GeXcube Radeon 9600XT Extreme

Introduction

Info-Tek Corp. Taiwan, and its GeXcube brand of graphic products are still little known in Russia, and it makes sense telling a bit more about them. The company was established in 1990 as a developer and supplier of solely video cards. To date, these are only cards built on ATI chips, and the company is not going to change such strategy.

22% of the company property is owned by IT specialists of GIGABYTE, with the major package of shares owned by a Taiwanese millionaire investor. The management, marketing, engineering and quality assurance services are based in a united building within the numerous techno-parks of Taipei, by the way, in the same building where AOpen are based. The manufacture of products is performed at one of the four factories of Info-Tek in China.

Anyway, something's going wrong with the company name - the website address is www.gecube.com.tw, the brand on the boxes says GeXcube, and the nameplate on the wall says it's premises by "GIGAXCUBE" that belongs to Info-Tek Corp and partly is a division of GIGABYTE. In short, the Chinese chaos :-)

You can get an idea of the full line of GeXcube video cards on the company's product page. That's already an official line currently being manufactured. Our "guinea-pig" is still not there.

Inside the laboratory, there was developed a secret project dubbed GeXcube Radeon 9600XT Extreme - an advanced version of 9600XT with increased memory operating frequencies. The thing is that ATI Radeon 9600XT in its reference design cannot compete on par with NVIDIA GeForce FX5700 Ultra, which we saw in the review GeForce FX5700 Ultra. The work of NVIDIA for the middle-end sector of the market proved to be much more powerful. So the only possible competition remains in the pricing pressure. GIGAXCUBE, a faithful follower of ATI, did not undergo such state of affairs and decided to rescue ATI's reputation on their own :-)

So, we'll find it out how they succeeded in managing the posed task.

GeXcube Radeon 9600XT Extreme

*1200x845; 163Kb

Package bundle:

• The graphics board itself;
• Splitter adapter;
• DVI-to-D-Sub adapter;
• A thick manual in 8 languages;
  Soft:
• Driver (modification of Catalyst 3.9)
• Black Hawk Down - full version;
• Power DVD XP 4.0 - full version;

Hardly ever hardware manufacturers specify the version number for the driver written on the CD. And it's really unprecedented if the version turns out to be of the latest available. The probability of that outcome is negligible. The discs are ordered as a batch for several lines at at time and get awfully outdates by the time they are distributed among the boxes. In our case, all the further tests of the board and its rivals were done on Catalyst 3.10, but that the driver version is shown saves the user's time exploring the CD contents. But we rummaged about it and found three ATI's demos for Radeon 9800 cards taking up 80 Mb, which is nice.

Design of GeXcube Radeon 9600XT Extreme

The wiring of the board has nothing to do with the formerly reviewed ASUS Radeon 9600XT - that's the first thing to note. The board has been seriously re-worked, with its circuit technology upgraded and the voltage regulation circuit enhanced.

*1200x833; 197 Kb	*1200x816; 252 Kb

The board offers 128 Mb of DDR memory, 2x/4x/8x AGP, and a standard set of output connectors: one analogous, one digital, and one TV-OUT. No VIVO chip is installed.

Memory:
The video card is equipped with 128 MB DDR memory packaged in 8 chips (4 chips on each of the sides - front and rear sides of the PCB) of 128-bit memory bus.

The memory is produced by Samsung (K4D263238E-GC25), offers a 2.5 ns access time, which is equivalent to approximately 400 MHz of memory operation (800 MHz), but the memory is set to 350 MHz (700 MHz). Therefore, there is a small margin for memory overclocking, which we'll definitely verify. Remember that Radeon 9600XT reference cards offer 2.8 ns access time memory running at 300 MHz (600 MHz).

But the graphic chip is intact and, as per the specifications, runs at its intended 500 MHz.

Cooling system:
The cooling system is a block whose dimensions are reduced to fit the dimensions of the higher-end Radeon 9800XT:

*1200x908; 141Kb

The construction material is aluminum, but the fastening leaves much to be desired. As you can see on the screenshot, the whole system is fastened with two plastic pins.

The very strange 1-mm thick spacers between the memory chips and radiator led us into a dead end! The construction material is sort of a rubber plastic whose thermal conductivity raises great doubts. What on earth is this fashion these days? In chase of manufacturability of the assembly process for video cards on the production line, it's much easier to botch rubber items than devoting more time thoroughly applying thermal paste. Nevertheless, this thermal interface did a good job in overclocking the memory, which even was surprising. You can of course tear off those rubbers ... if you apply a bit more efforts, but this results in a huge gap between the cooling plate and the memory chips, i.e. you would have to glue metal spacers on the spots of thermal contact with KPT paste. All in all, this a job for confirmed overclockers and does not take a few seconds. In exchange you will get speedy memory and a loss of guarantee servicing.

Another difference from 9600XT reference cards is additional cooling of memory on the reverse side.

Now it's time we estimated the performance versus the standard Radeon 9600XT and the rival FX5700Ultra (pages 4,5), and then determine its place in the whole line of modern video cards (p.6).

Test configuration

Test cards

Test software:

1. Synthetic benchmark:
   
2. 3DMark2003 v340;
3. 3DMark2001SE v320;
4. Village Mark (cut-off of hidden surfaces)
5. Codecreatures v1.0.0 (a DirectX 8.1 application, shaders on, Hardware T&L);
6. AquaMark 3 (DirectX 9.0, Vertex Shaders 1.1/1.4/2.0, Pixel Shaders 1.1/1.4/2.0, Hardware T&L, AquaMark3 Triscore mode);

   Old games:
   

7. Comanche 4 Bench
8. Unreal Tournament 2003 Demo (Direct3D, Hardware T&L, vertex shaders, Dot3, cube texturing.);
9. SSAM The First Encounter (OpenGL);

   New games:
   

10. Gun Metal Benchmark 2 v1.20s (a DirectX 9.0 benchmark, Vertex Shaders 2.0, Pixel Shaders 1.1, Hardware T&L);
11. X2: The Threat Demo (Direct3D, multitexturing, Dot3, running in the benchmark mode embedded in the demo version);
12. Final Fantasy XI Official Benchmark 2 (a benchmark for assessing the performance in the future game Final Fantasy XI. The developers haven't presented any data on the gaming engine);
13. HALO: Combat Evolved 1.2 (DirectX 9.0, Vertex Shaders 1.1/1.4/2.0, Pixel Shaders 1.1/1.4/2.0, Hardware T&L);

Overclocking

The video card was overclocked with the RivaTuner utility. The operation stability in overclocking was verified with the comprehensive benchmark Aquamark3. First, the memory was overclocked. The maximum attained value is 800 MHz, which is a standard operation mode for the 2.5 ns memory. Probably, such weak memory overclocking is a result of placing that 1-mm thick chewing gum as the thermal interface. Since during the normal cooling of this type of memory it works fine even at 860 MHz. (The record value attained was 920 MHz, but for that I had to use two fans, which is unacceptable in normal conditions).

Then, the most interesting occurred - with the memory set to 800 MHz, the core wouldn't run in the overclocked mode at all. That is, the system was overheating and then hung at the very start of the test. I even had to revert the memory speed to the original 700 MHz and try overclocking the core separately. In the end, I produced the following values:

Even the first step of overclocking the core (513 MHz) allowed to reach values attained through the maximum memory overclocking. Then the gap was only expanding...

Overclocking findings: the best combination for attaining the maximum performance of the card is overclocking the core leaving the memory as is. A curious finding, isn't it? Considering the operation margin for the memory.

By no means we urge our readers to overclock their cards and put their money at risk. All in all, we torture the specimens just to prevent you from doing that and warn you of not exceeding "some critical values". Otherwise, you will smell the so familiar scent of burnt chips, so God forbid..

The "overclocking" section in reviews is meant for one purpose only - to determine the work capacity of a video card recommended by the developers and thus its operational stability, longevity and reliability. The less the work capacity margin, the better. As you see, "GeXcube Radeon 9600XT Extreme" has of substantial performance margin.

All the further tests were performed under nominal operating modes of the core and memory of video cards participating the boards.

Synthetic benchmarks:

3DMark 2003 v340

As we can see, "GeXcube Radeon 9600XT Extreme" was able to catch up with "NVIDIA FX5700 Ultra". Remember that it was attained through the departure of the "canonical standard" defined by ATI for video cards of XT series by merely 100 MHz of the memory frequency.

Here is a detailed view of results produced in 3DMark 2003 v340:

3DMark 2001SE

Let's run a test imagining there is no DX9 at all. Then we get the following. FX5700 Ultra and Radeon 9600XT Extreme start up very close to one another, but the "margin of safety" in NVIDIA produce is greater - the use of DDR2 memory makes itself felt.

Codecreatures

The small performance boost in the Extreme version is achieved, which is definitely not enough to compete with FX5700 Ultra.

Village Mark

This is a benchmark measuring the efficiency of invisible surfaces cut-off. The result for GeXcube and standard XT is the same, and the ATI RV360 does a better job coping with this task than the NV36.

Aquamark 3

This benchmark was created in close cooperation with NVIDIA. But to some reason it's just this benchmark that shows not the best with their video cards.

Tests with real-world gaming benchmarks

Comanche 4 Bench

Wow! In the two most popular resolutions, the new kid "Extreme" and the champion FX5700Ultra are on par.

Among the advantages of NVIDIA cards is of course the less performance drop under aggressive quality settings.

Unreal Tournament 2003

This is a favorite game for most game fanciers. Our both competitors are about on par at that.

SSAM The First Encounter

This is an OpenGL test. NVIDIA takes a lead.

X2: The Threat Demo

The "Ultra Shadows" technology implemented on the NVIDIA GeForce FX 5700 Ultra chip does a much better job handling the stencil shades.

Final Fantasy XI Official Benchmark 2

In this benchmark, all depends mainly on the fill-rate. The "Extreme" version failed to catch up with GeForce FX 5600 Ultra, albeit added enough score points as compared to the traditional Radeon 9600XT.

Gun Metal Benchmark 2

A test of implementation efficiency for version 1.1 pixel shaders. NVIDIA takes a lead.

HALO: Combat Evolved

A test of implementation efficiency for version 2.0 pixel shaders. That's a surprise, isn't it? :) Our "guinea-pig" rushed forward and caught up with FX 5700 Ultra!

Final Words

Already after the first tests of GeForce FX5700 Ultra it became evident that NVIDIA decided to gain revenge at any price. Its direct competitor, Radeon 9600XT lost at all the benchmarks (except AquaMark 3 and Village Mark).

"Regain the reputation at any price!" - even at the expense of income drop per sales unit (GeForce FX5700 Ultra is much more costly to produce than Radeon 9600XT). The difference was such that it was hard to imagine a card on the base of RV360 can be created to approach the NV36. But GigaXcube did that. The card takes quite an honorable place in the common line thus shrinking the gap between the yesterday's high-end and today's mainstream:

But if the cooling system is modified, then the overclockers will get a true pleasure watching the results of their skilful hands.

The "GeXcube Radeon 9600XT Extreme" only in a day or two may hit the retail. Therefore, its "unofficial price" has not yet been announced so far. Reminding it that today's price for a standard Radeon 9600XT is somewhere within 200-210$.

Pros:

• Modified PCB design;
• Improved cooling system;
• Use of faster memory.

• Thermal "chewing-gum" instead of thermal paste.

The GeXcube Radeon 9600XT Extreme video was granted by GigaXcube
GeForce FX5700 Ultra - by NVIDIA
ASUS Radeon 9600XT was presented by ASUS

VGA Roundup `2003:

Read more on this topic: