Possible Radeon 390X / 390 and 380X Spec / Benchmark (do not hotlink images!!!!!!)

Final8y · 3 Jun 2015 at 10:03

ubersonic said:
Well, actually while my post you quoted doesn't change that "fact" it does prove it to be wrong and that they did not come first

If you are incapable of following the context that the TitanX and 980Ti came after the 980/970 then yeah.

r7slayer · 3 Jun 2015 at 10:05

eyetrip said:
I was under the impression they were launching Fiji with 4GB first, with 8GB some time later... If that is incorrect, well, consider me corrected!

Did I correct you? I said ppl sayin 4gb only but if u read it points otherwise... Where 8gb cards are comming. Not 4gb only... This means 4gb and 8gb not just one or the other. Lol

Martini1991 · 3 Jun 2015 at 10:05

TallPaul_S said:
Not when I'm running a 1.5GB GTX580, either single card or SLI (depending on the game/temps).

Also it's the same story every time, "oh the next architecture will blow everything away, 40/50/60% faster!!!11!"

When it actually arrives, it's the usual increase of 15-30%.

If I was to wait for the next set of cards, it would be for the 2nd gen when the full fat chips are released (at least for the nvidia cards), so that's 18 months at least.

That's not the usual increase, or at least it's not meant to be.
The 5870 basically doubled the performance of the 4870.

The 7970 basically doubled the performance of the 6970 too.

Jono8 · 3 Jun 2015 at 10:06

Final8y said:
The 980 4GB and the 970 3.5GB came first, talk about short memory.

8-9 months ago....

CAT-THE-FIFTH · 3 Jun 2015 at 10:06

Mtom said:
Well for calling it a fail lets just wait for benchmarks first.

BTW Its really funny that for many people the thing nv has an uper hand is above all. When amd has more vram no one cares, when nv has more vram its above all.
Same for power consumption....

Lots of fruit lovers I suppose?

Final8y · 3 Jun 2015 at 10:07

Jono8 said:
8 months ago....

Which still does not change the fact.

And if AMD was using DDR5 it would be 8GB and 4GB at release, they are using new tech there is nearly always a compromise to be made at first.

CAT-THE-FIFTH · 3 Jun 2015 at 10:09

Let's see the actual performance and pricing before saying its a GPU uber fail LOL.

I am still more worried about the midrange cards though.

Muz333 · 3 Jun 2015 at 10:10

Looking forward to seeing benches and what real world benefit HBM can bring.
I'll maybe jump on a new card but at the same time a die shrink next year is maybe the time to go all in and move up to 4K.

r7slayer · 3 Jun 2015 at 10:11

Bonjour said:
At least, once the 4GB version is released we'll get an idea of performance and whether it's worth waiting for the Fury 8GB.

Atleast I see some one being constructive and using their head instead! Its exactly the same thoughts as mine. See how Hbm does if 4gb is as much as we see at launch. Then wait for the 8gb if it performs well tho with Hbm im sure the more memory the more bandwidth so slightly more performamce? We will have to see.

Final8y · 3 Jun 2015 at 10:12

Bonjour said:
At least, once the 4GB version is released we'll get an idea of performance and whether it's worth waiting for the Fury 8GB.

Indeed.

ubersonic · 3 Jun 2015 at 10:13

Jono8 said:
8-9 months ago....

He didn't actually mean first he just meant before the TX/ti, he re-clarified his statement above.

fs123 · 3 Jun 2015 at 10:13

The revised Hawaii cores will be 8gb as standard IIRC. These are supposed to compete with the 970/980 cards. If priced right they will be the main sellers imo. The top end Fury will be enthusiast level performance and price.

Davedree · 3 Jun 2015 at 10:14

Martini1991 said:
That's not the usual increase, or at least it's not meant to be.
The 5870 basically doubled the performance of the 4870.

The 7970 basically doubled the performance of the 6970 too.

Unfortunately the 6970 was a massive disappointment in performance over a 5870.
:-)

so the 7970's scaling wasn't linear like 4870-5870.

Martini1991 · 3 Jun 2015 at 10:16

Davedree said:
Unfortunately the 6970 was a massive disappointment in performance over a 5870.
so the 7970's scaling wasn't linear like 4870-5870.

The 6970 was on the same fabrication as the 5870. That's why it was a poor performance gain.

What we're witnessing at the moment is unprecedented, three full generations (From both vendors) on the same fabrication is unheard of.

efish · 3 Jun 2015 at 10:17

Two weeks time I suspect what we get are two cards, one red one green around 980ti speeds and at attractive and competitive price points by the looks of it.

Whats not to like?

xtremedetailz · 3 Jun 2015 at 10:18

http://www.guru3d.com/news-story/computex-2015-exclusive-amd-fiji-gpu-die-photo.html

Gregster · 3 Jun 2015 at 10:18

Final8y said:
The 980 4GB and the 970 3.5GB came first, talk about short memory.

The 970 has 4GB as well.

Martini1991 said:
The 6970 was on the same fabrication as the 5870. That's why it was a poor performance gain.

What we're witnessing at the moment is unprecedented, three full generations (From both vendors) on the same fabrication is unheard of.

This is true. It makes me worry for the future and what will be happening over the next 10 years or so. 16nm next and then 10nm etc but if yields are poor, will we be on 16nm rehashes for 3-4 years?

Davedree · 3 Jun 2015 at 10:19

Martini1991 said:
The 6970 was on the same fabrication as the 5870. That's why we've got slow progress.

What we're witnessing at the moment is unprecedented, three full generations (From both vendors) on the same fabrication is unheard of.

I know 32nm half node was cancelled, just like half node.20nm.
But my linear point still stands. They had to.chop a lot out of the 6970 especially shaders to get it to work on 40nm. Humbug feels it's fine to use 28nm so it must be all ok.
:-)

Martini1991 · 3 Jun 2015 at 10:19

Watercooled and 2 8 pins doesn't scream performance per watt for me

Martini1991 · 3 Jun 2015 at 10:21

Davedree said:
I know 32nm half node was cancelled, just like half node.20nm.
But my linear point still stands. They had to.chop a lot out of the 6970 especially shaders to get it to work on 40nm. Humbug feels it's fine to use 28nm so it must be all ok.

But that's why the performance gain is small. Once a die shrink hits, performance will leap ahead again. But 15-20% gain certainly isn't the "usual".