by Tim Tscheblockov
10/13/2004 | 02:31 PM
The Taiwan-headquartered Gainward is known for its desire to produce non-standard graphics cards. You may recall the famous CoolFX card series on NVIDIA’s chips with a water-cooling system, as an example. Right now this series is represented by models on the whole spectrum of graphics processors, from GeForce FX 5700 Ultra to GeForce 6800 Ultra. Quite naturally all these devices feature high overclockability, but they are rather costly, too. Only a well-to-do overclocker can afford a top-end water-cooled graphics card from Gainward.
<%BANNER[article]%>So, if you’re not rich yet, you may want to look at “ordinary” solutions with more or less acceptable prices – Gainward offers products manufactured by the reference design, but with exclusive cooling systems from Gainward itself. Among these “standard” cards there are models the very name of which signifies their good overclocking properties or even initially overclocked frequencies – Gainward calls them Golden Samples.
Our today’s review is about one such product, officially named Gainward PowerPack! Ultra/2400 Golden Sample. Let’s see if this is a worthy choice for an overclocker.
Gainward’s nomenclature of graphics cards does not allow guessing what graphics processor a given card is based on. Our model, the PowerPack! Ultra/2400 Golden Sample, belongs to a new series of products; it is based on the GeForce 6800 GT graphics processor from NVIDIA, and this is indicated on the package:
The card comes with a standard set of accessories – drivers, utilities, manual and other stuff. There’re a couple of unusual things, though: a coupon for purchasing a power-supply unit at a discount and two DVI-I-to-VGA adapters (not one, as usual) because the card is equipped with two DVI-I connectors:
The Gainward card sticks to the reference design, but features a non-standard cooling system. Instead of the ordinary fan of the GeForce 6800 GT, which is placed near the GPU heatsink and is blowing air through the heatsink ribs, this card has a large aluminum plate with two fans – it covers the heatsinks of the GPU and memory chips.
This cooling system is rather bulky, but the Gainward engineers took it easy – they just made a dual-slot bracket. Some people may kick up a fuss about that because the neighboring PCI slot is lost but there are really few computer systems where the dual-slot design poises any serious problems.
The fans on the PowerPack! Ultra/2400 Golden Sample are highlighted with red LEDs – they are not as bright as the fashionable blue LEDs, but still visible even in a mildly-lit room.
The heatsinks of the cooling system resemble those of the reference card, but with minor differences: the GPU one is made of copper, while the heatsink on the memory chips doesn’t use a heat pipe, but is a simple aluminum bar with “combs” glued up to it.
Paste and pads are the thermal interface to the GPU and memory chips, respectively:
The fans are placed almost symmetrically relative to the heatsink on the GPU. This design is the opposite of the GeForce 6800 GT’s standard one where air is directed from the fan to the heatsink ribs with the help of a casing.
In this card, however, the two fans have to fight each other blowing air at the ribs of the heatsink. Is this solution effective? We’ll test it later on.
Like all GeForce 6800 GT-based graphics cards, the card from Gainward has a single power connector. The next snapshot shows this connector as well as an additional heatsink over the power elements of the voltage regulators of the GPU and memory chips.
The graphics card is based on the NVIDIA GeForce 6800 GT processor:
And carries 256 megabytes of 2.0ns GDDR3 memory manufactured y Samsung:
The nominal frequencies of the graphics card are 350/1000MHz, like in the standard GeForce 6800 GT design. On installation of the ExperTool utility, enclosed with the card, the frequencies grow to 400/1100MHz, i.e. to the frequencies of the GeForce 6800 Ultra. ExperTool allows setting the GPU and memory clock rates up to 420/1200MHz, and the card is perfectly stable at them, suggesting an even higher overclockability.
Besides controlling the frequencies, ExperTool can adjust the fan speeds. The utility offers two modes: dynamic and manual control. In the dynamic control mode the driver adjusts the speed of the fans basing on the GPU temperature data. In the manual mode you set up the fan speeds for 2D and 3D modes manually. More exactly, you set up the voltage the fans receive – changing it from 50% to 100% stepping 10%. The fans are rather loud at their full speed, but this noise becomes barely perceptible at 50%. The dynamic control is the most annoying mode – the tonality of the noise is constantly changing, getting on your nerves very soon.
Throughout my tests I will be trying to determine if the fan speed affects the temperature of the card and if overclocking is possible at the minimal fan speed.
But let’s get back to overclocking. People from Gainward must have limited the overclocking capabilities of ExperTool to ensure a higher stability, but I hate to be within any limits. So, I remove ExperTool and install RivaTuner and get to my tests.
The testbed was configured as follows:
The settings of the ForceWare driver were left at their defaults; the graphics quality settings of 3DMark05 were default, too. The graphics quality settings in the games were the highest, too. I also carried out tests with 4x full-screen anti-aliasing (selected in the settings of Doom 3, Far Cry and 3DMark05) and 16x anisotropic filtering (selected in the 3Dmark05 settings and forced through the driver for Doom 3 and Far Cry). My testing went in an open testbed and without any additional air cooling; the room temperature was about 24..25°C. The temperatures of the GPU and the graphics card were read with the help of RivaTuner.
So, the standard frequencies of the Gainward PowerPack! Ultra/2400 Golden Sample are 350/1000MHz. They are set to 400/1100MHz on choosing the Enhanced Mode Settings in the ExperTool utility. The card was stable at these clock rates at any fan speed, but we certainly should not consider the latter frequencies as the device’s limit.
The maximum frequencies the card was stable at when overclocked by RivaTuner were 430/1200MHz (with the standard cooling system). That’s very good for a GeForce 6800 GT. What’s remarkable, the card went through the whole cycle of tests at such frequencies both at 100% and 50% of the fan speed.
Let’s see what temperatures of the graphics processor and of the PCB RivaTuner reports when the card is working at its regular frequencies. There were two test modes: “Idle” (Windows’ Desktop on the screen, no applications running, the temperature values are written down after half an hour of the system’s being idle) and “Burn” (3DMark05 is running five times in a row with its default settings, then the temperatures are written down). So, here are the results:
The GPU temperature only increases by 14°C in the Burn mode when the fans are working at their maximum speed. If the fan speed is reduced to 50%, the temperature grows by 18°C – that’s very well for a noiseless mode.
When the dynamic control over the fan speed is selected, the Idle temperatures almost equal those at the 50% fan speed, while the Burn temperatures – those at the 100% fan speed. That’s quite natural since the GPU temperature is low when there’s no load and the driver automatically keeps the fan speed low. When there’s load, the core temperature starts to grow, making the driver speed up the fans.
Now, the same test, but the frequencies of the card are set to 400/1100MHz, i.e. to the frequencies of the GeForce 6800 Ultra:
Overclocked to the frequencies of the GeForce 6800 GT, the card became just a little hotter: 3°C more of the GPU temperature and 2°C of the PCB temperature. The cooling system evidently does its job right even when the fans are working at half of their maximum speed.
Now, let’s try the card at the maximum clock rates (430/1200MHz):
The temperatures have grown by a single degree, not more. So, the temperatures of the graphics card have only grown by 4°C at worst during overclocking! That’s an excellent result – you shouldn’t have any overheat-related problems with the Gainward PowerPack! Ultra/2400 Golden Sample.
So, are these frequencies – 430/1200MHz – the maximum you can squeeze out of this Golden Sample from Gainward? Let’s check it out. I remove the plate with the fans and do the same with the GPU heatsink, replacing it with a GPU block from the Thermaltake Aquarius II water-cooling system:
This allows me achieving 450/1200MHz frequencies – 20MHz more of the GPU clock rate. With the stock cooling system on, the heatsink on the memory chips was cooled by the standard fans. Removing the cooler I didn’t specially provide for cooling of the memory chips – they became hotter of course, but still remained stable at the same 1200MHz frequency.
The following diagrams show you the temperature measurements with the water-cooling system installed. First, at the standard frequencies of the card:
The temperature of the GPU is of course smaller than it was with the standard cooling system, while the PCB became 10°C hotter – heated up by the memory chips and the power elements of the voltage regulators.
At the maximum overclocking (450/1200MHz) the temperatures of the GPU and PCB were:
The temperature of the GPU increased by 2°C and of the PCB by 3°C at the maximum frequencies. Those 54 degrees centigrade are rather hot, but as you remember, I performed my tests in an open testbed without installing any additional blowers. In a good system case the graphics card will be cooled down by the system fans, and if you have a bad case, just install one low-speed fan to blow at the card.
I checked out the performance of the graphics card at the increased frequencies in Far Cry (my own demo record on the Carrier level), Doom 3 (the standard demo1 record) and 3DMark05.
I offer you the results of the Gainward PowerPack! Ultra/2400 Golden Sample in four modes: at the standard frequencies, at the frequencies of the GeForce 6800 Ultra, at the maximum frequencies with the stock cooling system and at the max frequencies with the water-cooling system. I won’t compare this card to others, as there are a lot of such comparisons around the Web. Knowing the value of the performance gain you can easily relate the results of the Golden Sample to other graphics cards.
In 1024x768 resolution, without anti-aliasing or anisotropic filtering, the graphics card of the GeForce 6800 GT class easily handles the graphics load, so it is the CPU that becomes the bottleneck. That’s why the fps rate grows but slightly during overclocking.
The performance bonus is more easily felt in the other modes. Moreover, the value of the bonus suggests that the game speed depends on the GPU frequency rather than on the memory clock rate even with FSAA and AF enabled.
It’s similar with Doom 3: the performance growth is small in the easiest modes, as the central processor becomes the main bottleneck in the system. Doom 3 is overall more indifferent to overclocking than Far Cry: the graphics memory bandwidth must be important for this game, and the memory chips on our sample of the graphics card had worse overclockability than the GPU.
The results of 3DMark05 are mostly determined by the speed of DirectX 9 shaders and of shadow maps generation, i.e. by the speed of the graphics processor. So, the rather low overclockability of the graphics memory doesn’t limit the performance growth you get through overclocking here.
It’s the same in the second game test – the relative performance growth is big enough.
The rather low overclocking potential of the graphics memory chips doesn’t prevent the card from providing a high performance gain due to good GPU overclocking.
In the hardest mode – 1600x1200 with full-screen anti-aliasing and anisotropic filtering – the ratio of the scores and GPU frequencies in the standard mode and at overclocking coincides to about a couple of percent, and this indicates that the results of 3DMark05 are mostly determined by the GPU speed, at least with the GeForce 6800 GT graphics card. Good overclockability of this GPU is most helpful here.
So, the PowerPack! Ultra/2400 Golden Sample graphics card from Gainward did well in my overclocking tests – the good overclockability of the graphics processor seems to have not been hindered by the rather average overclockability of the memory chips. Our thanks go to NVIDIA who developed a graphics processor that can optimally use the available bandwidth of the graphics memory bus. Gainward too receives our thanks for culling the best GPU chips for the Golden Sample series and for installing an efficient cooling system that frees you from bothering about the temperatures of the graphics card’s components at overclocking.
The performance of GeForce 6800 GT graphics cards allows running the most advanced computer games, but may be not enough to run the hardest games like Far Cry or Doom 3 at their highest settings. In this case overclocking may help.
The diagrams below show the performance gain I received by overclocking the graphics card in all the tests:
Overclocked to the level of the GeForce 6800 Ultra, the graphics card enjoys a performance bonus of about 10-15%. At the maximum stable frequencies the card performed about 25-30% faster. By the way, these maximum frequencies are above those of the GeForce 6800 Ultra by about the same value as the frequencies of the Ultra are higher than those of the GT.
So, the card from Gainward is good almost in everything, there’s only one downside: top-end products on graphics processors of the new generation are still rare and cost a lot. A sky-high price can spoil all the fun from any graphics card.