Review AMD Radeon R9 Fury X reviewed!

PlayfulPhoenix

Founder of SFF.N
SFFLAB
Chimera Industries
Moderator
Gold Supporter
Feb 22, 2015
1,049
1,960
I'm actually rather disappointed in this, all-told. For $650, you're getting a card that performs slightly worse than the 980Ti (on average), and has mediocre overclocking of ~10%, while on water (so, comparing to the 980Ti on air). And this is with the water cooler AMD bragged was made for 500W parts :(

All the while, it needs 25W more power, making dual-GPU solutions much more challenging on SFX. And, worse yet, you only get 4GB of RAM, which is almost certainly going to be a performance bottleneck in the near future.

If this was a $579 card, I'd be quite happy. If this was a $599-629 card, I'd still be a little happy. But I don't think this is very competitive as a $650 card at all, with what these two reviews seem to be saying.
 

MJVR1

SFF Lingo Aficionado
Jun 10, 2015
92
55
This card only traded blows with the 980 Ti which is a bit disappointing to be honest. But this card definitely has its place in the market. ITX users are gonna love this card because of its smaller form factor and special cooling solution. Because of that, if I had $650 today, this would be my card. Though like Phoenix said, this card would be a killer deal IF it had a lower price point. $599 would have taken the market by storm tbh.
 

rawr

SFF Lingo Aficionado
Mar 1, 2015
137
10
Yeah, this isn't the killer that everyone anticipated. As seen, it's really just on par with the 980 ti or slightly worse.

But for AMD, that's ok. They may not get outstanding sales, but the kicker for them is their next series.

The Fury won't sell badly, and if this release has given AMD some room to play around (with new tech like HBM) so that their next series is executed perfectly, then that's a success in my eyes.

However, if they fail the next series, it will likely be near the end for AMD.
 

PlayfulPhoenix

Founder of SFF.N
SFFLAB
Chimera Industries
Moderator
Gold Supporter
Feb 22, 2015
1,049
1,960
This card only traded blows with the 980 Ti which is a bit disappointing to be honest. But this card definitely has its place in the market. ITX users are gonna love this card because of its smaller form factor and special cooling solution.

Pretty much. Which isn't nothing, either - and AMD did a lot of good work with the shroud, and other things. It's a good card, in balance, and with respect to its place in the market.

Really, I think what they did poorly was set expectations - nVidia was superb at getting people to be wowed by Maxwell's energy efficiency and overclock-ability, followed by the 980Ti's price point later on. When reviewers got their hands on the 980 (non-Ti), for example, they had expectations that were generally exceeded by the performance of the part. Overclocking, in particular, saw words like "jaw-dropping" and "incredible" tossed around.

AMD, conversely, really seemed to hype up the efficiency and the cooler on Fury X, only to have reviews (so far at least) not quite live up to the "nVidia killer" card we were all hoping for, or maybe even expecting to an extent.
 

Phuncz

Lord of the Boards
Editorial Staff
Moderator
Gold Supporter
May 9, 2015
5,250
4,586
Yeah it was a little underwhelming but considering the entire picture, including power, temperature, noise, performance and price, it's a nice card. Let's not forget HBM is new tech and games need to make proper use of it. This card also seems to be at home at 1440p and preferably 2160p.

While it's not "the best card", it's still a good alternative to the 980Ti. Still, the current 28nm cards are a stop-gap, 14nm can't come soon enough. Let's not forget this card is very interesting for SFF and what it might achieve.

Also relevant for SFF:


Dual Fiji :eek: While it is technically cheating with a dual-GPU, this could fit in many SFF cases with the future SX700-LPT and the R9 295X2 is still king of the hill when looking at single cards (single or dual GPU).
 

PlayfulPhoenix

Founder of SFF.N
SFFLAB
Chimera Industries
Moderator
Gold Supporter
Feb 22, 2015
1,049
1,960
Also relevant for SFF: Dual Fiji :eek: While it is technically cheating with a dual-GPU, this could fit in many SFF cases with the future SX700-LPT and the R9 295X2 is still king of the hill when looking at single cards (single or dual GPU).

The problem is, power consumption is just too high. Reviews have seen ~275W at load and ~290W peak for the Fury X - so a two-GPU solution will be 550-600W, just for it. A good 10+% higher than the R9 295x2, which is hardly SFF-friendly even irrespective of its size.

EDIT: Do we know what the length of the dual-GPU card is, actually?
 
Last edited:

Phuncz

Lord of the Boards
Editorial Staff
Moderator
Gold Supporter
May 9, 2015
5,250
4,586
It shouldn't reach 550W, the 295X2 also didn't:



686W total system power on the system below. At the wall, so this is about 624W internally (91% efficiency).
CPU: Intel Core i7-4960X @ 4.2GHz (easily over 130W)
Motherboard: ASRock Fatal1ty X79 Professional
Power Supply: Corsair AX1200i (around 91-92% efficient at that load)

So it's certainly possible if you're not overclocking, on a 700W PSU. I expect the dual-Fiji card to not go (much) over 500W because of the twin 8-pin sockets.
 

Vittra

Airflow Optimizer
May 11, 2015
358
85
I did a 4770K and a 690 in the M1. Asking a 2x120 to dissipate both a CPU and Dual-GPU's heat was more an exercise of "sure it can be done" than something I would find any comfort finding practical application in. You end up with high temps or a lot of noise.

I think it's more if an interesting opportunity in something like the Kimera though, where you have additional rad space available and can reclaim other pci-e slots for different purposes (multiple SSDs, capture cards, sound cards, etc)
 

Phuncz

Lord of the Boards
Editorial Staff
Moderator
Gold Supporter
May 9, 2015
5,250
4,586
Indeed ! The Ncase M1 does seem to reach its comfort-zone limit above 500W builds, but I'd guess getting a dual-GPU card running properly in the Kimera shouldn't be too much of a challenge.
 

4RTEX

Trash Compacter
Feb 24, 2015
51
11
This card is brilliant so far, from construction point of view. Reminds me of FHD OLED TVs, which are same price tag as 4K TVs. Technology is there and it's going to be developed. It's AMDs first shot at these drivers as well. I mean I don't know much about drivers development, but it's got to be different than GDDR. Let's give them some time and I think they will get much much better. I'd buy one today if I had spare cash :p

Edit:
Do you guys know how thick is that radiator and fan?
 
Last edited:

veryrarium

Cable-Tie Ninja
Jun 6, 2015
144
44
Has anyone found a review of Fury X where the reviewer bothered to note the fan speeds at idle and at full load? I've only checked several English reviews so far (TPU, PCPer, HardOCP, guru3D, hexus) and they show noise levels but not fan speeds. Interestingly the noise level at load is more or less only about 1 db higher than at idle, measured at various distances depending on which reviews, making me curious about the fan speeds both at idle and at load and also wonder why AMD decided to go with the 3000rpm version of GT if the fan speed turns out to be low even at full load.
 

PlayfulPhoenix

Founder of SFF.N
SFFLAB
Chimera Industries
Moderator
Gold Supporter
Feb 22, 2015
1,049
1,960
It shouldn't reach 550W, the 295X2 also didn't...

Problem is, that's but a single benchmark that only looked at the power consumed by their entire system - which is a measure that will be intrinsically fuzzy/imprecise and prone to high variability. TechPowerUp, comparatively, measures power consumption directly at the PCI slot and power connectors (emphasis mine):

For this test, we measure the power consumption of only the graphics card via the PCI-Express power connector(s) and PCI-Express bus slot. A Keithley Integra 2700 digital multimeter with 6.5-digit resolution is used for all measurements. Again, the values here only reflect the card's power consumption as measured at its DC inputs, not that of the whole system.


For maximum power draw (so, worst-case, as they run FurMark), it's not very promising. The card can pull a massive amount of power when it wants to, even just at stock clocks:


For the more reasonable "peak" power draw, though (which is just the highest captured power consumption during game play alone), it's still quite high:


It's tempting to assume that, just as the 290X scales favorably to the 295x2, the Fury X could scale well with a second GPU added. But we just don't have any confidence in that assumption for a variety of reasons:

  • The 290X runs very hot on air, while the 295x2 runs a lot cooler on water. With the Fury X, the GPU is already on water.
  • With the 295x2, AMD used binned chips that were more efficient than those used in the 290X. With a dual-GPU card on Fiji, which uses a brand new manufacturing process and architecture, AMD's yields and ability to do this sort of binning could be limited.
  • The power/efficiency scalability of Fiji in general is currently a complete unknown to us - all we know for sure is what the Fury X does in comparison to the 290X.

Basically, we can't be totally confident about much of anything (with respect to the performance of a dual-GPU Fiji card), but there are more than a few signs that indicate that it will likely be very challenging to use on SFX.

Indeed ! The Ncase M1 does seem to reach its comfort-zone limit above 500W builds, but I'd guess getting a dual-GPU card running properly in the Kimera shouldn't be too much of a challenge.

The airflow scheme within Nova is much more conducive to multi-GPU builds, and we've seen that even just a 140mm side fan can sustain positive pressure that clears out heat pretty quickly. Although you'll generally have cards and components see moderately warmer temperatures than in a much larger case, SLI and CrossFire are certainly feasible if (bluntly put) you aren't stupid about what you put in.

And, of course, cards with blocks and rads - such as the Fury X - are an interesting proposition in their own right.
 

Vittra

Airflow Optimizer
May 11, 2015
358
85
This card is brilliant so far, from construction point of view. Reminds me of FHD OLED TVs, which are same price tag as 4K TVs. Technology is there and it's going to be developed. It's AMDs first shot at these drivers as well. I mean I don't know much about drivers development, but it's got to be different than GDDR. Let's give them some time and I think they will get much much better. I'd buy one today if I had spare cash :p

Edit:
Do you guys know how thick is that radiator and fan?

Believe it's a 30mm rad. 25mm on the fan, it's a Gentle Typhoon. That "shroud" part is supposedly extra reservoir for fluid.

Has anyone found a review of Fury X where the reviewer bothered to note the fan speeds at idle and at full load? I've only checked several English reviews so far (TPU, PCPer, HardOCP, guru3D, hexus) and they show noise levels but not fan speeds. Interestingly the noise level at load is more or less only about 1 db higher than at idle, measured at various distances depending on which reviews, making me curious about the fan speeds both at idle and at load and also wonder why AMD decided to go with the 3000rpm version of GT if the fan speed turns out to be low even at full load.

My understanding is they are using an AP-14. 1450rpm.

The black fan blades don't necessarily mean AP-29 (3000+ rpm) anymore, Nidec can paint them black if the person asks - Dazmode ordered a whole bunch of AP-45's (2150rpm) painted black from them.
 
  • Like
Reactions: 4RTEX

Vittra

Airflow Optimizer
May 11, 2015
358
85
Hoping this is the norm moving forward with the next gen offerings.
 
Last edited:

veryrarium

Cable-Tie Ninja
Jun 6, 2015
144
44
My understanding is they are using an AP-14. 1450rpm.
The black fan blades don't necessarily mean AP-29 (3000+ rpm) anymore, Nidec can paint them black if the person asks - Dazmode ordered a whole bunch of AP-45's (2150rpm) painted black from them.
No, it's defintely the 3000rpm version of GT, just look at the rated current (0.22A) on the label.
GT 3000/5400rpm versions have fewer blades (7) and a support ring connecting the blades, the two features that distinguish those high speed versions from the 500/800/1150/1450/1850/2150rpm versions (9 blades, no support rings.)

ETA: I forgot to include the 4250rpm version for the high-speed 7 blade version.
 
Last edited:

Vittra

Airflow Optimizer
May 11, 2015
358
85
I can't see that photo. Is it the same one linked in post #6 of this thread?

No, it's defintely the 3000rpm version of GT, just look at the rated current (0.22A) on the label.
GT 3000/5400rpm versions have fewer blades (7) and a support ring connecting the blades, the two features that distinguish those high speed versions from the 500/800/1150/1450/1850/2150rpm versions (9 blades, no support rings.)

Yes, you are correct. I wonder if the PWM header wire was soldered in? It's a very strict profile at any rate since this is the case, there would have been significant complaint from reviewers if the fan was running anywhere near 3000, as 2150 is already quite loud - I have a couple, one in particular I use with a resistor on the C12P.