The generational fight: the GTX470 vs the GTX570
Many of you are going to be looking at the title and thinking this is a pointless article due to us being on the verge of a new generation of graphics cards. You may be correct, but only about the impending release of new graphics cards. Not everyone is able to spend two, three or maybe more if really setting your sights high, hundred pounds on a graphics card and so they must hunt high and low for bargains whether they are second hand or new stock that needs to be shifted. With the new cards in mind, it’s obvious that enthusiasts and those in the position to do so, will be upgrading at the earliest chance, casting their old graphics cards to the second hand market. Retailers may even join in with the need to clear old stock for bargain prices, netting you some incredible value for money.
More on the cards we’re pitting against each other then. First off we have the MSI GTX470 Twin Frozr II which, as a case in point of my opening paragraph, I purchased for £140 when the 5xx series had been out for around a month. The GTX470 utilises NVIDIA’s GF100 Fermi architecture with 448 CUDA cores, 1280MB of GDDR5 memory and a TDP of 215W. This card, by default, has a core clock speed of 607MHz, a shader clock speed of 1215MHz (locked in a 2:1 ratio with the core) and an effective memory clock speed of 3348MHz. Sitting on top of this and keeping it cool is MSI’s Twin Frozr II cooler which shows up how bad NVIDIA’s stock cooler on the 4xx series cards is by doing wonders for operating temperatures and even managing to leave headroom for overclocking.
Against it we have Gigabyte’s GTX 570 OC which uses NVIDIA’s refined GF110 chip to ship with the same number of CUDA cores as the GTX480. With 200 million less transistors than the GTX470, the GTX570 has 480 CUDA cores, a TDP of 219W and the same amount of vRAM as the 470. As is suggested by the name, Gigabyte have factory overclocked this graphics card and not just by 10MHz like the Asus GTX570. Instead, we see the card shipping with a core clock of 780MHz (whereas stock is 732MHz) but no bump to the memory, so it remains at its stock speed which is an effective speed of 3800MHz. Keeping it all cool is Gigabyte’s Windforce 3X cooler favouring a vapour chamber to keep the already cooler running GPU nice and frosty, again leaving some headroom for overclocking.
Now, onto overclocking them both and while it can vary as to how well various cards can be overclocked, the majority of enthusiasts graphics cards can go a fair way and these two are no different! Shipping with a stock voltage of 962mV the MSI GTX470 has a fair amount of headroom left for voltage and thanks to the Twin Frozr II cooler, temps aren’t a problem either. Setting the core’s voltage to Afterburner’s maximum allowed (1.087V) gave the extra power to push the core to 800MHz which is an amazing 32% overclock. Overclocking the memory is often seen as being less beneficial overall but the 470’s memory was able to be pushed to an effect 3600MHz. This overclock proved stable during benchmarks, gaming and Kombustor’s burn in mode with a maximum temperature of 74 degrees Celsius and a fan speed of no more than 60%. Anymore than this resulted in the NVIDIA driver crashing or a BSOD but it’s still impressive and it shall be interesting to see how much is gained by this. Overclocking the GTX570 in a similar fashion saw a final core clock speed of 950MHz with the memory at 2000MHz (4000 effective). This was reached with a core voltage of 1.1V and proved stable under the same testing as the 470. Temperatures never broke 70 degrees C with the fan speed, again, not going over 60%. This meant that the 570 was overclocked by 22% when compared to its factory overclock and 30% compared to NVIDIA’s stock settings.
Starting off with 3DMark 11, the stock 470 sees an extreme preset result of 1401. This breaks down to a graphics score of 1277, a physics score of 3608 and a combined score of 1690 – for the results of each test please see the table below. Overclocking the GTX470 sees quite an improvement, with the overall score jumping up by nearly 400. This, as expected, is largely due to an increased graphics score but what is interesting is that the physics score also increased. Every test benefited to some extent with the 32% overclock providing a 27% increase in overall score. This is not a performance increase to be sniffed at and it will be interesting to later see how it continues during games. Trying the factory OCed 570 with 3DMark 11 shows the overclocked 470 as punching well above its weight and there seeming to be very little difference. The 570 managed a 7% increase in overall score, again with every test benefiting and curiously the physics score continuing to go up. When compared to a stock GTX470 the difference is quite a respectable 36% improvement. Letting the overclocked 570 now run through the test, the gaps really start to open up. It managed to return a score of 2245 which is an 18% increase on itself, 26% and 60% when compared to the 470 overclocked or not respectively. Between the stock 470 and the overclocked 570, the physics score has now gone up by 496 marks which is one that’s quite hard to explain. The physics tests purpose is to test the CPU and so should be as little to do with the graphics card as possible. Looking at the differences, the card to benefit the most from its overclock here is the 470, with it not really being beneficial in upgrading unless the GTX570 you get can also manage to overclock fairly well. If talking purely about factory settings then the difference is more worth it.
Onto in-game benchmarks now and all of these will be run at 1080p with the highest possible settings and on the games that support it, PhysX or NVIDIA GPU only features will be enabled and set to their highest possible setting. This does push out the AMD/ATI users a tad but as these are both high end NVIDIA cards I believe they should be tested to their fullest. This means you see the absolute worst performance as you could always turn down settings to make the game playable or smoother. Where anti-aliasing is concerned, I used 32xCSAA where possible or the highest the game engine would allow. Our first game to talk about is Just Cause 2, being a bit more specific The Dark Tower benchmark. The MSI GTX470 at its stock settings managed to return an average of 33.05 which for the vast majority of people is perfectly playable. Next, trying the overclocked 470 gives a nice boost of 8FPS with it posting an average framerate of 41.18. This is a difference of 25% which has dipped compared to the 3DMark result but is still a respectable increase. Returning a result of 39.92, the stock 570 shows the overclocked 470 can punch well above its weight and shows that unless you overclock it, coming from a highly overclocked card may not be worth it. Thankfully for the 570, overclocking does benefit it quite largely with it achieving on average 50.07 frames per second. With a 25% increase over itself and a51% increase over the stock 470, Just Cause 2 seems to benefit quite well from the cards being overclocked.
Next we’re onto Mafia II and it’s another game which makes use of NVIDIA’s PhysX for increased tire smoke, debris (e.g. from walls being damaged) and advanced cloth simulation. PhysX may not be a deal breaker but the features it offers add that extra level of immersion and something to play with when you’re bored. Interestingly, Mafia II shows a marginal gain between the 470 at its stock settings and when overclocked. Overclocking it takes the result from an average of 26.9 frames per second to 28.4, a rather poor increase of 6%. The GTX570 tells a very similar story with it only managing an average of 29.9. Oddly enough in this testing, we see no difference between being at the 570’s stock settings and being overclocked. I will return to these results later with a different CPU just to make sure the result is not an odd one off as well as to check that the CPU is not holding the GPU back in any way.
Moving on we have Batman: Arkham Asylum. While the sequel has now been released, NVIDIA haven’t said half as much about Arkham City’s use of PhysX as they have for Arkham Asylum. Starting at the top this time, the overclocked 570 posted an average frame rate of 76 FPS and a minimum of 43. Running at stock settings causes no change to the minimum frame rate and a mere 6 FPS drop to 70 for the average. Putting this as a percentage, the 22% overclock is giving a 9% performance increase in this case. Returning an average frame rate of 55 and 64 FPS between respectively running at stock clock speeds and overclocked, the GTX470 also does fairly well and with the minimums never dipping under 32 we’re seeing perfectly playable results. The MSI card manages to best itself by 16% when overclocked with the Gigabyte card winning by no more than 38% when comparing the 570’s overclocked result to the 470 at its stock speeds.
The final game tested here is Metro 2033 and for those who have not had the chance to play it, it is a Crysis-esque game in how heavy on the system it is so to expect playable settings here is quite a big ask. However, we shall see how they do with everything maxed. Starting off by setting the bar fairly low, the stock 470 returns a minimum frame rate of 5.2 and an average of 13. An overclock of 32% helps boost performance by what at first seems quite a small improvement but when broken down into a percentage works out as greater than a 1:1 ratio of the overclock vs the improvement. With a difference of 36%, the rather low 17.67 FPS average is quite impressive but with a minimum frame rate of 6.85 it’s still not comfortably playable for anyone. Trying with the Gigabyte GTX570 running at its factory defaults, we see quite a large step up in performance with an average of 27.00 but still a rather low minimum of 9.92. This means that the stock 570 is, as far as averages are concerned, 53% faster than the overclocked 470 and just over double the stock 470. A staggering result indeed but to what new heights can an overclock push the GTX570? Unfortunately, not as high as you may expect; the average posted by the overclocked 570 is a mere 3.67 frames per second higher than its stock settings with the minimum frame rate reached remaining disappointing at 7.95 FPS. Not to take away the 570’s victory as it may have only improved on itself by 14% but compared to the 470 it is between 74 and 136% better!
I mentioned earlier that some results didn’t sit correctly with me as they seemed rather low compared to what may be expected given the large overclocks put to the test. To check that these results were true I ran the same tests with the same graphics drivers (285.62) on an AMD FX-8150. To my surprise, they confirmed my results with some even being exactly the same as when first run on my Phenom II x4 805. Any variances were minute and not enough to warrant further testing, worry or really even a mention. For example, the overclocked GTX570 with the FX-8150 managed an average of 30FPS in Metro 2033 which was 0.67FPS less than with the Phenom II. The 3DMark 11 score was thrown out a bit by the fact that the CPU tests could benefit from the more threads the 8150 is able to execute but when comparing the graphics card tests there was a marginal difference seen. To clarify a bit, by marginal I mean 0.1 or 0.2 frames per second difference. The table below shows these results so you can see for yourself.
To wrap this up then, both cards are impressive. They both offer great value for money and it just goes to show that great gains can be had from overclocking them. I must however, say that not all manufacturers or cards take too kindly to overclocking and you do so at your own risk – check your warranty and its small print to see if it is covered or not. From these quick five tests it’s pretty fair to say that the move from a GTX470 to a GTX570, even with no overclocking on either card, can give some rather large gains. But as is also demonstrated it must be understood that not all games or applications will see such large benefits. Comparing them against each other one final time we end up with the following table:
This shows that unless you’re coming from a heavily overclocked GTX470, going to a GTX570 with the intention of overclocking it then you’re not going to gain much. In every other situation you have quite a lot to gain. In terms of stock vs.overclocked, it’s clear that overclocked is better but what do you get for your work and the improvement what’s translated to pure performance? The GTX470 had a 32% overclock and upon average saw nearly a 28% gain from it. With the GTX570, it was only able to manage a 22% overclock compared to its stock settings but still managed to average just under an 18% performance increase.