Gigabyte GeForce 6800

Author: Tom K
Editor: Howard Ha
Publish Date: Tuesday, December 28th, 2004
Originally Published on Neoseeker (
Article Link:
Copyright Neo Era Media, Inc. - please do not redistribute or use for commercial purposes.

Update (12/30/2004): The review sample as shipped to us by Gigabyte made use of a faulty BIOS which was enabling only 8 of the 12 pipes of the 6800 core. We were made aware of this by a kind member of our forums, and we promptly re-flashed the card with the latest BIOS (F1 -> F3). As such, we have re-run all of our benchmarks and updated the charts. The previous scores of the 8-pipe 6800 are also provided on some of the charts for comparison. We are quite disappointed at Gigabyte's failure to mention such a significant piece of information. Had we not been informed of it by a 3rd party, some of our previous poor benchmark scores which turned out to be good could have definitely influenced someone's buying decision.

There is practically no price point at which a consumer cannot find a GeForce 6 family product - NVIDIA announced that they would be creating a top-to-bottom lineup of products, all based on the NV4x core. At the entry level is the 6200, in the mid-range is the 6600/6600 GT, and at the high-end is the 6800/6800 GT/6800 Ultra. When we say "mid-range", we don't just mean regular run-of-the-mill average performers -- the $200 price-point 6600's can hold their own, even surpassing last generation's high end parts like the 9800XT in many instances

In the 300$ range of NVIDIA's line up is the often looked over GeForce 6800. At the 400$ range is the very popular 6800 GT and at the 200$ mark is the 6600GT AGP and sometimes it seems like the 6800 has been neglected. Up for review today is Gigabyte's GV-N68128DH, a passively cooled AGP GeForce 6800 with 128 MB of memory, utilizing heat pipes and massive heat sinks on both sides of the card.


5700 Ultra 5950 6200 6600 6600 GT 6800 6800 GT 6800 Ultra
Architecture NV3x NV3x NV41 NV43 NV43 NV40 NV40 NV40
Manufacturing Process 0.13u 0.13u 0.11u 0.11u 0.11u 0.13u 0.13u 0.13u
Transistor Count 82 M 130 M ? 146 M 146 M 220 M 220 M 220 M
Pipelines 4x1 4x2 4x1 8x1 8x1 12x1 16x1 16x1
TMUs/Pipe 1 2 1 1 1 1 1 1
Clock Speed 475 475 300 300 500 325 350 400
Fillrate (megapixels) 1900 1900 1200 2400 4000 3900 5600 6400
Memory Interface 128-bit 256-bit 128-bit 128-bit 128-bit 256-bit 256-bit 256-bit
Memory Size 128 256 128 128 128 128 256 256
SLI-Capable No No No No Yes Yes Yes Yes

Looking at the last three columns, it is obvious that the plain-vanilla 6800 is merely a clock-reduced NV40 with four of its pipelines disabled. This core variation still manages to edge out the 6600 GT, except where fillrate and raw core clock speed are concerned -- at least on paper. Since the plain 6800 is still a 6800, it sports a full 256-bit memory interface giving the 6800 a fairly hefty advantage in the memory bandwidth department.

The Card

As was the case with the Gigabyte GV-NX66T128D, Gigabyte has opted to go with a reference design. The board layout has remained more or less identical, with the exception of a few more capacitors when compared to our BFG GeForce 6800 GT. The colour of the PCB is Gigabyte's trademark blue, and although the board is a reference design, the cooling solution is definitely not. The regular heatsink/fan unit of the 6800 series has been replaced with a large passive solution (the focus of this product). Other than that, we do not notice any other obvious deviations from the reference design. This board ought to be a run-of-the-mill plain-vanilla GeForce 6800 in all other respects.

Here is a photo comparison between the Gigabyte GV-N68128DH and the BFG Geforce 6800 GT:

Connectors on the card include a Molex connector for external power, a VGA connector, a DVI connector, and an S-Video Out connector. The card is very beefy, weighing approximately 2 to 3 times more than our BFG GeForce 6800 GT -- this is most definitely attributed to its passive cooling solution.

The passive cooling solution is very interesting indeed. On the GPU-side of the card, the heat sink spans just over half the length of card, centered over the GPU. After close inspection, it seems that the memory is not cooled by the heatsink. The GPU core has a copper heatspreader that is attached directly to the heatpipes. The majority of the heat dissipation seems to happen through the rear heat sink. Two heat pipes, which make direct and close contact with the copper heat spreader assembly on the GPU, loop around the card and take heat to this rear heat sink, which spans approximately 4/5 of the entire area of the card. . The sides of the copper heatspreader make contact with the lower heatsink but the rest of it does not make physical contact with any of the components on the rear of the card -- instead, it is suspended close to the edge entirely by the heat pipes.

A close up of the copper heatspreader covering the GPU on the left - the shiny part on top is the heatpipe. Notice that only the edges of the spreader touch the bottom heatsink

The Bundle

Our Gigabyte GV-N68128DH came to us in an attractive retail box advertising the key features of the product, as well as the bundled software. The bundled hardware consisted only of a DVI-to-VGA adapter, and an S-Video-to-RCA converter. We would have liked to see something more, such as a video dongle like the one provided to us with the GV-NX66T128D, since this is after all a fairly high-end product.

Gigabyte has bundled a fairly impressive software collection with this product. Along with the de-facto driver CD, we also found three games -- Joint Operations: Typhoon Rising, Thief 3: Deadly Shadows, and Spell Force: The Order of Dawn, all of which are games from this year. PowerDVD 5 is also provided on the driver CD. It is nice to see fairly recent (and full) versions of games.

Last but not least, we have the manual. While most of us rarely even open the manual, these are very useful for the newbie who has never installed a video card. At a total of 36 pages, it is a commendable and much-appreciated effort in an era where instructions seem more like an afterthought.


Our benchmark system consisted of the following:


Both the Gigabyte GV-N68128DH and NVIDIA GeForce 6600GT AGP were tested with Forceware 66.93 drivers. We will be doing a direct comparison between the plain 6800 and 6600 GT as they should have similar performance characteristics and they are not too different in price. We are curious to see how much better (or worse?) the passive-cooled 6800 is compared to the mid-range 6600 GT. The ATI 9800XT is tossed in there also - the 9800XT is in the same general price range as the 6800 as ATI does not currently have a R4xx based AGP card at the 300$ mark.

*edit The N68128DH came with BIOS revision F1 which for some reason only has 8 pipes enabled. We will be correcting the benchmarks shortly but keep in mind that the 6800 as is currently represented, is crippled. Sorry for the inconvenience.


3DMark 2005

3DMark 2005 is one of two synthetic tests in our series of benchmarks. It is an extremely heavy user and abuser of the latest Direct3D technologies. The scores 3DMark gives are really only useful when comparing theoretical performance between two cards - they may or may not necessarily be representative of "real-world" performance.

Here were compare fill rate, pixel shader performance, and vertex shader performance:

The 6800 has the fastest single-texturing fill rate, and is within a hair of matching the 6600 GT in multi-texturing. For pixel shaders, the 6800 beats the 6600 GT by 7 FPS, coming out on top. The 9800 XT is put to shame in this test by the 6800, pulling in almost half the speed. For vertex shaders, the 6800 and 6600 GT are virtually neck-and-neck, with the 6800 pulling ahead by a few FPS. They are both bested by the 9800 XT, which makes a comeback.


Aquamark is the second synthetic benchmark in our series of benchmarks. We ran two sets of benchmarks at resolutions of 640x480, 800x600, 1024x768, 1280x1024, and 1600x1200, with one set utilizing trilinear filtering and the other 4x anti-aliasing/8x anisotropic filtering.

The 6800 really kicks some butt here, coming out on top in every combination of settings. The 9800 XT, which it seems was top of the line just yesterday, falls in last place when AA/AF are not in use, but manages to beat the 6600 GT by a miniscule margin with AA/AF. All of the cards scale almost identically as the resolution and AA/AF levels are increased.

Call of Duty

The Quake 3 engine is archaic by today's gaming standards, being superseded by the Doom 3 and Half-Life 2 engines, but it still powers many popular games, such as Call of Duty, Jedi Knight 2, and Wolfenstein: Enemy Territory. These games push this engine far beyond what we originally saw with Q3A.

The Gigabyte 6800 throws us some very impressive numbers in Call of Duty, even at 1600x1200, pulling ahead of every other card by a sizable margin. The 9800XT just gets a plain kick in the pants. Going from 1024x768 to the aforementioned resolution yields a decrease that is lower than we would have expected, which seems to indicate that this benchmark is more CPU bound when AA/AF are not in use. The story changes under 4xAA/8xAF, however. The 6800 leaves the competition literally in the dust at 1024x768, and then increases that lead by another mile by pulling in over twice the framerate than the tightly-knit group behind it at 1280x1024. When 1600x1200 comes around, the 6800 hits a brick wall. Performance declines so severely that it is then on-par with the 6600 GT, and the 9800 XT manages to beat both the two NVIDIA cards by about 5 FPS, due to its 256 MB of memory. For users looking to play at high resolutions with AA/AF, it is recommended that they look carefully at a 256MB card.

Jedi Knight 2: Jedi Academy

The 6800 gives the other competitors a run for their money, and pulls in an impressive showing even at 1600x1200 with 4xAA/8xAF. Under those AA/AF settings, the Gigabyte 6800 consistently manages to beat both the 6600GT and the 9800XT in all resolutions on DM_TASPIR, with an average margin of 25 FPS over the 6600GT. That margin is lower for non-AA/AF settings, but it is still there for all resolutions.

Taking all of these results into account, it is reasonable to deduce that Call of Duty is a much more CPU-bound benchmark than is Jedi Knight 2. Scaling of scores as we went up the resolution ladder was fairly linear on all of our video cards in JK2, while in CoD the 6800 seemed to flatten out slightly. When we turned on the AA and AF, it got shot down at 1600x1200.

All in all, the Gigabyte GeForce 6800 is definitely sufficient for these two games, as long as you don't crank up the resolution too high while keeping AA/AF settings at a maximum. The sixty frames per second at 1600x1200/4xAA/8xAF in JK2 is definitely playable, but that could very possibly drop to unplayable frame rates in heavy firefights. Nevertheless, an impressive showing.

Unreal Tournament 2004

Unreal Tournament 2004 is the latest game in the UT series, adding vehicles, new game types, and new maps to the very popular UT2003. It is a fast-paced shooter that makes heavy use of Direct3D.

The general trend we saw with CoD and JK2 seems to change at this point. All of the cards start out in a tightly-knit pack at 1024x768 and no AA/AF, and then fan out slightly at 1600x1200, with the 6800 leading the pack. With 4xAA/8xAF, the NVIDIA cards all scale in the same fashion, with the 6800 besting the 6600 GT by an average of 6 FPS. The 9800 XT starts gaining momentum at 1280x1024, and finally finishes almost 30 FPS ahead of the competition at 1600x1200. The 256 MB of memory really pays off in UT2004 when AA/AF are in use. Performance on the 6800 remains at playable levels on all resolutions with no AA/AF, and manages to stay playable up to 1280x1024 with 4xAA/8xAF. At 1600x1200, 4xAA/8xAF are not recommended for any of the NVIDIA cards.


Halo, while not being the newest of games, still manages to bring some video cards to their knees when paired with a high resolution and AA/AF.

Overall, not too shabby a showing in this game for the 6800, except where 1600x1200 is concerned. It's interesting to note that the 6600GT scores ~10 FPS higher on average with 8xAF than the 6800 with no AF. Resolutions of 1024x768 and lower with 8xAF ought to be very playable, but anything higher and performance may drop to sub-playable levels under heavy action.


Here we have another popular Direct3D benchmark.

There really seems to be no concrete trend: the 6800 is beaten by the 6600GT with no AA/AF, but the tables turn once again when AA/AF are turned on. The plain 6800 and its low clock speed just doesn't do it any good at no-AA/AF performance. None of the three cards did too well at 1600x1200 with 4xAA/8xAF, with all of them falling below 30 FPS.

Splinter Cell

Splinter Cell has also grown to become quite a popular Direct3D benchmark.

The Gigabyte 6800 is consistently faster than the 6600GT by ~5 FPS on Tbilisi 1_1_1 at all resolutions, and ~10 FPS on Tbilisi 1_1_2. The 9800XT falls behind both cards.

Doom 3

We all know what Doom 3 is. We also know that it is the most advanced OpenGL game and benchmark to date, probably rivaled only by Valve's "Source" engine and Far Cry's Crytek engine, both of which are Direct3D-based. Doom 3 and Half-Life 2 are the de-facto benchmarks for modern graphics cards today, as the engines that power these two games will continue to power games for the next few years.

The 6800 is beaten by the 6600GT in non-AA/AF performance, but comes out on top when 4xAA/8xAF are enabled. Memory bandwidth, something the 6800 has an abundance of, is more important when it comes to anti-aliasing and anisotropic filtering, and that helps the 6800 overcome its otherwise hurting core clock speed. All performance figures are at playable levels for this game, with the exception of 1600x1200 4xAA/8xAF, which is at that "under heavy action, it'll get bad" level.

Half-Life 2

As mentioned above, the recently-released Half-Life 2 is a de-facto benchmark for Direct3D performance, as the engine, along with the Doom 3 engine, will be powering games for years to come.

The 6800 pulls in absolutely awesome scores in this game, beating every other card in every test combination, with the exception of 1600x1200 resolutions + AA. Neither of the NVIDIA cards have a chance of beating the 9800 XT at such settings, as they simply do not have the memory.


What overclocking?

We weren't surprised to find that the Gigabyte GV-N68128DH has absolutely no overhead which we can use to overclock the core or memory. It is literally at the edge of its limits, and raising the core clock speed even 5 MHz caused lock-ups in 3DMark. After a session benchmarks at regular clock speeds, board temperatures as reported by a utility application exceeded 70C. That is hot.

Gigabyte has pushed this puppy as far as it will go. On the temperature discussion note, we put the Gigabyte through a more rigorous test than usual - we mounted the Gigabyte in a closed case for the duration of testing with no case fans - only the OCZ Modstream and a Gigabyte 3DCooler. We're happy to report that we did not see stability problems. Ambient room temperature was a balmy 25C.


Before we get to the meat of the conclusion, let us just say one thing:

We had to get down closer to the ground and check whether our test system was turned on when we fired it up with the GV-N68128DH, our OCZ ModStream 520W power supply, and the Gigabyte 3D Cooler-Ultra GT at the lowest setting. Hearing, or rather, not hearing, the system was a major shock, but also quite a pleasant surprise. Are you quiet-computer enthusiasts out there listening?

Gigabyte has done an outstanding job in creating a video card that, while still managing to play the latest and greatest games at very acceptable levels of performance, is completely inaudible. This product is a downright impressive addition to that silent computer you've been yearning for. Coupled with a few other products, such as the OCZ ModStream or Thermaltake Silent PurePower, and an adjustable CPU cooler like the Gigabyte 3D Cooler-Ultra GT, you'll be able to comfortably sleep at night while avoiding the pounding headache the next morning from the jet-engine computer under your desk.

During our laborious "testing", we comfortably played through two hours of Half-Life 2 at 1024x768, 4x AA, and 8x AF while noticing only a slight slow-down during a heavy firefight. This product is not the latest and greatest screamer, but if it can comfortably play virtually every game out there, Doom 3 and Half-Life 2 included, with all of the details turned up while producing absolutely no noise, it is a good compromise for those who are not interested just in flat out performance.


Copyright Neo Era Media, Inc., 1999-2014.
All Rights Reserved.

Please do not redistribute or use this article in whole, or in part, for commercial purposes.