New AMDs benchmark for Star Citizen ! (GPUs)

[VoA]

New AMDs benchmark for Star Citizen ! (GPUs)

Star Citizen 1.1.3 Updated Graphics Card Benchmark & VRAM Consumption

 

By Steve Burke Published June 26, 2015 at 11:30 pm

 

•  

 

 

 

GAME BENCHMARKS

Regardless of how its mechanics pan-out, Star Citizen is slated to claim the throne as one of the most graphically intense PC games in recent history. This is something we discussed with CIG's Chris Roberts back when the Kickstarter was still running, diving into the graphics technology and the team's intent to fully utilize all tools available to them.

We've been trying to perform frequent benchmarks of Star Citizen as the game progresses. This progress monitor comes with a massive disclaimer, though, and is something we'll revisit shortly: The game isn't finished.

The recent launch of the GTX 980 Ti, R9 Fury X, and AMD 300 series cards almost demands a revisit to Star Citizen's video card performance. This graphics benchmark looks at GPU performance in Star Citizen's 1.1.3 build, testing framerates at various settings and resolutions.

Previous Star Citizen Content

We've worked with Star Citizen's team at nearly every major convention and, as such, it's difficult to list all the related content. Here are a few samples:

• Previous Benchmark: Star Citizen 0.9.2.2
• Richard Garriott & Chris Roberts Discuss Zero G Physics
• PAX East – FPS Module Release
• Procedural Generation Explored with Chris Roberts
• Character & Ship Customization in Star Citizen
• All Star Citizen Content

The Obvious Disclaimer: No Drivers, No Optimization

Star Citizen is not a complete game. For a great number of reasons, we generally avoid benchmarking an incomplete product; the tremendous interest behind Star Citizen and the number of times we're asked if “my computer can play Star Citizen” means that we make an exception. Even excepting the game to the rule, the same disclaimers apply.

An incomplete game means a couple critical items are absent. Drivers are the main one – AMD and nVidia (and Intel, to some extent) go through great efforts to prepare drivers that bolster performance for specific games. Each major game release sees a driver update (see: GTA V, Witcher 3). The hardware vendors tune their technology to better render graphics for each game, and this can create profound performance gains from one driver set to the next. Because Star Citizen isn't done, there are no game-specific drivers that improve its performance. The game runs with effectively no hardware-side optimization as it stands now.

Then there's the development side of the coin. Without software-side optimization performed by the game developer, graphics will run at full bore and could beat-down hardware in a manner which is not representative of the final product. Star Citizen currently has one quality setting with only four options (low, medium, high, very high). It is highly likely that the end product will offer a more detailed options page in the menus, but for now, we have limited control and CIG has performed minimal tuning.

All this is to make it abundantly clear that these tests are not representative of the final product. They might not even represent performance in the next patch. These tests strictly represent the game in its present state, and should help determine whether systems will play it as is at this moment. As the game iterates and drivers ship, it is almost a guarantee that performance will increase. That stated, the art assets also aren't final and could impact performance in either direction.

Test Methodology - Hangar Only!

We tested using our updated 2015 GPU test bench, detailed in the table below. Our thanks to supporting hardware vendors for supplying some of the test components.

The latest GeForce 353.3 drivers were used during testing. AMD Catalyst 15.15 was used for the 300 & Fury series cards, with 15.5 used for 200 series cards; 15.6 was not functional at the time of testing. Game settings were manually controlled for the DUT.

VRAM utilization was measured using AIDA64 and then validated with MSI's Afterburner and GPU-Z. FPS measurements were taken using FRAPS and then analyzed in a spreadsheet.

Each game was tested for 30 seconds in an identical scenario on the two cards, then repeated for parity. We strictly tested the hangar for this Star Citizen benchmark. Flight was tested briefly and proved significantly more framerate friendly due to the nature of reduced polygon density; given time constraints, we opted to only test the most intensive aspect of the game. As of now, the racing and battle sim aspects are less intensive. These benchmarks are not linearly comparable to our previous benchmarks because we changed our course methodology.

GN Test Bench 2015	Name	Courtesy Of	Cost
Video Card	Sapphire AMD R9 390 Sapphire AMD R9 380	Sapphire	$330 $220
CPU	Intel i7-4790K CPU	CyberPower	$340
Memory	32GB 2133MHz HyperX Savage RAM	Kingston Tech.	$300
Motherboard	Gigabyte Z97X Gaming G1	GamersNexus	$285
Power Supply	NZXT 1200W HALE90 V2	NZXT	$300
SSD	HyperX Predator PCI-e SSD	Kingston Tech.	TBD
Case	Top Deck Tech Station	GamersNexus	$250
CPU Cooler	Be Quiet! Dark Rock 3	Be Quiet!	~$60

Average FPS, 1% low, and 0.1% low times are measured. We do not measure maximum or minimum FPS results as we consider these numbers to be pure outliers. Instead, we take an average of the lowest 1% of results (1% low) to show real-world, noticeable dips; we then take an average of the lowest 0.1% of results for severe spikes.

Cards Tested

Video Card	Price	Courtesy Of
AMD R9 Fury X Reference	$650.00	GamersNexus Reader [Loaner]
Sapphire R9 390	$330.00	Sapphire [Loaner]
Sapphire R9 380	$220.00	Sapphire [Loaner]
Powercolor R9 290X	-	CyberPower
XFX R9 285	-	AMD
AMD R9 270X	$150	AMD
GTX Titan X Reference	$1,000.00	NVIDIA
EVGA GTX 980 Ti Hybrid	$770.00	EVGA
GTX 980 Ti Reference	$650.00	NVIDIA
GTX 980 Reference	$500.00	NVIDIA
EVGA GTX 960 4GB	$230.00	EVGA
ASUS Strix 960 2GB	$210.00	ASUS
GTX 780 Ti Reference	-	NVIDIA
GTX 770 Reference	-	NVIDIA

Note: Anomalous Performance & SLI / CrossFire

Some performance exhibits anomalies resultant of the incomplete product and drivers. AMD's R9 285, for example, underperforms considerably even compared to weaker AMD devices. The R9 380 – effectively an R9 285 – performs better than the R9 285 counterpart.

We also see the GTX 780 Ti outperforming the 980 and 980 Ti (and Titan X), something that is not consistent with any of our other tests; this suggests that Star Citizen is presently seeking raw horsepower over anything else. Maxwell's memory optimizations will likely yield superior performance in the final leg of development, but for now, the 780 Ti's raw power from Kepler puts it ahead.

As another example, the brand new Fury X barely outperforms the GTX 770 – something that is far from the truth in nearly any other game we've tested.

We attempted one SLI configuration (2x 980 Ti) and one CrossFire configuration (2x Fury X) with no luck. In both instances, Star Citizen operated at effectively the same framerate as a single card for either of those configs.

Star Citizen 1080p / Very High GPU Benchmark

At 1080p with Very High settings, Star Citizen proves difficult to play with any of the hardware tested. The 780 Ti produces a better framerate than even the Titan X – something that won't happen with finalized content. The 980 Ti Hybrid's performance gain over the Titan X is actually consistent with our testing, though, and is a result of the bolstered clockrate on the Hybrid version of the 980 Ti.

AMD's first card to appear on the chart is the Fury X, with just 32FPS. It is possible that CIG has performed more nVidia optimizations at this point, though we're unsure of the root cause for the unreasonably low performance.

Star Citizen 1080p / Medium GPU Benchmark

1080p / Medium is more playable. Everything from the 290X and upward sits at a decent FPS, though the 770 and 960 are tolerable at these framerates given the hangar's lack of accuracy-dependent gameplay.

Star Citizen 1080p / Low GPU Benchmark

For the most part, there's a small performance gain in low vs. medium. Observe the two 960 cards (EVGA 4GB, ASUS 2GB) carefully, though, and it becomes clear that video memory is bottlenecking performance at higher quality settings.

Star Citizen 1440p / Very High & Medium GPU Benchmark

Performance struggles at higher resolutions, though the high-end hardware pushes nearly 60FPS at medium.

 

Star Citizen 4K / Very High & Medium GPU Benchmarks

4K is effectively unplayable at this time. Users of 4K displays would be better off dropping to a lower resolution while retaining higher quality settings.

 

Star Citizen VRAM & System Memory Consumption

We logged system memory consumption at 4637MB peak (this is a delta value – we subtract the background process memory consumption from this measurement, isolating Star Citizen). This was extracted from a 20-minute gameplay session at Very High / 1080p.

	VRAM – Dynamic + Used
VRAM / 1080 Very High	4639MB
VRAM / 1080 Medium	4197MB
VRAM / 1080 Low	4052MB
	System Memory Delta
System Memory – Very High	4637MB

VRAM consumption exceeded 4GB at all times when using the Titan X, which hosts 12GB of VRAM. This allows us to see the amount of memory consumed without limitation from the hardware. There is a measurable difference from “low” to “very high” of 600MB.

Conclusion

We normally publish benchmarks with a great deal of confidence, feeling certain that our results are representative of the whole. In this instance, given the unfinished state of the game and lack of driver support, it is possible that performance could vary greatly on other configurations. We were unable to get CrossFire or SLI functional, but we also only tested two high-end configurations – if you've gotten it working, let us know your setup below.

As it stands now, Star Citizen requires a serious amount of graphics processing power to operate at playable framerates with high settings. Anomalous output in some instances confirms the lack of optimization, as shown most obviously when switching between the 285 and 380– two effectively identical GPUs with disparate performance. One thing that is clear, though, is Star Citizen's consumption of video memory. There is a clear advantage afforded to video cards with greater than 2GB of VRAM; the 2GB vs. 4GB GTX 960 test illustrates this, especially when looking at the jumps between 'low' and 'medium' settings.

We'll continue to update with new benchmarks as the game makes progress.

- Steve “Lelldorianx” Burke.

[Fintz]

Also  not to forget that Dx12 or Mantle haven't been used  

[Perditor]

I know I only run 1440x900 but my reference 7970 does ~14fps at very high in the hangar. Something must be really screwed up with the drivers for those newer cards. I was surprised to see the 780ti at the top...

[GeraldEvans]

How did the 780Ti do so well?  On the other hand for the rest of them I feel very good about my incoming 980Ti.

[Automated]

Man i thought that my R9 270 OC would push this game quite well.. boy was i wrong, i can play the arena commander vandul modes and races at around 25-40 fps but the first tutorial mission about taking off completely destroys my machine! lol

[Emphatic]

Yea I dunno about these benchmarks I run a single R9 290 with a I5-4590 and I get almost double the frames he listed on the 290x in hangar i get a solid  60 fps (capped) on very high and in game i get 50+

I even went so far as to run Elite dangerous while runnign fraps and SC and I still got 38-40fps, while in FSD in elite, I would highly recommend you take these results with a grain of salt as my pc is slightly worse than theirs in every regard yet I got vastly better numbers than they did.

Here is a video of a guy benchmarking a 290 on very high in an early version of AC, as you can see hes capped at 60 fps and its sits there pretty consistently

[Fintz]

I haven't checked my R9 280x Toxic base min-max fps at "Ultra hight"

[Alex]

Yea, those results look iffy. Something is wrong with their methodology of benchmarking.

I'm pretty sure my R9 280 will do better at Very High than their 285. *will have to test it

[Pagan-Tempest]

I'm going to have to retest my 290X in the hangar apparently. The results he's getting are far far lower than the results I get in the hangar. Perhaps it's just an issue with 1.1.3. I'll do another test when I get home.

[Wyvyrias]

Yea I dunno about these benchmarks I run a single R9 290 with a I5-4590 and I get almost double the frames he listed on the 290x in hangar i get a solid  60 fps (capped) on very high and in game i get 50+

I even went so far as to run Elite dangerous while runnign fraps and SC and I still got 38-40fps, while in FSD in elite, I would highly recommend you take these results with a grain of salt as my pc is slightly worse than theirs in every regard yet I got vastly better numbers than they did.

Here is a video of a guy benchmarking a 290 on very high in an early version of AC, as you can see hes capped at 60 fps and its sits there pretty consistently

How? My R9 290X is able to provide capped but stable 60 FPS on low, but it's already getting quite hot at this point (80 to 90°C) on 1920x1080.
Around 70 to 80°C when I lower the resolution one or two steps.

[Pagan-Tempest]

How? My R9 290X is able to provide capped but stable 60 FPS on low, but it's already getting quite hot at this point (80 to 90°C) on 1920x1080.Around 70 to 80°C when I lower the resolution one or two steps.

For the sake of my own curiosity, what model of 290X do you own?

[Wyvyrias]

For the sake of my own curiosity, what model of 290X do you own?

http://www.asus.com/us/Graphics_Cards/R9290XDC2OC4GD5/

Probably the cooler of this one really has some issues. Not sure about that.

[GeraldEvans]

Well that hangar is hardly the nice ones we have now. That's a very old build.

[Emphatic]

Tho old hangars were more resource intensive and had memory leaks all over, the hangars for the most part are much more optimized now then they were. Having seen first hand personally that my 290 gets much better numbers than were advertised in that article on the version he tested I was getting no less than double the numbers he listed, the video I posted was just a reference and was the only one i could find with a similar setup.

And another thing to note is if you go back and look at his previous SC benchmarks he got better numbers with a 290x one a much earlier version of the game with a slightly worse PC, by about 30% average frames...to me that seems highly suspect. but then this is why you don't do benchmark reviews on games that are still in alpha as optimization changes from build to build. Regardless my 290 has never performed as bad as his 290x did and like i said his PC is better.

Here is a video of a more up to date benchmark again, as you can see the results are much higher than the ones in the article.

It should be noted that he is comparing a 290X to a 980 here, not really a fair comparison as the 980 was a much newer card, by that logic you could grab a R9 390x for the same price as a 980, but you are still gonna get about a 5-10 fps gain if you run a 980, whether or not driver support can make up that gap later who knows but for now 980 is king, closest competitor is 295x2 but when it does out perform the 980 its only by about 5% and the price really doesn't justify it and the 980 still outperforms the 295x2 on most tests. Intel+Nvidia = just better.

Edited August 19, 2015 by Emphatic