The release of the 8800GT chipset from nVidia has seen the market go crazy for what is one of the best bang-for-your-buck video cards available, and many
manufacturers are offering factory overclocked versions to give you even more performance, but are they worth the money? We look at the Inno3D’s normal 8800GT and their Overclocked version to find out.
The standard 8800GT chipset from nVidia operates with a core speed of 600MHz and a 900MHz memory clock (often referred to as 1800MHZ due to DDR memory – GDDR3 memory actually) and as of the time of writing this review, the Inno3D (aka Innovision) 8800GT (standard version) sells online for around AU$385. Inno3D’s Overclocked version has pushed the core speed up to 650MHz and the memory speed up to 950/1900MHz and sells online for around AU$425.
So from those figures we have the overclocked version offering an 8.33% core speed increase and a 5.5% increase in memory speed (and thus bandwidth) for around a 10% increase in price.
Both the standard and overclocked versions come with the same standard cooling setup, which utilises the reference nVidia single slot cooling system. This of course consists of a heatsink covering the whole card including the memory chips and an active blower fan just like every other 8800 series card that came before it. However unlike all the other 8800 cards the 8800GT and its cooling system only take up the space of a single slot, which is great for saving space, but it also means all the hot air is expelled from the cards inside the case, not externally like all the other 8800’s, which is sure to make the inside of your case nice and warm.
There’s really nothing fancy or special about these cards themselves, both are the basic reference design, standard Dual DVI-I ports on both (1 VGA connector included per card) and TV-Out with HDTV Composite break out cables. Both cards even come with the same game – Ghost Recon 2 Advanced Warfighter. So there’s really no difference between the two other than the clock speeds, and of course the price. Both have the same GDDR3 memory, same 256Bit memory interface, 112 stream processors each etc...
To see whether the OC version is a good investment we’ve run them through a series of benchmarks, of both DirectX9 and DirectX10 games and artificial benchmarks to see what the real performance difference is between the two. They will be tested as single cards, as well as in an SLI configuration.
Performance:
For the performance tests we will be performing our standard DirectX9 tests of 3DMark06, Prey and Quake 4. We will also be testing DirectX10 performance with the Lost Plant Demo and Crysis, both of which will of course be performed under Windows Vista (Ultimate), while the DirectX9 tests will be performed under Windows XP.
The Test Setup:
- Intel Core 2 Duo E6600 (2.4GHz) CPU, running @ 2.93GHz
- ASUS P5N32-E SLI, nForce680i Motherboard
- Corsair TWIN2X 2GB PC2-8500 Memory Kit (2 x 1GB)
- Seagate Barracuda 7200.10 250GB SATA-II Hard Drive
- Enermax Liberty 620W Power Supply
- Windows XP Pro SP2 for DirectX 9 Tests
- Windows Vista Premium for DirectX 10 Tests
We’ve never run our tests with an overclocked CPU before, however with our first test, which was 3DMark06, we found there was next to no difference between a single card and when they were run in SLI mode. Initially we thought SLI wasn’t operating, even though the nVidia control panel reported it as running, and everything else behaved as though it was operating normally, we eventually realised that the CPU really didn’t have the power to properly test a pair of 8800GT’s in SLI mode and had only enough processing power to push one card to the limit, but with two there was almost no difference. As we didn’t have access to be a better CPU at the time, we decided a simple overclock of the current CPU was the best option and gave us the difference (i.e. improvement) we were looking for to best demonstrate the power of these 8800GT’s.
3DMark06:
Our first cab off the rank with our benchmark results is 3DMark06, which performs a range of tests based on DirectX9.
3DMark shows a performance increase of around 3.5% for the Overclocked 8800GT over the standard version when in single form, which is somewhat of a decent increase in the 3DMark world, however in SLI mode there is an almost negligible difference between the two speeds of cards, showing around a 1% increase, hardly worth writing home about.
Quake 4:
Our Quake 4 tests were run on a range of resolutions starting at 800x600 and going up to 1600x1200, with normal settings (No Anti-Aliasing), 8x Anti-Aliasing and 16x Anti-Aliasing. Each setting/resolution combination was tested 3 times to ensure the results were accurate.
The No-AA test gave some quite unexpected results (yes they are accurate, I had to check them half a dozen times to convince myself before I would accept it), with the single cards outperforming the SLI setups and with each setup maxing out at an almost pre-defined FPS (frames per second). We have seen this before with the Quake 4 tests, though usually maxing out at a lower number and we figure the cards are running as fast as the CPU’s processing power will allow, and in this case we have the SLI cards putting slightly more strain on the CPU and thus maxing out at a lower number to the single cards.
Only at the max resolution of 1600x1200 do we see the performance of any of the cards start to drop at all, and even then it’s not by much, hopefully the tests involving Anti-Aliasing can provide a more accurate picture of these card’s performance potential.
The 8X Anti-Aliasing test with Quake 4 shows a more predictable result, but we still see only minor improvements for the OC version of the 8800GT over its standard counterpart, with the biggest difference being in the highest resolution, which gives us a 9.5% increase in performance. This is actually a little more than what we expected, but it’s hard to get an accurate result as a percentage at such low frame rates. The difference (in percentage terms) stays fairly consistent as the resolution drops for the single cards, however for the SLI cards, at low resolutions there is a much smaller difference, as low as 1.2%, and only a 7-8% difference in the highest resolution.
The 16X Anti-Aliasing test shows no difference between the two single cards in the high resolutions, and a much smaller difference in the lower resolutions than the 8X-AA test was giving us. However with the cards being really pushed we start to see at least some difference between the SLI setups, however it’s more so in the lower resolutions this time.
It’s good to see that when the cards are really pushed with 16X-AA and high resolutions, there is a big difference between the single and SLI setups, up to 81% increase, so those people who invested in an SLI setup know they haven’t (completely) wasted their money.
Prey:
For our Prey tests we ran them on standard settings (No Anti-Aliasing) as well as 4X Anti Aliasing which is the max setting the benchmarking software we’re using allowed for, thankfully they gave us some very useable results even on such low video settings.
For the No-AA tests we see a difference between the standard and OC cards of around 4.2% in the top resolution, though this drops as the resolution comes down, similar to some of the Quake 4 results. The is only a minor difference between the two SLI setups, and only at the very top resolution, a whole 2% difference – Whoopty do!
The 4xAA tests show a bit more of a difference between the cards, but not a whole lot, the biggest difference is around 6% between the single cards at 1600x1200 and 1280x1024, the SLI setups showed an increase across the range for the OC cards, but only VERY minor in the low resolutions up to a maximum difference of around 4.1% up high.
Lost Planet – DirectX 10
Our first DirectX 10 benchmark is using the Lost Planet Demo’s performance test, run at 3 different resolutions, starting with 640x480, 1280x720 and finally 1600x1200. In past reviews we have listed two different Lost Planet results, one called ‘Snow’ which has the camera running through a snowy outdoors environment, with lots of weather effects, explosions, motion blur etc. And another called cave which is inside an enclosed environment with lots of characters moving around. We have not included the results of the Cave test in this review, because every card setup, on every resolution reported the same result – 52 frames per second. Obviously this is another limitation of the CPU’s processing power, so rather than make the most boring graph in human existence we chose to leave it out altogether.
The Lost Planet benchmark shows almost identical results from all 4 setups in the low resolution test, as well as very similar results at 1280x720 amongst all 4 setups, but this time with noticeable improvements between the standard and overclocked versions of the cards. Similar differences again at 1600x1200 with the SLI setups really starting to come into their own in the high resolutions.
Crysis - DirectX 10:
We’ve updated our test software for this review to include the DirectX10 version of Crysis, which has to be one of the best looking and most graphically intensive games on the market, it’s certainly one of the most popular at the moment.
We tested Crysis on the highest quality level, with No Anti-Aliasing and with 4X Anti-Aliasing. The game engine supports much (MUCH!) higher AA levels, but the numbers were getting low enough already with only 4X, so we just left it at 4X for this review.
With the single cards we’re seeing an improvement of around 5-6% across the range for the overclocked versions, which considering the mild level of overclocking they have over the standard cards, it’s about as good an increase as you can expect.
The SLI setups have shown similar improvements for the OC cards versus the standard items, so at least it’s a consistent improvement, however there’s very little improvement in some areas for the SLI versus Single card setups, showing that this game is very CPU intensive and could really benefit with as much processing power as you can afford to throw at it.
Conclusion:
The test results here have shown that the performance increase you can expect for an OverClocked Inno3D 8800GT over a standard version is anywhere from 2-8%, with an average increase closer to 5% or so. Considering the price increase for the OC version over the standard item is in the order of 10% it’s a bit hard to justify. If it was my money on the table I think I’d stick with the standard version and save myself a bit of cash for something else.
The cards themselves were rock solid performers so if you’re after a good quality reliable 8800GT I’d have no hesitation recommending the Innovision/Inno3D brand, but again, I’d stick with the standard version is it was my own money.
These products can be purchased at www.techbuy.com.au (Standard Version) (Overclocked Version)
Crysis with a single card @ 4xAA has a really poor frame rate. Must have been very jerky.
What processor/MB would you recommend to get a single 8800GT or GTS (512) to work with higher frame rates?