The market for graphics cards got huge boost last year when ATI released the first of the DirectX 11 graphics cards, numbered, unsurprisingly, the 5 series. Demand for the latest and greatest easily outstripped supply and even now, some months later, it's still very difficult to have a wide selection of cards to purchase if you're loyal to a certain manufacturer.
Naturally the first out the door of the ATI labs was the, at the time, top of the range 5870. This was quickly followed by the 5850 and rumours persisted about the possibility of a 5870X2, which eventually reached us named the 5970. At least the adjustment of the naming conventions should make it easier for the average consumer to know where in the range their card fits.
As with all high-end graphics cards though, not long after the more value orientated models started to drip into the marketplace. Anyone who remembers the amazing value for money that the 4770 produced in Crossfire will be delighted to know that a 5770 has been made available and these four cards make up today's huge test.
With so many cards and combinations to test this is also a first for OC3D in that two of our reviewers, myself Tom Logan and Bryan Waters, are collaborating to bring you this huge amount of information as concisely as possible.
Naturally with demand so high obtaining these cards is quite a feat, so we were delighted when ASUS stepped in to offer us an amazing 6 of their excellent 5 series cards to enable us to run the full battery of tests.
We know you're all as itching as we are to see some numbers and find out which would be the best bang for buck, so let's cut to the chase.
Initially it would appear that the ASUS cards are reference based designs, utilising the stock ATI cooling solution and in the de rigueur black and red the current ATI design is based upon. However as is often the case, under the hood things are surprisingly different.
ASUS have loaded their ATI 5 cards with a voltage modifiable BIOS. This is a far cry from the old days of having to get out your pencil and should aid overclocking tremendously. But don't think that this is some special for a review site as the versions available on the retail market also come with this handy feature that should enable some very stable core overclocks.
Naturally with only the reference cooling there is very little to say about these cards. The main points of interest are the very tasteful colour scheme which will match a lot of peoples systems. Most high-end motherboards are starting to be produced in black these days, and a lot of the extreme performance memory has a red and black colour scheme. So it's great to see that it's possible to create a cohesive internal look without having to spend yet more money having everything rebuilt into certain colours.
Secondly ASUS have resisted the temptation to plaster the cards with some random fantasy figure, lending an air of class to the whole thing with a small logo in the bottom left corner and on the fan the only branding visible.
To aid in identification we've labelled the cards for you, although basically the higher the card, the bigger it is. The 5770 is 220mm long, the 5850 242mm, the 5870 is 280mm and the 5970 a quite staggering 305mm long. Suffice to say the 5970 is definitely not for those with a mid-tower case and even some of the smaller or poorer designed full-towers might struggle.
Moving on you'll notice the plethora of outputs available on all the models. ATI's 5 series of cards come with Eyefinity, a technology that enables triple-monitors to be used for an all-around gaming experience that has to be seen to be believed. With the need to output to three monitors for this it stands to reason that there are so many outputs. For those still on a single monitor setup, which we'd imagine to be nearly everyone, having such a choice ensures that no matter what monitor you have there is a compatible output.
We see a lot of exceptionally desirable hardware come through our offices, but there is something about the four cards all together that made even the most jaded of us get a little warm and fuzzy. Don't they look lovely.
It's time to see our test setup and try some overclocking.
Naturally testing something as potentially insanely powerful as the 5970 requires all the beef it's possible to get to push the cards as hard as we can. Although it would be possible to go really mad and get the LN2 out, we always feel that the results we obtain at OC3D should be easily replicated in the real world by our readers and so we've stuck to a sensible overclock that should be enough.
CPU : i7 965 overclocked to 3.6GHz
RAM : 6GB of OCZ Blade
Audio : ASUS Xonar Audio
Motherboard : ASUS Rampage II Extreme
PSU : 1000w OCZ Gold
GPU : ASUS HD5970, HD5870 and HD5870xf, HD5850 and HD5850xf, HD5770
OS : Windows 7 Ultimate 64
Drivers : Catalyst 9.12
Many early reports about the 5850 said that it could achieve much higher clockspeeds than those they ship with by default, and that the 5970 voltage and speeds were intentionally kept low by ATI to keep the card under the 300w threshold. As the ASUS cards make it so easy to adjust the voltage and push for those high GPU clocks it had to be done.
Firstly please note that the only overclocks were done to the GPU clock as the GDDR5 clock speeds are already pushed high at stock and any overclocking to the memory produced negligible increases, so we concentrated solely upon pushing for the best GPU overclock we could obtain. Stability was determined by a two hour gaming session with the details maxed out. As long as there were no artifacts observed then we considered the overclock to be stable.
Starting with the ASUS HD5770 which actually ships with the highest default voltage of any of the cards on test today we managed to increase the core speed from 850MHz up to 950MHz by upping the voltage from 1.124v to 1.25v.
The HD5850 is the card we were most interested in as it had the potential to produce the biggest gains, and sure enough the default clock speed of 725MHz was managed to be increased to a mind-bending 1GHz core with voltage upped to from 1.15v to 1.275v. A staggering increase.
The ASUS HD5870 ships with 850MHz core and as this is the highest performance single card available we weren't expecting a large increase yet once again the ASUS voltage mod BIOS enabled us to reach 950MHz on the core with the voltage increased from 1.15v to 1.225v.
Finally the HD5970. When ATI first announced this card they took pains to point out that the clock speeds have been kept low to stay within the specifications demanded of a PCIx slot. However as we don't need to stay within an arbitrary limit the default 725MHz at a very low 1.049v swiftly became 900MHz at 1.275v. A very handy increase although we're really stretching the abilities of the 5970 thermal solution.
Remember that we ensured that all the cards were stable over a long period of gaming before declaring it successful overclocks and therefore we didn't think we'd see the egg-frying temperatures some of the early DirectX 10 cards experienced and sure enough we didn't.
The main things to take from this graph are :
1) For all the single chip cards they have lots of headroom for overclocking. Whilst 70c is quite high, it's only similar to the stock results of the last generation, proving the evolution of the cooling solution employed.
2) The 5850 clearly has insane amounts of headroom. To get a 33% increase in clockspeed for only a 5c increase in temperatures is nothing short of outstanding.
3) Finally, it's clear that although the 5970 is perhaps conservatively volted and clocked, it clearly is right at the limit of how much heat the fan and heatsink can expel. We really pushed it to its limits for our testing and it was loud, but it was stable. We'd definitely suggest accepting the extra noise that 100% fan speed brings to get those temperatures down if you plan on running a 5970 full-time.
I hope you're all sitting comfortably, because there is a wealth of data and graphs upcoming. So get your comfy chair and a brew and settle in.
As we recently documented within these pages we've made some changes to the games we're going to be using, and so from the excellent Modern Warfare, we move on to Modern Warfare 2. As with all of our tests today we're going to be running at 1920x1200 with the maximum settings available in game. As Modern Warfare 2 is designed with good gameplay on consoles as a vital component we're lucky because us PC gamers can really crank the image quality up high and still very playable frame-rates.
This is clearly demonstrated in our graph. Even with 8xAA only the stock 5770 is incapable of supplying 60fps average, but it can do it with the overclock the ASUS voltage BIOS allows. However as soon as we move into cards we'd expect to be used for gaming the Infinity Ward game gets quickly chewed up and spat out. A stock 5850 providing nearly 100fps average and not dropping below the magical 60fps barrier is only the beginning. Our Crossfire 5870 setup giving mind blowing frame-rates at a level that makes the 5970, even overclocked, look a bit wheezy.
Crysis eh. When it was first released the general opinion was that the graphics were lush, the gameplay pretty ropey, and it needs some mythical future computer to be able to run. Well, here we are with a 3.6GHz i7 Extreme and more graphical horsepower than could have been dreamt about a year ago. How does it fair?
Not well. We had to leave the anti-aliasing off entirely to stop the 5770 grinding to a halt, and although we wondered if the power of the other card might necessitate two runs using different settings we can see that even the 5870 Crossfire setup doesn't hit 60fps. Thankfully an overclocked 5850 provides a sufficient average frame-rate to give a smooth gaming experience, but the dependence of the Crytek engine on Physx, plus some pretty shoddy optimisation, means that nothing yet can give us that 60fps we need.
Ok so we've killed a few people, let's put the pedal to the metal.
Codemasters Dirt 2 is infamous for being heavily delayed to enable the introduction of DirectX 11 code paths and features. Thankfully the final product is amazing with great handling, a deep career mode, lots of rewards, a robust online component and, yes, those graphics. The Codemasters EGO engine making a complete mockery of the Crysis effort by giving us all the glitz and glamour we demand from our modern games coupled to the single greatest water effect even produced by silicon.
The main things to note from our Dirt 2 tests are firstly, and most obviously, how close the minimum, maximum and average framerates are to each other. This ensures a very smooth experience at all times even down at the 5770 range of the graph. Secondly is something that doesn't fully come across in this test, but from other tests we've done, is how scaleable the engine is. Naturally with such behemoths we had to use 8xAA to stop the 5870xf not blowing the graph apart, but this has meant that it takes until the 5870 to reach the 60fps bracket. However, we need to point out that reducing the anti-aliasing to only 4x means that it easily hits 60fps on all but the 5770.
If anyone doubts the potential of DirectX 11, take the car through the puddles on Battersea or Malaysia and then get back to us. And this is one of the first games to utilise it. We can't wait.
The oldest game on test today is a standby of OC3D and still one of the most enjoyable racing games around. It's probably at the end of it's life as a useful way to push graphics hardware hard but, with so many of our reviews and our readers using it, we do enjoy taking it out for a spin and seeing some of the ludicrous frame-rates it's possible to achieve.
Even at this high resolution and level of anti-aliasing the little 5770 never dips below 68fps and the clear winner is the stunning 5870 Crossfire setup that very nearly broke the 400fps maximum mark. However before we move on to our final game it is worth taking a moment to compare a couple of results.
The 5970 is roughly the features of a 5870 at the clockspeeds of a 5850, but the 5850 Crossfire setup spanked even the overclocked 5970 and would save you about £100 in purchase price. Ok at these frame-rates the difference between 180 and 200 is impossible to notice, but that £100 in your wallet will be easily spotted.
Finally let's have a look at NFS:Shift and the ever important Vantage results before wrapping this up.
Need for Speed : Shift
EAs venerable Need for Speed series has come a long way since it's first incarnation as a demonstration title for the Panasonic 3DO technology. It initially was street driving, quickly diverged into it's more familiar cops/robbers and street racing styles, before being reborn as NFS : Shift. We here at OC3D applaud the new direction of the series that has breathed new life into the sagging franchise by being graphically excellent and enjoyable to play. Although it is not an element on test today the engine sounds are as good as any game in the world.
When it was first released there were a lot of technical issues on ATI cards, especially with 4870x2 performance, so it's great to see those have been eliminated with patches and ATIs always speedy driver updates. Once again everything was maxed out and we applied 8xAA for some exceptional visuals.
As is becoming quite a theme throughout our testing the 5770 made a good fist of it, but came up a little short even in its overclocked state. Once we moved on to testing the 5850 we broke the 60fps average we all seek to obtain and things only improved. Perhaps most noticeable is that even the crossfire 5870s which so far have swept all before them couldn't really give us the big gains in frame rates we'd expect for the price differential.
Love it or hate it, there is no escaping it. Since it's origins a decade ago the Futuremark (or Mad Onion as it was) benchmark suite 3D Mark has been the only number anyone cares about. We all have different tastes in gaming. We all have different image quality levels we're willing to accept. But everyone knows their 3D Mark score.
Normally we put the synthetic benchmarks at the beginning and see if the real-life results bear this out. Today we decided to change it around and so we didn't have any preconceptions about the performance of a certain card or combination. Anyone looking at the P-Score below would instantly dismiss the 5770 as not worth the time yet our gaming tests showed that, whilst it's no silicon behemoth, it can easily cope with most games providing you're sensible with the anti-aliasing.
However, if you want a graph that shows how cards scale synthetically, this certainly is it. More money = higher score and that appears to be it. Most noticeably though this is the first time the 5970 has looked like coming close to the 5870 Crossfire setup and beating out the 5850 Crossfire system by some margin. Make of it what you will.
Phew. Let's condense this epic test into a few pithy paragraphs for the low-attention span types in the audience.
So what conclusion can be drawn from todays testing?
Firstly I'm sure a lot of you have come into todays review with a preconceived notion of exactly what card you'd like. In previous times the general consensus was to purchase the highest priced card you could afford and stretch if possible to the next one up. For single card purchases that would lead you into looking at the 5970 and pretty much stopping there. So let's start at the top.
If our tests show anything it's that the £540 HD5970 is a poor purchase regardless of how much money you have available. Firstly it's the hottest card on test when run at stock settings. The only hotter card is the overclocked 5770. So straight away you have more heat to dispel and a high power draw. Secondly it's enormous at just over a foot long. There aren't a lot of cases around that will take such a behemoth, although we'd imagine anyone with over 500 notes to spend on their graphics will have a suitable case. Finally, and most tellingly, it just doesn't add up on a pure performance level. An overclocked HD5870 is close in most tests and £200 less, and the HD5850 Crossfire setup at stock speeds easily keeps up with even the overclocked HD5970 and is still £100 cheaper. An overclocked HD5850xf setup would blow it away. So if it suits neither the performance junky nor those looking for value for money, what should you buy instead?
Subscribe to OC3D's YouTube channel here
The HD5870. This is an amazing card. Truly outstanding. As a single card solution it is the equal of the last-generation king the Radeon 4870X2 at stock settings but with the power savings and heat reduction you'd expect from a single chip. It also clocks incredibly well. We managed to get the core to 1GHz on air and it coped although it was a little too hot for stability in our tests. Nonetheless being £200 cheaper than a HD5970 you could spend half of that saving on a water-block and then the sky is nearly the limit.Truly one of the best ways to spend around £330 you can find.
In Crossfire the HD5870 is far and away the fastest graphics solution we've tested. It reached, Crysis notwithstanding, the magical plateau where you actually wished for even higher settings because no matter how hard you pushed the game settings, or the AA/AF levels, it just laughed and carried on crunching numbers. A 4870X2 still stands up as an exceptional performing card and so with a HD5870 Crossfire setup being so brutally fast and its performance CPU limited, if you can afford the arse-clenching £650 price tag then it will serve for many years to come. Amazingly even that price doesn't seem as silly as it appears because if you brought a new high end single card every couple of years then you'd spend that anyway.
Subscribe to OC3D's YouTube channel here
Moving down the list to the HD5850 this was the little train that could. It's around £100 cheaper than a HD5870 at £230 and performance is around where you'd expect for the price. But my word does it clock. We hit a 1000MHz on the core and could probably have gone a little further. In its overclocked state it just about kept up with the HD5870 so if you're on a bit of a budget then it's easily worth considering. Again price comes into play when looking at Crossfired HD5850s. For a decent Crossfire setup you'd be looking around the £450 mark and the performance slots into about where you'd expect. Better than the £100 cheaper HD5870, about the same as the £100 more expensive HD5970 and quite a bit slower than the £200 more HD5870 Crossfire. Certainly on a performance standpoint the HD5870xf setup is where it's at, but the HD5850xf performs admirably by keeping up with the ever-more disappointing HD5970.
Subscribe to OC3D's YouTube channel here
Finally the little engine that could, the HD5770. Naturally it doesn't give the immense performance of its bigger brothers, but conversely one can be had from around £115. Playable frame-rates at such high resolutions and detail whilst still allowing for immense AA settings it starts to look like it will be the bargain the 4770s were in the last generation. You could get Crossfire overclocked HD5770s for the same price as a vanilla HD5850 and suddenly we're all stroking our beards and nodding at the possibilities. If "Bang for Buck", or I guess as we're in the UK "Excitement per Euro" or "Pleasure Per Pound", is an important consideration then this is probably the best of the lot. Certainly if you're building a system to a reasonable budget then it has to be considered.
Subscribe to OC3D's YouTube channel here
Phew. I hope you enjoyed that all as much as we did. Making a definitive "buy this" statement isn't easy at all as there are some people who want the best regardless, others who want value for money, still others who just want a cheap card that can game a bit and finally those who want future potential.
Generally our advice can be whittled down to the following :
1) HD5870 Crossfire is as good as it gets, with a price to match. Water-cooled these would be mind-blowing in an OC Crossfire setup.
2) HD5850 is very impressive especially in Crossfire and best for those who can't justify a months wages on eye candy.
3) HD5770 is quite a pocket rocket and we look forward to testing a Crossfire setup, coming soon to OC3D.
4) HD5970 is over-priced and under-performing. HD5850 Crossfire is as quick for less, HD5870 Crossfire annihilates it for £100 more.
Whichever of the three single GPU cards we tested today you decide upon, you'll be pleased with the performance.The ASUS voltage modified BIOS certainly makes clocking of these cards a joy and we very quickly reached massive gains in a stable manner. If anything we reached the thermal limits of the cooler long before we reached the performance limits of the chip.
We really can't rate the BIOS enough to enable us to hit mega core speeds with stability and a total lack of artifacts. For this reason we happily give it the innovation award.
Once some third party cooling solutions hit the market the ATI Radeon 5 series will truly shine and the Nvidia GF100 needs to be very special indeed to sway consumers.
Many thanks to ASUS for the cards on test today. Discuss this in our forums.