AMD Radeon R9 Fury Review: Sapphire Tri-X Overclocked

[FormatC]

[
@Igor What I really want to see are the Fury results run consistent with your ultra/max settings and factory overclock. Didn't you mention getting a Fury for personal use?

In maxed out settings Fury looks not worse, mostly better against the 980 vanilla. The limited memory might be a bottleneck, but only in a few, more or less special cases (game mods, hi-res texture packs etc.). Take a look at my fresh UHD-benches in the power consumption part. On slide 3 you can see the performance results for both cards in this (especiallly for my measurings hand-picked) benchmarks - but the trend is always visible.

It really makes sense to compare this factory-oc'ed Fury from Sapphire with factory-oc'ed GTX 980 to keep price-point and performance in an objective balance. I have here in my archive a lot of 980's with up to 17% more performance as the temperature-limited 980 vanilla. In this case the Fury looks not so sexy because it is simply too expensive.

If other AIBs will also launch their Fury cards - if - we can better decide, what is the best bang for the buck. But I know from a few AIBs that they will not launch any Fiji Pro as custom Fury card. I know the reasons too but this is not for public. We will se after official launch a handfull of models, not more. And I don't know if I'll get a personal sample. May be I have to buy one, again and from my own money. This is not really media-friendly but the reality

 

[Gurg]

@AMD 300 series launch vs. this launch

There were a few reasons why we decided to change the benchmarking setup for the AMD 300 series "launch" story in comparison to the setup we used for most of the other GPU launches - which were mostly high-end cards, mind you.

- The primary focus of the 300 article was different: We mainly wanted to see how much the 300 series cards changed in comparison to the 200 series cards (and the HD 7850 that uses the same Pitcairn GPU as 270 and 370). The Nvidia cards were just in there for comparison and to show a "frame of reference" if you will.
- The 300 series launch article also included "smaller" cards (HD 7850, R7 270, R7 370) that simply can't really cope with specific high-end workloads. It made no sense to use the standardized setups for these - at least from my point of view. After all you wouldn't use the same criteria to compare a family car with a Formula 1 rocket, right?
- This article was done under extreme time pressure. We got the new AMD launch driver two days before launch date. These days it's getting harder and harder to do real in-depth analysis due to these late-term intrusions and the fact that test samples are getting more and more scarce.
- We didn't use reference cards at all but instead benchmarked exclusively retail models in the 300 series article. All those cards where benched completely new for this specific "event".
- We used a slightly different games portfolio to benchmark gaming performance, yes. We benchmarked BF 4, FC 4, GTA V and Mordor like we usually do. We didn't do Metro LL (too demanding for the smaller cards) and Tomb Raider (we didn't have the specific save file/sequence available) but included AC Unity (was asked for by our readers) and Bioshock (as it's not very demanding and gives us a frame of reference usable in APU articles as well. And like we wrote: We prefered to deliver the performance in three different resolutions - 1080p, 1440p, 2160p - to playing around with more games/settings. Plus: Time pressure, remember.

==> All in all you can't compare the 300 series launch article with this specific one.


@Conclusion Fury vs. 980/980 Ti

Please remember that Kevin did benchmark the reference specs (1000 MHz GPU clock) as well. And we included the pricing for the non-OC model as well. So just take these values and this price, then.
Btw: The German Fury launch article's conclusion is different from the US/UK one as the MSRPs and the market prices are totally diferent over here.


- Gerald Strömer, EiC Tom's Hardware Germany

Igor's review of the MSI390x vs MSI980 with the extreme settings and factory overclocks was perfect. I personally would never buy a plain reference card and want to see aftermarket cooler cards at the factory overclock settings as that is what I will evaluate for purchase. The only time I ever want to see reference card results after an initial launch is to see how much improvement the aftermarket brands have achieved.

The problem is with the Fury reviews! What I would like to see is Igor's gaming charts updated to include the Fury and FuryX run at extreme/ultra/max settings with the factory overclock enabled and at the three resolutions 1080-1440-2160.

 

[Dark Falz]

An overclocked 980 (~15%) is about equal to an overclocked R9 Fury, has been out for a year and is $50 USD cheaper. Also chews less power. I guess it's nice to have competition though.

 

[FormatC]

An overclocked 980 (~15%) is about equal to an overclocked R9 Fury, has been out for a year and is $50 USD cheaper. Also chews less power. I guess it's nice to have competition though.

I'm just testing here in Germany seven GeForce GTX 980 Ti custom, nice single reviews with a separate, roundup-style landing and result page after the 3rd published review (with continuous updates). I will compare this later also with Fury X. And if some other Fury cards (Fiji Pro) will enter the market I will do the same with Fury the GTX 980's. I have a lot of this cards in an older, but large roundup with 9x GTX 980 and 6x GTX 970. It is all only in German but the charts are always self-exploring without Google translator.

Just trying all my contacts in Asia to catch a living Fury Pro for my battlefield and archive. But not only Gigabyte will not launch a custom Fiji Pro due the low profit (and other reasons).

BTW:
I know, it is really off-topic, but I've tested exclusive together with Gigabytes R&D guys a new silent BIOS for their GTX 980 Ti Gaming G1, based on my suggestions and fan curves. Lower power consumption, similar performance and finally up to 4 dB(A) less noise. The temperatures are also similar. If someone need it, read the German sister page at Monday. This BIOS was proofed now from NV and is ready for mass production

 

[cmi86]

I would love to see AMD drop the Fury X down to 980 prices and the Fury down to 970 prices and smash nvidia to pieces in the realistic enthusiast segment. Let Nvidia have the ultra high end/bragging rights snob crown, I won't ever be buying a 980Ti/Titan-X/Fury X (currently Priced) and neither will 99.9999999% of the other gamers on this planet. Price the cards where they can dominate, not just compete and increase revenue through increased sales volume. More cards in the market and devs start to favor that architecture and the same card gets even faster so more people buy in. It is the same exact thing nvidia has been doing since the GTX 4XX stuff. AMD needs to start doing this instead of trying to cash in on ultra high end users and leaving the average enthusiasts stuck with slower hardware. Drop the fast stuff down and let green have the snob crown, who cares.

 

[cmi86]

It is funny that when Metro LL dropped it was SO heavily optimized for NV that most AMD users couldn't even play the game until a month or so later when a patch was released turning off all the NV crap that was built in to cripple AMD cards seemingly on purpose. Now here we are some years later and AMD cards are actually faster in Metro LL. Old game, doesn't really matter just goes to show that we should take all reviews with a grain of salt. All it takes is a little monetary motivation for said company to flip a few switches inside the game and that is the difference between a good and a bad review.

 

[Gillerer]

Slapping on a standard tri-cooler onto this short board is effing dumb.

No, on the contrary, its very clever. With the cooling fins overhanging the PCB, the third fan can blow air straight thru the card upwards to the top of the case, without being obstructed by the card. I'd bet it increases the cooling capacity and does its part in keeping noise down.

 

[rvalencia]

I hope you can add the 5.7 driver results.

15.7 is a feature/WHQL driver, not a performance driver. The performance %s that AMD gives on the website are "since Omega".

Nvidia Titan X reduce image quality cheating...

The root of the problem is the image quality setting in Nvidia control panel. Check the difference (and proof) here:

http://hardforum.com/showpost.php?p=1041709168&postcount=84










Screenshots taken from the game

I'd say that is a pretty big difference. Where did the AF go in the first screenshot?

+ there seems to be around 10% perf. drop after setting the quality to highest.




Somehow, I think if the situation was reversed and AMD's default produced worse image quality, these same sites will jump all over it and sensationalize it to the max! The double standards are blatant.







http://hardforum.com/showpost.php?p=1041709226&postcount=100

On AMD, default = higher quality, but it hurts performance.

On NV, default = worse quality, but it runs faster (10% according to BF4 tests).

Do a detailed analysis in recent games, image quality comparisons on default driver settings (since its what you test on), same settings in-game, same location. Then do it forced on quality in CC/NVCP.

You will get a ton of site hits from everyone interested in this issue. If it turns out to be true, then basically NV has "optimized" their way to extra performance for years at the expense of IQ.

 

[alextheblue]

Considering the temperatures being reached with even this large of a cooler, I completely disagree with you there.

That's with the fans running silently in real-world use. Even at default slow automatic settings it still doesn't exceed ~80-81, much cooler than many Hawaii cards and very close to the competition. I like the cooler but it is a bit overbuilt until you have the means to boost voltage - then you could adjust the fan speeds a bit and actually make better use of it.

Then again it's only a $20 premium, and as someone else mentioned most decent gaming cases (even relatively small ones) have made provisions for cards of this length.

i had a 17 inch 1600x1200 monitor. i can tell you right now, that i never once used it at 1600x1200 outside of playing 720p video, that monitor was always at 1024x768... but here is something you are failing to remember, crts did not look like hell outside of native resolution like lcds do...

Yeah that alone kept me from upgrading to LCDs for some time. Drop resolution and it's completely native. A good CRT looked perfect running different "recommended" resolution/refresh combinations. Those were the days. Well, minus the size and weight, haha...

 

[Reynod]

Nice article ... well balanced.

 

[Dodecahedron]

Nvidia Titan X reduce image quality cheating...

The root of the problem is the image quality setting in Nvidia control panel. Check the difference (and proof) here:

http://hardforum.com/showpost.php?p=1041709168&postcount=84

It appears that this was just a mistake in settings and other users could not reproduce the problem, see http://hardforum.com/showpost.php?p=1041717506&postcount=252.

Somehow, I think if the situation was reversed and AMD's default produced worse image quality, these same sites will jump all over it and sensationalize it to the max! The double standards are blatant.

AMD has had image quality problems in its texture filter for a long time. In the Radeon 5000 series there were some obvious banding problems. They were supposed to have been fixed in the 6000 series, but then users noticed texture shimmering, i.e., aliasing. The first GCN products were supposed to address that problem, too. In reality, I still observe both problems on a Radeon R9 280X based on the first generation GCN chip Tahiti. The banding problem is not very noticeable in practice (except for artificial test cases), but shimmering is a real issue in some games. The only way to deal with the shimmering is supersampling but it kills performance. On Nvidia cards (based on Fermi and Kepler) I have not seen banding problems at all and shimmering is also reduced compared to the Radeon.

Far from sensationalizing, the issues have been almost completely ignored. The only places where I have found any information about the problems are discussion forums, some of them in German language.

Do a detailed analysis in recent games, image quality comparisons on default driver settings (since its what you test on), same settings in-game, same location. Then do it forced on quality in CC/NVCP.

You will get a ton of site hits from everyone interested in this issue. If it turns out to be true, then basically NV has "optimized" their way to extra performance for years at the expense of IQ.

This kind of a test I would actually like to see. I have no idea what the results would be with the most modern hardware (such as AMD Fiji and Nvidia GM200), but very likely some differences could still be found.

 

[JackNaylorPE]

In almost every test, Sapphire’s R9 Fury Tri-X outperformed Nvidia’s GeForce GTX 980.

Correction:

"In almost every test, Sapphire’s overclocked R9 Fury Tri-X outperformed Nvidia’s reference GeForce GTX 980."

Again, why no apples and apples comparisons ? Since most users have Afterburner open right after installing drivers, wouldn't such comparison's be relevant ? Is the single digit % overclocked fps results on the Fury X / Fury not relevant compared to the 31% we see on the 980 Ti or the 23% on the 980 ?

 

[somebodyspecial]

@kcarbotte

One more point, even the amazon clicked link, shows new from 509, and it has zero shipping for that price. Direct from amazon is $524.99 but another seller on there has it 509 for SC version also. So again, amazon or newegg both show $509 or less. With the newegg link I checked the ONLY NEWEGG option so those are really newegg pricing direct for $499 on all those, not some funky dealers.

http://www.amazon.com/gp/product/B00NT9UT3M/ref=olp_product_details?ie=UTF8&me=
Even my amazon price was high...$507 for EVGA SC ACX2.0 Base Clock: 1266 MHZ Boost Clock: 1367 MHz Memory Clock: 7010 MHz Effective

"Fury fits nicely between the GTX 980 and 980 Ti in both power and cost. "

OK...LOL. Whatever. In stock and sold by amazon, regular $569, but currently $507. Hmmm...This is just pricing issues, but saying you don't have a Factory OC card in house is confusing anyway. You can't raise the one you have to said clocks to simulate them? Can't just take old benchmarks from a OC review and throw them in the charts? pfft.. So old drivers and bad prices, pretty much nullifying the conclusion of the review IMHO. Also you need to jack up the details like hardocp etc, who turns stuff down on $550+ cards? I guess I'm more interested in highest playable settings/details these days and how cards perform at those settings as that is what I'd be doing at home anyway. Odd choices in reviews these days here.

Also, bring back charts of cards for temps, watts, noise please for easy comparisons.

- you linked the price of a GTX 980 that is cheaper than Fury cards. GTX 980 Ti are still more expensive. - I said the Fury sits right between the two in price and performance - how is that not accurate?

- I don't have a GTX 980 - regular, overclocked, or Ti. - TH has three different Graphics card reviewers that share the workload. - I have never tested a GTX 980 reference card, and the only overclokced card I've ever tested isn't here anymore.

- the numbers that I did take were from old reviews, hence the old drivers. - Do you want old chart data, or new drivers? It can't be both.

- as I've noted previously, the settings used were taken directly from the GTX 980ti review that was published in May. We've never benched with max settings, so if we started doing that we'd have no data to compare to. - there are plenty of places to find the max settings per game, we compare performance from card to card at set settings (not sure why the 390x review was different)

- charts never went away. - Initial reference reviews are done with the detailed tests that Igor added to this article. - we used this card as that reference review since there is no reference cooler, and this is the card we had access to.
The next Fury review will have the familiar charts that custom boards typically recieve.

What I linked to was ALL Superclocked prices, which again highlights how different things would look had one been used for comparison purposes. You are saving massive amounts while getting superclocked 980's, which was the point. Perf of a superclocked 980 would not fit your comment saying fits nicely regarding perf/cost.

In the article you wrote:
"Fury X, aimed more at the Nvidia GeForce GTX 980, at least from a pricing standpoint."
Just proved that pointless when superclocked stuff is far less already at $499 etc.

"In almost every test, Sapphire’s R9 Fury Tri-X outperformed Nvidia’s GeForce GTX 980. Even at reference clock rates it's able to keep up. Fury fits nicely between the GTX 980 and 980 Ti in both power and cost."

I should have given your whole quote, as my comment makes more sense with the pricing I pointed out and getting superclocked cards for FAR less. I'm thinking perf becomes a WASH with these superclocked 980 cards @$500, and then the whole line you wrote needs some fixing right?

"AMD plans to sell the Radeon R9 Fury for $549, which is $10 to $20 over most GTX 980s, undercutting a majority of overclocked 980s."

Again...I could go on but you get the point. Is it so hard to check pricing real quick before posting the article to make sure you aren't posting ludicrous data/pricing comparisons?

"Sapphire smartly keeps the price of its reference clocked Fury Tri-X at $549, and offers the overclocked model for $569, putting it in the path of many overclocked 980 cards."

Yeah if they weren't $479-507 tops. You'd be right with another $60 added on all 980 SC cards. Really if this is the process and you intend to NEVER spot check to adjust a few lines in the articles, your articles pricing info is utterly useless, and in this case PERF too when considering SC 980's are $500. It isn't that hard to check this, and it only affects a few lines in the article that really make a huge difference to buyers. And yeah, when you are that far out of date with drivers, I think the guy who still has the REQUIRED card (980/980ti here) should at least run a few of the benchmarks to put an update at the bottom or top of the article showing how massively things have changed. I mean if extremetech is correct and at least Metro LL (which many sites used) is up to 25% faster that would make reviews look massively different, yet this isn't being shown. How many other games went up even 10% with the last few revs you're not testing? Drivers are a BIG part of the story. It was pretty clear NV had seem Metro LL benchmarks and at the very least attacked that game with a driver update (probably others too considering they had AMD's fury chart saying the won a dozen games vs. 980ti for ages before reviews). My guess is if the may31 driver didn't do it, the june22nd drivers were correcting some games they felt might be slow (June22 is the one extremetech discussed).

Buyers face whatever the situation is when the article HITS, not when it was written, so you guys should spot check info on price just before the articles hit the web. I mean written with the pricing (and perf of a Superclocked 980 at $500) your whole article would be massively different. It certainly wouldn't win in "almost all games" and pricing is just whack period.
http://www.pcgameware.co.uk/reviews/graphics-cards/zotac-geforce-gtx-980-amp-review/
1400+ core/1500+ boost. You can hit some pretty amazing speeds on these and this was a feb 2015 article on one (just as an example). Even out of the box it's a free 10%+ on almost all of them, never mind what you can do yourself when tweaked (8.2ghz on mem also in there). People might settle lower for less heat/noise, but the point is an SC card was the real competition here not regular 980's when considering price or perf on SC versions.

 

[JackNaylorPE]

Listing an entire lengthy quote before responding makes it extremely difficult to read, especially when the response gets lumped into the quotes.

As the SC is basically a reference GPU with a nice cooler, it hardly represents the capabilities of the non-reference 980s.

 

[Scibbo]

I am very interested and exited to see where this new gpu technology leads for amd's the next generation gpu's.

With 1440p becoming more the accepted norm and 4k is out with 8k around the corner. it is a very interesting time to see how amd and nividia technology with adapt and transform over the next 18-24 months.

At 1440p and beyond, AA is beginning to become useless due to monitor pixel densities. Unless you have 20/20 vision, AA I believe now is deemed pointless.

It is going to be very interesting to see how new technology will adapt to 4k 144hz and 8k 60-120-144hz in the not to distant future.

 

[JackNaylorPE]

4k @ 144 Hz is not currently possible with current cable technology. And I don't think we'll cards being able to drive 4k @ > 60 fps in most games until Xmas 2016. 4k represents just 0.06% of the market (6 in 10,000) currently so I don't see it as being a breadwinner for either side in what I would all the "not to distant future".

With the Acer Predator (144 Hz / G-Sync / ULMB) removing all barriers to using IPS for gaming, 1440p is where it's at. The Freesync version is TN and Freesync is currently broken on most monitors requiring a factory firmware update, I'm anxious to see how that one looks, tho w/o ULMB, > 60 fps gaming quality will suffer. The Asus MG279Q (Freesync works) looked promising but again w/o ULMB performance > 60 fps suffers and the response time and lag was more than I expected.

The new 3440 x 1440 Freesync Predator is out but again no ULMB and no more than 75 Hz. Again, this will be curtailed until Display Port 1.3 arrives. The tftcentral review says:

http://www.tftcentral.co.uk/reviews/acer_xr341ck.htm

All in all, it's likely to be a fair amount of time before we see a 144Hz capable 3440 x 1440 screen sadly.

So, if 3440 x 1440 is a "fair bit off", suffice to say 3840 x 2160 is even further off at 144 Hz.

Starting with the 2016 Xmas generation of cards, I think we'll start to see HBM making a significant impact. Now that Freesync is fixed, I am anxious to see what AMD comes up with in the ULMB department as FreeSync / G-Sync start losing their luster over 60 fps.

 

[Blueberries]

If we had affordable solutions for 4k 144hz nobody would be in business.

 

[Gurg]

4k @ 144 Hz is not currently possible with current cable technology. And I don't think we'll cards being able to drive 4k @ > 60 fps in most games until Xmas 2016. 4k represents just 0.06% of the market (6 in 10,000) currently so I don't see it as being a breadwinner for either side in what I would all the "not to distant future".

With the Acer Predator (144 Hz / G-Sync / ULMB) removing all barriers to using IPS for gaming, 1440p is where it's at. The Freesync version is TN and Freesync is currently broken on most monitors requiring a factory firmware update, I'm anxious to see how that one looks, tho w/o ULMB, > 60 fps gaming quality will suffer. The Asus MG279Q (Freesync works) looked promising but again w/o ULMB performance > 60 fps suffers and the response time and lag was more than I expected.

The new 3440 x 1440 Freesync Predator is out but again no ULMB and no more than 75 Hz. Again, this will be curtailed until Display Port 1.3 arrives. The tftcentral review says:

http://www.tftcentral.co.uk/reviews/acer_xr341ck.htm

All in all, it's likely to be a fair amount of time before we see a 144Hz capable 3440 x 1440 screen sadly.

So, if 3440 x 1440 is a "fair bit off", suffice to say 3840 x 2160 is even further off at 144 Hz.

Starting with the 2016 Xmas generation of cards, I think we'll start to see HBM making a significant impact. Now that Freesync is fixed, I am anxious to see what AMD comes up with in the ULMB department as FreeSync / G-Sync start losing their luster over 60 fps.

People are talking about HBM as a done deal. But with the die shrinks coming next year, a safer more economical approach to generating substantially more performance. might be packing much more capacity and speed into the the same imprint chips and cards without increasing the heat and throttling.

 

[Eggz]

Great showing at this price point!

 

[computertech82]

The arrows on the charts work fine for me. As for the video cards, looks like fury and 980 TI are pretty close to each other. Will probably boil down to price.

 

[blazorthon]

The arrows on the charts work fine for me. As for the video cards, looks like fury and 980 TI are pretty close to each other. Will probably boil down to price.

The complaint isn't about how well the arrows function. They block part of the charts because they're too large and opaque.

 

[JackNaylorPE]

If we had affordable solutions for 4k 144hz nobody would be in business.

There are no solutions at any price for 4k at 144 hz. Two 980 Ti's struggle to get to 60 in most of the games in TPUs test suite. Until Display Port 1.3 drops, ULMB grows into that realm, and we slough through two generations of GFX cards, it just ain't happening.

 

[blazorthon]

People are talking about HBM as a done deal. But with the die shrinks coming next year, a safer more economical approach to generating substantially more performance. might be packing much more capacity and speed into the the same imprint chips and cards without increasing the heat and throttling.

A smaller fab process isn't enough for GDDR5 to catch HBM and if you shrink GDDR5, then why not shrink the HBM memory for even more performance per watt? Sure, if you want 16 or 32 chips then GDDR5 is the answer, but that's ridiculous right now, especially with the smaller fab process where HBM should easily scale to 8GB or 16GB.

 

[jimmysmitty]

If we had affordable solutions for 4k 144hz nobody would be in business.

There are no solutions at any price for 4k at 144 hz. Two 980 Ti's struggle to get to 60 in most of the games in TPUs test suite. Until Display Port 1.3 drops, ULMB grows into that realm, and we slough through two generations of GFX cards, it just ain't happening.

I think Pascal and AMDs Greenland will both get us closer to 144Hz 4K gaming. Only because we have been on 28nm for quite a while and we have gone pretty far. A Fury X is quite a bit more powerful than the first 28nm GPU, the HD7970, and as is a GTX 980Ti. A die shrink alone should lower heat/power and increase performance by adding cores. As well both will be on HBM 2 and have other enhancements.

To top that all off we should have games using DX12 which should provide a bit of a boost in some cases.

 

[blazorthon]

I think Pascal and AMDs Greenland will both get us closer to 144Hz 4K gaming. Only because we have been on 28nm for quite a while and we have gone pretty far. A Fury X is quite a bit more powerful than the first 28nm GPU, the HD7970, and as is a GTX 980Ti. A die shrink alone should lower heat/power and increase performance by adding cores. As well both will be on HBM 2 and have other enhancements.

To top that all off we should have games using DX12 which should provide a bit of a boost in some cases.

They may bring us closer, but close enough? I doubt it. DirectX 12 is not for making the game run MUCH faster as much as it is for reducing overhead. That will help a lot, but mostly on the CPU side and with multi GPU setups. If it's like a 20% gain or even 30%, which I believe is stretching it, that's still not going to make 4K 144Hz a reality in today's most intensive games.