AMD’s Radeon RX 460 graphics card reviewed

Man, AMD doesn’t leave us much new to talk about when we review graphics cards these days. The company already took all the wraps off its Radeon RX 460 graphics card and its Polaris 11 GPU a couple weeks ago. The only open question at the time was how much RX 460 cards would sell for, and we now know that the card carries a $109.99-and-up suggested price. With that sticker, AMD is squarely taking aim at the budget space currently dominated by Nvidia’s GeForce GTX 750 Ti and its ultra-efficient GM107 Maxwell GPU.

The Polaris 11 GPU. Source: AMD

  Base

clock

(MHz)

Boost

clock

(MHz)

ROP

pixels/

clock

Texels

filtered/

clock

(int8/

fp16)

SP

TFLOPs

Stream

pro-

cessors

Memory

path

(bits)

Memory

transfer

rate

(Gbps)

Memory

bandwidth

(GB/s)

Peak

power

draw

R7 260X 1100 16 56/28 2.0 896 128 6.5 96 115W
R9 265 925 32 64/32 1.9 1024 256 5.6 179 150W
RX 460 1090 1200 16 56/28 2.2 896 128 7 112 < 75W
GTX 750 Ti 1020 1085 16 40/40 1.4 640 128 5.4 86 60W
GTX 950 1024 1188 24 48/48 1.8 768 128 6.6 106 90W

For an idea of where this chip fits into AMD’s product line, the version of Polaris 11 on board the RX 460 most closely resembles the Bonaire chip in the R7 260X. It has 896 stream processors enabled (of a possible 1024 on the chip), a 128-bit path to memory, and 56 texture units. The move to 14-nm FinFET production lets AMD clock the chip at 1200MHz boost speeds, though, and it occupies a vague “less than 75W” thermal envelope instead of the 115W board power typical of R7 260X cards. AMD also pairs Polaris 11 with 2GB or 4GB of 7 GT/s GDDR5 memory, up from 6.5 GT/s on the 260X.

Here are some theoretical peak numbers to give an idea of how the RX 460 stacks up with a range of Radeons and GeForces:

  Peak pixel

fill rate

(Gpixels/s)

Peak

bilinear

filtering

int8/fp16

(Gtexels/s)

Peak

rasterization

rate

(Gtris/s)

Peak

shader

arithmetic

rate

(tflops)

Memory

bandwidth

(GB/s)

Radeon R7 260X 18 62/31 2.2 2.2 104
Radeon RX 460 19 67/34 2.4 2.2 112
Radeon RX 470 39 154/77 4.8 4.9 211
GeForce GTX 750 Ti 17 43/43 1.1 1.4 86
GeForce GTX 950 29 57/57 2.4 1.8 106
GeForce GTX 960 38 75/75 2.4 2.4 112

These all may sound like mild bumps, but Polaris 11 inherits all of the same under-the-hood improvements that first showed up on the Radeon RX 480 and RX 470. If you haven’t already read our RX 480 review, you should head over there now to brush up on what AMD has improved with the fourth generation of its GCN architecture. In short, though, the most important bits of Polaris are its delta-color-compression facility for more efficient use of memory bandwidth, as well as a primitive-discard accelerator for more efficient geometry processing. Polaris also offers support for forward-looking standards like DisplayPort 1.3, HDMI 2.0b, and ultra-high-definition content like wide color gamuts and HDR video.

AMD sent over Sapphire’s handsome Nitro Radeon RX 460 card for our tests. This manifestation of Polaris 11 has 4GB of RAM on board, and it lists for $139.99. Sapphire’s board design has a six-pin power connector on board, so this isn’t the card to get for folks dreaming of plugging an RX 460 into a slot and going. (For that, you need this Asus or this MSI card, among others.)

Sapphire bumps the RX 460’s boost clock speed to 1250 MHz, and it rates the board for a 72W total power draw. While that figure might tempt the brave to try and run the card off the PCIe slot regardless, my test system wouldn’t run the card without the six-pin PCIe plug occupied. It’s also interesting that Sapphire only uses a PCIe 3.0 x8 connection for the RX 460. Given the copious bandwidth that bus offers, however, eight lanes of PCIe are probably sufficient for a chip with Polaris 11’s vitals.

Sapphire spruces up the Nitro card with an LED-backlit Nitro logo on the PCB’s back side. The LEDs behind this logo are a light blue (almost sapphire) color, but the yellow tone of the circuit board causes the light to appear greenish from the back side.

Undoing a few screws and flipping over the cooler reveals the Polaris 11 GPU itself. Sapphire uses a high-quality heatsink with a pair of heatpipes and a copper insert in an aluminum base plate. The plastic fan shroud comes off separately from the heatsink itself, making this cooler one of the easier ones to disassemble that I’ve ever had my hands on.

Here’s the Polaris 11 die. It’s tiny! Let’s see how it runs.

 

Our testing methods

As always, we did our best to deliver clean benchmarking numbers. Our test system was configured as follows:

Processor Intel Core i7-6700K
Motherboard ASRock Z170 Extreme7+, BIOS version 3.10
Chipset Intel Z170
Memory size 16GB (2 DIMMs)
Memory type Corsair Vengeance LPX

DDR4 SDRAM at 3200 MT/s

Memory timings 16-18-18-36
Chipset drivers Intel Management Engine 11.0.0.1155

Intel Rapid Storage Technology V 14.5.0.1081

Audio Integrated Z170/Realtek ALC1150

Realtek 6.0.1.7525 drivers

Hard drive OCZ Vector 180 480GB SATA 6Gbps
Power supply Corsair RM850
OS Windows 10 Pro with Anniversary Update

We used the following graphics cards and drivers to represent the GeForces and Radeons chosen for our benchmarks:

  Driver revision GPU base

core clock

(MHz)

GPU boost

clock

(MHz)

Memory

clock

(MHz)

Memory

size

(MB)

Sapphire Radeon RX 460 Radeon Software 16.8.1 beta 1250 1750 4096
Asus Strix Radeon R7 370 Radeon Software 16.8.1 beta 1050 1400 4096
XFX Radeon R7 360 Radeon Software 16.8.1 beta 995 1400 2048
Asus GeForce GTX 950 2G GeForce 368.81 1026 1190 1653 2048
MSI GeForce GTX 750 Ti N750TI-2GD5 GeForce 368.81 1058 1216 1750 2048

This review marks the first time we’ve formally gathered DirectX 12 and Vulkan frame-time data for our Inside the Second testing methods. To make this happen, we’ve set aside Fraps in favor of PresentMon, a utility that monitors an application’s calls to the Present method using a facility called Event Tracing for Windows. PresentMon gives us a nice file full of frame times that we then plug straight into our data-digestion tools.

For the moment, we’re not running a three-frame moving average on this data before passing it into our internal tools, as we have with all of our Fraps frame-time info. PresentMon can be used to capture information from DirectX 11 and OpenGL applications, as well, so we’ve used it to gather all of the data you see on the following pages.

As you can see from the table above, we use custom-cooled, factory-boosted graphics cards in our testing. Whenever you see “GeForce GTX 950” or “Radeon R7 370” in our tests, for example, just remember we’re referring to these hopped-up cards rather than a reference design.

The Asus GTX 950-2G goes commando

No graphics card review is complete without some healthy competition. To represent the GeForce GTX 950, we picked up one of Asus’ GTX 950-2G cards. This is one of the much-ballyhooed GTX 950s that does without a six-pin power connector, and it’s available on Newegg for $134.99 right now.

See, I figured we would be getting a power-plug-less RX 460, as well, given the card’s frugal board power, and I guessed there’d be no better way to pit apples against apples than to get a similarly-provisioned GTX 950. That plan didn’t work out, of course, but we still have a nice GTX 950 to work with for the trouble.

Since this card can’t rely on external power to support a gratuitous factory clock-speed boost, Asus clocks it at near-reference 1026 MHz base and 1190 MHz boost speeds in its “gaming mode” clock profile. An “OC Mode” pushes those figures to 1051 MHz base and 1228 MHz boost speeds. We left the card in its Gaming Mode profile for our tests. Asus’ twin-fan cooler doesn’t spin down its fans at idle, but as you’ll see later on, that’s hardly an issue.

Now that we’ve set the stage, let’s get to testing.

 

Doom (Vulkan)

Let’s kick off our next-generation benchmarking with a look at the Radeon RX 460’s performance in this year’s reimagining of Doom. To test this fast-paced FPS, I mauled some demons while retrieving a yellow security pass in the early stages of the game’s Foundry level. We created a custom blend of low and medium settings to make a playable framerate happen at 1920×1080 for these cards.


Out of the gate, the RX 460 takes a strong lead over the GeForce competition, and it even edges out the somewhat better-provisioned Pitcairn chip in the R7 370. The march of progess is a wonderful thing. The RX 460, R7 370, and GTX 950 cluster together in our latency-sensitive, 99th-percentile frame time result, while the Radeon R7 360 and GTX 750 Ti bring up the rear.


These “time spent beyond X” graphs are meant to show “badness,” those instances where animation may be less than fluid. 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. And 16.7 ms correlates to 60 FPS, that golden mark that we’d like to achieve (or surpass) for each and every frame.

None of the cards we tested spend any time past the 50-ms mark, an excellent result. The GTX 950 also turns in a perfect score at the 33.3-ms threshold. Both the Radeon R7 370 and the RX 460 spend barely any time past this post, either, while the 750 Ti and R7 360 struggle to varying degrees.

At the critical 16.7-ms threshold, though, the RX 460 spends a little over half the time on tough frames that the R7 370 does. Impressively, the pint-sized Polaris spends about a fourth of the time that the GTX 950 does past 16.7 ms, as well. The RX 460 was noticeably smoother than the rest of the cards I tested in Doom with Vulkan enabled. That’s excellent performance from a chip ostensibly meant to replace the Bonaire silicon inside the R7 360. Let’s see if this behavior carries over to the card’s OpenGL performance.

 

Doom (OpenGL)

To gather OpenGL numbers for Doom, we kept our graphics settings identical and flipped over to the OpenGL 4.5 renderer. You can switch between our Vulkan and OpenGL results by clicking the buttons beneath the graphs for easy comparison.




The conventional wisdom among enthusiasts is that Nvidia’s OpenGL driver is superior to AMD’s, and our numbers bear that out. Switching to Doom‘s OpenGL 4.5 renderer puts the GeForce GTX 950 on top of our average FPS measures, but only by about a 10% margin. The GTX 950’s 55-FPS average is practically identical to its performance in Vulkan, though, while the RX 460 loses a whopping 30% of its performance potential from the API change.

The other Radeons are similarly kneecapped by this move, and the R7 360 moves from “playable” to “marginal” thanks to its low average FPS and outsized 99th-percentile frame time. Stick with Doom‘s Vulkan renderer if you own a Radeon. This game wants to move fast, and 30 FPS just doesn’t cut it.



Here’s one instance where our measures of badness really show off their importance. The Radeon R7 360 struggles with frames that would drop frame rates below 30 FPS for over three seconds, compared to about one-tenth of a second with Vulkan enabled. All of our contenders struggle much more with frames that take more than 16.7 ms to render under OpenGL than they do with Vulkan, too. While Doom isn’t unpleasant to play under OpenGL, it’s just that much better with Vulkan enabled, and our numbers prove it.

 

Hitman (DirectX 12)
Hitman is one of AMD’s marquee titles for showing off what DirectX 12 can do on its cards. It’s a punishing title that tends to favor Radeons, so keep that in mind as we discuss our results. To test the game, we set up a blend of low and medium settings with ambient occlusion on. Like the rest of our tests, we ran Hitman at 1920×1080.


So that’s something. Flip on Hitman’s DirectX 12 mode, and the RX 460 takes a commanding lead in our average FPS measure, even beating out the better-provisioned Pitcairn chip on the R7 370 and the GM206 GPU that powers the GTX 950. Not only is the RX 460 the fastest thing going in this game, it also matches that high performance potential with the lowest 99th-percentile frame time of the bunch. In practice, that translates to a noticeably smoother gameplay experience than any other card we tested.


Given the class of card we’re working with, it’s probably best to start our “badness” analysis at the 33.3-ms mark. None of the Radeons spend a single millisecond past this point, so we can be assured they’re always delivering 30 FPS or better during our test run. The GeForces spend a bit of time working on frames past this point, but probably not enough to be significant.

Move to the time-spent-past-16.7-ms mark, though, and the RX 460 really shines. Of our one-minute test run, the flyweight Polaris card spends just four seconds working on frames that drop the average FPS rate below 60. The beefier R7 370 spends almost twice as much time churning on these tough frames, and the GeForces do worse still.

Even accounting for Hitman‘s apparently favorable disposition toward Radeons, this is remarkable performance for such a tiny GPU. Like we already noted, the RX 460 is the smoothest thing here by a wide margin, and it really shows. Let’s see if that performance continues in Hitman‘s DirectX 11 mode.

 

Hitman (DirectX 11)

To test Hitman‘s DirectX 11 performance, we used the same settings as we did for our DirectX 12 tests, just with DirectX 11 enabled instead of DirectX 12. To make comparing the two results easier, we’ve added switcher buttons beneath several of the graphs on this page for easy flipping between the APIs, just as we did with Doom and OpenGL.




Flip over to DX11, and the RX 460 doesn’t lose any performance potential. Its 99th-percentile frame time inches up a bit, though. The GeForce GTX 950 narrows the 99th-percentile frame time gap the RX 460 opened up with DirectX 12, but it can’t close it. The R7 370 maintains the same average FPS under DX11 that it did with DX12, but its 99th-percentile frame time gets a little worse. The R7 360 picks up a couple FPS, but its 99th-percentile time also suffers. The GTX 750 improves its 99th-percentile frame time, but its FPS average remains the same.

In layman’s terms, these numbers show that despite average FPS remaining largely the same between APIs, there are still meaningful changes in the perceived smoothness that a gamer will experience with a GeForce or Radeon running Hitman, all else being equal. Under DX11, these changes mostly favor the GeForces. Under DX12, they favor the Radeons. Simple, ish.

If we can toot our own horns for a moment here, these results show why our Inside the Second methods are vital in teasing out the differences between these cards. Were we only working with average FPS numbers, we might be tempted to say the cards perform similarly between Hitman‘s DirectX 12 and DirectX 11 modes, but our advanced frame-time metrics show us that such a blanket statement just isn’t true. With this data in hand, we can tell you exactly how performance varies for each card between these APIs—something average FPS data alone just can’t do. This stuff is really important, folks. Accept no substitutes.



Speaking of which, our “badness” measures let us further characterize the performance improvements the GeForces get from DirectX 11 and the debuffs the Radeons get under this API. Only the R7 360 and the GTX 750 Ti spend any time past the 33.3-ms mark, but the R7 360’s tally here is so small as to be insignificant.

Flip over to the 16.7-ms threshold, and we find the RX 460 still holds onto its smoothness lead here, even if it’s spending about 100 ms more on tough frames than under DX12. Even though the GTX 950 gets a big boost from the move to DX11, it’s still spending quite a bit more time on tough frames that drop the average frame rate below 60 FPS. The R7 370 has practically the same 99th-percentile frame time as the GTX 950, but it spends much less time churning on these difficult jobs.

 

Rise of the Tomb Raider (DirectX 12)
Rise of the Tomb Raider provides a demanding, beautiful setting for Lara Croft’s latest round of adventures. We ran the game on medium settings at 1920×1080 with some minor tweaks for better visuals than that preset offers.


If Hitman tends to favor Radeons, Rise of the Tomb Raider has shown a similar affinity for GeForces in our past tests. If we turn on DirectX 12, however, the RX 460 very nearly catches the GeForce GTX 950 in our average FPS measures. It also comes within a nose of tying the GTX 950’s 99th-percentile frame time. The Pitcairn-powered R7 370 hangs tight with the RX 460 and GTX 950, too. Not bad for a GPU that first debuted over four years ago.


No card in this test is consistently cracking 60 FPS, so it’s probably most instructive to consider our measures of “badness” at the 33.3-ms mark. The RX 460, R7 370, and GTX 950 all have no trouble maintaining at least 30 FPS over the course of our test. For what it’s worth, the RX 460 also spends the same amount of time working on frames that take 16.7 ms or longer to render as the GTX 950 does, although both of these cards spend about a third of our test run hard at work on those frames. Admirably, none of the cards spend significant amounts of time past the 50-ms mark, either. They’re not delivering face-melting frame rates, but at least they’re fairly consistent.

 

Rise of the Tomb Raider (DirectX 11)

Just as we did with Hitman and Doom, we merely flipped the API setting in RoTR‘s menus to DirectX 11 and repeated our tests.




Switch to DirectX 11 with Rise of the Tomb Raider, and all of our tested cards save the R7 360 get a small drop in 99th-percentile frame times, despite the outwardly similar frame rates between the APIs. The difference in 99th-percentile times isn’t huge, to be sure, but it does suggest that RoTR‘s DX11 rendering path delivers slightly smoother gaming performance in practice.



Once again, we’ll start our examination of “badness” using the 33.3-ms mark as our threshold. The GTX 950 turns in a near-sterling performance here, but the RX 460 and R7 370 aren’t far behind. Even the somewhat laggardly GTX 750 Ti doesn’t struggle much with difficult frames that take longer than 33.3 ms to render. Only the R7 360 spends a significant amount of time churning on these frames.

At least with these graphics cards and the settings we picked, Rise of the Tomb Raider seems to run smoothest under DirectX 11. We’ll have to explore whether that trend holds with more demanding settings on more powerful graphics cards at some point.

 

Grand Theft Auto V

Now that we’ve examined most of the titles that take advantage of next-gen graphics APIs on the market today, let’s go back to a good old DirectX 11 title. Grand Theft Auto V generally runs well on a wide range of hardware. We used the same settings for this test that we chose for our GTX 950 review, and we ran the game at 1920×1080, just as we did with the rest of our test suite.


Here, the GTX 950 stretches its legs, turning in the best average frame rate and 99th-percentile frame time of all the cards tested. The R7 370 and R7 360 both manage to get ahead of the RX 460 in GTA V, as well. All of the cards except for the GTX 750 Ti come quite close to delivering a solid 60 FPS thoughout our test period, though. The GTX 950’s 99th-percentile frame time even suggests near-perfect performance in that regard.


None of the cards spend any time past the 50-ms or 33.3-ms marks, so we can focus on their “badness” with regard to the critical 16.7-ms threshold. The GTX 950, R7 370, and R7 360 all have imperceptible blips past this point, but the RX 460’s record is marred a tiny bit by a quarter-second of time spent rendering frames that took longer than 16.7 ms to finish. Only the GTX 750 Ti’s result translated into issues I could feel, though. That card felt noticeably hitchier in practice than its competition.

 

Crysis 3

Here’s another DirectX 11 favorite of ours. Despite its age, Crysis 3 can still put the hurt on modern graphics cards. For this test, we adjusted the game’s settings to their medium presets and hopped into our nanosuits.


Here’s another game where the old guard takes the trophy from the young gun. The R7 370 and the GTX 950 open up a 17% lead on the RX 460, though the cards’ 99th-percentile frame times suggest there’s more to that number than meets the eye. The RX 460’s average FPS number suggests it has enough performance potential to hit that magical 60-FPS figure, even if its 99th-percentile frame time also warrants further investigation.


None of the cards we tested spend appreciable time past the 50-ms or 33-ms mark, so let’s look at that magical 16.7-ms threshold. The R7 370 and GTX 950 each spend between one and two seconds of the one-minute test period working on challenging frames—not perfect, but nothing to worry much about. The RX 460, however, spends about six seconds in the churn zone. That result might translate to a less smooth gaming experience than the card’s near-60-FPS average might suggest.

 

The Witcher 3

We’ll close out our purely DX11 test suite today by checking in with Geralt of Rivia. We ran The Witcher 3 at 1920×1080 on its High preset.


While that High preset may have been a little too ambitious for our contenders, it does put the RX 460 and the GTX 950 neck-and-neck. The R7 370 stays in the mix, as well. The GTX 750 Ti and the R7 360 can charitably be described as offering console-class performance at these settings. None of the cards produced especially impressive 99th-percentile frame times in The Witcher 3, suggesting the spikes we see in the frame-time plot above are having a significant effect on perceived smoothness during gameplay. Let’s dig further.


Our top three performers all spend a small amount of time past the 33.3-ms threshold, but not enough to be worrisome. The GTX 750 Ti is quite overwhelmed by these settings, though. It spends two seconds past the 33.3-ms mark, enough to translate into a noticeably hitchy gameplay experience. I’d much rather play The Witcher 3 on the Radeon RX 460, given the choice.

 

Power consumption

These power consumption numbers aren’t peak results. Instead, we run a real-world workload—Crysis 3—on our graphics cards and take measurements while that game is running.


At idle, the RX 460’s contribution to our system power draw is no worse than the GeForce competition. It’s also competitive with the 28-nm GeForces in our test stable. The real story here is that the RX 460 is delivering basically the same class of performance as the 28-nm Pitcairn chip in the R7 370 while allowing our system to consume almost 60W less power under load. That’s an impressive increase in performance-per-watt.

Noise levels

To test each card’s noise levels, we set up an iPhone 6S Plus running the Faber Acoustical SoundMeter app. The meter was placed 14″ from the graphics card on our test bench. The noise floor in our testing environment is 30 dBA with our test system running.

Since the Radeon R7 370 and RX 460 can both shut their fans off at idle, they’re no louder than the noise floor in our test environment. The Asus GTX 950 keeps its fans running at all times, but its 31-dBA result will be extremely hard to notice in the average home.

At load, the Sapphire cooler on the RX 460 is somewhat tonal in its noise character, but it’s quiet enough that it almost certainly won’t be bothersome in a case. The Strix cooler on the Asus R7 370 is superlative—its noise has no distinct character, and it’s barely discernable over the rest of our test system. The cooler on the Asus GTX 950 is also quite good—it barely moves the needle on our sound meter app under load.

GPU temperatures

While the RX 460 may get a bit warm, it has no problems sustaining the 1250-MHz boost clock that Sapphire dials in. That temperature isn’t accompanied by bothersome heat or noise, so between that and the card’s clock stability, we have no complaints.

 

Conclusions

Before we dive deep into our concluding thoughts on the Radeon RX 460, let’s have a gander at six of our infamous value scatter plots. Since we now have DirectX 11, DirectX 12, and Vulkan data living together under the same roof, we needed to present that information in a way that’s fair, and we settled on three charts each for average-FPS-per-dollar and 99th-percentile-FPS-per-dollar data.


The first set of charts accounts for our average-FPS-per-dollar data. The first chart shows performance with current-gen graphics APIs only, i.e. OpenGL and DirectX 11. The second shows next-generation API numbers blended with the three DirectX 11-exclusive titles in our test suite. The final chart is a “best-API” graph that lets each card take its best result from each API we tested, also blended with results from the three DirectX 11-exclusive games in our suite.

Going by performance with today’s APIs, the Radeon RX 460 has just about as much performance potential as the Asus GTX 950 we tested, and its slightly higher price tag gets buyers 4GB of RAM. That’s a pretty good deal for $140. Look at next-gen API performance in isolation, and the RX 460 narrowly edges out the GTX 950—but so does the Radeon R7 370, a Pitcairn-powered card that also comes with 4GB of RAM.

Surprisingly, the four-year-old Pitcairn is the price-to-performance champion in our final “best APIs” measure of potential performance with average FPS. The excellent Asus R7 370 I tested is going for $139.99 on Newegg right now, and a $30 rebate card brings that price down to $109.99. Pitcairn has a lot missing, though, like all of Polaris’ next-gen video and display support mojo. It also won’t run 4K or FreeSync displays. Still, gamers who don’t care about those things might still find Pitcairn a decent bargain. We would look to the future and nab an RX 460 anyway.


What about our latency-sensitive 99th-percentile-FPS-per-dollar metric? The GTX 950 is our smoothness champion for DX11 and OpenGL, while the GTX 950, the R7 370, and RX 460 all tie in our next-gen-APIs-only chart. Throw the best results from each API in a blender, and the GTX 950 comes out on top. Still, it’s pretty neat that AMD has narrowed what has long been a large gap —and an expanded lineup of DirectX 12 titles could eventually swing the pendulum in the RX 460’s favor. The fact that 2GB RX 460s exist for $110 to $120 can’t be ignored, either. A theoretical RX 460 2GB dot in the chart above is in a very appealing spot.

With the arrival of the RX 460, AMD has laid all of its 14-nm cards on the table for now. Truthfully, this is the next-gen Radeon that surprises and impresses me the most. We all lust after the shiny Ferrari behind the velvet rope, but most people get around in Honda Civics. With the RX 460, AMD has a Civic-like card that offers much better performance across the board than the evergreen GTX 750 Ti, both in average FPS and our advanced metrics. It also lays waste to the Bonaire-powered R7 360 that it replaces in the red team’s model lineup.

All together, Polaris’ underhood improvements and inherent smoothness come together to offer a quality gaming experience that used to cost a lot more than AMD’s $109.99 suggested price. I say “used to” because Nvidia apparently saw this baby bruiser coming, so it cooked up an aggressive set of rebate offers for the GTX 950 in response. The Asus GTX 950 we got our hands on for testing goes for just $105 after rebate on Newegg right now, and TR readers pointed out GTX 950s going for as little as $99.99 after rebate this week. Rebates are risky things at the best of times, but we might take a gamble for such juicy deals.

With competitively-priced GTX 950s elbowing in on its turf, the RX 460’s value proposition will depend on whether you play lots of DirectX 11 games now (like we imagine most people do) and you don’t see anything appealing on the DX12 horizon. If that’s the case, a cheap GTX 950 may be most appealing for its low frame times and quiet running. Those cards’ 2GB of RAM is already starting to feel claustrophobic to us in a lot of titles, though, while the RX 460 is one of the few cards in its price class that offers a 4GB option.

If your Steam library already contains next-gen titles like Doom and Hitman that take advantage of Vulkan and DirectX 12, though, the RX 460 can deliver surprising performance at 1920×1080. Doom especially feels like it’s running on a much more expensive graphics card when paired with the RX 460. Other DX12 titles, like Rise of the Tomb Raider, actually have a detrimental effect on the card’s performance. Still, the RX 460 offers plenty smooth gameplay in DX11 titles as a fallback case, even if its frame rates aren’t always best-in-class. Heck, even if the only next-generation game you play on this card is Doom, you’ll have lots and lots of fun for the money.

The most intriguing place we may yet see Polaris 11 is mobile devices. AMD has long indicated it’ll be working hard to get these chips into entry-level gaming notebooks, and if the RX 460 is any indication, we may actually end up with compelling yet affordable laptops that can deliver a smooth, solid gaming experience. If that’s the case, Polaris looks like it’ll continue shining brightly for AMD. We’re just curious what, if anything, Nvidia has in store to steal AMD’s spotlight at this price point. If the brisk pace of GPU releases from the green team this year are any indication, we won’t have long to find out.

Like this review? Support us by subscribing today or whitelisting The Tech Report in your ad blocker. We rely on your support to maintain our independence and to continue performing the dozens of hours of in-depth testing reviews like this one require. Thanks for pitching in.

Comments closed
    • OneShotOneKill
    • 3 years ago

    So… what is AMD going to do to compete with this generation’s cards?

    • MileageMayVary
    • 3 years ago

    Something I would like to see in these entry level comparisons is the inclusion of the CPU/APU’s integrated graphics since that is what you would be upgrading from.
    As PBCrunch pointed out, if you have a top end CPU I doubt you’d consider anything below a RX480 or GTX1060 but it might still be a little indicative.

      • flip-mode
      • 3 years ago

      I could see doing it for a single game just for laughs, but it would be telling you what you already know – IGPs are vastly inferior to cards like these.

    • ingerulol
    • 3 years ago

    Courtesy of Techreport.com

    12:07 AM on March 22, 2013
    Radeon HD 7790 1000 896 56/28 16 6.0 GT/s 128 85W
    12:22 AM on October 8, 2013
    Radeon R7 260X 1100 896 56/28 16 6.5 GT/s 128 115W
    2:49 PM on August 11, 2016
    RX 460 1090 1200 896 56/28 16 7 128< 75W

    • jokinin
    • 3 years ago

    Pardon my ignorance, but the R7 370 is sporting the same Pitcairn chipset as my HD7870? The performance would be about the same then?

      • Concupiscence
      • 3 years ago

      The 7870 is running Pitcairn XT. The Radeon 7850s run Pitcairn Pro, which is what the modern Radeon R7 370’s running (albeit at a lower 110W TDP). As far as I know your 7870 will be at least a little faster. I’d guess the manufacturing process is mature enough that the 370s hit decent turbo clocks at lower voltage, though.

        • willmore
        • 3 years ago

        So closer to my HD7850 OC to 1GHz?

          • Concupiscence
          • 3 years ago

          My [url=http://www.newegg.com/Product/Product.aspx?Item=N82E16814202152<]Sapphire R7 370[/url<] runs at a core clock of 985 MHz, so that's very, very close.

            • willmore
            • 3 years ago

            That’s what I was guessing. Looks like a 460 would not be an upgrade for me. That leaves the 480 or 470. Also, there’s the issue of availability…

            • CScottG
            • 3 years ago

            [url<]https://jet.com/product/SAPPHIRE-Radeon-RX-470-100407-4GOCL-4GB-256-Bit-GDDR5-PCI-Express-30-x16-HDCP-Re/fa93c89bd188406c8e64bbbd67077f2e[/url<] Didn't see anything about stock. -with their 1st time customer "coupon-code" it's a pretty good deal (..for about $30 off). Same with this one (..that looks better to me): [url<]https://jet.com/product/ASUS-ROG-STRIX-Radeon-RX-470-4GB-OC-Edition-AMD-Gaming-Graphics-Card-with-DP-14-/9f3d205624bf460e8329e8396eeb4f44[/url<]

    • DragonDaddyBear
    • 3 years ago

    As always, the review was thorough and worth the wait.

    I was talking to someone last night and I had a thought. This GPU is very likely to end up in an old system, the kind where we give it to a child or family member. Think Core i 2XXX, a C2D/Q or maybe a Phenom that was a factory-built system. I’m certain it would hold back the frames, but by how much? And how would it compare to the green team in the same system? I think it would be an interesting comparison for TR to check out, especially when NV releases their competitor to this.

    • Chrispy_
    • 3 years ago

    So it struggles to compete on every front with a cheaper GTX950 and will become completely pointless as soon as Nvidia releases GP108 which (I predict) will be a 75W card to match the GTX960 for about $149

    • Unknown-Error
    • 3 years ago

    The power consumption, GPU temps for the 14 nm process is just baffling. I guess this is not AMD’s fault but as usual, a Global Flounderies issue.

      • Airmantharp
      • 3 years ago

      Are you worried about temps, or watts?

      Temperature measured at one point is not a good indicator for how much power something is consuming and thus radiating as heat.

      • BaronMatrix
      • 3 years ago

      GCN 4 has a nearly CPU front end to use C++ pointers and pre-emption and cache-coherency… nVidia used Tile rendering which is more efficient… AMD built Polaris 11 for mid-range APUs and mobile…
      There is a pattern of newer games running much better on Polaris and we know Vulkan is worth 20% free perf…
      A year from now most games will have DX12 and Vulkan as Unreal has support for both now…

        • PBCrunch
        • 3 years ago

        Vulkan gives nothing for free. Developers have to work hard to get those gains. Three games run on Vulkan, but AMD advocates only want to talk about Doom. Talos Principle and Dota 2 currently give no performance benefit in Vulkan.

        Even in Doom on AMD graphics chips, Vulkan performance is only good when a very high end CPU is used.

        Vulkan performance is NOT FREE. It takes developer attention and prodigious CPU power to make it happen.

        [url<]http://www.hardwareunboxed.com/gtx-1060-vs-rx-480-in-6-year-old-amd-and-intel-computers/[/url<]

    • NimaV
    • 3 years ago

    So an overclocked 4GB 140$ 460 can’t even beat a stock slot powered 2GB 950. I didn’t expect that.

      • flip-mode
      • 3 years ago

      AMD sent what is probably the most expensive RX 460 to TR for testing. It is probably also one of the fastest RX 460s because of the clock bump, but only marginally. I doubt the difference between 2 GB and 4 GB amounts to anything at 1080p. So the unfortunate thing is that if AMD had sent over one of the stock-clocked $110 or $120 cards, the RX 460 would look a lot better than it does. It would not beat a GTX 950, but it honestly does not have to beat it, it just has to provide essentially the same experience.

        • NimaV
        • 3 years ago

        2 GB is not enough even in 1080p in all those games in this benchmark except Crysis 3 so it’s really surprising to me a 2 GB GTX 950 doing so good against 4 GB cards. and can you explain it to me why a brand new GPU based on New 14 nm architecture does not have to beat a card based on a 2 year old 28 nm Maxwell architecture!?

          • flip-mode
          • 3 years ago

          [quote<]2 GB is not enough even in 1080p in all those games in this benchmark except Crysis 3 [/quote<] Evidence for that? Here's evidence that says having more than 2 GB rarely matters: [url<]http://www.gamersnexus.net/guides/1888-evga-supersc-4gb-960-benchmark-vs-2gb/Page-2[/url<] [quote<]can you explain why a brand new GPU architecture does not have to beat a 2 year architecture!?[/quote<] Why do you think it has to? Yes, I can explain why is does not have to. A new product has to either beat the old product on performance or beat it on price - it does not have to do both, even if it's great when it does. And really it would not matter if it beat the GTX 950 by some token amount of 5% - 10%. It wouldn't matter; it wouldn't change the game play experience. If it beats it by more than 10% then hooray, but AMD probably would have priced it higher too. GTX 950 launched 11 months ago at $170. RX 460 launches right now at $120. The RX 460 is bringing GTX 950-class performance to a lower price point.

            • NimaV
            • 3 years ago

            I have a 2 GB card and I know 2 GB Vram is not enough in newer games from my own experience. the benchmark you linked is more than 1.5 year old and even then it shows huge difference in some of those games (ACU, BFHL) at 1080p.

            MSRP of GTX 950 was 160$ at launch and it was one year ago, the card in this review is available for 105$ after rebate and even this stock clocked, slot powered, 2GB, 105$ GTX 950 is faster than overclocked 4 GB 140$ RX 460 and you have to consider that GTX 950 has much bigger overclocking headroom.

            I think this card is overpriced and can’t compete even with GTX 950 so I don’t know how it can compete with upcoming pascal replacement of GTX 950.

            • flip-mode
            • 3 years ago

            Neither one of us is wrong here. For the 2 GB <–> 4 GB thing, the best advice is to do as you see fit for yourself. Gamers come in all shapes and sizes. For some people 2 GB won’t be enough, for other people 2 GB will be plenty. I posted benchmarks to actually put some meat on the table as opposed to the two of us throwing opinions at each other.

            The situation is always changing in terms of both price and competition. Prices will change over time. RX 460 prices will drop. GTX 950 prices depend on rebates, which some people find highly distasteful. The GTX 950 won’t be around forever either. But, if RX 460 has trouble competing with the GTX 950 then a GTX 1050 could only make the matter much worse, unless Nvidia prices it higher. Until then, the RX 460 is certainly in the same league as the GTX 950; for people that prefer Radeons and want something more current than Pitcairn, that’s a very good thing.

          • DrDominodog51
          • 3 years ago

          No. 2 GB isn’t enough to run all of these games at 1080×1920 [b<]on max settings[/b<]. For medium to low settings, 2 GB is enough.

            • Meadows
            • 3 years ago

            Why do you play games in portrait orientation?

            • DrDominodog51
            • 3 years ago

            Why not?

    • anotherengineer
    • 3 years ago

    Interesting, the die is not plain black like the RX 480 & 470.

      • willmore
      • 3 years ago

      Are you making a joke? It should be an almost perfect mirror. So, if it’s white or black depends on what it’s reflecting. Guess they used a white background for this die shot.

        • anotherengineer
        • 3 years ago

        Not what I mean at all. The die of the 470 and 480 there is nothing, however the 460 die has the amd logo etched into it, which you typically see from a TSMC chip.

        It’s that discrepancy I find odd between the 470/480 and 460.

          • willmore
          • 3 years ago

          Ahh, that makes way more sense. I was thinking “this guy normally knows his stuff, did he just take a dumb pill today for some reason?” Glad to know it was my misunderstanding and that you didn’t have a stroke or something.

          Now, I want to know the answer, too.

          • ronch
          • 3 years ago

          Nah, AMD just wants to be known for cheaper parts.

        • ronch
        • 3 years ago

        And flash will almost certainly make the die look black if it’s not aimed directly perpendicular to the die.

      • slaimus
      • 3 years ago

      It also says Made in Taiwan even though it is not made by TSMC–maybe just packaged there.

        • Meadows
        • 3 years ago

        Do we know for certain it’s not made there?

    • djayjp
    • 3 years ago

    Wow amazing job on the API and value front, TR! I couldn’t have asked for anything better 🙂

    Though could include the RX 470 for bonus points 😉

    • maroon1
    • 3 years ago

    LOL 28nm nvidia is able to compete with AMD 14nm in performance per watt

    • SlappedSilly
    • 3 years ago

    In the final graph ’99th percentile FPS per dollar – “best” api’ the R370 is placed at a different price point than it is the the other API groups.

      • Jeff Kampman
      • 3 years ago

      Oops. Fixed.

    • CScottG
    • 3 years ago

    I’d feel sorry for someone who bought this for gaming performance.

    When searching for video card deals:

    For about 15% less you can get a 950.

    For about 15% more you can get a 960.

    (..2 months ago you could have gotten a R9 380 for only 10% more).

    Of course ALL of the above is old architecture and process.

    IMO the ONLY thing truly worthy about these cards are their outputs – which this review devoted all of one sentence to.

    Yeah, I know:

    AMD (check)
    Negative comment (check)

    Let the down-votes commence!

    Edit:
    Good video with good comparisons only lacking 960. Please note the statements made at the beginning of the video..

    [url<]https://www.youtube.com/watch?v=-RtTzsUhwxY[/url<] also, [url<]https://www.youtube.com/watch?v=_HfizETJtsY[/url<]

      • SlappedSilly
      • 3 years ago

      Down-vote you for down-vote bait.

      … wait, did I fall for something here?

        • DrDominodog51
        • 3 years ago

        Nah.

      • anotherengineer
      • 3 years ago

      Depends what card you have, and depends where you live. Be a good upgrade from a 4850. Don’t really know how it would compare to my 6850.

      On CND newegg, it seems the vanilla 2GB RX 460 is typically $40 ish cheaper than the 2GB vanilla GTX 950.

      So like everything, it’s all relative.

        • CScottG
        • 3 years ago

        I would still feel sorry for someone who bought the card reviewed for $140 US (and what ever added taxes/fees that would accompany it).

        (..it’s not like I’d feel *less* sorry if they payed MORE for the damned thing.)

        The fact is – it just doesn’t perform that well.

        $60 more for a 470 (at even today’s inflated prices) seems like a bargain by comparison, and its level of performance is decent and the value (in spite of the price hike and vapor-ware 4GB 480 at $200) is also decent. I’d also look at the used market before this card. If the prospective purchaser couldn’t swing either option, I’d recommend waiting until they could.

        -again though, this is all with respect to gaming performance.

        (..I’d happily recommend the 2GB 460 for HTPC duty at $110 or less, or for someone who just wants a UHD display for typical desktop duty.)

        RX 470 performance:

        [url<]https://www.youtube.com/watch?v=CNNRGH3gMvg[/url<]

        • CScottG
        • 3 years ago

        [url<]http://www.newegg.ca/Product/Product.aspx?Item=N82E16814126090[/url<]

      • flip-mode
      • 3 years ago

      There’s more to the picture than you say, though.

      First of all, brand new products always have a price premium.

      The GTX 950 has been out for a while and of course there are excellent deals out there for that card.

      The RX 460 being a newborn, prices are as high as they’re ever going to be – it’s all down hill from here.

      Secondly, the MSRP for the RX 460 is apparently in the $110 to $120 range (2 GB model).

      The card tested here is a primo version of the RX 460, and priced much higher than MSRP, and comes with 4 GB of memory.

      Prices will fall from there and 12 months from now you’ll be able to get it for $80 to $100.

      Off topic: are we putting all the empty lines between sentences for a reason? to make the post look more substantial or something?

        • CScottG
        • 3 years ago

        I’m specifically looking at the model reviewed at the price reviewed and compared against current competitors (regardless of this particular RX 460’s current pricing – which sadly may not be inflated).

        Down-the-line? Sure, pricing might drop enough to where it’s competitive as far as value goes.

        Still, it’s not like it’s ever going to get better in performance – and in that respect I still wouldn’t recommend it (..unless it was dirt cheap and crossfire worked wonders with several cards and most games). To be fair though, I wouldn’t recommend the 950 either (..in this context). In fact about 2 months ago I argued against the 950 in favor of spending about $25 more for the R9 380 – just a better performer overall and a better value (both now and in the long term when compared to the 950).

        Off topic: I format for clarity – white space is important.

    • bjm
    • 3 years ago

    That penny is dirty.

    • Shobai
    • 3 years ago

    Jeff, your Doom [Vulkan] and Crysis 3 Frame Number charts are pretty much spot on; same scales between tabs, DUT on both tabs, all data within the chart space – thanks for working on this. The other Frame Number charts in the review would benefit from a bit of tweaking to bring them up to snuff – it really does improve your presentation.

    I note that, in the second last picture on the first page, there doesn’t appear to be a thermal pad between the black heatspreader over the VRMs and the flat section of the heatsink that would seem to mate with the heatspreader. Could you give us an idea of what the VRM temperatures were like during testing?

    • USAFTW
    • 3 years ago

    Excellent article Jeff, as always.
    The RX460 does a lot better in these tests than I have seen other review sites, especially from a po2er efficiency standpoint. Looks to me any board vendor, except Sapphire, may have turned up their power limits to get consistent boost clocks/higher clockspeeds, hence 100 W+ power draw for those.
    Not that I would actually buy a card of this tier, but why wouldn’t they release a fully enabled variant? Are the yields really that bad?

      • RAGEPRO
      • 3 years ago

      As other commentors have opined, it’s possible the fully-enable dies are harvested for a customer paying top dollar. Like, say, Apple, who would love a 14nm Radeon in the new Macbooks.

    • PBCrunch
    • 3 years ago

    I posted this in the GTX 1070 review thread, but long after the traffic on the thread had dried up. The same question applies here.

    Are frames per second (FPS) and 99th percentile (99) figures from individual games normalized when the FPS per dollar and 99 per dollar graphs are generated?

    It seems like these individual FPS and 99 values should be normalized before the averages are calculated, but I didn’t see any mention of this in the article.

    If individual game FPS and 99 values are used in their raw form, a single high FPS benchmark with a large raw difference between two cards could easily wash out a number of smaller differences in titles with lower FPS.

    Example:
    Let’s consider a benchmark suite of five hypothetical titles. Three of these titles are very punishing on the graphics cards being reviewed. One of them is older and less demanding. The last title is a super-high FPS competitive shooter.

    Demanding Strategy Title (DST):
    Card A1: 33.2 FPS
    Card A2: 34.1 FPS
    Card N1: 31.8 FPS
    Card N2: 33.1 FPS

    Demanding Open World Title (DOWT):
    Card A1: 35.2 FPS
    Card A2: 36.1 FPS
    Card N1: 33.8 FPSCard
    Card N2: 35.1 FPS

    Demanding Third Person Title (DTPT):
    Card A1: 34.2 FPS
    Card A2: 35.1 FPS
    Card N1: 32.8 FPS
    Card N2: 34.1 FPS

    Older Title (OT):
    Card A1: 64.2 FPS
    Card A2: 65.1 FPS
    Card N1: 62.8 FPS
    Card N2: 64.1 FPS

    High Framerate Shooter (HFS):
    Card A1: 219.5 FPS
    Card A2: 225.3 FPS
    Card N1: 251.8 FPS
    Card N2: 264.7 FPS

    If we consider these values, card A2 consistently outperforms the Nx cards in the first four titles by differences that are small in magnitude. The Nx cards outperform the Ax cards by a large margin in the last title, a competitive online shooter that delivers 30 FPS on a potato.

    To my eye, this means the card A2 is the best performer; it delivers superior performance in a wider range of (completely made up) titles. Furthermore, card A2 performs better in the titles where it really matters (the demanding ones).

    If you average the raw FPS figures, the both Nx cards come out ahead of card A2. The results in the last test, which are probably completely imperceptible, completely overrun the results of all other tests. By this metric, the Nx cards beat both Ax cards in average FPS, despite the fact that A2 takes the crown in four of five titles.

    Raw averages:
    A1: 77.3 FPS
    A2: 79.1 FPS
    N1: 82.6 FPS
    N2: 86.2 FPS

    However, if the average FPS for each title is normalized to between 0 and 1 (* formula below), where 0 represents the lowest FPS score for any card in that title, and 1 represents the highest FPS of any card in the same title we get results that are more representative of the true pecking order of these completely made up graphics cards:

    Normalized averages:
    A1: 0.49
    A2: 0.83
    N1: 0.14
    N2: 0.65

    When the data is normalized, card A2 is clearly the top dog, and card N1 is clearly at the bottom of the pack. Cards A1 and N2 are pretty close in the middle, which is a pretty accurate representation of their relative performance.

    Maybe you guys already perform this normalization. If so, it doesn’t seem to be mentioned anywhere. Normalization is a common early step in scientific comparisons, particularly in life sciences.

    (*) Normalization formula: (x_i – x_min) / (x_max – x_min)

      • Jeff Kampman
      • 3 years ago

      We use a geometric mean of our average-FPS numbers and converted 99th-percentile frame times to create the final indices of price-to-performance that get presented in our scatter plots. We don’t perform the kind of normalization you present, but the geometric mean should prevent any wild outliers from affecting the final results too much. Hope that helps.

    • gerryg
    • 3 years ago

    Jeff, you benched against R7 360/370’s but those don’t even show up on your specs tables in the introduction page showing “where it fits” and “how it stacks up”. What’s up with that?

    I noticed because both tables have an R7 260X, which I have in one PC, so I thought Oooh, can’t wait to see the RX 460 go up against my card in the benchmarks! But no luck, big sad face. 🙁

    Now I have to go back into the TR archives and pull up several other reviews in order to cross-reference specs and numbers in order to take a best-WAG comparison. I like to re-read all those TR classics at bedtime, anyway, so there’s a silver lining…

      • Jeff Kampman
      • 3 years ago

      The R7 360 is a slightly cut-down version of the Bonaire chip in the R7 260X. Your R7 260X will perform somewhere between the R7 360 and the RX 460 in our tests.

        • gerryg
        • 3 years ago

        Ah, there’s the connection, thank you very much!

        • Concupiscence
        • 3 years ago

        Jeff’s on the money. The 260X has 896 SPUs versus the 360’s 768, has 56 texturing units vs the 360’s 48, and operates at a very slightly lower clockspeed while consuming 15% more power on average. The 260X will be around 10% faster than the 360, and the RX 460 will cheerfully walk away from both of them.

      • Amien
      • 3 years ago

      I’m in the same boat. Read the 7790 and 260x reviews before reading this one. The 360 should be a bit slower from what I can tell…

      Edit: Just read Jeff’s comment. Thanks Jeff! Please get some rest ^^

    • Concupiscence
    • 3 years ago

    I almost regret snagging a 4 gig R7 370 a few months ago, but for my use case – hooked to a 1080p TV via HDMI, with no plans for 4K or need for H.265 or VP9 acceleration – it won’t make much of a difference. Great review, guys. I won’t hesitate to recommend an RX 460 if anyone I know needs a cheap & cheerful upgrade.

    • DrDominodog51
    • 3 years ago

    I find it interesting that the Radeon cards are buttery smooth in GTA here, but struggle in the 470 review. Looking at the settings used in the 470 review, it must be the MSAA and tessellation as always.

      • Jeff Kampman
      • 3 years ago

      There is a problem with our RX 470 numbers that I’m writing about at this very moment. We’ll be updating that review and publishing an article about what happened soon.

        • DrDominodog51
        • 3 years ago

        Oh… that latency issue affected the 470 review as well…

    • PBCrunch
    • 3 years ago

    RX 460 looks good as presented in this article, but every review I have seen for this card has the same flaws.

    $100ish graphics cards are seldom paired with high end i7 CPUs. Every RX 460 review I have seen uses an (usually overclocked) i7-6700K or an i7-5820K. The real domain of graphics chips like these is upgrading OEM systems with Haswell Pentiums or AMD APUs. AMD GCN cards tend to require lots of CPU power to make up for driver overhead. So far, no RX 460 review has attempted to investigate this.

    This CPU requirement is particularly evident in Vulkan Doom benchmarking performed at Hardware Unboxed. When RX 480 and GTX 1060 are compared on a Skylake i7, the RX 480 comes out ahead; when the same cards are paired with i5-750 or Phenom X4 955, the GTX 1060 is substantially faster than RX 480.

    Every review seems to fall all over itself to test using the latest and greatest $60 AAA game titles programmed using next-generation APIs, even though the majority of people buying these cards will likely play older discounted/F2P OGL/DX11 titles.

    League of Legends
    Dota 2
    Team Fortress 2
    CS:GO
    Rocket League
    Overwatch
    GTA V
    Skyrim

    How many of those older (but VERY popular) games have Vulkan/DX12 versions that outperform the older API version? Dota 2 has a Vulkan version, but the standard OGL version performs better.

    Real world pricing is not compared. GTX 950 is available all the time for $99 after rebate. The RX 460 reviewed here costs $140. Even in its best game (Vulkan Doom), the RX 460 does not deliver 40% better performance than GTX 950. Nvidia is giving away free games with GTX 950 and 960 now, as well.

    This review has the unique flaw of comparing a gimped bus-powered GTX 950 against a super-primo factory OC 4GB RX 460. Normal 6-pin GTX 950 generally outperforms bus-powered GTX 950 by 10% or more.

    In a budget system, playing the most popular games, paying the current street prices for these two cards, there is no doubt in my mind that GTX 950 offers an overwhelmingly superior price to performance ratio.

      • gerryg
      • 3 years ago

      I can understand keeping CPU (i7) and other things in the testbed the same for economic and apples/apples reasons, so doesn’t bother me. But I do agree that testing against AMD’s “target market” games would be worthwhile, e.g LoL, Dota 2, CSGO, Overwatch. Maybe not all of them need to be tested, but a couple would be nice.

        • DrDominodog51
        • 3 years ago

        Overwatch I can see, but TF2, CS:GO, DOTA, and League can be run well on pretty much any modern ~$100 GPU.

          • gerryg
          • 3 years ago

          Even at 4k?

            • DrDominodog51
            • 3 years ago

            At 4k, you wouldn’t be using this card in the first place because you could afford a far superior GPU.

            • chuckula
            • 3 years ago

            That’s not necessarily true.
            After buying that 4K screen you might be so broke that this is all you can afford.
            😉

            • gerryg
            • 3 years ago

            4k monitors are starting at just under $300 USD (FreeSync ones start at $400), which isn’t bad. I’ve head there is a bring-your-own PC eSports place that is upgrading half of their monitors to 4k by end of year. Portable desktop with this class of card is perfectly reasonable to evaluate. All I’m saying is would be good to have some data to prove good/not good and where the grey areas are. Anyway, I especially would only buy “just enough” for such a PC, in case it was stolen or damaged then I would be able to replace sooner, and a tiny card/low power makes the desktop a lot smaller and more portable, which is appealing if it could handle 4k. For home use I wouldn’t even consider anything slower than RX480.

            • RAGEPRO
            • 3 years ago

            Not necessarily! I know some folks who are casual gamers that play windowed, or in 1080p, on their 4K monitors that they have for productivity reasons.

            • DrDominodog51
            • 3 years ago

            It doesn’t matter then because you aren’t playing at 4k and this card would be able to handle it.

            • RAGEPRO
            • 3 years ago

            That’s true. I was really just pointing out that it’s very possible people could pair this board with a 4K monitor and then game on it. 🙂

            Also, I know that a hot-clocked GTX 950 can run TF2 in 4K at near-60 FPS. It seems likely this card will too.

            • I.S.T.
            • 3 years ago

            I’m legitimately surprised it can’t run TF2 at higher than 4K at locked 120. I heard it got an engine update recently, could that be playing a role?

            • Krogoth
            • 3 years ago

            TF2 is surprising demanding with all of the extra hats and particles effects, but I suspect the limitations and inefficiencies of the Source engine are creeping up here.

          • PBCrunch
          • 3 years ago

          I suspect Overwatch was left out because the game engine monkeys with the graphics options when the action gets intense in order to keep framerates steady.

          Still, AMD wants to market RX 460 as a card for MOBA gamers, why not hold their feet to the fire and test some MOBAs.

          Aren’t CS:GO players the ones saying they can tell the difference between 200 FPS and 400 FPS when playing on 144 Hz monitors?

        • DragonDaddyBear
        • 3 years ago

        I agree keeping a constant, high-end CPU gives a much better idea of what the GPU can do without limits. But it also does not represent a real world scenario.

        Maybe a “high end” and “low end” bench may be useful going forward?

      • LocalCitizen
      • 3 years ago

      while i agree with what you said about oddity paring i7 (or even haswell-e) with a budget graphics card, especially for the Vulcan Doom that make use of the extra cpu power. but it is only a question about a few games that will require deep investigation. i think Jeff needs some catch up sleep first.

      what you said about “unique flaw” is unfair. amd tout polaris 11 as a low power chip for laptops that runs <75w, so naturally a <75w NV card makes the best comparison. but since amd can’t keep the power under control, the AIBs had to supply extra juice to make the card competitive. when 460 without 6pin eventually come to market, expect worse performance.

      • LoneWolf15
      • 3 years ago

      The reason this is done is to prevent a CPU from being a bottleneck and affecting the benchmarks. This may not give you the exact FPS with a lower end card…but it gives you a true idea of what % a card is better than another card.

      • funko
      • 3 years ago

      there seems to be a market of people buying prebuilt dells, hps, acers, that are restricted by the stock power supply and case form factor that would buy these to give a gaming boost to the otherwise office work only PCs.

      I’m guessing this market stands (in terms of marketshare and market value) on its own compared to the hobbyists that (like many of us) that DIY their own gaming rigs. A lot of these rigs have core i7-6700s fully built for $600ish which arent too far off from these bench marks.

      …add in the flock of affordable freesync monitors that are starting to flood the market…. not hard to imagine people buying these

      • flip-mode
      • 3 years ago

      For more than a DECADE it has been explained that the fastest CPU is used to test GPUs because the goal is to remove all bottlenecks that can be removed. This is clearly sensible to do. And also, if not the fastest CPU, which one? CPU selection then becomes arbitrary and based on the reviewer’s whim.

    • Ushio01
    • 3 years ago

    How come your not showing the exact settings your testing the games under? you usually do.

      • Jeff Kampman
      • 3 years ago

      Working on it.

        • Ushio01
        • 3 years ago

        Thank You.

    • DrDominodog51
    • 3 years ago

    I believe there might be a mistake in the graph on the number of frames over 50 ms for the 750 Ti in Crysis 3.

      • Jeff Kampman
      • 3 years ago

      Excel is just buggy like that sometimes. The number is correct, even if the graph isn’t.

    • PBCrunch
    • 3 years ago

    The table on the second page says the GTX 750 Ti has 4096MB RAM, which seems suspect.

      • Jeff Kampman
      • 3 years ago

      Stupid mistake. Fixed.

    • chuckula
    • 3 years ago

    Minor correction on p. 2:
    [quote<]See, I figured we would be getting a power-plug-less RX 470, as well, given the card's frugal board power, and I guessed there'd be no better way to pit apples against apples than to get a similarly-provisioned GTX 950. [/quote<] Make that Rx 470 an Rx 460.

      • Jeff Kampman
      • 3 years ago

      Too many similarly-named graphics cards out there right now. Fixed.

        • gerryg
        • 3 years ago

        So we won’t see the RX 750 going against the GTX 460 anytime soon?

    • derFunkenstein
    • 3 years ago

    Edit 2: I am thoroughly surprised. Several other reviews showed the RX 460 getting [url=http://www.tomshardware.com/reviews/amd-radeon-rx-460,4707.html<]walloped[/url<] by the [url=http://www.pcworld.com/article/3104596/components-graphics/amd-radeon-rx-460-review-an-affordable-graphics-card-with-bleeding-edge-tech.html?page=5<]GTX 950[/url<] in the majority of their test suites. It's not like TR did any favors by using a stock clocked 950 either so I wonder... Jeff: Is it possible that since TR's review was delayed by logistics and the DPC latency issues that you wound up testing with other drivers? Maybe it's just the test suite, which rightfully leans on the emerging standards of DX12 and Vulkan. Whatever it is, great review, and a nice showing by AMD.

      • Jeff Kampman
      • 3 years ago

      Tom’s Hardware used a massively overclocked GTX 950 to gather its results, whereas the Asus card we tested is quite conservative with its clocks. Tom’s doesn’t share exact settings, so it’s hard to say exactly what else might have affected the balance of power between the cards.

        • derFunkenstein
        • 3 years ago

        Oh, right, I saw that and forgot. Dumb mistake.

        • SoundChaos
        • 3 years ago

        I was also tricked by tomshardware’s 950. EVGA FTW edition with 115mhz core overclock and 125w power rating($150-170 US market price), and it completely skewed my judgement. Glad to see you guys keep the playing field level!

        I’m still afraid that the gtx 1050 might offer a similar wattage while performing closer to an RX 470 or even 480. The 460 should have been able to beat out the 950 in power/performance 🙁

          • Hattig
          • 3 years ago

          It’s likely the GTX 1050, if based around GP107, will have 640 shaders at around 1.6 GHz.

          That puts its TFLOPS firmly around 2 TFLOPS, give or take. It may edge the 460 overall, but it’s not out yet, and there aren’t even any rumours about it coming out soon – so it’s probably at least a month off.

          The Nitro+ 460 is just a bit too expensive for my liking – maybe if it came with a game token or similar to add some bonus value it would be more enticing. Fact is the 470 isn’t out of reach at this price and is clearly vastly superior. Maybe it’ll drop to $130 at some point.

          A $99 2GB 460 would have been a killer deal, it’s a shame these are starting at $109 right now, but it’s a more affordable deal for a low-end gaming system builder (the sort of person often on a 3-5 year upgrade cycle, games casually, just got a 2014 game on sale and it’s struggling on their $120 2011 graphics card.

          What is reassuring is AMD’s constant graphics driver improvements over time, the good Vulkan and DX12 performance in games where these are coded well, and so on.

          I agree that it would be neat to see results when the CPU is older/weaker.

          • derFunkenstein
          • 3 years ago

          I have a feeling the 1050 is going to cost more than an RX 460, or else be severely cut down from the 1060 and therefore be closer in performance (although probably better performance per watt)

            • Concupiscence
            • 3 years ago

            Yeah, but I’ve got a feeling that a 3 gig 1050 will show its age pretty fast…

            • derFunkenstein
            • 3 years ago

            I agree, and the rumored 3GB 1060 moreso. Anything more than $150 you have to figure is going to last a while.

      • LocalCitizen
      • 3 years ago

      this is a low power gtx 950, handicapped by the lack of extra 6pin

        • derFunkenstein
        • 3 years ago

        Yeah, I saw that, and then I forgot about it. Dumb me.

        And since plug-less 460s seem to be clocked similarly to the review sample, hopefully the performance holds.

    • wingless
    • 3 years ago

    It’s a cheap card that isn’t a complete POS. I ain’t mad.

    • chuckula
    • 3 years ago

    Yay! Huge bonus for the frametime analysis under DX12 and Vulkan too!

    On balance it looks like the overclocked version of the Rx 460 is a decent match for the GTX 950 in DX12/Vulkan and a bit behind in OGL/DX11. We’ll see how a small-Pascal GP107 product performs when Nvidia launches it.

Pin It on Pinterest

Share This