AMD Radeon R9 Fury X Reviews

[UniversalTruth]

I say unknown because they are unknown.
Do you know actual power consumption? Do you know Retail Price? Do you know noise output? Do you know gaming performance? I don't, so I say unknown. Quite easy.

We are offtopic here.
Expected gaming performance is somewhere between Fury and 390X, probably faster enough than 390X.
Retail price - 450$.
Noise output is irrelevent for me, at this point.
Actual power consumption won't be so much different given that the form of the card.

So, now, would be please so kind to move to Pirate Islands discussion thread?

 

[Deleted member 2197]

R9 Fury X Reaches 1215MHz with Voltage Tweak
http://forum.overclock3d.net/showthread.php?t=71840

 

[gamervivek]

I seriously doubt that many game devs will bother to support Mantle when DX12 offers almost all of the benefits. Besides, Mantle's performance benefits are often pretty small in real games today. Those gains will be almost non-existent when DX12 comes around.

I meant is dx12 using mantle itself, as in it's based on that?

Why?

Why not?

To clarify: I meant in terms of people having systems that support DX12. I expect that will occur much faster than the ramp-up time for DX10. Hopefully this will influence game devs to roll out DX12 support sooner.

Considering the ease of dx11 while having all the features of dx12 and nvidia having enough of a leg up in dx11 right now to make AMD look distant second best at resolutions lower than 4k, I don't think it'd be that quick to be adopted.

R9 Fury X Reaches 1215MHz with Voltage Tweak
http://forum.overclock3d.net/showthread.php?t=71840

Around a stock 980Ti's boost clocks. Should've done some firestrike.

 

[Deleted member 2197]

Meh ... they are using LN2 to achieve these clocks. The people using the Voltage Tweak said the following about FireStrike. :

i will post fire strike scores soon! the card just wont run FS at any descent overclock without blackscreen...yet. i am working on getting a few runs in we tryed to ln2 the card today but discovered that it throttles to 300mhz when the core goes below 0c which we beleive is ocp or the cards bios simply does not know the difference between hot and cold and throttle because its out of temp range.

 

[gamervivek]

The video uploaded shows a temp. above zero, in 30-40s, so it doesn't seem to be ln2 only. The voltage is on the high side though with 1.45v, I doubt that is going to used except for benchmarking.

 

[lanek]

Meh ... they are using LN2 to achieve these clocks. The people using the Voltage Tweak said the following about FireStrike. :

What is saying the guys, is they have make some bench but ended with blackscreen ( certainly caused be the driver .. ) .... they have THEN.. try LN2, but the core clock seems throttles when running under subzero to 300mhz ... as for now, it is not practicable for them to a do run under LN2 .. ( specially LN2 is not cheap, so if it is for make a benchs run at 300mhz, lol not valuable )

 

[Deleted member 2197]

Maybe driver, maybe not ... the reviews had no problem with driver in their FS runs. In FS blackscreen is usually caused by memory issues.

 

[lanek]

Maybe driver, maybe not ... the reviews had no problem with driver in their FS runs. In FS blackscreen is usually caused by memory issues.

sorry but maybe we are speaking about different things.. the initial runs was not under Ln2.. the second was..... but they have discover problem.. basically the clocks boards seems to respond strangely ... and i find this really more than more strange verythin else..

Passed the subzero you could have 2 problems .. but one of the biggest is the "cold bug" ... for a cpu or gpu, its like the temperatures stop the chips to start ( for make it short )..

In their case what is really abnromal, it seems that a temperature under subezero, put the clock to idle speed .. ( this said, nothing to do with the initial run with the stock cooling )..

On the run posted at 1215 wit this "non offcial" tweaks ( not supported overvoltage ), they are not under subzero anyway ,, this said this not really impressive at all, and i got the feeling a new firmware need to be provided .

 

[gamervivek]

Fury Voltage scaling.

http://www.techpowerup.com/reviews/AMD/R9_Fury_X_Overvoltage/2.html

Pretty disappointing, barely reaches 1.2Ghz. The improvements.

Core - 1050 to 1215 15.7%
Mem - 500 to 560 12%

BF3 - 50.9fps to 56.5fps 11%

 

[Razor1]

beat me to it, yikes, definitely not an overclocker's dream

 

[Deleted member 2197]

 

[Deleted member 13524]

Fury Voltage scaling.

http://www.techpowerup.com/reviews/AMD/R9_Fury_X_Overvoltage/2.html

Pretty disappointing, barely reaches 1.2Ghz. The improvements.

Core - 1050 to 1215 15.7%
Mem - 500 to 560 12%

BF3 - 50.9fps to 56.5fps 11%

AMD seems to have this historical fixation on disappointing everyone with their reference cards, one way or another.
But here's this bit from that article:

In all these tests, GPU temperature barely moves thanks to the watercooling block. Going from 67°C at stock voltage to 71°C at +144 mV isn't worth mentioning. Heat output definitely increases, though. The watercooler just soaks up all the heat that will ultimately be dumped into your room.

Combine that with this:
http://forum.hwbot.org/showthread.php?t=142320&page=3

And 3rd parties could be making a Fury X with a custom PCB that allows some 1.3GHz on the core and >600MHz on the memory.

 

[gamervivek]

I doubt Fury chips will reach 1.3Ghz except for same rare ones. 1.2Ghz at acceptable level of power usage might be excellent considering hawaii's capabilities.

AMD have this clockspeed disadvantage since Kepler cards which could reach 1.4Ghz with more voltages but with way more power usage than Maxwells. But AMD's couldn't clock that high.

A quicker release of parts on newer node is the best they've got going for them regarding that. And of course, drivers to eek out a little more.

 

[3dilettante]

Somewhere around page 12 or so there was discussion about the bandwidth tests that evaluated the effectiveness of Fury's delta compression.

Some additional information was provided about some of the do's and don'ts for getting the most out of it.
http://gpuopen.com/dcc-overview/

The discussion about how AMD's scheme is implemented has happened sporadically across several product threads, but there are some interesting tidbits. There appears to be some special-casing for the most likely cleared states like 0.0 and 1.0, etc. The bandwidth tests use an all-black texture, which seems to play into this optimization for AMD. It makes sense, and might be similarly optimized for in other implementations. I could be interesting to test with a different color to know how big the difference could be.

Not writing back all channels is more problematic with compression, which seems understandable when the compressor needs to know what is in the rest of the format in order to compress things.

There was some discussion somewhere on where compression data may be, and how much back and forth in read/write traffic between the ROPs and TEX paths might be, and the linked article provides cases where the system becomes conservative in simultaneous read/write scenarios.

There are a number of other optimizations there.

 

[Deleted member 13524]

Somewhere around page 12 or so there was discussion about the bandwidth tests that evaluated the effectiveness of Fury's delta compression.

Eeerm.. I think maybe you replied to the wrong thread? How about this one?

 

[3dilettante]

Around the middle of this thread, there was discussion about delta compression and the bandwidth test run by Techreport. Maybe page 10 is a better point, posts were lost around that time and had to be re-entered, so it's a little less readable than before.

 

[Rys]

Entirely my fault in all cases. I'll take a look at that DCC page and see if I can make changes to the bandwidth tests to make them better analyse how it works. Should really buy a Fury, too, to test it all on!

 

[3dilettante]

Tonga should have a shared interest in these optimizations, although its launch predates Fury by so much that it wasn't tested in this way.
Should GCN benefit from changes in the test, would Nvidia see a difference as well or is it a sign that their implementation is able to remain in effect across more situations?

The memory subsystem isn't split like it is for GCN, which might remove some of the concerns like metadata and data being incoherent, or Nvidia's driver can manage it less conservatively.

 

[Deleted member 2197]

AMD Radeon Pro Duo Preview and Details

In our believe, this dual-GPU Fury was improperly marketed. Having the AMDs CEO holding a Dual-GPU PCB in front of tech press and the world in a livestream creates expectations.For weeks now AMD has been teasing photo's over facebook, showing the product in a VR setup and then at launch date ... they have nothing available for review. All that teasing created excitement and then at the actual launch they do not want to show the product or have it tested makes everybody's eye-browses frown. AMD stated that its Pro Duo is “specifically designed for gamers who create and creators who game”. We can understand why some are baffled and disappointed about this product release, ourselves included.

In the end, even if people cannot purchase the stallion of the product series, they want benchmarks plain and simple to feast their eyes and look at the awesome stuff that two FIJI GPUs can do. Some numbers have leaked out today though (courtesy of expreview) however even these numbers might simply be two Fury X cards in Crossfire set at 1000 MHz:

The Radeon Pro Duo announcement and release feels like a big marketing miss, AMD created hype, they went viral and all media outlets posted about it. Then at launch AMD doesn't deliver for reasons mentioned in this editorial. For the reader-base, the end-users and the consumers this is beyond dissatisfying to see.

http://www.guru3d.com/articles_pages/amd_radeon_pro_duo_preview,1.html

 

[silent_guy]

I'm calling it now: the Polaris introduction will be absolutely flawless. After so many failed attempts, statistics simply dictate that AMD marketing will get things right one way or the other.