[Rumor (Various)] AMD R7/9 3xx / Fiji / Fury

[Azix]

First of all, it's a false comparison. These cards shouldn't even be in the same tier, and the fact that they are is in itself a demonstration of AMD's weakness in the current dGPU market.

Tahiti (280, 280X): 352 mm^2, 4.3 billion transistors
Tonga (285, 380): 359 mm^2, 5.0 billion transistors
GM206 (GTX 960): 227 mm^2, 2.9 billion transistors

One of these things is not like the others. The GTX 960's GM206 chip has a far smaller die and a memory bus half as wide, yet that card trades blows with the Tonga-based R9 285. According to the latest TPU benchmarks, the 285 and 960 have almost exactly the same performance at 1080p (and neither can perform adequately at a resolution much higher than that in newer games).

Tonga is competitive in performance with GM206, and its performance per watt isn't that far behind when you compare the good chips in FirePro W7100 and R9 M295X instead of the trash silicon in the R9 285. But Nvidia managed to get similar performance out of a chip 125 sq. mm. smaller, with over 2 billion fewer transistors, and a memory bus half as wide. That's a technical achievement that AMD simply can't match. We're talking about nearly the equivalent of a full node shrink, simply from a more efficient design.

Oh, and GM206 does HEVC decoding in hardware, while none of the AMD cards do.



Hawaii is not "taking on" the GM204-based cards. In terms of price, it's down there with GTX 960. Yes, that's a second-generation Maxwell, but if AMD needs a chip 92% bigger with a memory bus four times as wide and twice as much RAM in order to compete with Nvidia, then AMD isn't going to be making much money. And those of us who live in Florida (tomorrow's high temperature: 95 degrees) aren't likely to want a 300W+ space heater in our rooms.

And I don't think DirectX 12 is going to rescue AMD. It may move us away from the need for game-specific driver hacks and multi-GPU profiles by putting this work back on the developers where it belongs, and it's true that AMD may benefit slightly more than Nvidia because AMD has fewer resources to do this stuff. But I don't see it making a huge impact.

Its only unfair if they aren't in the same market category. The AMD chips at the higher end have been consistently smaller and nobody made that an issue. Hawaii was over 100mm^2 smaller than the kepler cards and nobody cared.

It is impressive but I suspect its going to be very driver dependent and that will suffer later on as usual. Ultimately for us the price and performance s what we should be looking at, not die size and transistor count. I believe tonga was being designed for 20nm even

Saying hawaii is down there with the 960 is a matter of value. In performance it takes on gm204, in price its down there with the 960. Power consumption is similar to kepler. If you would pick a 960 over a 290 at the same price then there really is no helping you.

For dx12 I believe AMDs architecture is simply better to take advantage of it. I'd bet your money that the 290x will be doing better than the 980 when those games come out. I need some technical info but I am suspecting the aspects of the AMD chips that make them use more power will help with dx12 once async shaders kick in. If it is true that the issue comes from compute units that are on but not being used while gaming. Below maxwell 2 nvidia does not have that option. For things like tressfx and other compute tasks it could mean even less impact on performance.

http://www.redgamingtech.com/asynch...eir-role-on-ps4-xbox-one-pc-according-to-amd/

Get a 290(X) at the cheap prices before they sell out. Unless you are utilizing 2+ cards, a single 390X will probably not be powerful enough to run anything that requires more than 4GB of VRAM.

This is probably a myth spread when nvidia was pushing 2GB on their high end.

 

[geoxile]

It's pretty clear AMD's rebranding because they don't have the budget for it. Lisa Su, or some other exec maybe, said Zen was getting the biggest part of their budget. Along with FIji and HBM in the pipe these past few years they've no doubt had little to no money to spare for for updates to existing die configurations. Speaking of budgets, didn't someone claim 2016 will see a big change in AMD's GPU uarch?

 

[Azix]

It's pretty clear AMD's rebranding because they don't have the budget for it. Lisa Su, or some other exec maybe, said Zen was getting the biggest part of their budget. Along with FIji and HBM in the pipe these past few years they've no doubt had little to no money to spare for for updates to existing die configurations. Speaking of budgets, didn't someone claim 2016 will see a big change in AMD's GPU uarch?

It costs more to rebrand than to simply leave the older generation cards in the market like nvidia did last year.

 

[AnandThenMan]

It costs more to rebrand than to simply leave the older generation cards in the market like nvidia did last year.

This, which is why I don't understand what AMD is doing one bit. Either we are missing something or AMD has made a monumental blunder.

 

[jj109]

This is probably a myth spread when nvidia was pushing 2GB on their high end.

Ah now we have a pickle.

Is the extra 4GB useful on the 390X, and the 4GB limit on Fiji going to be a problem.

Or is the 4GB limit on Fiji enough, and the 8GB 390X is just a pointless cost addition on 4GB 290X.

Going to be difficult to eat both cakes.

 

[jpiniero]

It costs more to rebrand than to simply leave the older generation cards in the market like nvidia did last year.

OEMs want new products.... and new products is what they get. It's pretty much a necessity to keep things 'fresh' with new names even if it is a rebrand or a respin.

 

[Wreckem]

Well, do you want us to give you ALL the information so you can make your own decision, or lead you to your decision?

Not sure if you're of this world, but the moment you even hint "old" people lose their minds. Intermix "old" with "uses more power" "it's also slower" and end it with "but it's cheaper" their first question will be "how much cheaper." "Like $20-30."

Feel free to give your answer where they will go.



Because people will pay more money if they think a product is "all new!"

The problem here is at most price points, you are wrong with the "it's also slower" bit.

The R300 series has aggressive pricing and AMD will be better performance per dollar at most price points.


What is probably going to happen is

Fiji XT will be faster than the 980ti for around the same price.
Fiji Pro will be faster and around the same price as the 980.
The 390X will trade blows with the 970 for around the same price.
The 380 will trade blows with the 960.
The 370 will continue to beat the 750ti.

 

[Wreckem]

Ah now we have a pickle.

Is the extra 4GB useful on the 390X, and the 4GB limit on Fiji going to be a problem.

Or is the 4GB limit on Fiji enough, and the 8GB 390X is just a pointless cost addition on 4GB 290X.

Going to be difficult to eat both cakes.

The 390x with 8GB is clearly aimed at being the best bang for buck performance when it comes to running SLI/Crossfire. Nvidia currently has no competitor to 390x 8GB in crossfire. IE: Dual 390x 8GB is alot more future proof than dual 970 3.5GB.

No one should be buying the 390X unless they are buying 2+ of them for crossfire and planning to run them at higher resolutions. They would be advised to get a 295x2 while they still can...

 

[Azix]

When I read the stuff about dx12 its quite possible AMD is betting heavily on it and mantle pushing it forward could be a huge win for AMD. Its essentially what AMD has built the GCN chips for all these years and its about to become reality. Also the tech is integrated into popular engines already. I see why they'd have a PC gaming conference now. On the verge of something huge.

http://www.redgamingtech.com/infamo...rt-2-ps4-performance-compute-particle-system/

If they do it right PC gaming will beeee awesomer soon

Or not. meh

Ah now we have a pickle.

Is the extra 4GB useful on the 390X, and the 4GB limit on Fiji going to be a problem.

Or is the 4GB limit on Fiji enough, and the 8GB 390X is just a pointless cost addition on 4GB 290X.

Going to be difficult to eat both cakes.

For people who crossfire the 8GB will probably be epic. For people who have a single card they may go over 4GB but stay under 5GB. I don't expect it to be a massive benefit for single GPU folks

 

[LTC8K6]

right? right!

390x just need 10% performance over 980 even at the same power usage as 290x = win

apparently rebrand is now set in stone! amd E3 show? wth do we need it for? concrete info/performance/benchmarks? whatever do we need those for? :biggrin:

Well, that weird dude with the 390X wasn't impressed with the card, or it's performance.

It was a tie with a 970, apparently.

https://www.youtube.com/watch?v=6kEcbLRnbVI

 

[LTC8K6]

The 390x with 8GB is clearly aimed at being the best bang for buck performance when it comes to running SLI/Crossfire. Nvidia currently has no competitor to 390x 8GB in crossfire. IE: Dual 390x 8GB is alot more future proof than dual 970 3.5GB.

The 290X cards with 8gb didn't make any waves:

http://www.newegg.com/Product/Product.aspx?Item=N82E16814150723

 

[geoxile]

It costs more to rebrand than to simply leave the older generation cards in the market like nvidia did last year.

I really doubt rebranding will incur more than a trivial cost for AMD, their partners perhaps once they got briefed to update marketing material. AMD itself has been basically silent about this entire thing.

 

[maddie]

Ah now we have a pickle.

Is the extra 4GB useful on the 390X, and the 4GB limit on Fiji going to be a problem.

Or is the 4GB limit on Fiji enough, and the 8GB 390X is just a pointless cost addition on 4GB 290X.

Going to be difficult to eat both cakes.

In reality HBM and GDDR5 have different properties that might make a straight comparison inaccurate. Smaller page sizes being one, concurrent read/write operation with single bank refresh, another. It seems possible to save some space with HBM. Whether it will be enough is another question.

Answers soon.

 

[Azix]

Well, that weird dude with the 390X wasn't impressed with the card, or it's performance.

It was a tie with a 970, apparently.

https://www.youtube.com/watch?v=6kEcbLRnbVI

He used the drivers from the CD

 

[Head1985]

More 390x benchmarks
http://hardforum.com/showpost.php?p=1041663247&postcount=72
And confirmation that it is GCN 1.1
http://hardforum.com/showpost.php?p=1041663283&postcount=81
and you guys can try flash 290x to brand new 390x
http://www.overclock.net/t/1560231/r9-390x-bios-leaked-on-web#post_24034548

 

[railven]

The problem here is at most price points, you are wrong with the "it's also slower" bit.

The R300 series has aggressive pricing and AMD will be better performance per dollar at most price points.


What is probably going to happen is

Fiji XT will be faster than the 980ti for around the same price.
Fiji Pro will be faster and around the same price as the 980.
The 390X will trade blows with the 970 for around the same price.
The 380 will trade blows with the 960.
The 370 will continue to beat the 750ti.

R9 390 has already been leaked (and sold) costlier than GTX 970, so you might want to re-tweak your whole price structure for this argument.

Without the actual metrics to go on, all we know is that it cost more but brings 4GBs more VRAM. Assuming it is a straight rebrand, we know it will be on par with GTX 970, use more power, have older features, and now cost more.

If that is still a winner to some, there are not enough of you to offset the market share discrepancy.

 

[TechyGeek]

Leaves buyers like me in a pickle. I want a stop gap card for ~$350 with hdmi 2.0 to game until big pascal / 400 comes out. Sucks this is straight 290x rebrand.

 

[maddie]

The 390x with 8GB is clearly aimed at being the best bang for buck performance when it comes to running SLI/Crossfire. Nvidia currently has no competitor to 390x 8GB in crossfire. IE: Dual 390x 8GB is alot more future proof than dual 970 3.5GB.

No one should be buying the 390X unless they are buying 2+ of them for crossfire and planning to run them at higher resolutions. They would be advised to get a 295x2 while they still can...

Now that's a good way to increase market share. AMD should start marketing this.

 

[dark zero]

OK, AMD is dead now... and HBM is the last blow from them.
nVIDIA finally will get the GPU crown in the x86 place.... until Intel nerfs them and makes them DOA.

 

[LTC8K6]

http://hardforum.com/showpost.php?p=1041663247&postcount=72

http://www.3dmark.com/fs/5060799

Looks like the same performance.

 

[Azix]

R9 390 has already been leaked (and sold) costlier than GTX 970, so you might want to re-tweak your whole price structure for this argument.

http://cdn.overclock.net/1/1f/1fd87803_6aGGKLm.jpeg[/img

Without the actual metrics to go on, all we know is that it cost more but brings 4GBs more VRAM. Assuming it is a straight rebrand, we know it will be on par with GTX 970, use more power, have older features, and now cost more.

If that is still a winner to some, there are not enough of you to offset the market share discrepancy.[/QUOTE]

There was a pic showing the 970 as 370 or so at that place. But yeah, that price can go find a cliff. It may be OK but I don't like it. Might as well buy a 290x so why go through all this stress.

My strix 970 was 9000+ (maybe 9500 or so) stock. 10449 is faster

dx12 has to be it for their cards below fiji

[url]https://www.youtube.com/watch?t=103&v=NFEhA_L_VlU[/url]

That game demonstrates asynch shaders and suggests its promising.

They are likely going to couple the release with a new driver that might offer significant gains in performance. That conference will have a lot more worth noting than hardware I think

 

[crashtech]

It would make more sense if the 290X became the 390, leaving the 390X to be something a bit different, there's nothing that says they couldn't have some tweaked silicon that is pin-compatible.

 

[Wreckem]

The 290X cards with 8gb didn't make any waves:

http://www.newegg.com/Product/Product.aspx?Item=N82E16814150723

Your point?

It has a value if you willing to run dual cards and have a high rez monitor.

I mean, dual 290x's are faster than the 295x2, and the 295x2 is faster than the $1k Titan. Right now nothing Nvidia has challenges AMD's performance per dollar when it comes to dual 390Xs/290Xs or 295x2. The 8GB 390x is clearly aimed at people wanting to run crossfire setups.

 

[JDG1980]

What is probably going to happen is

Fiji XT will be faster than the 980ti for around the same price.

Faster than stock 980 Ti, or faster than a decent AIB card with factory overclock? You might say this is comparing apples and oranges, but there's a good chance that the watercooled Fiji will already be running pretty close to its limits, and it's not as if AIBs can improve much on a closed loop cooler. (What would they implement: Peltiers? LN2?)

The watercooled Fury X will have to beat the stock 980 Ti by >10% in order to score a victory here. This is because it will be competing for sales with AIB 980 Ti cards that are easily 10% faster than stock. If it can't do this, AMD will quickly find itself having to drop the Fury X down a tier in price (to $549) to secure sales.

Power consumption is another wild card. Granted, it doesn't matter as much at the ultra high-end, and if it tops out at 300W that will probably be acceptable. But if it goes to 350W then that's 100W more than GTX 980 Ti, and this is where the criticism will start to come in - especially since HBM was supposed to help with power efficiency.

Fiji Pro will be faster and around the same price as the 980.

A lot of that depends on the performance of the Fiji chip, which we currently don't know. Let's say that Fiji Pro is ~20%-30% faster than GTX 980. And it's shiny new technology. OK, that's reasonably competitive at GTX 980 pricing - if power consumption stays within the bounds of reason. I think AMD can get away with up to 250W; if they go much beyond that to Hawaii-like levels, then they're going to have some trouble.

And this is assuming Nvidia's prices remain stagnant. Nvidia could easily drop the GTX 980 to $399 and the GTX 970 to $299, which would really screw up AMD's pricing strategy. Remember, these chips aren't much bigger than Tonga, and have the same bus width and similar power requirements.

The 390X will trade blows with the 970 for around the same price.

Current leaks have the 390X more expensive than the 970. And if the two are "trading blows" at "around the same price" then AMD has already lost, because Hawaii uses about twice as much power, and most buyers aren't putting up with that. Not to mention that AMD and their AIBs are going to be making far worse profit margins with Hawaii's much larger die, higher power requirements (more and higher quality VRMs), memory bus twice as wide, and newly bloated VRAM size.

The 380 will trade blows with the 960.

And if they are priced similarly, then the 960 will be a far better choice because it offers lower power usage and hardware HEVC decoding. AMD can't expect to beat Nvidia at the same price point with a product that is in many ways inferior. Though at least Tonga isn't laughably outdated.

Why AMD didn't release the full Tonga instead of the cut-down chip for the R9 380 is a real mystery. Some sort of exclusive contract with Apple comes to mind, but AMD is selling the full Tonga chip to Dell for Alienware laptops, too. A full Tonga with 4GB of RAM at $199, preferably with the power usage cut down to 150W by better binning, would be needed to beat the GTX 960. AMD could and should have done this even with their limited R&D resources, and their failure to do so underscores how lazy this release cycle is.

The 370 will continue to beat the 750ti.

In raw performance? Sure. And the 750 Ti will continue to outsell the 370 by a massive margin, because the 750 Ti doesn't require an external power connector, which is a BIG deal for people trying to upgrade inexpensive OEM systems.

 

[LTC8K6]

Your point?

It has a value if you willing to run dual cards and have a high rez monitor.

I mean, dual 290x's are faster than the 295x2, and the 295x2 is faster than the $1k Titan. Right now nothing Nvidia has challenges AMD's performance per dollar when it comes to dual 390Xs/290Xs or 295x2. The 8GB 390x is clearly aimed at people wanting to run crossfire setups.

It's just that a lot of people seem unaware of the 290X 8gb cards. I have often seen posts about the "extra ram" on the 390X.