Metro: Last Light
Metro: Last Light (DirectX 11)
Beneath the ruins of post-apocalyptic Moscow, in the tunnels of the Metro, the remnants of mankind are besieged by deadly threats from outside – and within.
Mutants stalk the catacombs beneath the desolate surface, and hunt amidst the poisoned skies above. But rather than stand united, the station-cities of the Metro are locked in a struggle for the ultimate power, a doomsday device from the military vaults of D6. A civil war is stirring that could wipe humanity from the face of the earth forever.
As Artyom, burdened by guilt but driven by hope, you hold the key to our survival – the last light in our darkest hour…
Our Settings for Metro: Last Light
2 GPU Testing
NVIDIA's scaling in Metro: Last Light is exceedingly poor, while AMD's Fury X card is able to trumpet some impressive gains at both tested resolutions without huge increases in frame time variance.
3 GPU Testing
Huh. In our 3-Way testing, there are no winners. Only tie…ers.
| Metro: Last Light – 2560×1440 | ||||||
|---|---|---|---|---|---|---|
| Single GPU (FPS) | 2-Way (FPS) | Scaling | 3-Way (FPS) | Scaling | ||
| AMD Fury X | 72 | 103 | +43% | 103 | +0% | |
| GTX 980 Ti | 70 | 82 | +17% | 83 | +1.2% | |
| Metro: Last Light – 3840×2160 | ||||||
| Single GPU (FPS) | 2-Way (FPS) | Scaling | 3-Way (FPS) | Scaling | ||
| AMD Fury X | 40 | 68 | +70% | 68 | +0% | |
| GTX 980 Ti | 36 | 42 | +16% | 42 | +0% | |
Today's table is a little different than recent ones as we are not comparing directly between the AMD Fury X and GTX 980 Ti, and instead we are comparing single GPU to 2-GPUs and to 3-GPUs. The resolutions are separated to make things a big easier to digest and the "scaling" columns show you how much performance improvement you see going from 1-to-2 GPUs and then from 2-to-3 GPUs.
NVIDIA's results here are just plain bad – the best scaling rate we get is at 2560×1440 going from one card to two and even then we only see a 17% gain. AMD's 43% at those settings still isn't amazing but it's definitely usable. That improvement jumps up to 70% at 4K though so users with higher resolution panels will be able to take advantage of Fury X in CrossFire!
I’m a noob, what exactly is
I’m a noob, what exactly is scaling?
Measuing scaling tells us how
Measuing scaling tells us how much of an improvement adding a second (or third) GPU to your system can improve the gaming experience with SLI or CrossFire.
If you get 40 FPS with a single card, but 70 FPS with two cards, then you have a (70/40 = 75%) scaling rate, which is quite good.
Unfortunately the system you
Unfortunately the system you are using is not powerful enough to show this. This is why the fury is showing “better scaling”, becasue you are hitting the limit of your CPU/MOBO combo.
That’s both wrong and funnu
That’s both wrong and funnu
Well yes the the test show
Well yes the the test show that the cards scale.. but the final result is not correct
also this
also this http://imagescdn.tweaktown.com/content/6/7/6726_27_4k_showdown_intel_x79_vs_x99_with_asus_geforce_gtx_980_4gb_quad_sli.png
Your site is the worst among
Your site is the worst among the worst. Here is how you present scaling data correctly if we pick the Crysis 3 results you got:
Crossfire 20 -> 39 = 95% WTH PCPer!
Trifire 39 -> 57 = 47% WTH PCPer!
20*2 = 40 (100% 2 cards scaling)
39 / 40 = 0.97 * 100 = 97% scaling
20*3 = 60 (100% 3 cards scaling)
57 / 60 = 0.95 * 100 = 95% scaling
2 Way SLI 19 -> 35 = 84% WTH PCPer!
3 Way SLI 35 -> 51 = 45% WTH PCPer!
19*2 = 38 (100% 2 cards scaling)
35 / 38 = 0.92 * 100 = 92% scaling
19*3 = 57 (100% 3 cards scaling)
51 / 57 = 0.89 * 100 = 89% scaling
Then why do you come here?
Then why do you come here?
I am sorry but you are
I am sorry but you are calculating the wrong numbers, you are calculating total percentage of the graphics cards being scaled down too and not the scaling of each individual additional card.
Scaling = percentage of additional performance with 1 extra card.
2 way crossfire
(2 way – 1 way)/1 way = scaling
3 way crossfire (Trifire as you call it)
(3 way – 2 way)/2 way = scaling
According to your math if you have 20 FPS with 1 card and the same with 2 cards that would be 50% scaling when it should be 0%.
Yes the graphics are being scaled down 50% of the possible 100%. but that is not the same as how the 2nd card scales with the first card.
Same equations you did but correct.
2 way crossfire
(2 way – 1 way)/1 way = scaling
(40-20)/20=100% Perfect Scaling
(39-20)/39=95%
(35-19)/19=84%
3 way crossfire theoretically maxes out at 50% since you can only add 50% more performance. 2*50%=3
(3 way – 2 way)/2 way = scaling
(60-40)/40=50% Perfect Scaling
(57 – 39)/39 = 46.15%*
(51 – 35)/35 = 45.71%*
This makes perfect sense to me.
I hope this answers anyone’s questions about how scaling is calculated.
*PCPer has 47% & 45%. This is most likely due to them giving us rounded base numbers and rounding the final number.
PS. I know this is almost a week old but my OCD had me correct the math. Also I hope that people read it to understand better how the math in scaling works.
Welp, disappointing games
Welp, disappointing games list as usual. We can hardly glean worthwhile conclusions from such s tiny and old games list. No witcher 3? Come on. Also still no 390X review? What’s up with THAT?
Welp again PC Per didn’t get
Welp again PC Per didn’t get the memo about the new AMD driver so yet another back to the future article not worth reading.
take time to test and write
take time to test and write an article.
I wonder if these cards are
I wonder if these cards are constrained by the PCI-E bandwidth of the 3930k + X79 setup. That CPU only supports PCI-E 2.0, pretty sure those cards are PCI-E 3.0.
Happy to be corrected on this.
3960x rather.
3960x rather.
X79 does support PCI-E 3.0,
X79 does support PCI-E 3.0, but it's up to the driver if it's implemented. These were not bandwidth constrained, and there should be no bottlenecking issue using a CPU with this much headroom.
the CPU only support 2.0
the CPU only support 2.0
http://imgur.com/fwTANOp
http://imgur.com/fwTANOp Screenshot of my desktop with CPUZ and GPU-Z open. 3930k and a 780 classified running PCI-E 3.0
Look like you are right it’s
Look like you are right it’s just Intel who are to lazy to update their spec sheets and still list these CPUs as 2.0 only 🙁
http://ark.intel.com/products/63696/Intel-Core-i7-3960X-Processor-Extreme-Edition-15M-Cache-up-to-3_90-GHz
Expansion Options: PCI Express Revision 2.0
To be fair, in the first
To be fair, in the first batches (in my case the C1 revision), PCI-E 3.0 transfer rates were there; but nVidia decided to only officially support 2.0 speeds due to signal timing variance between different CPU and motherboard combinations.
This is a post on nvidia forums by a customer care rep explaining the details and how to enable on x79:
https://forums.geforce.com/default/topic/521828/geforce-drivers/geforce-600-series-gen3-support-on-x79-platform-updated-6-19-2012/
My running at PCI-E 3.0 speeds is legitimate, but also forced. I have also noticed no real world difference with one 780 at 4k.
Which 980ti was used in which
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 =
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
The nividia driver will
The nividia driver will always clock the SLI setup to the lowest card specs (so in this case reference settings)
I have heard the opposite, do
I have heard the opposite, do you have a source for that? I know someone with 980ti in SLI and they are able to OC them independently.
Even if the base clocks are the same, the non reference cards would presumably boost higher because of superior cooling.
Nvidia GPUs in SLI are able to scale independently based on what is needed of them:
http://hardforum.com/showthread.php?t=1685772
sure looks like it, guess its
sure looks like it, guess its only the base clock that will be the same, and boost is dependent on cooling.(but that also makes sense since they will have different temperatures in the stack)But actually the nvidia reference cooler is usually the most effective in a stack since it is not circulating hot air.
But again, don’t think any of this matters much since i think its pretty obvious from the results that the Motherboard + CPU is a bottleneck.
This site’s anti-spam thing
This site’s anti-spam thing sucks so I had to create an account..
Anyway, if the driver does in fact downclock to the lowest common denominator, that could make the scaling results even worse if a non-reference card is used first.
If zotac was used first:
(100 + 100*.75) / 110 = 59% effective scaling
If ref was used first:
(100 + 100*.75) / 100 = 75% effective scaling
Edit: This would not affect the actual FPS of the game, but it would affect the “% increase” values. It’s not a huge deal but I thought it was worth pointing out.
2 is the most ill ever do as
2 is the most ill ever do as I have done it for years now and never really have any problems like I guess most people do. Right now I am sporting to 780’s for my 1440 monitor and its great! but I cant wait to see that single card that just crushes what I have cause then I can upgrade.
Thanks Ryan for taking the
Thanks Ryan for taking the time to do a review.
I think that the conclusions we can draw are:
1. Adding a third card is a waste.
2. In most cases, adding a second card is actually a decent investment.
3. SLI currently does not scale as well as Crossfire. Reportedly the Fury X CF at 4k can on average, outperform a Titan-X at 4k, where the Fury X has enough VRAM.
Although it pains me deeply to say this (and I want AMD to succeed here because a monopoly for us consumers is bad), I’m inclined still to recommend the Nvidia 980Ti solution.
1. The overclocking headroom is much larger on the 980Ti, which negates some of the superior scaling. Power consumption is lower, allowing for more OC headroom.
2. The 6GB of VRAM could be much more “future proof”.
3. Custom PCBs could improve the OC headroom even more (an MSI 980Ti Lightning is expected and EVGA, Galax, along with Asus have OC variants).
The only things working in favor of AMD are:
1. Crossfire does scale better, resulting in a better FPS and more importantly, frame times in many games.
2. AMD Drivers have historically improved performance more. Compare say, the 7970 to the 680. Both were pretty close on launch, but now the 7970 has pulled ahead.
The Fury kind of negates the issue of no custom PCBs, but the OC headroom still isn’t as good as on the 980Ti, where overclocks of 1500 MHz+ are not uncommon.
You could make a case that a Fury CF would be a viable alternative as clock for clock, it’s not much slower (I think the 64 ROPs/8 ACEs are bottlenecking the whole GPU here), but it still suffers from most of the Fury X flaws.
Anyways, let me know what you think. Thanks again for taking the time to review.
Here’s a nice benchmark
Here’s a nice benchmark pitting the Fury card against the 980 TI 980, and FuryX:
https://www.youtube.com/watch?v=0jBjDIakZxc
Multi card support not worth
Multi card support not worth it……HA! Yeah okay….