AMD’s Radeon R9 series is growing at an alarming rate. Just over a month ago, we were treated to the R9 280X and R9 270X. AMD followed up with the top-of-the-line R9 290X in late October and the slightly less top-of-the-line (yet much more compelling) R9 290 last week.
Today, AMD pulls back the curtain on the R9 270, which extends the R9 family down to the $179 price point. This is the family’s most affordable member, and it’s also the least power-hungry. It gets by with a single six-pin PCI Express power connector with a “typical board power” of only 150W, by AMD’s count.
The R9 270 will face off against Nvidia’s fastest sub-$200 card, the GeForce GTX 660. Nvidia slashed the GTX 660’s price to $179 last month, although today, a Newegg search shows the card selling for $190 before mail-in rebates. AMD, then, has an opportunity to undercut Nvidia with a newer product. But will the R9 270 be better?
Pitcairn’s latest gig
The Radeon R9 270 is based on the same Pitcairn chip as the 270X and the older Radeon HD 7800 series. As you can see in the table below, the R9 270 features a fully enabled version of the chip, just like the R9 270X and the older Radeon HD 7870.
|Radeon HD 7850 2GB||860||–||1024||55||28||4.8 GT/s||256|
|Radeon HD 7870 GHz||1000||–||1280||80||32||4.8 GT/s||256|
|Radeon R9 270||??||925||1280||80||32||5.6 GT/s||256|
|Radeon R9 270X||??||1050||1280||80||32||5.6 GT/s||256|
In fact, as far as I can tell, the R9 270 only differs from the R9 270X in its GPU clock speed and power envelope. The R9 270 runs about 75MHz slower (though its memory is clocked at the same 5.6 GT/s), and AMD has cut typical power consumption from 180W to 150W. The trimmed power envelope allows for this:
The R9 270 only requires a single six-pin PCI Express power connector, which is good news if you’re stuck with a lower-wattage power supply. The R9 270X and Radeon HD 7870 both need two six-pin connectors, while the GTX 660 requires only one. So far, then, Nvidia has had a small flexibility advantage on that front.
|Radeon R7 260X||18||62||2.0||2.2||104|
|Radeon R9 270||30||74||2.4||1.9||179|
|Radeon R9 270 (Asus)||31||78||2.5||2.0||179|
|Radeon R9 270X||34||84||2.7||2.1||179|
|Radeon HD 7850 2GB||28||55||1.8||1.7||154|
|Radeon HD 7870||32||80||2.6||2.0||154|
|GeForce GTX 650 Ti Boost 2GB||25||66||1.6||2.1||144|
|GeForce GTX 650 Ti Boost 2GB (Asus)||26||69||1.7||2.2||144|
|GeForce GTX 660||25||83||2.0||3.1||144|
Based on our theoretical numbers, the drop in clock speed doesn’t put the R9 270 at much of a disadvantage versus to the R9 270X. The Asus version of the R9 270 that AMD sent us actually runs at 975MHz instead of the reference 925MHz, which helps narrow the gap even more.
Compared to the GTX 660, the R9 270 has, on paper, a higher pixel fill rate, a similar texture filtering rate, higher shader throughput, and much more memory bandwidth. Its only disadvantage is a rasterization rate about two thirds that of the GTX 660.
The R9 270 has another perk worth mentioning: it comes with a free copy of Battlefield 4. Or, at least, some versions of it do. Or, they’re supposed to.
Yeah, this card’s game bundling situation is clear as mud right now. AMD originally said that all R9-series cards would ship with BF4, but it now tells us that it will be up to retailers and vendors to decide which of their cards ship with the game. As of November 14, we can’t find a single R9 270 in stock at Newegg with BF4 included.
The GeForce GTX 660, by contrast, ships with Assassin’s Creed IV: Black Flag and Splinter Cell Blacklist, plus a $50 discount on the purchase of an Nvidia Shield handheld console. This deal applies to all GTX 660s listed at Newegg. I don’t know a lot folks interested in buying a Shield, and I expect serious PC gamers to log far more hours in Battlefield 4 than in the titles Nvidia offers. Still, an actual game bundle beats the promise of one.
But I digress. We’ve still got a whole suite of game benchmarks to show you—after a brief detour through our testing methods.
Our testing methods
We tested using our tried-and-true “inside the second” methods. Since we don’t have FCAT equipment up here at TR North, we used Fraps to generate all our performance numbers.
Fraps gives us information about things happening at the start of the rendering pipeline—not, as FCAT does, at the end of the pipeline, when frames reach the display. Having both sets of numbers would be better, but the Fraps data is largely sufficient for the kind of testing we’re doing here. We don’t expect there to be much of a discrepancy between Fraps and FCAT numbers on single-GPU, single-monitor configurations like these.
This time, we’ve run most of our Fraps numbers through a three-frame low-pass filter. This filter is designed to compensate for one of the side effects of triple buffering. It should smooth out irregularities in our frame time measurements that don’t actually affect when the frames are shown on the display. We didn’t apply the filter to our BioShock Infinite numbers, since that game is based on the Unreal Engine. Most UE games don’t use triple buffering, so the filter isn’t appropriate for them.
Whether filtered or not, our “inside the second” Fraps numbers are far more informative than the raw frames-per-second data produced by more conventional benchmarking techniques. Such data can cover up problems like latency spikes and micro-stuttering, which have a real, palpable impact on gameplay.
For more information about Fraps, FCAT, and our inside-the-second methodology, be sure to read Scott’s articles on the subject: Inside the second: A new look at game benchmarking and Inside the second with Nvidia’s frame capture tools.
As ever, we did our best to deliver clean benchmark numbers. Tests were run at least three times, and we reported the median results. Our test systems were configured like so:
|Processor||Intel Core i7-3770K|
|North bridge||Intel Z77 Express|
|Memory size||4GB (4 DIMMs)|
|Memory type||AMD Memory & Kingston HyperX
DDR3 SDRAM at 1600MHz
|Chipset drivers||INF update 184.108.40.2061
Rapid Storage Technology 11.6
|Audio||Integrated Via audio
with 6.0.01.10800 drivers
|Hard drive||Crucial m4 256GB|
|Power supply||Corsair HX750W 750W|
|OS||Windows 8 Professional x64 Edition|
|Driver revision||Base GPU
|AMD Radeon R7 260X||Catalyst 13.11 beta V9||1100||1625||2048|
|Asus Radeon R9 270||Catalyst 13.11 beta V9||975||1400||2048|
|XFX Radeon HD 7850 2GB||Catalyst 13.11 beta V9||860||1200||2048|
|XFX Radeon HD 7870||Catalyst 13.11 beta V9||1000||1200||2048|
|Asus GeForce GTX 650 Ti Boost 2GB||GeForce 331.40 beta||1020||1502||2048|
|Asus GeForce GTX 660||GeForce 331.40 beta||980||1502||2048|
Thanks to AMD, Corsair, Crucial, and Kingston for helping to outfit our test rig. AMD, Asus, Nvidia, and XFX have our gratitude, as well, for supplying the various graphics cards we tested.
Image quality settings for the graphics cards were left at the control panel defaults, except on the Radeon cards, where surface format optimizations were disabled and the tessellation mode was set to “use application settings.” Vertical refresh sync (vsync) was disabled for all tests.
We used the following test applications:
Some further notes on our methods:
- We used the Fraps utility to record frame rates while playing 60- or 90-second sequences 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 three times per video card in order to compensate for 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 P3 Kill A Watt 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 Crysis 3 at the same quality settings used for our performance testing.
We measured noise levels on our test system, sitting on an open test bench, using a TES-52 digital sound level meter. The meter was held approximately 8″ 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.
Testing multiplayer games involves a lot of finicky variability, so I stuck with Battlefield 4‘s single-player campaign. Benchmarking was done at the start of the Singapore beach landing, which features cataclysmic environmental effects as well as explosions, gunfire, and… well, see for yourself:
I tested at 1080p with every detail setting maxed out except for deferred antialiasing, which was left disabled. That mix of settings caused a negligible decline in AA quality, but it yielded substantially better performance than the “Ultra” preset on the R9 270.
The R9 270 is off to a nice start. It beats the stock-clocked 7870 and winds up with the highest frame rate, the lowest 99th percentile frame time, and the least time spent beyond 16.7 ms of all the cards we tested. (None of the cards spent a significant amount of time above our other thresholds, except for the R7 260X, which is a different class of product.)
Nvidia’s GeForce GTX 660 doesn’t fare terribly well here. It’s barely any faster than the Radeon HD 7850 2GB, which retails for a good $20-30 less.
You might notice one card missing from our comparison: the R9 270X. Alas, we had a short time window to produce this review, and despite our best efforts, we weren’t able to get a 270X up to our labs at TR North. That said, the only difference between our R9 270 sample and a reference 270X is a 75MHz reduction in core clock speed. It’s not hard to imagine how the faster card would fare: negligibly better than its slower sibling.
Irrational Games’ latest BioShock title uses a modified Unreal engine to deliver some truly stunning vistas. To test it, I ran around the “Raffle Square” area for 60 seconds, heading toward a large crowd before doubling back and roaming around empty streets.
Testing was conducted at 1080p using the game’s built-in “Ultra” detail preset.
The R9 270 earns itself another gold medal in BioShock Infinite—though here, it’s so close to the 7870 that there’s hardly any difference between the two.
Nvidia, meanwhile, is at a disadvantage again. The GTX 660 distances itself nicely from the 7850 2GB this time, but it’s still a ways behind the R9 270 and the 7870. The GTX 660 also spends a little more time than the Radeons above our 33-ms frame time threshold, but not much. We’re talking about 35 ms out of a 60-second run here, a nearly negligible span of time.
Developed by Crystal Dynamics, this reboot of the famous franchise features a young Lara Croft. I tested it by running around a small mountain village near the beginning of the single-player campaign.
Here, too, I went with the “Ultra” detail preset at 1080p. In Tomb Raider, that preset enables tessellation effects and a whole host of other goodies.
I think some kind of pattern is emerging here. Yet again, the R9 270 edges out the 7870 to snag the gold medal—and yet again, the GTX 660 doesn’t quite measure up.
There’s not much to say about Crysis 3, except that it’s the latest Crysis game, and it has truly spectacular graphics. To benchmark it, we ran from weapon cache to weapon cache at the beginning of the Welcome to the Jungle level for 60 seconds per run.
Testing was done at 1080p using the medium detail preset with “very high” textures and medium SMAA antialiasing. I thought these cards would be able to handle the “high” preset, but the game just didn’t feel smooth at that setting.
So, this is interesting. With all the cards, the upper 5% of frame times are much higher than the rest. Our 99th percentile results reflect this, with all the cards neck and neck around 31-32 ms. That means the last 1% of frames are rendered at the equivalent of 31-32 FPS or less. We’re not seeing a lot of particularly large spikes in frame rendering times, as evidenced by our beyond-a-threshold data, but the game definitely feels less smooth than those FPS averages suggest. Why? Quite likely because of the constant variance from relatively low frame rendering times of around 10 ms to relatively high frame times around 35 ms. This problem may be caused by the difficult workload in this section of the game. There’s a tremendous amount of geometry detail in the grass here.
Oh, and if you’re wondering which card wins here: excluding the R7 260X, which clearly spends more time above 20 ms than the rest, it’s kind of a toss-up—at least in this particular part of Crysis 3.
Far Cry 3
Far Cry 3 Blood Dragon is a little too technicolor for my taste, so I tested Far Cry 3, which is based on the same engine. I picked one of the first assassination missions, shortly after the dramatic intro sequence and the oddly sudden transition from on-rails action shooter to open-world RPG with guns.
The game was run at 1080p using the “Ultra” detail preset. HBAO was enabled, as well, but MSAA was left disabled.
A-ha! Finally, the GeForce GTX 660 scores a victory, albeit a small one. The Nvidia card is on par with the R9 270 and 7870 in Far Cry 3 in terms of the frame rate average and in our latency-focused “time spent beyond X” metrics, which attempt to quantify the “badness” or lack of smoothness in each card’s performance. However, the GTX 660’s 99th percentile frame time is slightly lower. A look at the full frame latency curve shows that the GTX 660’s frame times don’t rise quite as much after the 95th percentile mark compared to the Radeons. In other words, with the toughest frames to render, when animation smoothness is most threatened, the GTX 660 performs best.
Given how the R9 270 performs in our other games, though, this showing isn’t quite enough to vindicate the GTX 660. The R9 really is just about as good in this game.
For some reason, the R9 270 draws more at idle than our 7870 card—an XFX Black Edition model underclocked to match reference specs. The R9 270 is more power-efficient under load, however, and it’s about the same when the display is switched off. As we saw on the previous pages, that reduced power consumption under load doesn’t coincide with reduced performance; on the contrary, the R9 270 generally matches or beats the 7870.
Noise levels and GPU temperatures
Save for the R7 260X, all of these cards have nice and quiet dual-fan coolers. Their noise levels are pretty much on par with one another—not just according to our decibel meter, but subjectively, as well.
The R9 270 draws less power than the 7870 under load, and it’s outfitted with a nice Asus DirectCU II cooler. No wonder it runs so cool.
We’ll once again wrap things up with a couple of value scatter plots. In both plots, the performance numbers are geometric means of data points from all the games we tested. The first plot shows 99th-percentile frame times converted into FPS for easier reading; the second plot shows simple FPS averages. Prices were fetched from Newegg, the GPU vendors, and the card makers, depending on what was appropriate.
The best deals should reside near the top left of each plot, where performance is high and pricing is low. Conversely, the least desirable offerings should be near the bottom right.
Well, I think this is pretty clear-cut. The Radeon R9 270 outperforms the GeForce GTX 660 overall, and it does so while drawing roughly the same amount of power at idle and under load—and while sipping fewer watts at idle with the display powered off.
The Asus version of the R9 270 we tested may be hot-clocked, but it’s priced at the same $179.99 as other R9 270 cards. That makes it arguably a better choice than the GTX 660, which starts at $189.99.
That is, as long as you don’t start accounting for game bundles.
All GTX 660s listed at Newegg come with Assassin’s Creed IV: Black Flag and Splinter Cell Blacklist plus a $50 Shield discount. AMD says some R9 270s are supposed to ship with Battlefield 4, but we can’t find any such bundles at Newegg right now. If freebies matter more to you than a little extra performance, then the GTX 660 may be the card for you. But if you somehow manage to score an R9 270 with a free copy of BF4, then I’d say that’s the better bargain—simply because Battlefield multiplayer should have much more replay value than the titles Nvidia offers.
The R9 270 has another thing going for it: Mantle. Many Mantle-enabled games are on the way, and the performance gains hinted at by developers sound tantalizing. It could be that the R9 270’s lead over the GTX 660 will grow significantly thanks to the new API. That’s another thing to consider.
As for the R9 270’s big brother, the R9 270X, well… folks with (slightly) deeper pockets may prefer to cough up the extra $20 for it, but given the small difference in clock rates between the two, I’m not sure I’d bother.