler -> Maxwell)

If Fiji is just minor enhancements + HBM, it wouldn't make much sense to scale it down for the rest of the market. HBM is almost certainly too expensive for budget SKUs so you'd be looking at significant cost for minimal gain.

 

[arandomguy]

What people would prefer is the mid-range card to bring last generation high end performance but current technology and features, as opposed to simply just selling last generations high end at a lower price.

Let's say AMD simply relaunched Cayman GPUs that were faster than what they did launch as Pitcarin GPUs. Would you have prefered Pitcarins or Caymans?

Also the issue with AMDs potential new lineup would be consistency. We could be looking at a GCN 1.1 -> GCN 1.0 -> GCN 1.2-> GCN 1.1 -> GCN 1.3 situation (or worse GCN 1.3 -> GCN 1.0 -> GCN 1.2-> GCN 1.1 -> GCN 1.3). That is just confusing as higher end may have impactful drawbacks other than price and power budget.

Let's look at Tonga and Hawaii as potentially being 370/x and 380/x. Hawaii is not simply just faster than Tonga, even the performance difference is rather inconsistent (for instance certain Gameworks effects enabled) due to certain advantages of Tonga's architecture. The larger internal differneces also bring into question the issue of ongoing support and just in terms of game performance possibly a larger issue going forward with lower level APIs (see Mantle performance inconsistentcy across different games and architectures).

Also most people do not care nor buy into the high end market. So while there is currently excitment with Fiji's HBM and how AMD might distrupt the high end against the Titan X and GTX 980 and some hypothetical card(s) in between that really doesn't do anything for me (and the majority of the market). Nvidia distrupted the pricing at these lower segements with the GTX 970. If AMD were to simply just relaunch current cards at current price points (or even higher as they have a tendency to do) that really isn't going to generate much excitement.

 

[digitaldurandal]

-A Tonga with +10% to 15% performance and 4GB Vram instead of 2? That would be a ROCK SOLID product. 370X Boom!

I don't disagree it'd be a great 370.

It is going to be a terrible 380 though.

 

[AtenRa]

Making a whole new line makes sense when you're:

A) On a new node and want to extract performance from that node, or
B) Have made major architectural advances (eg. G71 -> G80, Kepler -> Maxwell)

If Fiji is just minor enhancements + HBM, it wouldn't make much sense to scale it down for the rest of the market. HBM is almost certainly too expensive for budget SKUs so you'd be looking at significant cost for minimal gain.

Why do you believe that smaller chips must have HBM memory controllers ?? You can have a 550mm2 chip with HBM memory controllers, a 300mm2 same architecture scaled down chip with GDDR-5 memory controllers etc etc.

edit: Also, you can make a new lineup if next node will not be available for 10-15 months. We dont expect 14/16nm GPUs before H2 2016.

 

[HurleyBird]

Why do you believe that smaller chips must have HBM memory controllers ??

I didn't say that. What I said was that in the scenario where Fiji only has minor architectural improvements and the majority of its performance increase comes from HBM it doesn't make sense to come up with a completely new chip stack for those kind of minor gains as HBM isn't likely to be cost effective for budget chips. If Fiji houses significant architectural advancements, then it does make sense to make a new stack.

 

[Shehriazad]

I don't disagree it'd be a great 370.

It is going to be a terrible 380 though.

Not as Tonga XT. Keep in mind that the 285 was not full Tonga.

Tonga XT would fit in perfectly.


So far Tonga XT wasn't even available for customers unless they bought an iMac. (and I think that was an M version).


Waiting for the 300 series to release a full Tonga seems like a smart move to me....makes it look like they actually have more new stuff than they actually have. xD And why the fudge not? Tonga XT with a 384 Bit bus and 3-6GB Vram that card is just a golden solution...I have no clue about the performance... But I guess it would easily destroy the 960 and have more minimum Vram.
Since the 960 4gb version is wayyyy too expensive a 6G version of the Tonga XT chip could end up being cheaper.

For the 970/980 "gap" you have the 390 and to battle the Titan X you got the 390X...