Final Words

Let us preface this conclusion with a reminder that we have not been briefed on this part by NVIDIA. While we have seen internal roadmaps pointing to the 7800 GS, we still don't have any confirmation from the source on it. We don't have absolute certainty that this part will even make it out the door -- much of that seems dependant on the performance of the upcoming G73 and G72 GPUs. That being said, it would be really nice to see this part make it to the light of day. The fact that we've actually seen hardware is a hopeful indication of the future.

We really don't know anything about price at this point as NVIDIA hasn't even spoken with us about this part and it's still too far out for any online retailers to have a price yet. As always, value depends on the cost of the product, and we wouldn't want to leave that out of our analysis. Instead of talking about the value we have here, we are more inclined to talk about how much a part with this kind of performance should cost to become a competitive product.

Since the performance of the part falls between the 6800 GS and the 7800 GT, we took a look at what those parts cost in order to get a lay of the land. The 6800 GS is priced between $200 and $230, and the 7800 GT comes in at $320 to $370. If we look simply at averages, this would mean we should expect the 7800 GS to be sold between $260 and $300 if 7800 GT prices stay stagnant - but most likely lower if the 7800 GT prices move before the 7800 GS launch. NVIDIA roadmaps confirm that even though GeForce 6800 GS will reach sub $190 levels by the end of the year, the product will not actively be continued.

At the same time, this part performs in the same class as the X1800 XL. While the ATI card fairly consistently beats the 7800 GS in these benchmarks, the fact that that the 7800 GS keeps up at all (and will cost even less than the 7800 GT) means that it would be very difficult for this card not to have a lot of value. Of course, if this part debuts at the high end of our estimate, a cheap 7800 GT would definitely be a better way to go. There are some rebate and coupon offers that can get you a 7800 GT for $300 right now if you look hard enough.

We are quite excited about this part; the sooner it arrives the better. The 7800 GS will be a great part for people who can't decide between the 6800 GS and the 7800 GT. Hopefully the introduction of the 7800 GS will also allow the price of the 6800 GS to get pushed down a little more without leaving a huge gap between it an the next higher performing part.

We will definitely publish more news on the 7800 GS when we get more details.

Quake 4 Performance
Comments Locked

43 Comments

View All Comments

  • DerekWilson - Wednesday, November 23, 2005 - link

    Sorry, most of our tests were run with AF enabled (8x in most cases).

    The only test that did not include AF was the Splinter Cell: Chaos Theory test with out AA (the 4xAA test also included 8xAF).

    Thanks,
    Derek Wilson
  • Jep4444 - Tuesday, November 22, 2005 - link

    I'd like to see some benches of the 6800GT or 6800Ultra(preferably the ultra) included just to see how the G70 architecture compares to the Nv40 in terms of "improvements"
  • ShadowVlican - Tuesday, November 22, 2005 - link

    6800GT is almost exactly like 6800GS.. so it's a waste of time... most (if not all) reviews of the 6800GS would have already taken care of that comparison
  • Jep4444 - Wednesday, November 23, 2005 - link

    the thing about the 6800GS and GT being similar is that they arent uniformly similar, they traded off a few benches and overall the GS is slower

    what i wanted to see was the fact that they have the same pipeline setup and the 7800GS is clocked similarly so i wanted to see how much the G70 architecture is actually improved
  • nullpointerus - Tuesday, November 22, 2005 - link

    One thing I've been wonderingabout your benchmarking: Some time ago there was a question of whether average framerate should be the sole basis for comparison (IQ issues aside). The argument was that if a card really chokes on certain parts of a scene, having 10% higher average framerate wasn't exactly great. Would it be possible to do a graph of the frames so we could see how much they vary? Or is this not really an issue today?
  • DerekWilson - Tuesday, November 22, 2005 - link

    It does matter to some extent, but it is very difficult to do a good job of representing this data. Graphs of instantaneous framerates over time is next to useless in our opinion. Its too hard to actually tell what's going on and way too difficult to properly compare a number of cards without totally destroying readability.

    I keep voting for boxplots but there's still some debate about whether teaching our readers about statistical analysis is a good idea :-)

    I think something that can show outliers, min, max, lower and upper quartiles, and median in as much space as a single bar in a bar graph gets my vote as a good thing. The huge problem is that our graphing engine doesn't support anything nearly this robust. We have been trying to augment our bar graphs with some line graphs generated using a spreadsheet program to show resolution scaling, but I still haven't found a good easy way to generate nice looking boxplots.

    Anyone tips?
  • Leo V - Wednesday, November 23, 2005 - link

    Boxplots are an EXTREMELY useful idea, especially the lower-quartile framerate. In fact, the lower quartile may be more useful than the average framerate.

    The lower quartile shows you how smooth the game is in the "slower" half of the time -- which makes or breaks a game's playability and enjoyment.

    Suppose card A averages 60FPS, but the worst half of the time it only averages 20FPS (lower quartile). Carb B averages 50FPS, but is more efficient in relatively complex scenes -- 35FPS lower quartile. Most likely, you want card B, despite its lower "average" framerate. What good is card A when its best 50% of scenes average 100FPS while its worst 50% average 20FPS? A better choice is probably card B, which might average 65FPS in the best 50% and 35FPS in the worst 50%. But average framerates don't show this!

    Posting min and max framerates (box whiskers) is useless, since they will be meaningless (and probably random) freak outliers. Min SUSTAINED framerate is good, and it happens to be closely related to the lower quartile number.
  • Donegrim - Wednesday, November 23, 2005 - link

    I reckon if you started to introduce boxplots alongside your standard graphs, readers would catch on pretty fast. People who want the extra information will take the time to work out what it means, and people who don't care can just look at the standard graphs. Or just stick a massive arrow pointing to the mean and leave the rest of the diagram a little more faded.
  • nullpointerus - Wednesday, November 23, 2005 - link

    Maybe you could make the boxplots appear instead of the regular graphs on mouse over like you do with some of your IQ comparisons?
  • DerekWilson - Wednesday, November 23, 2005 - link

    that would be awesome ... the big problem is making the boxplot though.

Log in

Don't have an account? Sign up now