The Gigabyte GA-630.  New and Hot!
Observations on Three Banshee AGP Video Cards

As I mentioned in the news yesterday, I just received a Gigabyte GA-630 Series Banshee AGP card.  It’s the sixth Banshee card I have tried out of the dozens now on the market.  The Gigabyte card looks virtually identical to the Diamond Monster Fusion AGP card, with very similar chip layout and sporting a fan/heatsink combination on the graphics chip. The third AGP card I have, the Creative Labs 3D Blaster Banshee, has a different chip layout and no fan on the graphics chip. What caught my attention was the Quake II timedemo score I got after installing the Giga and its drivers.  67 FPS is the highest timedemo (demo 1) score I have observed with a card straight out of the box.  With that encouragement I decided to put the card to the test and compare it to the two other AGP cards I have currently.  This is not a full review.  I will focus on clocking issues to examine the speed of this new card in some 3D benchmark and game results with more about the card to follow later. 

Test system
Abit BX6 motherboard, Celeron 300A CPU at 450MHz (100MHz FSB), AGP set 2/3 (66MHz), 128MB PC-100 SDRAM, Maxtor 17.2GB UDMA drive running under Win98 SR1 beta.  All cards used 3Dfx Release 1.0 drivers.  Desktop set to 1280x1024x16-bitx75Hz.  All screen modes set to 75Hz.  Vsync was disabled for all tests.  There were a few background tasks running, but I have determined they do not impact results.

First up is 3DMark 99, a new but popular benchmark.

3DMark 99 (Lite)
I started with “default” clocking using Entech's Powerstrip and worked my way up a step at a time until the cards failed.  I took big steps at first with the Diamond and Gigabyte cards because they tolerated higher clocking on the desktop.

The Creative card would not even exit Powerstrip above a memory clock setting of 126, locking up the desktop.  Failure with this card during 3Dmark99 consisted of corruption of the first graphic test, a driving scene.  Corruption consisted of white specks and dark, flashing lines scattered around the screen.  The test would run but the graphic artifacts were unacceptable and not seen at any time with the other cards. I re-ran all tests for the Creative card using a 486 fan attached to the heatsink. This allowed a 7 MHz increase in the graphics clock, but no change in the memory clock, as you might expect.

You can see that the Gigabyte and Diamond cards were able to run at the maximum available memory clock setting of 133MHz.  The Gigabyte tolerated higher graphics clock settings, I believe due to its slightly different fan and heatsink design.  The Diamond card has a sharp looking fan sunk into heatsink, whereas the Gigabyte fan sits on top of the heatsink in a more traditional design.  The effect of the Diamond design is that there are almost no heat-radiating fins on the heatsink or around the fan.  It’s like a copy of the TennMax Lasagna design, except that the TennMax has an array of fine fins around the fan.  The bottom line is that I don’t think the Diamond design cools the graphics chip as well.  However, you can see that the impact of the earlier graphics clock failure with the Diamond card is negligible – less than a 1% difference.

You may ask what is the impact of the fan? When the fan disconnected from power, the Gigabyte locked up at 117 instead of 123MHz graphics clock setting in 3DMark99.  So, it adds about 5% the to the graphics chip's clocking ability.  However, this adds only about 1% to the overall performance of the card.  The Creative card received a bit more benefit from an added on fan.

Now, let’s look at game performance.  I report frame rate results at the highest settings that each card ran successfully without screen artifacts or lock ups.  The clock settings that ran all resolutions successfully are in parentheses after card name.

Forsaken Demo

The Creative card displayed minor artifacts in the menu that worsened and spread to the game if the settings were any higher.  The added fan allowed the Creative to run 9 additional MHz for the graphics clock, but there was no benefit in Forsaken scores.

Incoming

Incoming was a bit more challenging for all the cards.  It required a step down from Forsaken settings at resolutions higher than 640x480.  Again, the successful settings are in parentheses.  The Creative card developed noise (dots) while the other cards locked up if pushed higher.  With a fan, the Creative card was able to run a 7% higher graphics clock, including a modest increase in memory clock, but the performance gain was negligible.

Turok

Turok demo’s benchmark really bothered the Creative card, requiring several setting reductions before it could even get past the 3Dfx logo.  Again, adding a fan to the Creative card boosted clock rates and provided a 7% performance boost.

Quake II timedemo demo1

Quake 2 demo1 ran at the 3Dmark settings at 640x480 but required a step down at higher resolutions to avoid lock ups.  The Creative card even locked up in the console prior to running the demo.  The 10MHz boost in graphics clock when a fan was added to the Creative card did not translate into any real speed increase.

Quake II timedemo massive1

All the Banshees handled Brett Jacob’s massive1demo very well, posting results nearly as good as with demo1.

Unreal timedemo

Unreal did not like the Creative card at all, requiring the lowest clocking of any of the games tested.  I just did a few runs through the timedemo benchmark.  This may not be as challenging as the game itself.

Total game score
How about a total score?  Adding up all the frame rates for all resolutions we get:

Conclusions
So my conclusion is that the highest practical settings for these cards are:

The usable clock settings for the Creative AGP card do not differ noticeably from the PCI version, although it uses slower SDRAM.  If you do not play Unreal, you may get away with a few points higher on the memory clock.  Also, since for practical reasons you should tune your system to the lowest common denominator, if you play Unreal or something equally unkind to the Creative Banshee, your overall performance will have to suffer in other situations.  The difference in total game score above for the Creative card is about 10% (8% with a fan).  In the proper perspective, this is not very significant.

Until I tested the Diamond, and now the Gigabyte, AGP cards, I saw no solid reason to prefer AGP over PCI in the case of the Banshee. It is clear that with good design and proper cooling, the SGRAM version can run away from the other.  The speed difference is larger than I expected, although still not a noticeable amount outside of benchmark numbers.

Did putting a fan on the Creative card bring it up to the other two in terms of settings?  Not in this evaluation.  Even though the Creative memory is rated at the same 8ns spec as the other AGP cards, it does not appear to be able to perform at that level. The fan did allow a 7-10MHz boost in the graphics chip clock.  Although there was a measurable benefit in the 3DMark 99 results and Turok, there was none elsewhere, and the overall performance increase was 2%.

In conclusion, the new Gigabyte card is a real pleasant surprise, beating the reigning king of the Banshee AGP cards, the Diamond Monster Fusion.  Although the margin was very small, this is a notable accomplishment for one of Gigabyte’s first entries into the video card market.  Gigabyte has applied their board-building know-how and quality control expertise to a new area and has hit the target right off the bat. 

Look for the Gigabyte GA-630 Series to begin entering the retail channel any day.  Suggested retail price is about $110 USD.  Go get it!

Perspective
This review is about overclocking and the stability of several cards under various levels of stress.  The ability to adjust clocking of video cards has become a standard feature, as most vendors now include such adjustments in their display utilities.  I'm not sure they actually like doing this, but the market seems to demand it, and it is reviews like the one I wrote above that encourage it, I suppose.  So, I want to try to put this in a broader perspective. 

Although it is remarkable to see a card like the Gigabyte handle elevated clock rates so well, recognize that the real benefit is hard to see. In the same vain, it is disappointing to see the Creative card, which sports a clocking slider in its utilities, run below memory spec.  However, in a blind test, I am certain no one would be able to tell one card from another in actual use, without a frame rate read out.  Eight to 10% is simply not discernable.  Even in Unreal, which was the most demanding, and gave the top cards a 12 to 18% advantage, no one would know the difference were it not for the frame rate readout.

I am writing this perspective a few days after doing the review because although I do overclock chips, I am not a fanatical proponent of it.  Re-reading my review, I found it would be hard to determine that.  The review developed this way because I was struck by the benchmark speed of the Gigabyte card, and the clock rate capabilities were the explanation.  I took the opportunity to explore this and learn more about the effect of clock rates on Banshee performance.  It certainly distinguishes the Gigabyte card, but when I received a question from a reader as to whether he should replace his Creative card with a Gigabyte card, I realized I had painted the wrong picture. NO!  First off, the real benefits would be very small and not discernable, as I said.  Secondly, there are lots of things to look at besides the overclockability that may be more important. 

For one, I have found the 2D Windows image of the Creative cards I have to be extremely sharp and clear at high resolution - perhaps clearer than the others cards.  The is VERY important because with these 2D/3D combo cards you have to live with the desktop everyday, unless all you do is play games.  The problem is that this is VERY hard to document objectively.  People use descriptors like sharp, crisp, vibrant, clear, good text legibility, etc. to convey what they see, but there are no obvious benchmarks that I know of.  But the differences among video cards can be obvious, especially at high resolution (over 1024x768).

Recently, PC Magazine published a terrific review of 2D/3D cards and the online version included some fascinating signal quality oscilloscope traces and comments about them (most of this was unfortunately left out of the print version in the December 1, 1998 PC Magazine).  They pointed out considerable variability in signal quality among the various Banshee cards that they tested, and linked this to visual anomalies that they observed, apparently the result of different circuitry and components.  The 3Dfx reference design allows for a lot of vendor choices and they don't really have to follow the reference design, for that matter.  Neither the Creative or Gigabyte card were in this review, but the Diamond Monster Fusion card received overall very high marks for signal quality.  Other Banshee cards, including the Guillemot, Quantum3D, and Meyabyte cards, did not fare so well.  Many of the "defects" may not be noticeable outside of a high-resolution CAD or photo-editing environment.  Also, as with ALL video card reviews, including mine, these results are based on a sample of one.  Despite the precision of modern electronics, this must make you look at reviews as just bits of information and not conclusive evidence.

So, take my review for what it is and look at the card choices you make based on lots of input and evaluation.  Don't let one variable sway you too much unless that is an extremely important variable to you or the only thing that distinguishes the cards from one another.

Published 01/08/99
Revised 01/09/99 (added data and comments about Creative card with added fan)
Revised 01/010/99 (added test system description and perspective)

Back to 3Dfx Banshee TroubleShooting and Game Guide

This page is last updated Friday, June 09, 2006
[home]

Created solely by William D. Ball © 1998-9
All rights reserved.
Site design by Ian Chiu