by Tim Tscheblockov
07/15/2004 | 12:55 PM
Recently, our test lab got a graphics card from PowerColor based on the least expensive and speedy of the new GPUs from ATI – the RADEON X800 Pro. Its frequencies are lower than those of the RADEON X800 XT Platinum Edition (475/900MHz against 520/1120MHz) and it has fewer pixel pipelines (12 against 16).
<%BANNER[article]%>The first question an overclocker may ask about the X800 Pro is if it’s possible to turn its disabled pipelines back on and then overclock it to the level of the RADEON X800 XT Platinum Edition? Then, will the X800 Pro be able to beat its immediate rival from NVIDIA, the GeForce 6800 GT, which is a cheapened version of a top-end model, too?
I will do my best to try to answer these questions by this overclocking report. But first, let’s examine the graphics card the RADEON X800 Pro came into out test lab on.
PowerColor’s card ships in this pretty bag:
A curious bug: although there’s a RADEON X800 Pro card inside, the paper band on the bag’s front pocket says it contains a RADEON X800 XT-based product. The many pockets, small and big, are full of the accessories PowerColor added to the device. By the way, the bag itself is a real transformer changing its shape into a back pack like this:
The accessories include a set of cables and adapters, a user manual, CDs with drivers and utilities, and a computer game called Hitman.
In spite of the original packaging, the graphics card itself is a very ordinary RADEON X800 Pro, designed according to the reference etalon; they only glued another sticker on the cooler, on top of ATI’s own picture:
The cooler’s heatsink consists of two parts: a thick copper base that takes heat away from the GPU, and a copper “accordion” that’s attached to this base with heat-conductive glue. The whole contraption is covered with a plastic casing, which directs the air stream along the heatsink ribs. The memory chips, like on ATI’s reference boards, have no special cooling. The PowerColor card is equipped with 256MB of GDDR3 – the 2ns chips are manufactured by Samsung.
The regular frequencies of the PowerColor RADEON X800 Pro are 475/900MHz. The heart of the card is the RADEON X800 Pro GPU from ATI:
Note that the bridge that had been cut with a laser is soldered up in the snapshot. That was a simple operation, but it didn’t help me convert this RADEON X800 Pro into an X800 XT, i.e. turn all the 16 pipelines on. I’ve already met a number of off-the-shelf graphics cards on the RADEON X800 Pro, but none of them allowed turning on all the 16 pipelines by any means. So I have come to this disappointing conclusion: there’s no way to do this modification on off-the-shelf RADEON X800 Pro graphics cards, at least at the moment.
So, we have only overclocking left. Let’s get to it right now!
The GPU power regulator on RADEON X800 Pro graphics cards is based on the FAN5240 chip from Fairchild Semiconductor. The output voltage of the regulator is determined by a digital code, applied to the chip’s VID0-VID4 inputs.
Unlike graphics cards on GPUs from NVIDIA, RADEON X800 Pro cards don’t control the GPU voltage during work: the core voltage is always 1.4v. Curiously, the digital code on the inputs of the controller chip corresponds to an output voltage of 1.0v. This discrepancy arises as the graphics card uses resistances, deviating from the typical connection scheme. For example, the resistance of the R6 resistor (the numeration according to the typical connection scheme) is not 1,000 Ohms, as indicated in the chip’s documentation, but 576 Ohms, as my measurements say.
This resistance should be changed to increase the core voltage. Here, we must increase this resistance, rather then otherwise. So, instead of just soldering up an additional resistor to shunt the R6, I had first to unsolder the R6 resistor and replace it with another. I used a 1,000 Ohms variable resistor as a replacement, soldering it instead of the R6 on wires:
The maximum output voltage of the power regulator is achieved at the maximum resistance of the variable resistor. It is about 1.7v if you use a 1,000 Ohms resistor.
So, the graphics processor is ready for my experiments. Now, I will deal with the graphics memory.
Increasing the memory voltage, I found that it practically didn’t affect the overclockability of the memory chips on the PowerColor card. However, the memory chips became much hotter (well, I wouldn’t call them cool at their ordinary frequency, either). Without any cooling of the memory chips, such overclocking threatened to damage the card utterly.
That’s why I chose not to tamper with the memory voltage – the possible reward doesn’t outweigh the risk of destroying the GDDR3 chips with their internal termination circuitry and much higher sensitivity to any overstrain compared to ordinary DDR SDRAM chips.
Anyway, you may be luckier than me with memory on a particular RADEON X800 Pro card. So if you feel like experimenting, examine the snapshot of the internal circuitry and the I/O buffers of the memory chips:
These regulators are based on the standard and widespread ISL6522 chips from Intersil, so there’s nothing extraordinary to increasing the graphics memory voltage on this card.
The core voltage being higher, the GPU consumes more power and generates more heat, nearly in direct proportion to the squared voltage increase. It means that the standard cooling system of the graphics card may not be able to handle the excessive heat during overclocking.
So I replaced the compact, but not very efficient cooler of the PowerColor card with a water-based solution from Thermaltake – Aquarius II:
Now I can be sure the GPU is not going to die of overheat – the card is ready for the tests.
I’m going to compare the PowerColor with the WinFast A400 GT TDH graphics card from Leadtek:
Besides the card itself, its package contained cables and adapters, a user manual, CDs with drivers and utilities and Splinter Cell: Pandora Tomorrow.
If the printed-circuit board of the card from PowerColor hadn’t been red-lacquered, it would turn red with envy looking at the cooling system of the Leadtek product:
I haven’t met such a solid copper design for quite a long time. This cooler looks more impressive than the notorious FlowFX from NVIDIA, but is much quieter at work. As a finishing blow at the heart of an impressive overclocker, the Leadtek engineers put a massive copper heatsink at the backside of the board, too. It is the same as on the GeForce 6800 card we reviewed recently.
The Leadtek graphics card features the GeForce 6800 GT graphics processor from NVIDIA…
…and comes equipped with 256MB of 2ns GDDR3 memory from Samsung:
The card turned to have a good overclockability – no wonder with such a cooling system! Its standard frequencies of 350/1000MHz grew to 420/1200MHz with the standard cooling system and to 450/1200MHz with water cooling.
So, we’ve got a worthy competitor to the PowerColor card. Let’s hear what the benchmarks have to say about this pair.
The testbed was configured as follows:
Software:
Considering the overall highest performance level of the graphics cards, I performed all the tests in the 1600x1200 resolution. In lower resolutions, even our quite powerful testbed often limited the performance of the graphics card proper: the CPU or the system at large would become the bottleneck.
The optimizations of tri-linear and anisotropic filtering on the two cards were enabled: NVIDIA offers them by default, and ATI’s control panel even doesn’t allow adjusting these optimizations – they are always enabled.
Before benchmarking the card, I first performed a few experiments, steadily moving from its nominal frequencies to the maximum possible. I wanted to check out how the higher core voltage and the installation of a water cooling system affect the overclockability of the graphics card. To question the card’s stability at overclocking, I ran Unreal Tournament 2004 Demo, Halo and Far Cry, one by one, without pauses. The results follow:
So, I achieved 630/1150MHz frequencies at extreme overclocking.
The maximum frequency of the graphics core was 530MHz with the standard cooler. By mounting the water system from Thermaltake, I pushed the ceiling up to 560MHz. Increasing the voltage by 0.05v, I increased the frequency by 10MHz more. After that, the maximum GPU frequency at overclocking was increasing practically linearly, according to the voltage.
The maximum voltage I supplied to the graphics core was 1.68v. Higher voltages led to the graphics card’s turning off in the hardest stretches of the tests. Yes, it did turn off, rather than hang from overheat (we’ll talk about overheat shortly). The controller chip of the GPU power regulator can keep track of the consumption current of the GPU and it just stops supplying any power after this consumption exceeds an acceptable range.
So, the maximum frequency of the graphics core was 630MHz (32.6% above the nominal frequency) at 1.68v voltage (20% above the nominal). Not bad, actually.
The die of the RADEON X800 graphics processor has an integrated thermal diode, and the card from PowerColor has a monitoring circuitry, so I checked what temperature the GPU had at extreme overclocking. To load the graphics card, I started Unreal Tournament 2004 Demo in the 1600x1200 resolution with enabled 4x full-screen anti-aliasing and 16x anisotropic filtering. Before taking the temperature data down, I was catching the bots on the Torlan map for about 20 minutes. The following diagram shows you the results:
With the standard cooler, the GPU temperature (the red line) was over 60°C even at the regular frequencies! This may not be very dangerous for the graphics card, but it is too much in comparison with the water cooling system. Interestingly, I never noticed the fan of the PowerColor card to accelerate to its maximum speed. Even when the card was overclocked, it would work at half of its capacity. Probably, that’s the way of ATI’s engineers to care about our sensitive ears – the graphics card is practically silent – but anyway I don’t think it is right to endanger the GPU.
The “ambient” temperature (the orange line in the diagram) corresponds to the sensor on the PCB. The ambient temperature goes down suddenly with the water cooling system and never gives any reason for worries afterwards, even at the highest frequencies and GPU voltage.
So, you can overclock the RADEON X800 Pro without any fear after installing a water cooling system. The GPU temperature thus remains low even after voltage adjustments. It is quite possible that higher voltages (in the range of the acceptable, of course) combined with extreme cooling systems would allow reaching higher frequencies, but I doubt the 12-pipelined RADEON X800 Pro can soar up to the level of the RADEON X800 XT.
A simple calculation tells that you should overclock the RADEON X800 Pro to 693MHz to reach the same pixel performance as the RADEON X800 XT Platinum Edition has. This may be possible as a short-term experiment, but I doubt the graphics card will live long working like that. You’ll use up the liquid nitrogen or the graphics processor too soon. :)
The cards are equals in Far Cry 1.1: the Leadtek is a little faster in the “pure performance” mode, while the PowerColor – with full-screen anti-aliasing and anisotropic filtering. This arrangement holds on at overclocking, too.
The Leadtek card, based on the 16-pipelined GeForce 6800 GPU, outperforms the RADEON X800 Pro in Halo. Even overclocking is no help for the RADEON here.
Graphics cards with ATI chips have traditionally been strong in Unreal Tournament 2004 Demo, but the 12-pipelined RADEON X800 Pro can only defeat the GeForce 6800 GT in the FSAA+AF operational mode.
The graphics cards perform similarly in the pure performance mode, where their speed was limited by the CPU, as well as in the hard mode with full-screen anti-aliasing and anisotropic filtering enabled.
There are no complex DirectX 9 pixel shaders here, on which the RADEON X800 Pro would show the advantages of its speedy architecture, and the GeForce 6800 GT wins the test. Extreme overclocking helps the PowerColor to outperform the rival but the overclocked GeForce 6800 GT is still beyond its reach.
It’s all the same in Prince of Persia: a high fill rate and a fast execution of simple pixel shaders matter much. As a result, the PowerColor RADEON X800 Pro cannot compete with the GeForce 6800 from Leadtek even during extreme overclocking.
The RADEON X800 Pro loses both modes of this test. The loss in the hard mode is rather strange, since the new chips from ATI are usually more efficient at doing FSAA and AF than the new GPUs from NVIDIA.
The graphics cards are equals in Lock On: the Leadtek is ahead in the pure speed mode, but the PowerColor is better at giving us the “eye candies”.
Homeworld 2 doesn’t use complex pixel shaders, and the NVIDIA GeForce 6800 GT is faster here. At overclocking, both cards produce the same fps rates since they are limited by the performance of the central processor. Regrettably, we couldn’t run Homeworld 2 in the “hardest” mode – NVIDIA GeForce 6800 GT cannot enable forced anisotropic filtering in this game.
So, my extreme overclocking of the RADEON X800 Pro was quite rewarding: the frequency grew by 32.6% over the regular clock rate. This means that ATI’s chips have a wide “safety margin”, at least the RADEON X800 Pro finds it no problem to work at the frequencies of the RADEON X800 XT Platinum Edition.
However, to get the performance of the RADEON X800 XT Platinum Edition, it is necessary to either enable the missing pipelines (and that seems unattainable so far), or boost the GPU clock rate to 693MHz, and this is hardly doable at home.
In comparison to the GeForce 6800 GT, which differs from the top-end GeForce 6800 Ultra in the frequencies only, the cut-down version of the X800 core, the X800 Pro GPU, looks disadvantageous: roughly, in two games out of ten it is a little faster than the competitor; in three games more they were equal; and in five games the RADEON was more or less slower than the GeForce 6800 GT.
The GeForce 6800 GT easily overclocks to the frequencies of the GeForce 6800 Ultra and more; that is, it provides the same performance as the top-end model – I can’t say that about the RADEON X800 Pro…
However, the appeal of the reviewed graphics cards will largely depend on their retail prices. A highly overclockable product but which costs much more than its competitor is unlikely to win a user’s heart. So I advise you to check the prices first and only then make up your mind about your future graphics card!