Nvidia’s GeForce GTX 950 graphics card reviewed

So far, 2015 has given us a bounty of high-end graphics cards. If you want to spend somewhere north of three hundred bucks on a graphics card—and heck, maybe three times that—then you have plenty of shiny new options from which to choose. I’ve spent a borderline scandalous amount of time testing graphics cards in 4K resolutions in the past six months.

Playing with the fancy toys is nice, but it’s not very realistic. The vast majority of PC gamers play at resolutions of 1080p or below, and for the most part, you can get away with using a much more affordable graphics card when you “only” have a couple of megapixels to paint.

Gigabyte’s GeForce GTX 950 Xtreme Gaming

That’s pretty much the rationale behind the GeForce GTX 950, a new entry in Nvidia’s graphics card lineup that promises a nice mix of price and performance. The GTX 950 isn’t especially revolutionary in technology terms. It’s based on a somewhat hobbled version of the same GM206 graphics chip used in the GeForce GTX 960. So no new silicon here. But the GTX 950 does bring a full Maxwell 200-series feature set to cards priced well under $200, and its performance is more than credible. In fact, as Nvidia points out, this card is “faster than any current console,” with the obvious targets being the Xbox One and PlayStation 4. As a result, the GTX 950 ought to be a popular choice for an awful lot of PC gamers, especially those folks with 1080p displays.

If you keep on top of these things, you may be aware that the GTX 950 has been on the market for over a month now. Our review has been lingering deep in the bowels of Damage Labs for . . . reasons. However, our delay has been somewhat fortuitous. Not only do we have our hands on three different versions of the GTX 950, but we also have a couple of variants of the competing card from the red team, the Radeon R7 370. We should be able to give you a bit deeper look not just at these mid-range GPUs but also the particular cards you might be buying.

The table above offers a look at how the GTX 950 fits into the current GeForce lineup. The mighty-mite GTX 750 Ti has dropped to $119.99 in order to make room for its new big brother. Since it’s based on a larger chip with more of everything, the GTX 950 represents a nice step up at $159.99.

At just 90 watts, the GTX 950 technically would only require a single six-pin aux power input to do its thing. Board makers like Gigabyte, whose Xtreme Gaming version of the 950 is pictured above, have taken things up a notch, as they tend to do. The cards we have from both Gigabyte and EVGA feature an eight-pin input. The extra juice might ensure more overclocking headroom. The Xtreme Gaming also happens to look like Batman’s hovercraft, which I find strangely appealing.

EVGA offers no less than four versions of the GTX 950, and the FTW edition shown above is the most extreme, with base and boost clocks of 1203 and 1405MHz, respectively. (Incidentally, that’s exactly the same rated clock speeds Gigabyte has assigned to the Xtreme Gaming.) The FTW edition is selling for $179.99 at Newegg, a price that may be justified by the presence of a “full-sized” board and cooler combo that measures just over 10″ in length.

Asus has taken a bit of a different approach with some of its mid-range graphics cards recently, and the Strix GTX 950 follows suit. This card has a capable dual-slot cooler, but it sports a six-pin power input and has somewhat more conservative clock frequencies of 1165/1355MHz. Don’t discount the Strix’s performance based on those specs just yet, though. We’ve found that actual clock speeds—and thus performance—with today’s GeForces depend quite a bit on cooling and board power delivery, and Asus has a good record on this front.

Since the Strix was the first example of the GeForce GTX 950 to arrive in Damage Labs, we’ve used it to represent this GPU in the bulk of the performance results on the following pages. That said, we have tested the various flavors of the GTX 950 against one another, as well.

The Radeon R7 370
This look at the GTX 950 allows us to devote some attention to its closest competitor, the Radeon R9 370, which also made its debut in recent months. The R7 370 is fascinating because it’s a reasonably competitive modern graphics card based on the AMD Pitcairn GPU, a chip first introduced aboard the Radeon HD 7870 and 7850 in March of 2012. Pitcairn also had starring roles aboard the Radeon R7 270X, R7 270, and R7 265.

Three and a half years seems like a long time in the high-tech realm, and Pitcairn shows its age by not supporting some features introduced in newer Radeons. For instance, the R7 370 doesn’t have the TrueAudio DSP for the acceleration of in-game sound effects, and its video decoder and encoder hardware isn’t ready for 4K data rates or encoding types. Even more notably, the R7 370’s display hardware isn’t capable of working with FreeSync variable-refresh displays, one of the niftiest innovations in PC gaming in recent years. (The GTX 950 can work with variable-refresh displays based on Nvidia’s G-Sync standard.)

One other notable feature Pitcairn lacks is delta-based color compression, which recent GPUs have used to squeeze more throughput out of their given memory bandwidth. The R7 370 makes up for this shortcoming the honest way: by using a 256-bit-wide path its GDDR5 memory, double the width of the GTX 950’s memory interface. In fact, the R7 370 sports wider, more robust hardware in almost every respect compared to its competition—and, in a classic AMD move, its starting price is ten bucks cheaper than the GeForce, too.

Oddly enough, the tables above will tell you that the R7 370 uses a slightly cut-down version of the Pitcairn GPU. The full chip has 1280 stream processors and 80 texels per clock of filtering power. Evidently, AMD prefers to disable the weaker parts of the Pitcairn GPU and crank up clock speeds instead. The R7 370 is almost entirely identical to the Radeon R7 265, yet the 370 has a 50MHz higher boost clock and slightly faster memory. AMD says it has implemented “new features at the micro-code level which enable higher overall performance.” I suspect most of the changes have to do with the PowerTune dynamic clocking algorithm, which the firm has refined incrementally over time.

Speaking of clock speeds, MSI has taken things even further with its R7 370 Gaming 2G, slapping one of its formidable coolers onto Pitcairn and raising the boost clock to 1050MHz. MSI is asking $159.99 for the Gaming 2G at Newegg.

Sapphire has chosen a more compact dual-slot cooler and a more conservative 985MHz boost clock for its Nitro R7 370, but this card has an ace that the MSI card lacks: 4GB of GDDR5 memory. Cards in this class have gotten along quite well with 2GB of RAM to date, but doubling up to 4GB could give the Nitro a bit of future-proofing. The Nitro is going for $169.99 at Newegg, but Sapphire also sells a 2GB version for 20 bucks less. Similarly, MSI offers an R7 370 Gaming 4G for $179.99.

Our testing methods
Most of the numbers you’ll see on the following pages were captured with Fraps, a software tool that can record the rendering time for each frame of animation. We sometimes use a tool called FCAT to capture exactly when each frame was delivered to the display, but that’s usually not necessary in order to get good data with single-GPU setups. We have, however, filtered our Fraps results using a three-frame moving average. This filter should account for the effect of the three-frame submission queue in Direct3D. If you see a frame time spike in our results, it’s likely a delay that would affect when the frame reaches the display.

We didn’t use Fraps with Civ: Beyond Earth or Battlefield 4. Instead, we captured frame times directly from the game engines using the games’ built-in tools. We didn’t use our low-pass filter on those results.

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

Thanks to Intel, Corsair, Kingston, 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.

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.

Sizing ’em up
Multiply the peak clock speeds by the peak per-clock rates of these graphics cards, and you end up with this:

Compare the peak rates for the R7 370 and GTX 950, and you’ll notice the disparities right away. The GTX 950 has substantially higher peak rates for pixel fill, fp16 texture filtering, and rasterization. The two cards have comparable shader arithmetic rates. Meanwhile, the R7 370 has a rather cavernous advantage in terms of memory bandwidth.

Of course, those are just theoretical peak rates. Our fancy Beyond3D GPU architecture suite measures true delivered performance using a series of directed tests.

The GTX 950’s advantage in the fill rate department is even larger in practice than in theory, likely because of its color compression storage capability. The R7 370’s ROP throughput is probably limited by memory bandwidth, a bottleneck that newer GPUs appear to have sidestepped thanks to compression.

By the way: GPU nerd alert. The GTX 950’s victory over its bigger brother, the GTX 960, suggests something odd about how Nvidia is disabling units in the GM206 in order to create the 950. Based on what we know about the Maxwell architecture, we’d expect the GTX 950’s six SM units to be limited to 24 pixels per clock of peak throughput, four pixels per clock for each SM. Instead, the GTX 950 is clearly hitting 32 pixels per clock of throughput in this test. Something unexpected must be happening.

Could the firm be disabling half of an SM across four different SMs, killing off the texturing and ALU hardware while retaining the full datapaths to the ROPs? That setup would be unconventional, but it would explain things. I’ve asked Nvidia what’s up and will post an update if we get an answer.

This bandwidth test measures GPU throughput using two different textures: an all-black surface that’s easily compressed and a random-colored texture that’s essentially incompressible.

The R7 370 fares pretty well here thanks to an abundance of raw memory bandwidth, even though it shows no evidence of any color compression at all. The GTX 950, however, nearly matches the R7 370’s throughput when compressing a texture of a single color.

The GTX 950 and R7 370 are very evenly matched when dealing with texture formats that are eight bits per color channel and 32 bits per color channel. When filtering a 16-bit format, though, the GTX 950 is roughly twice as fast as the R7 370. How much this difference matters in practice will depend on the texture formats used by individual games.

Nvidia has long had an architectural advantage in polygon throughput, both in terms of raw rasterization rates and DX11-style tessellation with geometry expansion. That situation persists here, with the GTX 950 doubling the performance of the R7 370 in TessMark.

There’s remarkable parity in shader processing power between the GTX 950 and R7 370. Other aspects of AMD’s GCN architecture are starting to look dated, but its shader arrays remain very respectable.

GTA V
Forgive me for the massive number of screenshots below, but GTA V has a ton of image quality settings.

Here’s our first look at performance in a real game, and none of the GPUs show signs of the big spikes and slowdowns we often see. Rockstar has done a nice job with GTA V on that front. As a result, the FPS averages and the 99th-percentile frame times tend to match up pretty closely.

Beyond that, the GTX 950 is pretty clearly faster than the Radeon R7 370 here. Truth be told, both cards are quite competent to run GTA V at these settings.

Those nice, flat frame-time plots above produce equally flat percentile curves. Even the most difficult-to-produce frames don’t cause much trouble for most of these cards. Look back a couple of generations, though, and the GeForce GTX 650 is overmatched; it’s consistently slow in this scenario.

These “time spent beyond X” graphs are meant to show “badness,” those instances where animation may be less than fluid—or at least less than perfect. The 50-ms threshold is the most notable one, since it corresponds to a 20-FPS average. We figure if you’re not rendering any faster than 20 FPS, even for a moment, then the user is likely to perceive a slowdown. 33 ms correlates to 30 FPS or a 30Hz refresh rate. Go beyond that with vsync on, and you’re into the bad voodoo of quantization slowdowns. 16.7 ms correlates to 60 FPS, that golden mark that we’d like to achieve (or surpass) for each and every frame, and 8.3 ms is a relatively new addition that equates to 120Hz, for those with fast gaming displays.

On our measure of “badness,” the GTX 950 and R7 370 are nearly equal. Neither card spends much time working on frames that take longer than 16.7 milliseconds to produce, which means they they both offer an almost flawless and steady 60 frames per second.

The Witcher 3

The frame-time plots in this game are a little less consistent, which is more typical of PC games generally. The worst offender here looks to be the GeForce GTX 750 Ti, which shows the sort inconsistency we’ve also seen from older, Kepler-based GPUs in this game. The R7 370 struggles with some 50-60-ms spikes, as well. Those are large enough to feel as hiccups during gameplay.

The GTX 950 is generally faster than the R7 370, as reflected in the FPS average, and it’s quicker when dealing with more difficult frames, as the 99th-percentile result reflects. In fact, in this case, there’s little practical advantage for the GTX 960 over the 950, as the percentile results show.

With the exception of ye olde GTX 650, all of these cards do a credible job of avoiding the worst slowdowns: frames that take longer than 50 ms to produce. The R7 370’s few spikes do push it past that threshold a bit. The real difference comes at 33 ms, a mark the GTX 950 exceeds only for a single millisecond, while the R7 370 spends a more than a third of a second working on frames that take longer than 33-ms—and are thus slower than a constant 30-FPS rate.

Civilization: Beyond Earth
Since this game’s built-in benchmark simply spits out frame times, we were able to give it a full workup without having to resort to manual testing. That’s nice, since manual benchmarking of an RTS with zoom is kind of a nightmare.

Oh, and the Radeons was tested with the Mantle API instead of Direct3D. Only seemed fair, since the game supports it.

The GTX 950 maintains its comfortable lead over the R7 370 here—and notably, once again, the 950 stays consistently below the 30-ms threshold with each frame produced, while the R7 370 does not.

Battlefield 4
BF4 appears to run better on the R7 370 (and on other Radeons using recent drivers) in DirectX 11 rather than Mantle, so the results for the R7 370 below come from the game’s DX11 mode.

Both the GTX 950 and the R7 370 handle BF4 on Ultra pretty well, with 99% of frames produced in less than 33 ms—or more than 30 FPS. The persistent reality is that the GTX 950 is somewhat faster, though, and that advantage translates into better performance according to every one of our advanced metrics.

Ashes of the Singularity
Here’s our first look at the Ashes benchmark, which lets us test performance in both DirectX 11 and 12 using code from a pre-release version of this upcoming game. DirectX 12 is a new graphics programming API built into Windows 10 that promises lower CPU overhead, better threading, and fuller use of the GPU’s potential. (We’ve also tested the GTX 950 and Radeon R7 370 in the Fable Legends benchmark right here.)

Click through the results below to see frame time plots from both versions of DirectX.

All of the cards appear to hit a tough spot early in the benchmark where there are one or more really slow frames. The slowdown appears to be a worse on the GeForces than on the R7 370, and the problem is further heightened on the GeForces in DX12.

Although these spikes are huge, over half a second long in some cards, they’re surrounded by enough other frames that they don’t drag down FPS averages. If you’ve seen FPS-only results other places, rest assured, they’re probably not terribly misleading as a result of this hiccup. Ahem.

Forgive me for not sorting the results in the bar graphs; doing so would require some extra work in Excel, and let’s face it: I’m not doing that.

Overall, the GeForces perform a little slower in the DX12 version of this benchmark than in DX11. Meanwhile, the R7 370 benefits from the switch to DX12. Still, the Radeon remains slower than the GTX 950 overall. And, truth be told, none of these cards are producing frames quickly enough over time to provide a good gaming experience. We might have to lower the display resolution or dial back the image quality to achieve acceptable performance.

A look at the various cards
Since we have three examples of the GTX 950 and two of the R7 370 on hand, I figured we should take a moment to look at the individual cards. Board makers tend to put a lot of effort into making their products stand out from the crowd these days, and these five cards illustrate how that trend plays out.

GeForce GTX 950 cards from Gigabyte, Asus, and EVGA

These three cards offer tremendous variety in terms of looks, cooling capacity, and physical dimensions. As I’ve noted, the Asus is the most compact member of the group, while the EVGA offers a cooler of gratuitous length and clock speeds to match. The Gigabyte Xtreme Gaming would represent a middle point between the two, except it matches the EVGA’s GPU clocks and throws in a ~100MHz higher memory clock, as well.

Sapphire’s Nitro R7 370 and MSI’s R7 370 Gaming 2G

MSI’s take on the R7 370 is easily the largest card we’ve encountered in this class, with a cooler that looks like it belongs on a much larger, power-hungrier GPU. I really dig the looks of these MSI coolers, though, so why not slap one on Pitcairn?

Sapphire’s Nitro looks understated by comparison, but its dual-fan cooler ought to be more than sufficient for its mission. The Nitro has a potential advantage lurking under that cooler in the form of extra memory capacity—4GB rather than 2GB.

Here’s how the cards stack up.

Gigabyte’s Xtreme Gaming takes the top spot among the GTX 950s, and MSI’s R7 370 leads the Sapphire Nitro. We’re mostly looking at small differences here, especially among the GeForces.

Power consumption
Please note that our “under load” tests aren’t conducted in an absolute peak scenario. Instead, we have the cards running a real game, Crysis 3, in order to show us power draw with a more typical workload.

The fact that the Maxwell-based GTX 950 is more power-efficient than the Pitcairn-based R7 370 probably shouldn’t be a surprise to most folks. We’re talking distinctly different generations of products. Still, the gap between the two cards, in terms of total system power draw when installed, is relatively minor at just 10-15W while running a game. We’ve already seen that the GTX 950 tends to outperform the R7 370, though, so its power efficiency is undoubtedly superior overall.

I suppose it should also come as no surprise that the fastest individual GTX 950 and R7 370 cards are the ones with the highest power draw under load. For instance, Gigabyte’s Xtreme Gaming beats out the EVGA GTX 950 FTW in the benchmarks by an eyelash, and it also requires a few more watts to do its thing.

Noise levels and GPU temperatures

Don’t put too much stock into the differences between the cards at the 30-31 dBA range. That’s near the noise floor in Damage Labs, and those differences are more likely attributable to fluctuations in ambient noise than anything else. In fact, I believe all of the GTX 950 and R7 370 cards simply stop their fans from spinning when idling at the Windows desktop. This semi-passive fan policy is a recent innovation, and everybody now seems to be using it, which is most excellent.

Meanwhile, ye olde GeForce GTX 650 just seems to keep its fan running at a constant speed, which is kind of sad.

These new mid-range cards all fall within a range of a couple of decibels under load, and none of them are particularly loud. Sapphire’s R7 370 is the noisiest of the bunch, thanks to its relatively small cooler and the Radeon’s somewhat higher power draw (and thus heat production). Gigabyte’s take on the GTX 950 registers just a bit higher on the sound level meter than its competitors, almost surely because it’s aggressively tuned to keep the GPU cool, as our peak temperature readings indicate.

In the end, none of the cards run all that hot. The highest temperature we saw was only 69°C, well below the temps in the low 90s that we’ve seen in past products.

I think these results attest to the fact that board makers are serious about producing solid products in this price range. If anything, some of the bigger coolers may be a little excessive—not that I’m complaining.

Conclusions
As usual, we’ll sum up our test results with a couple of value scatter plots. The best values tend toward the upper left corner of each plot, where performance is highest and prices are lowest. We’ve converted our 99th-percentile frame time results into FPS, so that higher is better, in order to make this layout work. These overall numbers are produced using a geometric mean of the results from all of the games tested. The use of a geomean should limit the impact of outliers on the overall score.

Since it’s an early automated benchmark based on a not-yet-released game, I decided to leave the Ashes of the Singularity results out of our overall performance index.

Our game-by-game results were pretty consistent in demonstrating that the GeForce GTX 950 outperforms the Radeon R7 370. The gap is a little larger when we focus on advanced metrics like the 99th-percentile frame time than it is with raw FPS. That’s a testament to the driver optimization work Nvidia has done in recent years in order to ensure smooth gameplay.

If you’re looking for a video card in this price range, the GeForce GTX 950 is an obvious choice. Heck, I’d have a hard time justifying paying more for a GTX 960, given the results on the preceding pages. Not only is the GTX 950’s value proposition strong, but this GPU itself lands in a nice spot by offering fluid gaming at decent image quality levels at 1080p.

That value proposition gets even stronger when you consider the excellence of the board offerings from Asus, Gigabyte, and MSI. Heck, I’m not sure I could pick a favorite among them. If forced to do so, I’d probably single out the Gigabyte Xtreme Gaming card, which looks great and is the fastest overall for only ten bucks more than the Asus. However, the Gigabyte board wants an eight-pin power input. If your PSU will only accept a six-pin input, the Asus Strix GTX 950 is a very safe bet.

As for the Radeon R7 370, this GPU can’t quite keep pace with the GTX 950, and AMD has built in a discount in recognition of that fact. I’m just amazed by how credible an option the R7 370 looks to be given that it’s based on a three-and-and-half-year-old chip. The GPU’s performance is generally decent at 1080p, and both MSI and Sapphire offer cards with reasonable power consumption and noise levels. Frankly, I think the biggest omission from the R7 370 is support for FreeSync-style variable-refresh displays. To see one in action is to want one.

I have to admit: I am a little perplexed about why AMD didn’t give the R7 370 a fully enabled version of Pitcairn with more stream processors and texturing capacity. AMD could have closed the gap in performance with the GTX 950 by shipping that config. Not doing so seems strange to me.

Also, I wish we had time to address the question of whether slapping 4GB of memory onto this class of card, as Sapphire does with the Nitro, offers tangible benefits. As you may have noticed, the 2GB cards we tested handled the games and settings we used quite well, with few major hiccups. We may have to explore the 2GB-versus-4GB issue in a future article with a more challenging set of games and image quality options.

For now, I’d like to end by pointing out something that may be obvious to many of our readers—but not all of them. The games we used in this review tend to be pretty demanding, as these things go. Quite a few popular titles will run even smoother when paired with the GPU power in this class of product. I had hoped to test Dota 2 to illustrate that fact, but the game got “reborn” with a massive update after I’d collected 95% of my results. I chose to exclude it rather than to re-test, but my preliminary results showed both the R7 370 and the GTX 950 maintaining a near-steady 60 FPS throughout our test scenario at 1080p. This was a fairly grueling section of gameplay, as Dota 2 goes. The FPS averages were in excess of 100 FPS for both cards.

Other MOBAs, like League of Legends, are even less demanding than Dota 2.

I suppose what I’m saying is, for most gamers playing the most popular games, this class of GPU hardware offers ample power. More is better, of course, and I love me some fast graphics chips—but I wouldn’t be shy about recommending a GTX 950 or R7 370 to most folks. You really can’t go wrong.

Enjoy our work? Pay what you want to subscribe and support us.