The Best ?100 Graphics Card: Radeon HD 6870 Review

Choosing your next PC upgrade is always something of a balancing act - a fine art in extracting the maximum gaming performance at the best possible value: the highest FPS per GBP, if you will. In putting together the Digital Foundry PC, we were looking to get the nucleus of great computer, a taste of the high-end for entry-level money. In choosing our parts, the Radeon HD 6770 looked to offer the best trade-offs in price vs. performance, but recent adjustments in the AMD range have made us rethink. The classic Radeon HD 6870 - for a long time the enthusiasts' favourite, and still a very powerful piece of kit - is reaching its "end of life" and if you keep your eyes peeled, you can pick up a superb GPU at an irresistible price.

Having bossed the £120-£140 price bracket for some time, the 6870 is starting to get some serious price-cuts, to the point where - at the time of writing - the card can now be picked up brand new on eBay for under £100, or at similar price-points from certain e-tailers. And this is just the beginning - as price-cuts begin to kick in on AMD's replacement 7xxx line of GPUs, we would expect more mainstream online suppliers to follow suit. Catch one while you can.

If you're running an older PC that's running short of horsepower on the latest games, at this price, a card with the pedigree of the Radeon HD 6870 looks like a bit of a no-brainer. And for Digital Foundry, it was a chance to re-evaluate our recently constructed custom PC. Our initial spec produces high frame-rates at 1366x768, but we were effectively limited to 1080p gaming at 30FPS. Could a straight swap from the 6770 to the 6870 elevate us to the ideal of 60 frames per second gaming? Curiously, while the 6870 has dropped significantly in price, our initial 6770 has actually gone up by around £5 to £10, making the extra expense a mere £20 to £25 extra overall.

Upgrading the Digital Foundry PC

On paper, the swap appears to justify the extra expense. When it comes to hitting higher resolutions and allowing for additional anti-aliasing, the HD 6870 benefits from its peak memory bandwidth of 134.4Gbps, compared to the 78.8GB per second available on the HD 6770. Each card boasts a matching 1GB of GDDR5 RAM, but the higher bandwidth we see here comes as a result of the 6870's wider 256-bit memory bus, which doubles that of the 6770's. The upgrade also sees the core clock go up from 800MHz to 900MHz, while the memory clock suffers a downgrade from 1200MHz to 1050MHz. However, taken as a whole this represents a big step forward.

Construction and specifications of the HD 6870 can vary by the vendor, but the XFX version we have for test here matches the physical design and clock speeds of AMD's reference model to a tee. Unlike the HD 6770, the pricier card is enclosed in a plastic case that squares off at the end with several vents, though the fan is contained in a matching position. It's much wider than the 6770 as a result, and its length, at 24.8cm, makes it one of the voluminous cards available. However, for the Casecom MA-1199 chassis we're using for our build, there's plenty of room near its bottom-front to allow us to angle one end of the card in first, before settling the back half down into the PCIe 2.0 slot.

1 of 5

Caption

Attribution

Due to the higher advertised TDP of the card - going from 108 watts on the HD 6770 to 151 watts on the upgrade- the HD 6870 requires two 6-pin PCIe connectors as opposed to just the one. The Corsair 430W power supply we're on our build using can only supply one of these connectors, however, meaning special arrangements need to be made for the other. Many card manufacturers pre-empt this need by boxing in a molex-to-6-pin PCIe converter cable- often requiring two molex connectors to provide the necessary power. Luckily, our PSU has three of these at the ready; though one is already used to charge the rear fan, the rest are fair game for GPU duties.

Switching the unit on, we find that neither heat or noise levels have changed much in the transition. The GPU's sensor reads no higher than 60 degrees Celsius in HWMonitor during our stress tests, suggesting the ventilation to and from the card is just fine. Meanwhile, we notice the fan doesn't make quite as much noise as the HD 6770's when under heavy load, though they're both remarkably close in this regard, and ultimately neither cards' decibel levels distract during gameplay.

Performance Analysis: Benchmarks

So let's see if it was worth the extra in practise. As is customary, we start with some synthetic tests to get a perspective of the card's strengths and weaknesses. We use 3DMark 11 to gauge each cards' proficiency at handling DirectX 11 features such as tessellation and compute shaders; here, the test is divided up into isolated graphics and physics segments, culminating in a single, combined test. To keep things easily repeatable, we run the Performance preset at its default 1280x720 resolution, and then the Extreme preset at 1920x1080.

Futuremark 3DMark 11: Tested on standard Performance and Extreme settings.

"3DMark11 scores suggest that the jump in performance from the Radeon HD 6770 to its more powerful sibling is something in the region of 50 per cent. Impressive."

The overall scores show a 50 per cent increase in favour of the newer card, though this doesn't tell the whole story. Because the test gives the PC a workout as a whole, the physics scores are being determined by the CPU in this case, where little changes across the board regardless of settings or the card installed. This indicates a bottleneck on the processor which, as you'll see in our later gameplay analyses, can impact certain games at points regardless of the upgrade.

Nevertheless, the boost in the graphics department stands present and correct, as predicted. We see the graphics scores for each test taking a larger margin of victory than the overall score in fact, though for the combined section it's not quite so clear-cut. Major improvements here are only really evident on the Extreme setting. This is largely how performance during gameplay pans out too, where the HD 6870's higher memory bandwidth permits much higher resolutions, and avoids the severe level of stuttering we perceive during the HD 6770's combined test.

To showcase more on the graphical side, we take Batman: Arkham City for a spin using its built-in benchmarking tool. The focus here isn't just on getting the Unreal Engine 3 to run at a fair clip, but to show how frame-rate is impacted when differing methods of anti-aliasing are applied. As usual, we disable PhysX and v-sync for both DirectX 9 and 11 tests, and run the game on the high graphical preset with ambient occlusion checked. The big variable here is between FXAA High, a post-processing method, and the more graphically taxing 4x and 8x MSAA settings.

Batman: Arkham City Benchmark: DX9, 1366x768, high detail, PhysX off, all other settings enabled.

"Curiously, the Batman DX9 benchmarks suggest virtually no difference at all in performance, with the extra power of the 6870 left completely untapped."

Just as with our previous tests on the HD 6770, we see very minimal variation in the average frame-rate on DirectX 9 with the new card. Point for point, we experience largely comparable results when we're guided through a heavily fortified enemy camp and around Arkham's floodlit skyscrapers. The superior maximum frame-rate is always attached to the HD 6870, which accounts for the very few instances during the test where light sources and enemies are kept to a minimum. Otherwise, these results are uncannily close, and simply aren't what we had bargained for.

To see if we can settle the score more conclusively then, we try this again with DirectX 11 settings enabled, such as tessellation. The resolution remains at 1366x768, though the higher MSAA settings should in practise put a strain on each cards' memory bandwidth to make the difference known.

Results with this newer API puts the HD 6870 in a very positive light. Running this benchmark with FXAA enabled places both cards in a similar ball-park, however we see the previous card trail behind on average values once we engage the more intensive MSAA methods. Even at 8x, the HD 6870 can potentially carry a perceptual 60FPS gameplay experience, and the maximum scores indicate that this could go even higher during the general run of play- particularly around indoor environments.

Batman: Arkham City Benchmark: DX11, 1366x768, high detail, PhysX off, all other settings enabled.

"The DX11 features of Batman: Arkham City drive a wedge in performance between the HD 6770 and its meatier sibling, the gulf increasing as we scale up anti-aliasing quality."

By comparison, the HD 6770 shows its trademark weakness when it comes to combining high fill-rate tasks with DirectX 11, which is indicative of how full 1920x1080 results might also pan out. For those willing to bite the bullet and cap the game at 30FPS, this shouldn't pose a problem at all, if playing this series at higher resolution settings is an imperative. In short, these are the figures we wanted to see here.

To round up our series of benchmarks, we run the upgraded DFPC through three passes of Metro 2033's dedicated Frontline benchmarking tool. This gives us the chance to stress each DFPC configuration in terms of lighting, specular effects, and approximate ambient occlusion. To stay consistent with our other tests, we leave this on the medium quality preset- more than enough to tax the hardware- and alternate between 4x MSAA and the 4A Games' own custom technique, known as AAA.

This is the most gruelling test of the bunch, and we can only barely stay afloat of the 30FPS average mark with our original HD 6770 setup. This is particularly true of the opening 30 seconds, where the highest number of enemies, tanks and exploding barrels appear in quick succession. For going through all the motions of the upgrade, the HD 6870 manages to best this by a matter of 10FPS on both AA presets, though this is still not quite the revolution we had hoped for given the strengths of the card on paper.

Metro 2033 Frontline Benchmark: DX11, 1366x768, medium detail, PhysX disabled.

"The Metro 2033 Frontline benchmark is a punishing workout for any GPU and here we see the HD 6870 pulling ahead with a 25 per cent advantage over the cheaper HD 6770."

At their peak, FPS values favour the HD 6870 by a wide margin when around the closing section, featuring fewer enemies and more derelict trains. During actual gameplay, we find Metro 2033's performance is much more palatable than these benches would suggest, where even a DirectX 11 playthrough can hold out at almost 60FPS during the earlier levels.

We also run several passes of the very same benchmark at 1920x1080, and find this places performance on the upgraded PC strictly in 30FPS territory. By comparison, an equivalent test on the original card gives us some unbearably choppy 10FPS motion, which brings to light the benefits of the HD 6870 offering double the memory bus width.

To follow this up, we put the DFPC through a gamut of gameplay tests to pinpoint precisely where each card thrives and flounders. Just as with our Face-Off features, this involves capturing each game in a lossless video format, and then feeding these like-for-like clips into our FPS analysis tools. V-sync is engaged at all times here to ensure image integrity, and this time we make 1920x1080 at 60FPS the ultimate target for both cards. We also provide 1366x768 comparisons for each game as well - just a click away beneath each video.

Gameplay Analysis: Battlefield 3

The first game we tuck into is Battlefield 3; famous for giving DICE's much-anticipated Frostbite Engine 2 its official debut, and being one of few games to make DirectX 11 support mandatory. The game facilitates all manner of CPU configurations, and fortunately scales up and down according to the number of cores it has at its disposal - in our case, the two 2.8GHz cores present in the DFPC's Pentium G280. Advanced features, including chip-physics on walls and other environmental cover, tessellation on geometry and horizontal-based ambient occlusion, are in support of an already visually arresting game.

Jumping straight in at 1080p, it's immediately evident that 60FPS just isn't quite in the running here for our upgraded rig. That is, at least not when it matters most, such as in the heat of battle. When running the game at medium graphical preset on both setups we find a difference of around 10FPS between the HD 6770 and 6870 holds true for much of the test. In the final battle sequence, however, there's little to distinguish the two when there's a long road filled with re-spawning waves of enemy militia and tanks. It's a mess of readings that favours neither party.

"Battlefield 3 results see only a small improvement between the HD 6770 and the HD 6870, proving that CPU power is growing increasingly more important on higher-end engines."

• Alternate Analysis: Battlefield 3, medium settings, 1366x768, Radeon HD 6770 vs. HD 6870

Results inevitably tend to favour the updated card, but it's clear that our CPU of choice is bottlenecking performance. For indoor shoot-outs this isn't quite so evident, but once gunfire is directed across open courtyards filled with destructible vehicles, the frame-rate takes 50 as its baseline, and then stutters downwards with each successive explosion. On the other hand, for the traversal-based sections of play, such as the ascent to the top of a building in the Operation Swordbreaker stage, the game runs without any struggle to speak of. These sections run at very similar speeds regardless of the card being used.

Our 768p comparison narrows these differences down further still, and the lead is largely interchangeable throughout. The HD 6770 even commands a surprising advantage during one indoor sequence. Otherwise, here we see 60FPS is a reality for much of the time on both setups, save for the occasional two-second dips when transitioning between areas. In summary, this isn't the improvement we were hoping for given the additional expense for this GPU, and we still need to rely on a 30FPS cap to enjoy these higher resolutions with a consistent refresh.

To put this CPU bottleneck into perspective, we insert the card into another quad core PC to pinpoint its limitations. Our higher-end setup packs an Intel i5-2500k processor overclocked to 4.2GHz, coupled with 8GB of 1600MHz RAM, which should for all intents and purposes show us the extent to which the HD 6870 is being anchored down by our budget PC's CPU. In practise with Battlefield 3, the difference at 1080p with the same graphical settings gives us precisely the results we had hoped for. It's pure 60FPS at this level, and at its very worst drops down by 5FPS briefly on the walk around the very first corner of the Operation Swordbreaker stage. Pushing for higher graphical settings, such as high and ultra, makes for an average of 40FPS-30FPS gameplay respectively in the same section - but the MSAA-free medium setting is a winner.

Gameplay Analysis: Skyrim

The Elder Scrolls V: Skyrim shows much greater promise in delivering on our 1080p60 goal. We use the introductory sequence for our tests, where our as-yet unnamed hero is transported downhill to Helgen village for a good old-fashioned beheading. Due to the number of NPCs bunched together on-screen once we hit the village, this sequence shows the game at peak levels of stress. This journey also illustrates the impact of object, foliage and character draw distances, where ours is set to the high quality preset, with 8x MSAA also enabled by default.

At last, we see a consistent advantage. For every step of the way, the HD 6870 upgrade grants us around 20FPS headroom or more over the original DFPC, where we usually find the frame-rate reading resting at the top of the graph. Controller feedback is noticeably smoother, even when it does take a tumble. Approaching the gates of Helgen drops this down to 40FPS at worst, but this soon picks up once we dismount from the cart and enter the first tutorial dungeon.

"Skyrim shows a very pleasing boost in 1080p performance with the HD 6870 installed, but curiously 768p frame-rates remain completely unchanged."

• Alternate Analysis: Skyrim, high settings, 1366x768, Radeon HD 6770 vs. HD 6870

In effect, we can smoothen out these drops by tweaking the high graphical preset slightly. Specifically, dropping 8x MSAA in favour of FXAA does wonders for this initial sequence, and more to the point, exploring the game's world rarely gives us any issues once we're free to roam it. The draw distance settings are evidently well matched with our hardware in this case too, and the pop-in of distant details rarely coincides with any manner of hiccup.

In our original DFPC configuration this was not possible. Previously, we were required to lower the settings to medium in order to achieve similar results, or simply leave the resolution at 1366x768. Comparing the two setups at this lower resolution once again puts the HD 6870 in the lead for much of the first leg of the journey, but interestingly, the route around the village gives us practically identical readings.

We also investigate performance of the HD 6870 with our higher-spec quad core Intel i5 system. Unlike Battlefield 3, this game is only optimised for two cores, meaning much of our CPU's potential power is going to waste here, and some of the inherent advantage lost - the only real gains will come from the i5's much higher clocks. Nevertheless, we have perfect performance at these same settings at 1080p, though being ambitious and ratcheting this up to the ultra setting leaves us with some uneven 45FPS playback instead.

Gameplay Analysis: Crysis 2

Next up, we turn our magnifying glass to Crysis 2's early submarine level, where we take to the very high graphical preset with DirectX 9 and v-sync engaged, while the resolution is set to 1920x1080. The crash in this first sequence does a sterling job of showing Crytek's technical wizardry in an enclosed space, where their water shaders and alpha effects dominate the spectacle and place high demands on our original DFPC.

Slotting the Radeon HD 6870 in for testing lands us results which show the biggest overall improvement over the former card. This time, we see a consistent lead of almost 30 frames per second for prolonged sections of this initial level, where the upgrade holds out at the maximum 60FPS with only the occasional dip nether-wards.

"Crysis 2 offers us great 1080p gameplay with the HD 6870 installed, but switching to a quad core CPU really lets the card strut its stuff, offering a relatively consistent 1080p60 experience."

• Alternate Analysis: Crysis 2, very high settings, 1366x768, Radeon HD 6770 vs. HD 6870

This lead isn't here to stay though. At its worst, we find surveying the long stretches of New York's urban sprawl in the second stage pulls the frame-rate down to an average of 50FPS. We still see a considerable performance boost here, though engaging in direct battle can often-times lower this further downwards to 40FPS. Overall, the frame-rate advantage remains convincing, and the 6770 DFPC rarely catches up in precisely like-for-like set pieces. But considering each level gets progressively more vociferous in its demands on the GPU, it's questionable that this processing power can be considered enough to sustain 1080p60 by the end of the game. When the time comes, lowering geometry and post-processing settings become a necessary step in lightening the burden.

At 768p, meanwhile, the original DFPC can sustain 60FPS with some drops, and many of these are significantly reduced by the HD 6870 upgrade. This takes the edge off many of the frame-rate nosedives we see with the HD 6770 installed, particularly while swimming around the flooded submarine in the prologue stage. Achieving a perfectly butter-smooth frame-rate across the board is never quite possible even at this resolution, alas; our take-away is that we'd walk the 1080p high road if given the choice, given that the frame-rate is liable to drop down to 45FPS during battle sequences in either case.

As before, curiosity takes precedent and we swap the HD 6870 across to a PC with a much faster quad core CPU to see how it fares. Crysis 2's results fall in line with what we're seeing on Battlefield 3 and Skyrim, where running at precisely the same settings allows 60FPS to be entirely tenable with this configuration. The 1080p60 dream is realised here, and even with ultra settings enabled, general playability is perceptually smooth despite the occasional odd dip. By comparison to playback on the budget PC's Intel G840, this is a huge leap forward for these higher settings. The conclusion we take is that the HD 6870 offers so much power you may well end up with a GPU that is "under-fed" by a lower end CPU.

Gameplay Analysis: The Best of the Rest

Of course, it would be unfair to call it a day at just these three high-end games. There have been plenty of top-notch PC releases over the last two years which don't require your hardware to bend over backwards to achieve great results. Among these are Call of Duty: Modern Warfare 3 and Portal 2, which even on our original DFPC configuration can run entirely maxed out at 1080p60.

Naturally, the window for improvement is fractional when these games are v-synced to a typical monitor's 60Hz refresh. The ceiling has been hit. Even when raising the stakes further by increasing the MSAA from 2x to 4x, we still enjoy an unerring 60FPS response during the first stages of each of these games. DiRT 3 suffers from the odd stutter when too many AI cars take the lead on a single-player Cross Rally, resulting in the average FPS landing at 57.2 for our original DFPC. With the upgrade, these creases in the graph are straightened out entirely, and maximum performance is easily attainable. Success.

Blizzard titles such as StarCraft 2 and Diablo 3 also benefit from the upgrade, albeit to wildly different degrees. With the former, running at 1080p on High settings makes our 4v4 in-engine replay run with a marginal advantage during the big, all-in showdowns. It's during these climatic stretches that the FRAPS average is dragged down to 55.6FPS, whereas all build-up macro play is essentially perfect in motion. It's an improvement overall, even if it is slight one.

Gameplay Performance Matrix: 1920x1080 resolution, high quality settings, 2x MSAA enabled, v-sync engaged (unless otherwise stated)

"The Radeon HD 6870 has room to flex its muscle at 1080p with high settings, but it's surprising how competitive our HD 6770 remains on a great deal of titles. The v-sync 60FPS cap is a great leveller."

Diablo 3 fans, on the other hand, can enjoy in excess of a 100 per cent increase in frame-rate when running global high settings. We ran our standard test, hacking and slashing en-route from New Tristram to the first waypoint. Originally, we opted to avoid smoothing shadows for this test due to the colossal performance hit it incurs- and at a very meagre benefit to the presentation - but we find adding it on our upgraded rig only lowers the frame-rate to 82.8FPS. This means that engaging v-sync pins performance at 60FPS regardless of how high you crank the settings. Between the HD 6770 and 6870, the difference basically boils down to being able to run with this smoothing enabled, and to have the assurance that the frame-rate won't be throttled down to 30.

The remaining two games, Rage and Saint's Row: The Third, aren't able to get close to this sky-high level of performance. We use the buggy ride through the wastelands to test the first, where we enable texture detail, have anisotropic filtering set to high, and have large texture cache selected to see the worst case scenario. Sticking to the 1080p60 master plan causes some egregious stuttering throughout the trip which lowers the average to just over 50FPS, though fortunately, dropping the resolution to 768p and disabling texture cache does at least help uphold our 60FPS target. Sadly, the resulting pop-in of its textures does sully the visual impact to a large extent.

To round off, we put Saint's Row: The Third through its zany paces during a complete playthrough of the first level, with high graphical settings and DirectX 11 enabled. The lengthy roof-top showdown with a chopper makes for the least flattering results out of all our selected tests. Even factoring in the 6870 upgrade, performance here averages out at just under 40FPS. We find this is directly linked to the DirectX 11 settings in effect for lighting and shadows, however, and toggling these to the next rung down is all that's truly necessary to optimise for 1080p60.

The Radeon HD 6870 Upgrade: The Digital Foundry Verdict

"The Radeon HD 6870 represents phenomenal value at the £100 price-point. It offers a surfeit of power to the point where you may need a CPU upgrade to see the full extent of its capabilities."

In review, the results inevitably favour the beefier, more watt-ravenous circuitry of the HD 6870, but the key point here is pricing. A gaming PC with this card bundled in, as opposed to the lightweight HD 6770, gives us a tallied total of just under £320 for a performance boost to every game we've tested. This ranges from the considerable to the slight when it comes to our entry-level dual core gaming PC. We're left with the feeling that the extra expense is backed up by the performance, but that the 6770 is probably a closer match to the dual core CPU we specced in our original parts list.

For the likes of Crysis 2 and Battlefield 3, we're surprised to find that the HD 6870 upgrade doesn't go quite far enough in making 1080p60 attainable during shoot-outs for either campaign. The general run of play does favour Crytek's shooter though, being the more stealth-based of the two, and offering the largest disparity in the "before-and-after" upgrade comparisons. It's interesting here that the same card in a quad core system does so much more - in effect, an entry-level dual core CPU simply can't "feed" the 6870 with enough data quickly enough. Otherwise, the most impressive performer of the trio would undoubtedly be The Elder Scrolls V: Skyrim, which is able to sustain 60FPS for lengthy durations in our tests - especially when the 8x MSAA setting is replaced with FXAA.

Specifically, we find titles like DiRT 3 and Diablo 3 now join the ranks of Modern Warfare 3 and Portal 2 in their ability to run entirely maxed out at 1080p60. Meanwhile, StarCraft 2 and Saint's Row: The Third need nudging down on the graphical settings, from high to medium in either case, in order to achieve the same class of performance with ease - again, there's a sense we're CPU bound at the higher level settings here. In the case of id Software's Rage, the lack of scalability in the graphics settings menu does mean we can't tone down texture asset quality or the complexity of the lighting model to optimise gameplay in a similar fashion. In the end, this means that the only ways to achieve a fluid 60FPS response with this game is to sacrifice either full 1080p resolution, or to disable texture detail.

In short, the level of power the HD 6870 represents at the sub-£100 price level is remarkable. Newer cards like the HD 7850 and NVIDIA's forthcoming GTX 660 will of course become the new enthusiasts' choices but they're in a different price category altogether and it's unlikely that a card approaching double the cost will translate into anything like double the performance - in this sense, the HD 6870 offers a remarkable sweetspot in price vs. performance. If you're looking to update an old PC, the 6870 at this price is a no-brainer - but just be aware that the more challenging games could well require a CPU upgrade to unleash the full power this card has on tap.