Well done AMD, well done.. HOWEVEEEEEEEER!

REEEEEEEEEE

>>60706151
Who was in the wrong ?

>>60706444
What are you talking about?

It's likely the right one is a cheaper part at lower clocks/shaders or still a ES

>>60706151
What did they mean by this?

>>60706151
all I see is the Frontier Edition having more RAM and a higher clock than a prototype.

>>60706151
How much bandwidth is 2 stacks of HBM2 at 945MHz?

>>60706151
What I'm supposed to see here?

>>60706475
Fury X at 1600MHz and unspecified IPC improvements.

Slower than a Titan XP at 2000MHz

I just want vegan to come out already so all this speculation can end.

If people could stop at thinking it's at least a fury at ~1500+mhz and be surprised when it's not. People spouting it's going to be 50% faster per clock is probably horse shit, some improvement sure but not magical amounts.

>>60707010
It's 60% higher clock than Fury X, that's a really good baseline before any IPC improvements.

And for performance at those clocks, we only got that LN2 of a Fury X at 1400+ MHz

>>60707106
That's my point. Fury +50-60% is a easy metric. Add in the uncrippling of the Fiji front end. Will probably only help it perform near its top potential more often (as opposed to fury having varied performance from doom down to project cars) some more from second generation hbm controller and that's it probably. The die size puts it near the 1080ti and that's it really.

People spouting hbcc and 11ops per clock improvements are not likely going to mean that much in the end. I'll eat my hat if it's more that 15% faster than the 1080ti. And I really do want a Vega nano regardless.

>>60707190
>I'll eat my hat if it's more than 15% faster than the 1080ti
Most people are expecting it to be 1080 performance at this point so it would be a pleasant surprise regardless.

>>60707212
Those people are probably shit posters.

Die size alone puts it near the Ti

>>60707287
The shitposters are expecting it to be sub-1070 performance.

>>60707212
Most people are dumb as shit.
You got clockspeeds, you got die size, you got transistor count.

That's all it takes to make a good estimate on performance.

>>60706151
It would be fucking amazing if it can actually pull another 200MHz.
And if the FE is clocked that high, which should be 220-250W, I'm getting real interested in RX.
AMD would be dumb to not offer a RX version clocked to the wall, because a lot of people don't overclock and it looks fucking nice on benchmarks as Nvidia nicely does it, the 1080ti has a 1582 boost clock officially but it actually goes from 1600-1800MHz in games on the FE cards, that's clearly far over its rated boost clock, and so people think they're getting 30% more clockspeed when they're overclocking it, which is not the case.

>>60706151
10% faster than Fury X is still slower than 1070

Vega is DOA

>>60707301

>>60707704
It's like clockwork.

>>60707714
Reminder to report shitposters.

>>60707704
not really sure how thats possible considering the fury x is on par with a 1070

>>60707977
Vega is just shit

>>60707992
This

>>60707992
>>60707840

So how fast would be Fury X at 1600MHz?
Both have the same shader count, but one is a lot higher clocked.

Ignoring any architectural changes of course, there are a lot.

the hardware looks ready but I bet they're having huge driver issues and aren't anywhere near being done with them

at least according to Raja's recent tweets

File: fury_ln2_03.png (646KB, 1414x1095px) Image search: [Google]

646KB, 1414x1095px

>>60708393
Like this.

Just that core clock is lower than 1600 while HBM is overclocked a lot(take off 10-15% from the score)


It's not 1600MHz, but it's our best baseline.

>>60708468
Do take note that Fury X's memory controller was funky and couldn't actually use 30% of the bandwidth, so that's another variable.

>>60708512
So uh.. close to a 1080ti without any architectural changes?

>>60708777
Yes?

>>60708777
Shitga won't TOUCH the 1080ti

>>60709546
>>60707840

>>60709546
Fucking Fury X at these clocks touches it. Even with shitty memory controller and front-end choking everywhere.

Titan X performance at 1080ti prices.

>>60709649
Big Vega is probably $550.

>>60706461
Or, Frontier Edition and a Vega powered 680

>>60710494
Both have the same amount of shaders, so no.
And you're not getting a midrange cards with HBM

>>60710527
>no ultra tiny single slot half height 150W cards that can actually be cooled
REEEEEEEEEEEEEEEEEE

>>60710527
well, they could use normal RAM for the mid-class / hevily cut versions.

File: dota2-linux.png (238KB, 759x1867px) Image search: [Google]

238KB, 759x1867px

>>60707977
Radeon R9 Fury isn't even close to the 1070.

>>60709662
I doubt you'll get a VEGA for $550 any time soon. AMD is targeting the high-value high-end market segment. My guess is closer to $900 initially. They will just release their best performing card this round and only later will mid and low-range cards that replace the RX570 and the RX580 appear.

>>60712378

>OpenGL not Vulkan

>Not even 4k

What an insecure faggot.

>>60712495
Testing Vulkan isn't possible in the rather interesting GPU test I had open in the tab next to /g/ when I posted this,

http://www.phoronix.com/scan.php?page=article&item=51-phoronix-13&num=1

...check it out yourself. It's kind of cool.

There are some Vulkan tests somewhere on that site too, if you're particularly interested.

>>60712378
1080ti is loads behind 1060,1070 and 1080. Sure, 10/10 believable metric.

>>60712880
Yes, that's the situation, the 1080ti really is behind the 1060, 1070 and the 1080.

0/10 on your reading skills there, this benchmark is free software. You can install Ubuntu 17.04 and NVidia's drivers version 381.22 and run the phoronix test suite and verify the results.

Should the 1080ti be way behind the 1080? No, of course not, but buying this card really is a total waste of money if you're running GNU/Linux. Perhaps NVidia will fix whatever it is that makes it run slower than the 1060 on GNU/Linux or perhaps they just (correctly) figure that nobody's going to buy that card for a GNU/Linux box.

>>60712378
>closer to $900 initially.
So you think they'll release one that outperforms 1080ti only, and just keep the worse binned ones instead of selling them as cheaper versions?

>>60713096
The point isnt that the benchmark is wrong, just irrelevant