GTX 480 vs. HD 5870, 8x AA Performance Analysis, Part 3
NVIDIA has released its long awaited GeForce GTX based on its brand new Fermi DX11 GF100 architecture just over five weeks ago. We were able to bring you Part One of our GTX 480 series, “NVIDIA’s GTX 480 Performance Testing” here. We also brought you Part Two, “Overclocking”, here. We have had five weeks hand’s on experience testing our GTX 480 versus the HD 5870 and we have learned quite a bit more that we would like to share with you. We will now explore the relative performance hit of each card when we increase the anti-aliasing (AA) from 4x to 8x.
In Parts One and Two, we learned that the GTX 480 does indeed overclock. For Part Two, we explored the performance of our overclocked GTX 480 (825/1100MHz) to our overclocked HD 5870 (975/1300MHz) and we added more benchmarks over Part One. In this Part Three, we will look at the relative performance hit of 8xAA over 4xAA on each card, and we shall continue to add games as we progress through this series of GTX 480 vs. HD 5870. Part 4 is going to be after this coming week’s analysis of our overclocked “new” Powercolor HD 5870 PCS+ vs. our “old” reference Diamond HD 5870 and we shall upgrade to the latest drivers. Currently we are still using Catalyst 10-3a as it has been a long time between AMD’s graphics driver releases and Catalyst 10-4 was just released this week.
In Part One, we compared the relative performance of 5 GPUs – GTX 480, GTX 280, HD 5870, HD 4870-X2 and HD 4870. GTX 480 won most of the benchmarks just over HD 5870, followed by HD-4870-X2. The HD 4870 and our GTX 280 were left in the DX10 dust while the DX11 cards showed their advantages over a dual-GPU flagship card of the last generation. In Part Two, we compared relative overclocked-to-the-max performance of the GTX 480 to our overclocked HD 5870, and the GeForce pulled away even further from the Radeon.
We expect that NVIDIA will shortly launch it’s own entire DX11 line-up based on their GF100 “Fermi” architecture. We expect to see GTX 460 and 450 launch within a few weeks or months and we need to see what this new GF 100 Fermi architecture brings over their GT200b series besides DX11 and a smaller process. Increasing the anti-aliasing for our GTX 480 may give us some idea of GF100 scalability as NVIDIA brings out their entire line up. Applying 8x AA was a weakness of the GT200 series – at least compared to their competitor, the Radeon 4000 series. Let’s see how Fermi architecture fares with higher anti-alaising.
To properly bring you this review, we purchased a Diamond HD 5870 from NewEgg and put it through its paces with the very latest performance drivers – Catalyst 10-3a. AMD is quite proud of this driver set as it brings sold performance increases over Catalyst 10-2 and over even the last WHQL drivers, Catalyst 10-3. Also, remember that AMD has had a long time to mature their drivers and that the latest GTX 480 GeForce 197.41 drivers that we are still using are their first WHQL drivers and they should leave some room for further performance improvement by NVIDIA’s GeForce driver team in the months to come.
Today you will see us pit our Diamond reference design HD 5870 which is now stock clocked against the new reference GTX 480 which is also at stock-clocks. We will benchmark sixteen modern games and two synthetic benchmarks ranging from 1680×1050 to 1920×1200 to 2560×1600 resolutions and with details fully maxed and with 16x anisotropic filtering (AF) applied. Our main focus is the performance comparison between 4xAA and 8xAA on each video card.
In Part One of our GTX 480 performance evaluation we declared the GTX 480 the performance winner even though the performance was relatively close. Part Two gave us a much clearer picture and we could see that the GTX 480 has superb overclocking headroom and awesome architectural potential. This review is continuing on as a series and we believe that we have a much clearer picture by examining the performance hit of 8xAA over 4xAA. Later on, we will also look at the individual features of each video card to see what else the new NVIDIA GPU brings to the table including CUDA applications such as 3D and GPU-assisted video encoding (as with Reveal/vReveal) as well as NVIDIA’s 3-panel Surround and other new features including 3D.
Here is AMD’s not-so-secret weapon to fight the new GTX 480 and please notice the free down-loadable copy of Call of Duty, Modern Warfare 2 bundled in with the PowerColor HD 5870 PCS+ as an incentive. The PowerColor HD 5870+ is a mildly overclocked version of our reference HD 5870; 875/1225 MHz; up +25 MHz each on the core and memory clock. Worthy of note especially for smaller cases, the PowerColor HD 5870 PCS+ PCB is shorter and wider than reference and the cooling solution is also both quieter and more effective than the reference cooling solution. This will be our upcoming review for next week and we shall compare our overclocked Diamond reference version against this one to see if our reference HD 5870 was limited in any way when we got to its most extreme overclock in Part Two of our series.
We already pointed out in Part 1 that it is practical to upgrade from HD 4870/GTX 280 class – which includes GTX 260, 275 and by extension, GTX 285. We also discovered that it is also logical to upgrade to DX11 from a HD 4870-X2 or HD 4870 CrossFire, or by extension, GTX 260 or 275 SLI or even a GTX 295 which is a bit more powerful than our HD 4870-X2. Since we do not want any chance of our CPU “bottlenecking” our graphics, we continue to test both of our graphics cards with our Intel Core i7 920 at 3.80 GHz (3.97 GHz effectively with the 21x multiplier in turbo mode), 6 GB Kingston DDR3 and a Gigabyte X58 full 16x + 16x PCIe CrossFire/SLI motherboard.
Later on we plan to also test our AMD DX11 video cards on AMD’s Dragon platform. We also acquired a brand new ECS black label A890GXM-A CrossFire motherboard which is a nice performance upgrade from our current Gigabyte 790X motherboard and we posted that review last week.
Please take into consideration that nVidia uses a different version of AntiAliasing starting 8x and up, therefore comparisons are henceforth limited at best. Sadly I don’t have a direct link right now, but please take it into consideration before drawing (final) conclusions.
We took special care to make sure that identical AA settings were applied in all of our benchmarks including for Crysis. We even noted that in the full retail game, Just Cause 2, that we observed the benchmark results showed the Radeon was running at 8xCSAA while the GeForce was 8xAA.
However, we have since learned from AMD that the benchmark results are wrongly identifying 8xMSAA as CSAA. The Radeon is actually running 8xMSAA and this minor issue will be addressed in a future patch.
Everything we test is “apples to apple” unless it is specified in the review.
Nice article man. Cheers
good