[PcPer] 2- and 3-Way Multi-GPU performance review (Fury X, 980Ti)

[IEC]

"They" as in PCper didn't stack the odds or anything dumb like that in favor of the 980 ti's. They simply used cards that were sent to them to review. Why are people so dramatic?

It's not an apple-to-apples, baseline-to-baseline comparison.

At minimum, you have reference card versus top-of-the-line factory overclocked cards.

Then add in an open bench testing setup which would help IMMENSELY with cooling for air-cooled SLI.

Despite this, the Fury X still scales a whole lot better, and I'm going to wait to see how the Fiji X2 card fares. Still got some time until my next upgrade in the fall

 

[pj-]

I don't see how you could call this comparison biased. The first thing you see when you load the article is the 3 different 980TIs, they aren't trying to hide anything, and the results speak for themselves. Fury X scales better.

According to the comments the author even accidentally left out the skyrim page which AMD didn't do well in. BIAS AGAINST NVIDIA

Edit: I just had a thought and asked a question in their comments.

Which 980ti was used in which role? Single = x, SLI = x,y? Seems like scaling results could be significantly affected by which card was first and which was second. E.g. Reference -> Ref + Zotac would have better scaling than Zotac -> Zotac + Ref

Rough example

Reference = 100% performance
Zotac = 110% performance

Situation 1: Reference solo compared to Ref + zotac, assuming 75% scaling

(110*.75 + 100)/100 = 83% effective scaling

Situation 2: Zotac solo compared to Zotac + Ref, 75% scaling

(100*.75 + 110)/110 = 68% effective scaling


I'm sure the math doesn't exactly work like that, but the % increase would definitely vary depending on which card was used when.

 

[IEC]

I don't see how you could call this comparison biased. The first thing you see when you load the article is the 3 different 980TIs, they aren't trying to hide anything, and the results speak for themselves. Fury X scales better.

According to the comments the author even accidentally left out the skyrim page which AMD didn't do well in. BIAS AGAINST NVIDIA

Bias by definition has nothing to do with whether preferential treatment is or isn't out in the open.

noun
noun: bias; plural noun: biases

1.
prejudice in favor of or against one thing, person, or group compared with another, usually in a way considered to be unfair.

Systematically skewing test setups to compare apples to oranges isn't very professional.

Accidentally forgetting to include meaningful results just further demonstrates the lack of professionalism of PCPer.

 

[pj-]

Unprofessional and biased are different. The dude in the video seems pretty even handed and I don't see how anyone could read that article and come away thinking 980TI SLI is a better idea than Fury X CF. If they are biased against AMD they are hiding it deep between the lines of the article I read.

 

[SniperWulf]

I swear, people are funny. Review sites can't win for losing because someone always gets butthurt over whatever the results are.

The bottom line is this:
Single Card at 4K, it's a wash, Both with some, both lose some
Single Card at 1440p, 980Ti whips up on Fury pretty good

Dual GPU at 4K, Fury takes it all except for GTA5 based on these results (and only one review)
Dual GPU at 1440p, Both with some, both lose some

All that said, just buy what you want and call it a day! I bought a Fury X because I wanted to continue using FreeSync. Am I dissapointed with my purchase? Nope. Not in the least. Am I concerned about what my friend and his 980Ti are doing? Not in the least (although I do have to endure some [redacted] talking from time to time).

Profanity isn't allowed in VC&G

-Elfear

 

[moonbogg]

No one could pass a blind test IMO when asked if FuryX Crossfire or 980TI SLI was running a game. In order to tell the difference, you'd need nothing less than two empty GPU boxes sitting there on the floor next to the test rig.

I'd still take the 980TI's. They are more desirable honestly for quite a few reasons. The most important listed below.

Faster at stock and single card performance
OC beasts
More Vram
Gsync
Plus Sexy

 

[railven]

No one could pass a blind test IMO when asked if FuryX Crossfire or 980TI SLI was running a game. In order to tell the difference, you'd need nothing less than two empty GPU boxes sitting there on the floor next to the test rig.

I'd still take the 980TI's. They are more desirable honestly for quite a few reasons. The most important listed below.

Faster at stock and single card performance
OC beasts
More Vram
Gsync
Plus Sexy

I'd imagine the resale value will be higher too (at least I'm banking on that haha.)

 

[amenx]

Hardware.fr CF vs SLI review:

http://www.hardware.fr/focus/111/crossfire-radeon-r9-fury-x-fiji-vs-gm200-round-2.html

14 games tested, only in 4k.

 

[Pariah]

It's not an apple-to-apples, baseline-to-baseline comparison.

At minimum, you have reference card versus top-of-the-line factory overclocked cards.

Why should Nvidia be handicapped because AMD doesn't allow aftermarket Fury X's, nor despite publically bragging otherwise (read...lied), the Fury X is a completely junk overclocker. Base clock comparisons are largely useless to the enthusiast market who rarely spends $650 (or $1300) on video cards to run them at stock clocks.

 

[Majcric]

I'd imagine the resale value will be higher too (at least I'm banking on that haha.)

You and me both. And this should be even more true should 4gb Vram start to become an issue.

 

[Silverforce11]

Did the 780Ti hold its resale value really well? The thing is slower than a R290X these days, in newer games, often slower by a lot. When Pascal comes.. I wonder whether we'll see a repeat of history with Maxwell no longer being the focus for optimizations.

 

[Zanovar]

Did the 780Ti hold its resale value really well? The thing is slower than a R290X these days, in newer games, often slower by a lot. When Pascal comes.. I wonder whether we'll see a repeat of history with Maxwell no longer being the focus for optimizations.

.:hmm:.I wouldnt be confident.

 

[raghu78]

Did the 780Ti hold its resale value really well? The thing is slower than a R290X these days, in newer games, often slower by a lot. When Pascal comes.. I wonder whether we'll see a repeat of history with Maxwell no longer being the focus for optimizations.

The fact that Nvidia Kepler cards are significantly slower to the competing GCN cards points to a careless attitude from Nvidia towards its past customers.

http://www.computerbase.de/2015-07/sapphire-radeon-r9-fury-tri-x-oc-test/4/

In fact it feels more like a strategy to get customers to upgrade to Maxwell by neglecting Kepler driver improvements.

 

[SPBHM]

move 1 year earlier and the support from Nvidia is way better (Fermi vs anything pre GCN), so it's not that simple... this time around AMD was more stagnant with their architecture than Nvidia, and the optimization for GCN 1.1/1.2 worked better with GCN 1.0 than Maxwell to Kepler I think.

 

[3DVagabond]

move 1 year earlier and the support from Nvidia is way better (Fermi vs anything pre GCN), so it's not that simple... this time around AMD was more stagnant with their architecture than Nvidia, and the optimization for GCN 1.1/1.2 worked better with GCN 1.0 than Maxwell to Kepler I think.

GCN is just a better all around performer.

 

[Jacky60]

Right like going from 1200-1300 cherry picked review / top of the line 980 ti cards to throttled 1000 MHz boost clocks in a case?

Imagine the 980 TI SLI throttling losing up to 20-30% of those speeds, plus the latest driver if it gives a performance boost. The situation would be considerably further in favor of the fury x's.

Just another example of NV Per using NV's best of the best scenario to downplay the advantage AMD has over them atm. I bet you can imagine the test case if AMD had a bad reference card anymore.

I was a relatively partisan AMD supporter as my posting history attests. However my reference Zotac 980tis in SLI boost to 1380 in a case with a relatively aggressive fan profile with ease. I'll try and do an extended gaming session today to see if I can fry them, ambient is 26c and the overclock makes a big difference compared to my 295x2s in quadfire which refused to do more than 1050 core. Yes the 980ti SLI is a bit louder but I have a fan in here anyway to cool me and I've had 290 trifire before (reference models ) so a bit of 'ambient' noise doesn't bother me. Finally I think I have a superior image quality to the 295x2 quadfire. 980ti SLI overclocking has made a real world difference to me with 10-20% faster in games than at stock. The 50% more Ram to begin with makes Arma 3 a much better experience. I can get 2xMSAA and just as fast fps in GTA5. A better resourced driver team will have time to optimise more games more often. If they can't make the hardware do any better than it's physical limitations idle Nvidia engineers can think of creative ways to 'win' in game benchmarks. I'm finally officially happy with my 980ti SLI. I'm concerned that AMD simply haven't the resources to properly support their high end in driver terms despite having often had equivalent or superior hardware. I desperately wish Samsung or someone with deep pockets would buy them to give them the resources to compete again.

 

[Keysplayr]

GCN is just a better all around performer.

Sure, if you throw enough shaders at the problem. I realize that an AMD shader is different than an Nvidia shader, but how different could they actually be? Both need to perform similar function. FuryX has 33+% more shaders than 980Ti. So, a sweeping statement like "GCN is just an all around better performer" doesn't really compute. Unless you're running certain compute functions. I made a PUN!! Not really. :biggrin:

 

[monstercameron]

Sure, if you throw enough shaders at the problem. I realize that an AMD shader is different than an Nvidia shader, but how different could they actually be? Both need to perform similar function. FuryX has 33+% more shaders than 980Ti. So, a sweeping statement like "GCN is just an all around better performer" doesn't really compute. Unless you're running certain compute functions. I made a PUN!! Not really. :biggrin:

And most devs are heading towards compute shaders, this might be a bit removed from the pc space but take a look at media molecules dreams
https://forum.beyond3d.com/threads/...cons-as-used-in-ps4-title-dreams-spawn.57006/

Also look at smaa and other shader based aa solutions. It is only a matter of time before we travel back in time to more software based solutions .

 

[therealnickdanger]

You think its hyperbole because... i'm right?

I was referring to your use of the words "stomp" and "pwn" and the accusation that PCPer was intentionally trying to give NVIDIA the edge instead of just doing GPU testing the way they usually do it. The Fury X can hold its own in some games and clearly CF setups do better in many games, but I wouldn't consider it a total and complete victory - not enough to justify using words like that.

Now when we get the tools to modify BIOS/voltage settings, then I think we'll see something closer to the stomping/pwning you're talking about. I'd really love to see unrestricted Fury X vs unrestricted, water-cooled 980Ti and see what happens.

 

[Sabrewings]

.:hmm:.I wouldnt be confident.

I wouldn't either. Nvidia put a lot of work into tearing apart Kepler and putting it back together as Maxwell. However, I would be confident that the architectural difference from Maxwell to Pascal are much less than Kepler to Maxwell. After spending that much time and money squeezing Maxwell's design for the most you can due to a node freeze, it makes no sense to start over again on the next node. They learned a lot on Maxwell and most of it will easily carry forward into Pascal.

 

[Shmee]

Hmm, I guess this just shows that Nvidia needs to implement their own version of XDMA SLI soon, will it be coming with Pascal? If pascal has a good SLI solution that may be my next upgrade, though it depends if AMD has another gen out around then as well.

 

[Subyman]

This is why I water cool dual card setups. Without water-cooling, dual cards are hard to OC and loud. Fury X is the optimal setup here because of the AIO, that is obvious. Would be neat to see dual card EVGA AIO 980TI vs the Fury Xs.

 

[Kenmitch]

Hmm, I guess this just shows that Nvidia needs to implement their own version of XDMA SLI soon, will it be coming with Pascal? If pascal has a good SLI solution that may be my next upgrade, though it depends if AMD has another gen out around then as well.

They tend to not follow in AMD's footsteps. Swallow pride and copy them? Spin it as something totally different I guess is another option.

 

[Sabrewings]

They tend to not follow in AMD's footsteps. Swallow pride and copy them? Spin it as something totally different I guess is another option.

I wouldn't call it following in their footsteps as much as following a smart business model. I think AMD knew they were going to be short on R&D funds further in the future and adopted an architecture update model to facilitate less work on their part. Nvidia wasn't ready to call it quits with their architecture development and have been pushing ahead through various evolutions of their architectures to pretty impressive power and performance efficiencies. GCN is showing its age in this respect. Nvidia is meeting or beating their performance with fewer cores, fewer transistors, fewer watts, and less bandwidth.

Regardless, Nvidia will have to follow this model at one point or another. You can't keep recreating the wheel every couple years and getting performance improvements. At some point you will have to let your engineers play with an existing architecture for longer while more engineers have lots of time to come up with your next big leap. This is where I think AMD is at and Nvidia will follow suit some day.

 

[railven]

Sure, if you throw enough shaders at the problem. I realize that an AMD shader is different than an Nvidia shader, but how different could they actually be? Both need to perform similar function. FuryX has 33+% more shaders than 980Ti. So, a sweeping statement like "GCN is just an all around better performer" doesn't really compute. Unless you're running certain compute functions. I made a PUN!! Not really. :biggrin:

Don't forget to turn off specific options too, I mean, you don't want to give any side an unfair advantage do you