The Holiday Stop Gap: GeForce 6800 GS
by Derek Wilson on November 7, 2005 9:00 AM EST- Posted in
- GPUs
The Card, The Test and Power
The 6800 GS cards will all sport NVIDIA's NV42 core. Previously, NV41 and NV42 cores were used on vanilla 6800 boards. The main difference between the two parts is that the NV41 is run on IBM's 130nm process (and was NVIDIA's first native PCIe part), while NV42 uses TSMC's 110nm process. And where there is a process shrink with no other major changes, higher clock speeds are more accessible.
Thus, the 6800 GS is physically the same as a vanilla 6800 with a higher core clock speed (425) and paired with GDDR3 at a 1000MHz data rate. From the 6800 GT, we see a decrease of 25% in pixel pipeline, but a ~21.4% increase in core clock speed. The net result is a theoretical core performance decrease of only 9%, but since NV42 doesn't require a bridge chip the results may be even closer than that. This puts the 6800 GS in the same class as the 6800 GT (but at a much lower cost).
The card itself is not remarkable looking compared to any of the previous generation 6800 cards out there. We see the familiar SLI connector on top of the card, stock HSF, and combination of one analog and one DVI port. The 6800 GS really does seem to be the result of someone realizing that the vanilla 6800 with an NV42 under the hood and some GDDR3 could revamp NVIDIA's midrange with little added cost.
The main reason this card will cost less than the 6800 GT is die size. Being manufactured as a 12 pipe / 110nm chip, NV42 will definitely be smaller and more cost efficient than the chip powering the 6800 GT. As far as we know, 6800 GT still uses the NV45 which features 16 pipes and is manufactured on a 130nm process. Combine this with the fact that NV45 is bridged from AGP to PCIe on package, and there is no question about the cost difference between silicon solutions.
What remains to be seen is how the current market will react. Our initial probes indicate that the 6800 GS will be a very limited offering with its cycle ending in Q1 06. With the potential to perform as well as a 6800 GT for the price of a vanilla 6800, the 6800 GS seems to warrant a longer shelf life than our sources indicate. Either way, the 6800 GS and ATI's X800 GTO parts point to limited run cards with excellent value growing in popularity. As with any good part, we would like to see the products linger a little longer, but the introduction of these late blooming parts is good for consumers in the market for a midrange card no matter why NVIDIA and ATI created them.
For our performance tests, we used this system:
Test Hardware
We measured power draw at the wall for each card as well. Load was generated by running our Splinter Cell: Chaos Theory benchmark and observing the maximum power draw. The 6800 GS falls somewhere in the middle of the pack in terms of power draw.
The 6800 GS cards will all sport NVIDIA's NV42 core. Previously, NV41 and NV42 cores were used on vanilla 6800 boards. The main difference between the two parts is that the NV41 is run on IBM's 130nm process (and was NVIDIA's first native PCIe part), while NV42 uses TSMC's 110nm process. And where there is a process shrink with no other major changes, higher clock speeds are more accessible.
Thus, the 6800 GS is physically the same as a vanilla 6800 with a higher core clock speed (425) and paired with GDDR3 at a 1000MHz data rate. From the 6800 GT, we see a decrease of 25% in pixel pipeline, but a ~21.4% increase in core clock speed. The net result is a theoretical core performance decrease of only 9%, but since NV42 doesn't require a bridge chip the results may be even closer than that. This puts the 6800 GS in the same class as the 6800 GT (but at a much lower cost).
The card itself is not remarkable looking compared to any of the previous generation 6800 cards out there. We see the familiar SLI connector on top of the card, stock HSF, and combination of one analog and one DVI port. The 6800 GS really does seem to be the result of someone realizing that the vanilla 6800 with an NV42 under the hood and some GDDR3 could revamp NVIDIA's midrange with little added cost.
The main reason this card will cost less than the 6800 GT is die size. Being manufactured as a 12 pipe / 110nm chip, NV42 will definitely be smaller and more cost efficient than the chip powering the 6800 GT. As far as we know, 6800 GT still uses the NV45 which features 16 pipes and is manufactured on a 130nm process. Combine this with the fact that NV45 is bridged from AGP to PCIe on package, and there is no question about the cost difference between silicon solutions.
What remains to be seen is how the current market will react. Our initial probes indicate that the 6800 GS will be a very limited offering with its cycle ending in Q1 06. With the potential to perform as well as a 6800 GT for the price of a vanilla 6800, the 6800 GS seems to warrant a longer shelf life than our sources indicate. Either way, the 6800 GS and ATI's X800 GTO parts point to limited run cards with excellent value growing in popularity. As with any good part, we would like to see the products linger a little longer, but the introduction of these late blooming parts is good for consumers in the market for a midrange card no matter why NVIDIA and ATI created them.
For our performance tests, we used this system:
Test Hardware
CPU: | AMD Athlon 64 FX-55 (2.6GHz) |
Motherboard: | EPOX NF4 SLI |
Chipset: | NVIDIA nForce4 SLI |
Chipset Drivers: | nForce4 6.70 |
Memory: | OCZ PC3500 DDR 2-2-2-7 |
Video Card: | ATI Radeon X800 XL ATI Radeon X800 GTO NVIDIA GeForce 6600 GT NVIDIA GeForce 6800 GS NVIDIA GeForce 6800 GT NVIDIA GeForce 7800 GTX |
Video Drivers: | ATI Catalyst 5.10a NVIDIA ForceWare 81.87 |
Desktop Resolution: | 1280x960 - 32-bit @ 60Hz |
OS: | Windows XP Professional SP2 |
We measured power draw at the wall for each card as well. Load was generated by running our Splinter Cell: Chaos Theory benchmark and observing the maximum power draw. The 6800 GS falls somewhere in the middle of the pack in terms of power draw.
70 Comments
View All Comments
Matrices - Tuesday, November 8, 2005 - link
I have great respect for this review and this website, but I have to emphasize I think the SLI section was extremely inadequate. You can't just bench one game barely anyone plays and then declare the issue closed. This is important becaues this card you can buy for just $200 now on the internet. So if anandtech did a thorough test of SLI and it turned out SLI beat out or tied a single GTX in Q4, D3, BF2, etc., that would be highly revealing.After all this card is on part with 6800 GT and 6800 GT SLI does best a single 7800GT. The difference between the GT and the GS is that you can get a GS for 200, not 270+ like a GT, so this could turn out to be a real steal of a deal. Unfortunately beyond Tech Report's more comprehensive SLI analysis I have nothing to go on...
AtaStrumf - Tuesday, November 8, 2005 - link
On another note, a comparison of a overclocked GTO @ XT PE and a overclocked GS would be very interesting. :)I ditto that. I have an X800GTO2 and it unlocks and OCs to X850XTPE easy. That thing eats 6800GT for breakfast and is cheaper and widely available. The only problem is that it's limited edition.
Jedi2155 - Tuesday, November 8, 2005 - link
Did anyone notice this on the power "consumpion" comparions ;).Should be consumption :).
Calin - Tuesday, November 8, 2005 - link
On the end of the last page there is:"You can buy the 6800GS today in its eVGA forum at several vendors."
I suppose you wanted to tell "its eVGA form"
Matrices - Tuesday, November 8, 2005 - link
The SLI results you got here are the exact opposite of the ones reported by Tech Report:http://techreport.com/onearticle.x/8993">http://techreport.com/onearticle.x/8993
Those folks saw a 100% increase in every benchmark with SLI using this card...
I wonder what the heck is going on here?
DerekWilson - Tuesday, November 8, 2005 - link
I did mention that sli can give almost 2x performance ... but the problem is that it doesn't happen in every game. You get consistently fast perforamnce for your money from the 7800 gtx. That was my point.BenSkywalker - Wednesday, November 9, 2005 - link
Can you try and come up with a better line? Your anti-SLI rants are looking fairly comical when looking at the 6800GS. It is less expensive then the 7800GTX by a decent amount even including the SLI mobo premium and it is almost always faster at resolutions high end users look for. I can understand you avoided running the SLI benchmarks as it makes your comments for some time invalid, but shouldn't you be able to clarify that by saying that the price/performance balance has changed in this singular instance?Zoomer - Wednesday, November 9, 2005 - link
I feel SLI is overrated, too.Higher mobo costs, higher psu costs, higher heat management costs (more heatsinks, fans, etc). And if you want to upgrade from that, you'll have to sell both cards.
What is wrong with selling your current card and getting a higher end version? Is really more economical to you? Or does it just line nvidia's pockets?
Remember, nvidia makes from the nf4 and from the extra chip sales that would otherwise be lost.
JarredWalton - Tuesday, November 8, 2005 - link
The SLI tests here are only from one game (Splinter Cell: Chaos Theory). Hopefully, we can get a few more titles benched in the near future.Zoomer - Monday, November 7, 2005 - link
Page 2:"The 6800 GT falls somewhere in the middle of the pack in terms of power draw."
Doesn't make sense, it had the highest and 2nd highest power draw, did you mean 6800 GS?
http://www.anandtech.com/video/showdoc.aspx?i=2593...">http://www.anandtech.com/video/showdoc.aspx?i=2593...
And I guess this tells us that 500W power supplies are redundant. A properly rated, quality 430 watt unit would be more than sufficient.
On another note, a comparison of a overclocked GTO @ XT PE and a overclocked GS would be very interesting. :)