Home Nvidia’s GeForce GTX Titan reviewed
Reviews

Nvidia’s GeForce GTX Titan reviewed

Scott Wasson Former Editor-in-Chief Author expertise
Disclosure
Disclosure
In our content, we occasionally include affiliate links. Should you click on these links, we may earn a commission, though this incurs no additional cost to you. Your use of this website signifies your acceptance of our terms and conditions as well as our privacy policy.

You may have noticed that, in 2013, things have gotten distinctly weird in the world of the traditional personal computer. The PC is besieged on all sides by sliding sales, gloom-and-doom prophecies, and people making ridiculous claims about how folks would rather play shooters with a gamepad than a mouse and keyboard. There’s no accounting for taste, I suppose, which perhaps explains that last one. Fortunately, there is accounting for other things, and the fact remains that PC gaming is a very big business. Yet the PC is in a weird place, with lots of uncertainty clouding its future, which probably helps explain why Nvidia has taken so long—nearly a year after the debut of the GeForce GTX 680—to wrap a graphics card around the biggest chip based on its Kepler architecture, the GK110.

The GK110 has been available for a while aboard Tesla cards aimed at the GPU computing market. This chip’s most prominent mission to date has been powering the Titan supercomputer at Oak Ridge National Labs. The Titan facility alone soaked up 18,688 GK110 chips in Tesla trim, which can’t have been cheap. Maybe that’s why Nvidia has decided to name the consumer graphics card in honor of the supercomputer. Behold, the GeForce GTX Titan:

Of course, it doesn’t hurt that “Titan” is conveniently outside of the usual numeric naming scheme for GeForce cards. Nvidia is free to replace any of the cards beneath this puppy in its product lineup without those pesky digits suggesting obsolescence. And let’s be clear, practically anything else is Nvidia’s lineup is certain to be below the Titan. This card is priced at $999.99, one deeply meaningful penny shy of a grand. As the biggest, baddest single-GPU solution on the block, the Titan commands a hefty premium. As with the GeForce GTX 690 before it, though, Nvidia has gone to some lengths to make the Titan look and feel worthy of its asking price. The result is something a little different from what we’ve seen in the past, and it makes us think perhaps this weird new era isn’t so bad—provided you can afford to play in it.

GK110: one big chip


The Titan’s GK110 laid bare. Source: Nvidia

I won’t spend too much time talking about the GK110 GPU, since we published an overview of the chip and architecture last year. Most of the basics of the chip are there, although Nvidia wasn’t talking too specifically about the graphics resources onboard at that time. Fortunately, virtually all of the guesses we made back then about the chip’s unit counts and such were correct. Here’s how the GK110’s basic specs match up to other recent GPUs:

ROP
pixels/
clock
Texels
filtered/
clock
(int/fp16)
Shader
ALUs
Rasterized
triangles/
clock
Memory
interface
width (bits)
Estimated
transistor
count
(Millions)
Die
size
(mm²)
Fabrication
process node
GF114 32 64/64 384 2 256 1950 360 40 nm
GF110 48 64/64 512 4 384 3000 520 40 nm
GK104 32 128/128 1536 4 256 3500 294 28 nm
GK110 48 240/240 2880 5 384 7100 551 28 nm
Cypress 32 80/40 1600 1 256 2150 334 40 nm
Cayman 32 96/48 1536 2 256 2640 389 40 nm
Pitcairn 32 80/40 1280 2 256 2800 212 28 nm
Tahiti 32 128/64 2048 2 384 4310 365 28 nm

Suffice to say that the GK110 is the largest, most capable graphics processor on the planet. At 551 mm², the chip’s die size eclipses anything else we’ve seen in the past couple of years. You’d have to reach back to the GT200 chip in the GeForce GTX 280 in order to find its equal in terms of sheer die area. Of course, as a 28-nm chip, the GK110 packs in many more transistors than any GPU that has come before.

Here’s a quick logical block diagram of the GK110, which I’ve strategically shrunk beyond the point of readability. Your optometrist will thank me later. Zoom in a little closer on one of those GPC clusters, and you’ll see the chiclets that represent real functional units a little bit more clearly.

In most respects, this chip is just a scaled up version of the GK104 GPU that powers the middle of the GeForce GTX 600 lineup. The differences include the fact that each “GPC,” or graphics processing cluster, includes three SMX engines rather than two. Also, there are five GPCs in total on the GK110, one more than on the GK104. Practically speaking, that means general shader processing power has been scaled up a little more aggressively than rasterization rates have been. We think that’s easily the right choice, since performance in today’s games tends to be bound by things other than triangle throughput.

Versus the GK104 silicon driving the GeForce GTX 680, the GK110 chip beneath the Titan’s cooler has considerably more power on a clock-for-clock basis: 50% more pixel-pushing power, anti-aliasing grunt, and memory bandwidth; about 50% more texture filtering capacity; and not far from double the shader processing power. The GK104 has proven to be incredibly potent in today’s games, but the GK110 brings more of just about everything that matters to the party.

The GK110 also brings something that has no real use for gaming: considerable support for double-precision floating-point math. Each SMX engine has 64 DP-capable ALUs, alongside 192 single-precision ALUs, so DP math happens at one-third the rate of SP. This feature is intended solely for the GPU computing market. Virtually nothing in real-time graphics or even consumer GPU computing really requires that kind of mathematical precision, so Nvidia’s choice to leave this functionality intact on the Titan is an interesting one. It may also explain, in part, the Titan’s formidable price, since Nvidia wouldn’t wish to undercut its Tesla cards bearing the same silicon. Nevertheless, the Titan may prove attractive to some would-be GPU computing developers who like to play a little Battlefield 3 on the weekends.

Double-precision support on the Titan is a bit funky. One must enable it via the Nvidia control panel, and once it’s turned on, the card operates at a somewhat lower clock frequency. Ours ran a graphics demo at about 15MHz below the Titan’s base clock speed after we enabled double precision.

Oh, before we go on, I should mention that the GK110 chips aboard Titan cards will have one of their 15 SMX units disabled. On a big chip like this, disabling an area in order to improve yields is a very familiar practice. Let’s put that into perspective using my favorite point of reference. The loss of the SMX adds up to about two Xbox 360s worth of processing power—192 ALUs and 16 texture units at nearly twice the clock speed of an Xbox. But don’t worry; the GK110 has 14 more SMX units on hand.

The card: GeForce GTX Titan

You can tell at first glance that the Titan shares its DNA with the GeForce GTX 690. The two cards have the same sort of metal cooling shroud, with a window showing the heatsink fins beneath, and they share a silver-and-black color scheme. Nvidia seems to know one grand is a big ask for a video card, and it has delivered a card with the look and feel of a premium product.

GPU
base
clock
(MHz)
GPU
boost
clock
(MHz)
Shader
ALUs
Textures
filtered/
clock
ROP
pixels/
clock
Memory
transfer
rate
Memory
interface
width
(bits)
Peak
power
draw
GeForce GTX 680 1006 1058 1536 128 32 6 GT/s 256 195W
GeForce GTX Titan 836 876 2688 224 48 6 GT/s 384 250W
GeForce GTX 690 915 1019 3072 256 64 6 GT/s 2 x 256 300W

One of our big questions about the Titan has been where the GK110 would land in terms of clock speeds and power consumption. Big chips like this can be difficult to tame. As you can see in the table above, Nvidia has elected to go with relatively conservative clock frequencies, with an 837MHz base clock and an 876MHz “Boost” clock, courtesy of its GPU Boost dynamic voltage and frequency scaling technology. That’s a bit shy of the gigahertz-range speeds that its siblings have reached. However, the Titan may run as fast as ~970-990 MHz, in the right situation, thanks to its new iteration of GPU Boost. (More on this topic shortly.)

Those clock speeds contribute to the Titan’s relatively tame 250W max power number. That’s low enough to be a bit of a surprise for a card in this price range, but then the Kepler architecture has proven to be fairly power efficient. The card requires one eight-pin aux power connector and another six-pin one, not dual eight-pins like the GTX 690. The Titan’s power draw and connector payload fits well with its potential to be the building block of a multi-GPU setup involving two or three cards. Yes, the dual connectors are there to allow for three-way SLI—you know, for the hedge fund manager who loves him some Borderlands.

One other help to multi-GPU schemes, and to future-proofing in general, is the Titan’s inclusion of a massive 6GB of GDDR5 memory. Nvidia had the choice of 3GB or 6GB to mate with that 384-bit memory interface, and it went large. Perhaps the extra RAM could help in certain configs—say in a surround gaming setup involving a trio of four-megapixel displays. Otherwise, well, at least the extra memory can’t hurt.

The Titan measures 10.5″ from stem to stern, putting it smack-dab in between the GTX 680 and 690. That makes it nearly half an inch shorter than a Radeon HD 7970 reference card.


Revealed: the blower and the heat sink fins atop the vapor chamber. Source: Nvidia


A bare Titan card. Source: Nvidia

Nvidia tells us Titan cards should be widely available starting next Monday, February 25th, although some may start showing up at places like Newegg over the weekend. Nvidia controls the manufacture of Titan cards, and it sells those cards to its partners. As a result, we’re not likely to see custom coolers or circuit boards on offer. Only select Nvidia partners will have access to the Titan in certain regions. For the U.S., those partners are Asus and EVGA.

Although those restrictions would seem to indicate the Titan will be a limited-volume product, Nvidia tells us it expects a plentiful supply of cards in the market. The firm points out that the GK110 has been in production for a while, so we shouldn’t see the sort of supply issues that happened after the GTX 680’s introduction, when GK104 production was just ramping up. Of course, the Titan’s $999.99 price tag may have something to do with the supply-demand equation at the end of the day, so we’re probably not talking about massive sales volumes. Our hope is that folks who wish to buy a Titan won’t find them out of stock everywhere. I guess time will tell about that.

Meanwhile, the Titan will coexist uneasily with the GeForce GTX 690 for now—you know, talking trash about micro-stuttering versus raw performance, throwing elbows when Jen-Hsun isn’t looking, that sort of thing. The two cards are priced the same, sharing the title of “most expensive consumer graphics card” and catering to slightly different sets of preferences.

GPU Boost enters its second generation
I said earlier that the PC market has gone to weird place recently. Truth is, very high-end graphics cards have been in a strange predicament for a while now, thanks to the tradeoffs required to achieve the very highest performance—and due to conflicting ideas about how a best-of-breed video card should behave. The last couple generations of dual-GPU cards from AMD have stretched the limits of the PCIe power envelope, giving birth to innovations like the dual-BIOS “AUSUM” switch on the Radeon HD 6990. Those Radeons have also been approximately as quiet as an Airbus, a fact that doesn’t fit terribly well with the growing emphasis on near-silent computing.

With the Titan, Nvidia decided not to pursue the absolute best possible performance at all costs, instead choosing to focus on two other goals: good acoustics and extensive tweakability. The tech that makes these things possible is revision 2.0 of Nvidia’s GPU Boost, which is exclusive to Titan cards. As with version 1.0 introduced with the GTX 680, GPU Boost 2.0 dynamically adjusts GPUs speeds and voltages in response to workloads in order to get the best mix of performance and power efficiency. Boost behavior is controlled by a complicated algorithm with lots of inputs.

What makes 2.0 different is the fact that GPU temperature, rather than power draw, is now the primary metric against which Boost’s decisions about frequency and voltage scaling are made. Using temperature as the main reference has several advantages. Because fan speeds ramp up and down with temperatures, a Boost algorithm that regulates temperatures tends to produce a constant fan speed while the GPU is loaded. In fact, that’s what happens with Titan, and the lack of fan variance means you won’t notice the sound coming from it as much. Furthermore, Nvidia claims Boost 2.0 can wring 3-7% more headroom out of a chip than the first-gen Boost. One reason for the extra headroom is the fact that cooler chips tend to leak less, so they draw less power. Boost 2.0 can supply higher voltages to the GPU when temperatures are relatively low, allowing for faster clock speeds and better performance.

Via Boost 2.0, Nvidia has tuned the Titan for noise levels lower than those produced by a GeForce GTX 680 reference card, at a peak GPU temperature of just 80° C. That’s somewhat unusual for a top-of-the-line solution, if you consider the heritage of cards like the GeForce GTX 480 and the aforementioned Radeon HD 6990.

I’m happy with this tuning choice, because I really prefer a quiet video card, even while gaming. I think a best-of-breed solution should be quiet. Some folks obviously won’t agree. They’ll want the fastest possible solution, noise or not.


EVGA’s Precision utility

Fortunately, Boost 2.0 incorporates a host of tuning options, which will be exposed via tweaking applications from board makers. For instance, EVGA’s Precision app, pictured above, offers control over a host of Boost 2.0 parameters, including temperature and power targets, fan speed curves, and voltage. Yep, I said voltage. With Boost 2.0, user control of GPU voltage has returned, complete with the ability to make your GPU break down early if you push it too hard.

As you can see in the shot above, our bone-stock Titan card is running at 966MHz. We had the “rthdribl” graphics demo going in the background, and our board was happy to exceed its 876MHz Boost clock for a good, long time. As you might imagine given the fairly conservative default tuning, there’s apparently quite a bit of headroom in these cards. You can take advantage of it,if you’re willing to tolerate a little more noise, higher GPU temperatures, or both. Nvidia tells us it’s found that most Titans will run at about 1.2GHz without drama.

Further tweaking possibilities include control over the green LED lights that illuminate the “GEFORCE GTX” lettering across the top of the Titan card. EVGA has a utility for that. And Nvidia says it will expose another possibility—somewhat oddly, under the umbrella of GPU Boost 2.0—via an API for use by third-party utilities: display overclocking. Folks have already been toying with this kind of thing, pushing their LCD monitors to higher refresh rates than their official specs allow. Nvidia will be enabling its partners to include display overclocking options in tools like Precision. We haven’t seen an example of that feature in action yet, but we stand ready and willing to sacrifice our 27″ Korean monitor for the sake of science.

Falcon Northwest’s Tiki
Our GeForce GTX Titan review card came to us wrapped inside of the system you see pictured below.

This is the Tiki, from boutique PC builder Falcon Northwest. The Tiki is one of a new breed of compact gaming systems that seems to be taking the custom PC builders by storm. Nvidia asked Falcon to send us this PC in order to demonstrate how a GTX Titan card might be used. Apparently, boxes from custom shops like Falcon have accounted for about half of GeForce GTX 690 sales, and Nvidia expects that trend to continue with the Titan.

Also, as I’ve said, things are getting a little weird in the PC market, and many folks seem to be expecting compact systems like this one to become the new norm. If that happens, well, cards like the Titan could still have a nice future ahead of them after the iPad-wielding reaper comes for our full-towers. I’ve included the gamepad in the shots above to offer some sense of scale, but I’m not sure the pictures do the Tiki justice. The enclosure, which is Falcon’s own design, is just 4″ wide and roughly 15″ deep and 15″ tall. The base is granite, if you can’t tell from the images, and the whole system feels like it’s chiseled from one block of stone. It’s heavy and dense, but it must pack more computing power per cubic inch than anything else I could name.

Pop open the side, and you’ll begin to get a sense of the power lurking within, which pretty much amounts to most of the best stuff you can cram into a PC these days. The motherboard is the Asus P8Z77-I Deluxe Mini-ITX board we recently reviewed, and it houses the top-end Ivy Bridge processor, the Core i7-3770K. Storage includes a Crucial m4 SSD, WD Green 2TB hard drive, and an optical drive. The CPU is cooled by an Asetek water cooler, and the compact PSU comes from Silverstone.

The Tiki is impressively quiet when idle, about as good as any well-built full-sized system of recent vintage, and its acoustic footprint doesn’t grow by much when gaming. With Titan installed, it’s a stupid-fast gaming PC that emits only a low hiss.

Working inside the Tiki isn’t easy, though. You can kind of see in the picture that the video card plugs into a spacer plugged into a PCIe header that comes up out of the motherboad and does a right turn. One must remove the plate holding the system’s storage in order to extract the video card.

I guess that’s my builder mentality talking. When we talked to Falcon Northwest’s Kelt Reeves, who designed the Tiki, about this issue, he likened working with the Tiki to building a laptop. That’s no big deal to Falcon, since, as Reeves put it, “We’re the people you call when you want someone else to scrape their knuckles and handle the tech support.”

We considered using the Tiki system as the basis for all of our testing, but unfortunately, that wouldn’t fly. You see, while the GTX Titan card will fit into this box, many other cards will not, including the GeForce GTX 690 and any sort of multi-card config. Instead, the Tiki became our go-to system for quick sessions of Borderlands 2 when taking a break from work.

The Titan’s competition
Our time with the Titan and proper drivers hasn’t been long, so we’ve tried to focus our testing on a narrow group of competing solutions. From Nvidia, the Titan’s closest siblings are the GeForce GTX 680 and 690, which we’ve talked about already. The Titan’s rivals from the AMD camp include the Radeon HD 7970 GHz Edition and, well, two of the same. Turns out two of AMD’s fastest cards in a CrossFire team will set you back less than a single Titan card.

Here’s a look at our 7970 GHz Edition CrossFire team mounted in our testbed system. These are reference cards that came directly from AMD. Heck, for the first time in ages, I believe every card we’re testing is a stock-clocked reference model. Purists with OCD, rejoice!

I should mention that there are other options with similar performance. For instance, although AMD never did launch its promised dual-GPU Radeon HD 7990, several board makers have offered dual-7970 cards for sale. Unfortunately, we don’t have one on hand, so we decided to focus on two discrete cards in CrossFire for testing. We’d expect the performance of the dual-GPU cards to be very similar to our dual-card team. Similarly, we had the option of testing dual GeForce GTX 680s in SLI, but our past testing has proven that they perform almost identically to the GTX 690. Just keep in mind those other options exist when it comes time to sum everything up.

Peak pixel
fill rate
(Gpixels/s)
Peak
bilinear
filtering
(Gtexels/s)
Peak
bilinear
fp16
filtering
(Gtexels/s)
Peak
shader
arithmetic
rate
(tflops)
Peak
rasterization
rate
(Gtris/s)
Memory
bandwidth
(GB/s)
GeForce GTX 680 34 135 135 3.3 4.2 192
GeForce GTX
Titan
42 196 196 4.7 4.4 288
GeForce GTX
690
65 261 261 6.5 8.2 385
Radeon HD 7970
GHz
34 134 67 4.3 2.1 288
Radeon HD 7970
GHz CrossFire
67 269 134 8.6 4.2 576

Here’s a brief overview of how the contenders compare in key theoretical peak graphics rates. One interesting feature of the numbers is the fact that the Radeon HD 7970 GHz straddles the space occupied by the GTX 680 and the Titan. The 7970 matches the Titan for memory bandwidth and nearly does for peak shader throughput, but it’s substantially slower in terms of ROP rates and texture filtering capacity.

Meanwhile, the multi-GPU solutions look very nice in these compilations of theoretical peak performance, but these numbers assume perfect scaling from multiple GPUs. As you’ll see soon, that’s a very rosy assumption to make, for lots of reasons. Yes, those reasons include problems with multi-GPU micro-stuttering, which our latency-focused game benchmarks are at least somewhat capable of detecting. We have expressed some reservations about the limits of the tool we use to capture frame rendering times, especially for multi-GPU solutions, but absent a better option, we still think Fraps is vastly preferable to a simple FPS average. We’ll have more to say on this front in the coming weeks, but for today, our usual tool set will have to suffice.

Our testing methods
As ever, we did our best to deliver clean benchmark numbers. Our test systems were configured like so:

Processor Core i7-3820
Motherboard Gigabyte
X79-UD3
Chipset Intel X79
Express
Memory size 16GB (4 DIMMs)
Memory type Corsair
Vengeance CMZ16GX3M4X1600C9
DDR3 SDRAM at 1600MHz
Memory timings 9-9-11-24
1T
Chipset drivers INF update
9.3.0.1021
Rapid Storage Technology Enterprise 3.5.0.1101
Audio Integrated
X79/ALC898
with Realtek 6.0.1.6662 drivers
Hard drive Corsair
F240 240GB SATA
Power supply Corsair
AX850
OS Windows 8
Driver
revision
GPU
base
core clock 
(MHz)
GPU
boost
 clock 
(MHz)
Memory
clock
(MHz)
Memory
size
(MB)
GeForce
GTX 680
GeForce
313.96 beta
1006 1059 1502 2048
GeForce
GTX 690
GeForce
313.96 beta
915 1020 1502 2 x 2048
GeForce
GTX Titan
GeForce
314.09 beta
837 876 1502 6144

Radeon HD 7970 GHz
Catalyst
13.2 beta 5
1000 1050 1500 3072
Dual Radeon HD
7970 GHz
Catalyst
13.2 beta 5
1000 1050 1500 2 x 3072

Thanks to Intel, Corsair, and Gigabyte for helping to outfit our test rigs with some of the finest hardware available. AMD, Nvidia, and the makers of the various products supplied the graphics cards for testing, as well.

Unless otherwise specified, image quality settings for the graphics cards were left at the control panel defaults. Vertical refresh sync (vsync) was disabled for all tests.

In addition to the games, we used the following test applications:

Some further notes on our methods:

  • We used the Fraps utility to record frame rates while playing either a 60- or 90-second sequence from the game. Although capturing frame rates while playing isn’t precisely repeatable, we tried to make each run as similar as possible to all of the others. We tested each Fraps sequence five times per video card in order to counteract any variability. We’ve included frame-by-frame results from Fraps for each game, and in those plots, you’re seeing the results from a single, representative pass through the test sequence.

  • We measured total system power consumption at the wall socket using a Yokogawa WT210 digital power meter. The monitor was plugged into a separate outlet, so its power draw was not part of our measurement. The cards were plugged into a motherboard on an open test bench.

    The idle measurements were taken at the Windows desktop with the Aero theme enabled. The cards were tested under load running Skyrim at 2560×1600 with the Ultra quality presets.

  • We measured noise levels on our test system, sitting on an open test bench, using an Extech 407738 digital sound level meter. The meter was mounted on a tripod approximately 10″ from the test system at a height even with the top of the video card.

    You can think of these noise level measurements much like our system power consumption tests, because the entire systems’ noise levels were measured. Of course, noise levels will vary greatly in the real world along with the acoustic properties of the PC enclosure used, whether the enclosure provides adequate cooling to avoid a card’s highest fan speeds, placement of the enclosure in the room, and a whole range of other variables. These results should give a reasonably good picture of comparative fan noise, though.

  • We used GPU-Z to log GPU temperatures during our load testing.

The tests and methods we employ are generally publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.

Texture filtering
We’ll begin with a series of synthetic tests aimed at exposing the true, delivered throughput of the GPUs. In each instance, we’ve included a table with the relevant theoretical rates for each solution, for reference.

Peak pixel
fill rate
(Gpixels/s)
Peak bilinear
filtering
(Gtexels/s)
Peak bilinear
FP16 filtering
(Gtexels/s)
Memory
bandwidth
(GB/s)
GeForce GTX 680 34 135 135 192
GeForce GTX
Titan
42 196 196 288
GeForce GTX
690
65 261 261 385
Radeon HD 7970
GHz
34 134 67 288
Radeon HD 7970
GHz CrossFire
67 269 134 576

In the past, performance in this color fill rate test has been almost entirely determined by memory bandwidth limitations. Today, I’m not so sure. None of the solutions achieve anything like their peak theoretical rates, but I think that has to do with how many color layers this test writes. The Titan’s certainly not 50% faster than the GTX 680, despite having literally 50% higher memory bandwidth.

The multi-GPU solutions, meanwhile, scale up nicely in these synthetic tests.

Among the single-GPU solutions, the Titan is very much in a class by itself here.

Tessellation and geometry throughput

Peak
rasterization
rate (Gtris/s)
Memory
bandwidth
(GB/s)
GeForce GTX 680 4.2 192
GeForce GTX
Titan
4.4 288
GeForce GTX
690
8.2 385
Radeon HD 7970
GHz
2.1 288
Radeon HD 7970
GHz CrossFire
4.2 576

The last couple generations of Nvidia GPU architectures have been able to sustain much higher levels of polygon rasterization and throughput than the competing Radeons. The big Kepler continues that tradition, easily leading the GeForce GTX 680, even though its peak theoretical rasterization rate isn’t much higher. Most likely, the GK110’s larger L2 cache and higher memory bandwidth should be credited for that outcome.

Shader performance

Peak
shader
arithmetic
rate (tflops)
Memory
bandwidth
(GB/s)
GeForce GTX 680 3.3 192
GeForce GTX
Titan
4.7 288
GeForce GTX
690
6.5 385
Radeon HD 7970
GHz
4.3 288
Radeon HD 7970
GHz CrossFire
8.6 576

These tests of shader performance are all over the map, because the different workloads have different requirements. In three of them—ShaderToyMark, POM, and Perlin noise—the Titan and the Radeon HD 7970 GHz are closely matched, with the 7970 taking the lead outright in ShaderToyMark. That’s a nice reminder that the 7970 shares the same peak memory bandwidth and is within shouting distance in terms of peak shader throughput.

I had hoped to test OpenCL performance using at least one familiar test, LuxMark, but it crashed when I tried it on the Titan due to an apparent driver error. Anyhow, I think the GPU computing performance of these chips deserves a separate article. We’ll try to make that happen soon. For now, we’re going to focus on the Titan’s primary mission: gaming.

Guild Wars 2
Guild Wars 2 has a snazzy new game engine that will stress even the latest graphics cards, and I think we can get reasonably reliable results if we’re careful, even if it is an MMO. My test run consisted of a simple stroll through the countryside, which is reasonably repeatable. I didn’t join any parties, fight any bandits, or try anything elaborate like that, as you can see in the video below. Also, all of my testing was conducted in daytime, since this game’s day/night cycle seems to have an effect on performance.

Yes, we are testing with a single large monitor, and we are “only” testing a single-card Titan config today. Inexplicably, I ran out of Red Bull and wasn’t extreme enough to test triple-Titan SLI across three displays. I have several kegs of energy drink on order, though, and am hoping to test Titan SLI in the near future. In the meantime, it turns out you can stress these high-end GPU configs with just a single monitor using the latest games at their highest image quality settings. Even an MMO like this one.


Frame time
in milliseconds
FPS
rate
8.3 120
16.7 60
20 50
25 40
33.3 30
50 20

We’ll start with plots of the frame rendering times from one of our five test runs. These plots visually represent the time needed to render each frame of animation during the gameplay session. Because these are rendering times, lower numbers on the plot are better—and higher numbers can be very bad indeed.

You can see some really intriguing things in these plots. Generally, the GTX Titan performs quite well, rendering the vast majority of frames in easily less than 20 ms. The table on the right will tell you that works out to over 50 FPS, on average. There are occasional spikes to around 40 ms or so, but that’s not so bad—except for the cluster of high frame times near the end of the test run on each GeForce. Subjectively, there is a very noticeable slowdown in this same spot each time, where we’re rounding a corner and taking in a new view. The Radeon HD 7970 single- and dual-card configs produce fewer frames at generally higher latencies than the Titan, but they don’t have the same problem when rounding that corner. (You’ll recall that AMD recently updated its drivers to reduce frame latencies in GW2.) You don’t feel the slowdown at all on the Radeons.

Interestingly, both the traditional FPS average and our proposed replacement for it, the 99th percentile frame time, tend to agree that the Titan and GTX 690 perform about the same here and that both are generally faster than the 7970 GHz configs. That one iffy spot on the GeForces is very much an outlier.

The 99th percentile numbers also say something good about how all of these solutions perform. They all render 99% of the animation frames in under 33 milliseconds, so they spend the vast majority of the time during the test run cranking out frames at a rate better than 30 FPS.

We can show the frame latency picture as a curve, to get a broader sense of things. The Titan tracks very closely with the GeForce GTX 690 throughout. Generally, these two cards are the best performers here, offering the smoothest animation. The shape of the curve offers no evidence that multi-GPU microstuttering is an issue on either the dual 7970s or the GTX 690.


However, we know about that one trouble spot at the end of the test run on the GeForces, and it shows up in our measure of “badness,” which considers the amount of time spent working on truly long latency frames. (For instance, if our threshold is 50 ms, then a frame that takes 80 ms to render contributes 30 ms to the total time beyond the threshold.) As you can see, that slowdown at the end pushes the three GeForce configs over 50 milliseconds for a little while.

As I’ve said, you’ll notice the problem while playing. Trouble is, you’ll also notice that animation is generally smoother on the GeForce cards otherwise, as our other metrics indicate. Lower the “badness” threshold to 16.7 ms—equivalent to 60 FPS—and the Titan and GTX 690 spend the least time above that mark.

So, I’m not sure how one would pick between the performance of the various cards in this one case. It’s pretty unusual. I’m happy to be able to show you more precisely how they perform, though. Totally geeking out over that. Let’s move on to something hopefully more definitive.

Borderlands 2
Next up is one my favorites, Borderlands 2. The shoot-n-loot formula of this FPS-RPG mash-up is ridiculously addictive, and the second installment in the series has some of the best writing and voice acting around. Below is a look at our 90-second path through the “Opportunity” level.

As you’ll note, this session involves lots of fighting, so it’s not exactly repeatable from one test run to the next. However, we took the same path and fought the same basic contingent of foes each time through. The results were pretty consistent from one run to the next, and final numbers we’ve reported are the medians from five test runs.



Oh, man. The truth is that all of the configs ace this test. With 99th percentile frame times at 20 milliseconds or less, all of these cards are almost constantly spitting out frames at a rate of 50 FPS or better. And none of them spend any substantial time on frames that take longer than 50 ms to render. We’re talking smooth animation all around, which matches my subjective impressions.

Sleeping Dogs
Our Sleeping Dogs test scenario consisted of me driving around the game’s amazingly detailed replica of Hong Kong at night, exhibiting my terrifying thumbstick driving skills.



Oooh. Look at the 7970 CrossFire plot, which looks like a cloud rather than a line. That’s potential evidence of the dreaded multi-GPU micro-stuttering. Here’s how small chunk of it looks up close:


Multi-GPU solutions tend to split up the workload by handing one frame to the first GPU and the next to the second GPU in alternating fashion. When the two GPUs go out of sync, bad things happen. Now, I should note that the 7970 CrossFire solution still seems to perform well here—even the longer frame times in the pattern are fairly low.

However, the FPS average doesn’t reflect the true performance of the CrossFire solution, which isn’t quite that far ahead of the pack, as the latency-focused 99th percentile metric tells us. The GeForces don’t do as well in the 99th percentile frame time metric as their FPS averages would suggest, either.

All of the GeForces struggle to some degree with the toughest 4-5% of the frames rendered, and we know from the plots that those frames represent spikes riddled throughout our test sequence.


Obviously the GeForce GTX 680 is the big loser here, no matter how you slice it. Its performance is just brutal. One of the surprise winners, in my book, is the GTX 690. When playing the game, the addition of a second GPU just like the GTX 680’s translated into vastly improved smoothness. The difference between the 680 and 690 in our “badness” measurement at 50 ms tells the story there. Multi-GPU solutions don’t always improve fluidity that effectively.

The Radeons are even bigger winners, though, by being straightforwardly quicker overall.

Assassin’s Creed III
This game appears to be a thought experiment centered around what would happen if the Quaker Oats guy had invented parkour in 18th-century Boston.

Since the AC3 menu doesn’t lend itself to screenshots, I’ll just tell you that we tested at 2560×1600 with environment, anti-aliasing, and shadow quality set to “very high” and texture quality set to “high.” I understand that the “very high” AA setting combines 4X multisampling with FXAA HQ post-process smoothing. This game also supports Nvidia’s TXAA, but Nvidia has gated off access to that mode from owners of Radeons and pre-Kepler GeForces, so we couldn’t use it for comparative testing.



Well, the strange news here is how Radeon HD 7970 performance drops when you add a second video card. That ain’t right. We’ve asked AMD what the problem is, but haven’t yet gotten an answer. A little googling around suggests we’re not alone in seeing this problem, though. This sort of thing happens sometimes with multi-GPU solutions, especially with newer games, but AC3 has been out long enough that this situation seems odd.

In other news, the Titan almost exactly mirrors the performance of the GTX 690 here. I sense a developing trend.

Hitman: Absolution
We’ve finally found a good use for DX11 tessellation: bald guys’ noggins.


We noticed latency problems with this game last year when we tested it on the Radeon HD 7950, and they’ve carried over here to the 7970 GHz Edition at higher resolutions. Adding a second 7970 card doesn’t banish the spiky frame time pattern, but the CrossFire rig renders twice as many frames at roughly half the effective latency, which gets the job done.


Our “badness” check provides a useful reminder that the stakes are pretty low here. None of the cards waste much time on truly high-latency frames. That comports with the subjective experience of playing on the single 7970 card. The animation is a little jumpy and unsettled, just like our Fraps data suggests, but it’s not terrible.

Far Cry 3



Here’s a nice example of the trouble with FPS averages. Have look at the average FPS for the Radeon HD 7970 CrossFire rig and then at the frame time plot for that setup. Yikes, right? Although the CrossFire rig slings out loads of frames, the animation it produces here is literally horrible and strangely jumpy. I seriously wondered if frames weren’t being delivered out of order. Even the 99th percentile frame time, which places the dual 7970 rig roughly on par with a single card, doesn’t capture the extent of it.

There’s some funkiness going on with all of the cards, though. Notice the big spikes to over 100 ms in the GeForce GTX 680 and 690 plots. Once almost every run, we hit a place where the game would just choke and then recover.

When I asked AMD about the CrossFire problems, they told me they were investigating multiple angles. When I asked Nvidia about this game, they said they believe there may be some problems with the application itself. So… yeah. In my experience, you can get Far Cry 3 to run fairly smoothly on these video cards, but you’ll have to back way down on the image quality settings.

The only non-loser here is the GeForce Titan, which didn’t encounter the big pauses we saw with the two GK104-based cards, although that could be a benefit of the Titan’s slightly newer driver revision. I think this one is still a work in progress all around.

The Elder Scrolls V: Skyrim
No, Skyrim isn’t the newest game at this point, but it’s still one of the better looking PC games and remains very popular. It’s also been a particular point of focus in driver optimizations, so we figured it would be a good fit to include here.

We did, however, decide to mix things up by moving to a new test area. Instead of running around in a town, we took to the open field, taking a walk across the countryside. This change of venue provides a more taxing workload than our older tests in Whiterun.

I should mention that Skyrim has some problems with extremely high frame rates, like most of these systems produced in this test. You wouldn’t want to play the way we’ve tested, with the frame rate cap removed, because the game’s AI kind of goes haywire, causing animals and NPC to move about erratically. We may consider testing with the game capped in the future. We can’t show differences in FPS averages that way, but we still could measure any frame latency spikes.



All of these graphics solutions are extremely fast in this test. The Radeon HD 7970 CrossFire config, which is the fastest in terms of FPS averages, does encounter a few quick latency spikes along the way. Those knock it back to third in the 99th-percentile standings. Still, we’re gonna need even higher resolutions or some mods for Skyrim to really stress any of these cards.

Power consumption

AMD’s nifty ZeroCore Power feature confers an advantage to the Radeons when the display drops into power-save mode. Their fans stop spinning and their GPUs drop into a low-power state. Even when idle at the desktop with the display turned on, the second card in the 7970 CrossFire team stays in ZeroCore mode, keeping idle system power lower than the GTX 690’s. Meanwhile, the Titan system draws less power when idling at the desktop than any other setup, which is quite the feat for such a large chip.

We tested power under load while running Skyrim, and we can do a rough estimate of power efficiency by correlating power draw and performance in this game, as our scatter plot does. The Titan-based test rig comes out looking pretty good overall.

Noise levels and GPU temperatures

At its default tuning, the Titan is true to its billing as quieter than a GeForce GTX 680. That’s a very nice accomplishment, since we know it’s drawing more power than the GTX 680 or the Radeon HD 7970, yet it dissipates that heat with less noise.

Speaking of the 7970, AMD’s reference cards are much louder than is necessary. We’ve seen better acoustics from 7970 cards with similar performance and non-reference coolers, like the XFX Black Edition we tested here.

Going scatter-brained
This page is a continuation of a little experiment with scatter plots that we started with our GTX 680 review. We’re looking at the correlations between various architectural features and delivered performance across our test suite. This time around, we’ve switched to our 99th-percentile frame time as the proper summary of overall performance, but we’ve converted it into FPS terms so that a higher score is better, to make the plots intuitively readable.

Yeah, the multi-GPU configs on the die size plot are a little iffy, since we’re just doubling the die area for dual-chip solutions. Anyhow, you can see that the GeForces tend to outdo the Radeon HD 7970 in terms performance per die area.

All of the theoretical peak rates we’ve included here look to be somewhat correlated to overall delivered performance. The weakest correlations look to be triangle rasterization rates and, oddly enough, memory bandwidth.

The correlations grow a little stronger when we match the results of directed tests against delivered game performance. I’d like to add more GPUs to these plots in the near future, though, before drawing any big conclusions about GPU architecture features and gaming performance.

Conclusions
As with many reviews of products this complex, we come to the end of a pretty extensive exercise with a long list of things we wish we’d had time to test, including overclocking and tweaking via GPU Boost 2.0, a broader selection of older cards, a nice range of GPU computing applications, and even more games, including the just-released Crysis 3. Oh, yes, and Titan multi-GPU configs. We’ll try to get to as many of those things as possible in the coming days.

For now, we can summarize our current results in one of our famous price-performance scatter plots. As usual, our prices come from cards selling at Newegg, and the performance results are a geometric mean from the seven games we tested. We’ve converted our 99th-percentile
frame times into FPS so that both plots read the same: those solutions closest the top-left corner of the plot area offer the best combo of price and performance.


If you were expecting the Titan to be a slam-dunk value winner, well, no card with a $1K price tag is gonna nestle up into the top-left position in one of these plots. And, if you look at raw FPS per dollar, the Radeon HD 7970 GHz CrossFire config looks to be the clear winner. However, we believe our latency-focused 99th-percentile frame time metric is the best measure of overall gaming performance, and the Titan acquits itself fairly well on that front, coming in juuuust below the GeForce GTX 690 at the same price.

You already know that the Titan is physically shorter, requires less power, and generates less heat and noise than the GTX 690. And it can serve as a basic building block in a two- or three-way SLI config, promising higher peak performance than even dual GTX 690s could achieve, especially with that 6GB of GDDR5 onboard. The Titan definitely has a place in the market, high atop Nvidia’s product stack, and we’re happy to see little extras like double-precision math support, a gorgeous industrial design, and excellent acoustic and power tuning as part of the mix. If you’re laying down a grand for a graphics card, you’re going to want the best of everything, and the Titan doesn’t skimp.

With that said, I can’t help but think most PC gamers would give up the double-precision support and accept a plastic cooling shroud and “only” 3GB of onboard memory in exchange for a price that’s several hundred bucks lower. That was essentially the deal when the GeForce GTX 580 debuted for 500 bucks—or, if you want to point to an even larger chip than the GK110, when the GTX 280 started life at $650. A premium product like the Titan is no bad thing, but we’re kind of hoping Nvidia follows up with something slightly slower and little more affordable, as well. At present, your best value-for-dollar proposition in this space from Nvidia likely comes from dual GTX 680s in SLI, which should perform very much like a GTX 690 at a lower overall price.

As for the competition, our dual Radeon HD 7970 GHz CrossFire team put up a very good fight, taking the top spot in multiple games. Looks to me like AMD has the GPU hardware needed to give Nvidia a truly formidable challenge—more so than it currently does. We saw CrossFire scaling issues in a couple of games, but those surely can be resolved with driver updates. And we’re still waiting on the rewritten memory manager for GCN-based chips that promises lower frame latencies in DX10/11 games. If AMD can deliver on the driver updates, a pair of 7970s could offer a compelling alternative to a single Titan, for less money—and with a very nice game bundle. Just know that waiting for driver updates to fix problems has become a time-honored tradition for owners of CrossFire rigs. And you’ll want to search out a board with a CrossFire-friendly custom cooler. The reference cards we tested are too darned loud.

Maybe we should get some quieter examples in the labs for some Crysis 3 testing, alongside two or three Titans, eh? We’ll get right on that.

People sometimes like to yell at me on Twitter.

The Tech Report - Editorial ProcessOur Editorial Process

The Tech Report editorial policy is centered on providing helpful, accurate content that offers real value to our readers. We only work with experienced writers who have specific knowledge in the topics they cover, including latest developments in technology, online privacy, cryptocurrencies, software, and more. Our editorial policy ensures that each topic is researched and curated by our in-house editors. We maintain rigorous journalistic standards, and every article is 100% written by real authors.

Scott Wasson Former Editor-in-Chief

Scott Wasson Former Editor-in-Chief

Scott Wasson is a veteran in the tech industry and the former Editor-in-Chief at Tech Report. With a laser focus on tech product reviews, Wasson's expertise shines in evaluating CPUs and graphics cards, and much more.