Every single day, your favorite vendors send in requests to have their products reviewed by Tom's Hardware. We typically try to maximize the number of components presented to you, our readers, by organizing round-ups. There's a lot more to be learned, we think, by comparing the strengths and weaknesses of many competing offerings. But occasionally, a unique specimen surfaces so unlike everything else that we devote an entire story to it. Today we're looking at Asus' Mars 760, a graphics card with no peer. To test it, we have to reach into our collection of cards for relevant data points.
This isn't the first board in Asus' Mars family, and if you're familiar with the brand, then you already know the ROG MARS760-4GD5 boasts two Nvidia GPUs (Asus reserves the Ares brand for its dual-Radeon creations). All semblance of suspense is relieved by the model number: 760 gives away the fact that we're looking at a couple of GK104 processors. And incidentally, they're overclocked, too.
| 2 x Radeon HD 7950 Boost | Radeon HD 7990 | Radeon R9 290X | Asus Mars 760 | GeForce GTX Titan | GeForce GTX 780 Ti | GeForce GTX 690 | |
|---|---|---|---|---|---|---|---|
| Shaders | 3584 (2 x 1792) | 4096 (2 x 2048) | 2816 | 2304 (2 x 1152) | 2688 | 2880 | 3072 (2 x 1536) |
| Texture Units | 224 (2 x 112) | 256 (2 x 128) | 176 | 192 (2 x 96) | 224 | 240 | 256 (2 x 128) |
| ROPs | 64 (2 x 32) | 64 (2 x 32) | 64 | 64 (2 x 32) | 48 | 48 | 64 (2 x 32) |
| Fab Process | 28 nm | 28 nm | 28 nm | 28 nm | 28 nm | 28 nm | 28 nm |
| Core (Boost) Clock | 850 (925) MHz | 950 (1000) MHz | (1000) MHz | 1006 (1072) MHz | 837 (876) MHz | 875 (928) MHz | 915 MHz |
| Memory Clock | 1250 MHz GDDR5 | 1500 MHz GDDR5 | 1250 MHz GDDR5 | 1501 MHz GDDR5 | 1502 MHz GDDR5 | 1750 MHz GDDR5 | 1502 MHz GDDR5 |
| Memory Bus | 384-bit | 384-bit | 512-bit | 256-bit | 384-bit | 384-bit | 256-bit |
| Memory Bandwidth | 240 GB/s | 288 GB/s | 320 GB/s | 192.2 GB/s | 288.4 GB/s | 336 GB/s | 192.2 GB/s |
| TDP | 400 W (2 x 200 W) | 375 W | 250 W | 300 W (per Asus) | 250 W | 250 W | 300 W |
| Amazon Price Range | $840 to $860 | - | $630 to $780 | $650 | $1000 to $1849 | $680 to $770 | $995 to $1350 |
| Typical Price | $840 | - | $650 | $650 | $1000 | $680 | $1000 |
Asus' packaging claims that the Mars 760 is about 7% faster than a GeForce GTX Titan, though at its $650 price point, it's really competing against the Radeon R9 290X and GeForce GTX 780 Ti. I didn't have a 780 Ti on-hand, so I'm tuning my Titan to approximate the more gaming-oriented card's performance.
Speaking of operating frequencies, the Mars 760 sports a 1006 MHz base and 1072 MHz typical GPU Boost clock rate, which is a little higher than the 980 MHz Nvidia guarantees its GeForce GTX 760 for. Asus uses 4 GB (2 GB per GPU on independent 256-bit buses) of GDDR5 memory at the same 1500 MHz as the reference 760.

The Mars, Ares, and Matrix graphics cards exist under Asus' Republic of Gamers (ROG) brand. Naturally, the first thing you'll notice about the Mars 760 is its distinct red metal-on-black shroud. It weighs 2.1 lbs (970 grams) and measures 11" x 5" x 1.5" (28 x 12 x 4 cm), making it large, but no bigger than single-GPU products like the Radeon HD 290X and GeForce GTX 780.

Plugged in and powered up, the word Mars slowly "breathes" thanks to red LEDs behind the name plate.
There is one SLI connector; Asus says it worked closely with Nvidia to enable four-way operation with a second card (since GeForce GTX 760s typically don't scale to four-way configurations).
Also, you'll need two eight-pin auxiliary power connectors. Nvidia cites a 170 W graphics card power figure for one GeForce GTX 760. Meanwhile, Asus says its card should fall closer to 300 W, with spikes higher if you run more taxing apps like FurMark. A PCIe slot is rated for up to 75 W; an eight-pin lead should be good for up to 150 W; a six-pin connector officially serves up 75 W as well. Although we've seen plenty of cards pull more power than their inputs were officially rated for, it's pretty clear why Asus felt two eight-pin connectors were necessary in this case.

A side view gives you a peek into Asus' DirectCU II cooler, which employs eight copper heat pipes (four per GPU). Two low-profile 85 mm fans blow through the sink's fins to keep the GPUs running efficiently. The company claims these are dust-proof, but we'll believe that after a year or two of real-world use.
Hidden under the cooler is a 12-phase VRM, which may play a role in overclocking. This will get tested later in today's story.

As part of its marketing message, Asus says the Mars 760 only uses Japanese Nichicon GT-series capacitors and other high-quality components like super alloy power chokes, MOFSETS, and POSCAPs.

The Mars 760 is equipped with two dual-link DVI-I, one dual-link DVI-D, and one mini-DisplayPort output. HDMI is also covered by a bundled adapter.

Asus keeps its bundle modest; it includes a DVI-to-HDMI adapter, a dual-six-pin-to-eight-pin power adapter, and a folder with the driver CD and manual. As a fun little extra, the company throws in a metal ROG case badge, which looks nice.

Aside from the physical accessories, Asus includes its GPU Tweak and GPU Tweak Streaming apps.
GPU Tweak Streaming doesn't actually have anything to do with altering the performance of your GPU. Rather, you can use the software with Adobe Media Encoder to stream your game play to services like Twitch.tv with overlaid pictures and text. That's in theory, though. The software is still in early beta form, and I couldn't get it working.
With this said, Nvidia's ShadowPlay feature works fine in conjunction with the Mars 760, and I'd prefer to use that anyway. ShadowPlay hooks in to Nvidia's NVEnc encoder, which offloads the processing-heavy compute workload.

GPU Tweak, on the other hand, is a more useful addition. Its interface is similar to MSI's Afterburner with just enough differences to throw off anyone already used to navigating the competition's app. Overall, Asus' implementation gets the job done, though we'd like to see the company expose memory voltage controls for the folks interested in dialing in the most aggressive overclock possible.
Again, I didn't have access to any of Chris' GeForce GTX 780 Tis, so I tuned a GeForce GTX Titan to approximate its performance in our benchmarks. With a 384-bit memory interface, the Titan offers the same amount of bandwidth as the 780 Ti with matching clock rates. Where you'd notice a difference is overrunning the 780 Ti's 3 GB of memory. The GPU overclock is trickier since the Titan's GK110 has one of its SMX partitions disabled. I tried to compensate with an approximate clock rate increase. At any rate, the match-up isn't exact, but it's close.
Otherwise, I have a GeForce GTX 690, a Radeon R9 290X, a Radeon HD 7990, and two Radeon HD 7950 Boost cards in CrossFire to show a broad range of competition for the Mars 760.
I'm generating two sets of benchmarks. In the first one, I have a single QHD panel with a native 2560x1440 resolution. The second set employs three FHD screens at 5760x1080.
| Test System | |||||
|---|---|---|---|---|---|
| CPU | Intel Core i5-2550K (Sandy Bridge), Overclocked to 4.2 GHz @ 1.3 V | ||||
| Motherboard | Asus P8Z77-V LX. LGA 1155, Chipset: Intel Z77M | ||||
| Networking | On-Board Gigabit LAN controller | ||||
| Memory | Corsair Performance Memory, 4 x 4 GB, 1866 MT/s, CL 9-9-9-24-1T | ||||
| Graphics | Asus Mars 760 1006 MHz GPU, 2 x 2 GB GDDR5 at 1501 MHz (6004 MT/s) GeForce GTX Titan 837 MHz GPU, 6 GB GDDR5 at 1502 MHz (6008 MT/s) also tested at overclock to simulate GeForce GTX 780 Ti: to 994 MHz GPU, 6 GB GDDR5 at 1750 MHz (7000 MT/s) GeForce GTX 690 915 MHz GPU, 6 GB GDDR5 at 1502 MHz (6008 MT/s) XFX Radeon R9 290X 1000 MHz Peak GPU Clock, 4 GB GDDR5 at 1250 MHz (5000 MT/s) 2 x Radeon HD 7950 Boost in CrossFire 850 MHz GPU, 2 x 3 GB GDDR5 at 1250 MHz (5000 MT/s) XFX Radeon HD 7990 950 MHz GPU, 2 x 3 GB GDDR5 at 1500 MHz (6000 MT/s) | ||||
| Hard Drive | Samsung 840 Pro, 256 GB SSD, SATA 6Gb/s | ||||
| Power | XFX PRO850W, ATX12V, EPS12V | ||||
| Software and Drivers | |||||
| Operating System | Microsoft Windows 8 Pro x64 | ||||
| DirectX | DirectX 11 | ||||
| Graphics Drivers | AMD Catalyst 13.11 Beta 9.5, Nvidia GeForce 332.21 WHQL | ||||
We've almost completely eliminated mechanical storage in the lab, and instead lean on solid-state drives to alleviate I/O-related bottlenecks. Samsung sent all of our offices 256 GB 840 Pros, so we standardize on these exceptional SSDs.
Naturally, discrete graphics cards require a substantial amount of stable power, so XFX sent along its PRO850W 80 PLUS Bronze-certified power supply. This modular PSU employs a single +12 V rail rated for 70 A. XFX claims that this unit provides 850 W of continuous power (not peak) at 50 degrees Celsius (a higher temperature than you'll find inside most enclosures).
| Benchmark Configuration | |
|---|---|
| 3D Games | |
| Metro: Last Light | Version 1.0.0.14, Built-in Benchmark, FCAT |
| Grid 2 | Version 1.8.85.8679, Built-in Benchmark Scene D6, FCAT |
| Assassin's Creed IV: Black Flag | Version 1.05, Custom THG Benchmark, 40-Sec, FCAT |
| Battlefield 4 | Version 1.0.0.1, Custom THG Benchmark, 90-Sec. Fraps |
| BioShock Infinite | Version 1.1.24.21018, Built-in Benchmark, FCAT |
Our first benchmark is run on one QHD display at its native 2560x1440 and the Ultra graphics preset.
Nvidia's FCAT tool set is not compatible with Battlefield 4, so the results we're presenting potentially include dropped and runt frames; this cannot be helped. There's nothing to worry about when it comes to single-GPU configurations, but dual-GPU cards and arrays won't necessarily be represented accurately if frame pacing is not working properly.


As far as the Fraps-reported frame rates go, none of these configurations drop below 30 FPS (aside from the GeForce GTX 690 at one point). For the most part, every data point is playable, though AMD's Radeon R9 290X serves up higher minimum frame rates than the rest of the field.
Fraps might not collect data late enough in the display pipeline to register certain display problems, it does at least maintain an accurate frame time record.


The results show us why you can't necessarily base your experiential evaluation on average frame rates. AMD's Radeon HD 7950 Boost cards in CrossFire fare the worst, but the GeForce GTX 690 and Asus Mars 760 exhibit some spikes as well. Curiously, the Radeon HD 7990 doesn't show the same issue as the 7950s. The single-GPU solutions perform impeccably.


At the highest detail settings with 2x MSAA and soft shadows enabled, the only AMD card that comes close to a 30 FPS minimum frame rate in Assassin's Creed IV is the Radeon R9 290X.
Asus' Mars 760, on the other hand, does very well, nipping at the GeForce GTX 690 and besting our overclocked GeForce GTX Titan. Presumably, that means it'd also beat a GeForce GTX 780 Ti).


While the Mars 760 does register a notable spike in frame time variance, it's both brief and low in amplitude. The rest of the cards perform admirably, including the slower Radeons.
Metro: Last Light can be brutally demanding, even for high-end hardware. We employ the Very High detail preset for our test at 2560x1440.


Other forces appear to be in play, as two GPUs don't benefit our Metro: Last Light benchmark as much as we'd expect (except for the Radeon HD 7990). As a result, Asus' Mars 760 is pushed to the back of the pack.


Most graphics configs report low variance. The Radeon HD 7950 CrossFire setup is the big exception. Interestingly enough, we saw hiccups throughout the benchmark on all cards, so there's observed behavior that doesn't correlate with the quantitative data.
While BioShock Infinite is an attractive-looking game, it's not particularly difficult for high-end graphics hardware to render at its top quality setting. Set to the Ultra detail preset, not a single card dropped below 48 FPS during the course of our benchmark.


With my frame rate over time chart capped at 60 FPS to focus on problematic slow-downs, there's very little to see aside from a small valley at the beginning of the sequence.


Only two Radeon HD 7950 Boost cards in CrossFire struggle a bit in our frame time variance measurement. The other cards fare well through the majority of this test.
As with BioShock, Grid 2 is a good-looking game that does tend to exacerbate processor and memory bottlenecks, but typically doesn't tax a graphics subsystem very hard. As a result, frame rates in this one tend to be pretty high, even with 8x MSAA and soft ambient occlusion enabled.


Even at the most demanding settings we can muster, a pair of Radeon HD 7950 cards in CrossFire is the only configuration that shows up under 60 FPS. All of the other setups sustain higher minimums and don't even show up when we chart frame rate over time.


The game facilitates low frame time variance numbers across the board, except for a few rare spikes. The Radeon HD 7950s post less attractive results, but aren't problematic.
Now we move on to the triple-monitor results using three 1920x1080 FHD displays, totaling 5760x1080. Keep in mind that, from this point on, AMD's frame pacing feature does not work. The company is purportedly on the verge of releasing its Catalyst 14.1 package, but it wasn't able to preview it to us for this story, and as of this writing, the software is still not available for download.
Once it is available, you should be able to switch on frame pacing at resolutions in excess of 2560x1600 and in Eyefinity, which will affect the Radeon HD 7990 and two Radeon HD 7950s in CrossFire.
Of course, because we didn't have a version of FCAT that was compatible with Battlefield 4, the frame rates for those two Radeon-based setups are going to be optimistic. The frame time variance figures should be right, though.


We had to drop the detail preset to High in order to get these graphics solutions cranking out playable performance. Even still, some of the minimum frame rates are flirting with our 30 FPS target.
Asus' Mars 760 does really well though, holding its own against the $1000 GeForce GTX 690.


Two Radeon HD 7950s in CrossFire make this chart look particularly painful, though none of the multi-GPU setups yield appealing frame time variance results in Battlefield 4.
Assassin's Creed IV is a challenging game to run at 5760x1080 (even if it looks great). So, we turned off 2x MSAA in favor of post-process SMAA. We also disabled soft shadows.


Once we factor out dropped and runt frames using FCAT, the outcome looks a lot different than what we saw in Battlefield 4 using Fraps, doesn't it? Suddenly, the Radeon HD 7990 and twin Radeon HD 7950s are unplayable.
The rest of the high-end graphics nobility float around 30 to 45 FPS, delivering playable performance.


The GeForce GTX 690 surprises us with the icky frame time variance line. But Asus' Mars 760, which is what we're here to test, fares much better.
Metro is a similarly-taxing title. To get it running well across three monitors, we dropped the detail preset from Very High to High and turned off tessellation.


Once again, when we're able to benchmark using FCAT, the dual-GPU Radeon-based configurations are reduced to unplayable frame rates.
The rest of the field yields stronger performance.


The Radeon HD 7950 Boost cards in CrossFire suffer horrible spikes in frame time variance; the Radeon HD 7990 doesn't do much better. In fact, these numbers take us all the way back to September of last year, when a crowd of game enthusiasts blindly singled out the GeForce GTX 690 for delivering a smoother experience than AMD's former flagship in Radeon HD 7990 Vs. GeForce GTX 690: The Crowd Picks A Winner. Almost five months later, and we're still waiting for AMD to resolve those issues.
With that said, even the GeForce GTX 690 registers some variance, while Asus' Mars 760 runs smoothly.
We don't need to turn the settings down in BioShock; using the Ultra preset, we still see playable numbers from a majority of high-end graphics configurations.


Frame rates naturally drop compared to our testing at 2560x1440, but most of the field continues plowing through this title at playable frame rates. The only configurations that suffer are the Radeon HD 7950s in CrossFire and the 7990, both of which need the fixes in AMD's upcoming Catalyst driver package.


We've come to expect the debilitating lag from AMD's dual-GPU arrays, though it's more surprising to see the smaller spikes from Nvidia's GeForce GTX 690.
Grid 2 rounds out our condensed benchmark suite. We're able to keep this game's detail settings maxed out with 8x MSAA and soft ambient occlusion enabled, despite the increase to 5760x1080.


AMD's dual-GPU configurations continue demonstrating problematic performance, while the other cards we're testing maintain more than our requisite 30 FPS.


As with any other test run across three monitors, two AMD Radeon GPUs struggle to render frames in a smooth, consistent manner. Asus' Mars 760 does post some unsavory spikes, but clearly has pacing better-handled.
It's typically more difficult to get two GPUs in CrossFire or SLI overclocked aggressively than it would be to run one processor at elevated frequencies. If anything, you're always going to be limited to the peak clock rates of whichever card is slower.
I was surprised, then, to get our Mars 760 card running with a 185 MHz GPU Boost clock rate offset and a 742 MT/s memory data rate boost. Even more impressive, the reported 1257 MHz Boost clock setting was exceeded in both of the apps I experimented with (FurMark and Metro: Last Light), where I saw 1306 MHz under heavy load. Achieving those numbers required the Asus GPU Tweak utility, a 12 mV offset on the GPU, a 105% power target, and a fan speed manually dialed in to 80%.
Now, there is absolutely such a thing as vendors cherry-picking their best GPUs and memory for review samples. It happens all of the time, and we totally get the desire to put a best foot forward. I'm not saying that's what's happening here. But I would also caution our readers to consider our results knowing that we're benchmarking a review sample, and not a retail card. Your mileage may vary.

The result of our overclock is significantly more memory bandwidth than a stock GeForce GTX 690, with comparable shader processing power.


At least in this one application, the overclocked Asus Mars 760 achieves a frame rate that is almost identical to the vastly more expensive GeForce GTX 690. And again, while we obviously can't guarantee that retail boards will overclock as well as our sample, Asus has to be doing some sort of binning to get amenable GK104 processors able to stay under a certain thermal limit anyway. From what we've seen and heard in the real world, the Mars 760 is garnering a good reputation for overclocking well as a result.
We measured power consumption through the Metro: Last Light benchmark run and recorded the following results:

The benchmark chart starts at 45 W, which is our test system's idle power use. We're trying to focus the graph on graphics consumption, though the platform undoubtedly still plays a role in the numbers we're reporting as load increases.
With that in mind, most of the configurations we're testing end up pretty close together, except for AMD's Radeon HD 7990, which peaks about 100 W higher than its competition.

Asus' Mars 760 leads the pack when we compare thermals. At the other end of the spectrum, AMD's Radeon R9 290X peaks and plateaus about 20 degrees Celsius higher than the rest of the field.

When it comes to noise under load, Nvidia's GeForce GTX Titan distinguishes itself as the quietest card in our high-end round-up of pricey contenders. The 690 is another card we know to be fairly quiet, and the Mars 760 isn't much different from Nvidia's reference effort.
On the other hand, we know from prior testing that the Radeon HD 7990 and 7950s in CrossFire can get obnoxiously loud. Partner cards are the only way to get around this.
Before we start drawing conclusions, let's look at the combined average performance across all of our benchmarks. The results are relative to Nvidia's GeForce GTX Titan at 100%. That black bar represents frame rates at 2560x1440, while the red bar is indicative of 5760x1080 using three monitors.

The chart's most glaring red flag is that AMD's frame pacing feature is not yet working in dual-GPU setups across multiple monitors. That's a long-standing issue, and it should be cleared up soon. AMD says its Catalyst 14.1 driver includes this functionality, though it'll be a beta release when it does finally emerge (Ed.: Not that proper frame pacing is going to change my opinion of the Radeon HD 7990...).
How about the Mars 760 against Nvidia's GeForce GTX Titan? Asus says the Mars board should be 7% faster, and the average at 2560x1440 across the games we tested lands right at 7%. It's easy to see why Asus' marketing team would want to draw its comparison to Nvidia's premium flagship. The Mars certainly looks like a stellar value next to that $1000 beast. No doubt this comparison helped spawn the creation of the Mars 760.
Times have changed since this thing was conceptualized, though. First, there's the Radeon R9 290X, which started at $550 and forced Nvidia to cut the price of its GeForce GTX 780. The Radeon might be closer to $630 now, but the impressive GeForce GTX 780 Ti is down from its $750 launch price, closer to $680. The overclocked GeForce GTX Titan benchmarks give us an approximate idea of the 780 Ti's performance, which is right on par with Asus' Mars 760.

Selling for $650 and able to keep up with a GeForce GTX 780 Ti, Asus' Mars 760 seems like a reasonable value next to some of Nvidia's heavy hitters. But there are other factors that tug us away from a dual-GPU solution, such as higher frame time latencies and a practical limit of 2 GB per GPU. Although Nvidia's SLI-based configurations perform notably better than CrossFire, single-GPU cards like the GeForce GTX 780 Ti retain an advantage (not to mention an extra gigabyte of GDDR5). But an even bigger value challenge comes simply from two GeForce GTX 760 cards in SLI, which sell for $500. So long as you have enough space on your motherboard and headroom from your power supply, that's probably a smarter choice.
Who might disregard those caveats and go with the Mars 760 anyway? Enthusiasts without room on their motherboards for two dual-slot boards, to start. Chassis dimensions come into play as well. Additionally, tuners will appreciate the Mars' willingness to exceed Asus' factory-set frequencies on the road to GeForce GTX 690-like performance (though we can't guarantee retail samples will scale as well as our press card). Finally, fans of Asus' one-off creations may find the style and exclusivity of this limited-production Mars 760 appealing.
No doubt, a $650 Mars 760 would have been a hotter-ticket item before Nvidia cut the price tag on its GeForce GTX 780 to make room for the 780 Ti. But with Nvidia's fastest single-GPU offering under $700 now, that card steals some of Asus' thunder. Today, it'd be easier to make a value case for the Mars somewhere under $600. This is clearly a special piece of equipment though, and it's fast enough that we doubt it'll last very long on Newegg. Even if it isn't the most attractively-priced graphics card, enthusiasts appreciate unique, and the Mars 760 does effectively integrate two very quick GK104 GPUs on a single PCB. Up until now, only Nvidia's reference-class $1000 GeForce GTX 690s did this.

