The Official GF 8600 & 8500 reviews thread

I think that considering its specks, 8600GTX performs rather well, the problem is the specks are below their own par.

 

Hobbled?
7900GS - Yes (G71-based, binned from non-functional 7900GT chips)
X1950Pro - No (RV570 was designed from the ground up, not binned from R580)

If you meant that the parts arrived later (refreshes) then the X1650XT arrived after the X1950Pro.

 

Like like 90+% of all Windows users, I don't run Vista. DX10 is only offered under that OS. If you don't run Vista, the DX10 checkbox is useless. And Vista doesn't require DX10 hardware, just DX9, for full effects even.

So when will game publishers be willing to pass up DX9 level gamers? Not until these cards are long past their "sell by" date.

They have to be judged on other merits and demerits. Compared to the last generation, the 8600GTS draws more power, performs worse and costs more (at this point in time) that the 7950GT and x1950xt. It may offer advantages in IQ, whether those settings are useful in practise remains to be seen - to be honest I'm sceptical that the presence or absence of 16xCSAA wlll ever affect its customers much.

At this point in time, it's difficult to see the trade-offs the 8600 cards offer as particularly attractive to consumers.

 

Actually, we use the highest settings we find playable, aiming to raise both resolution and AA. It also depends on the game, and we prefer to be able to have in-game settings maxed out. Take for example Flight Sim X, antialiasing greatly benefits that game. We would rather drop the resolution one level in that game in order to have 2X or 4X AA enabled. The 8600 GTS clearly has an advantage in that game though, with the same in-game settings we found 1280x1024 with 2X AA playable with the X1950 Pro but 1600x1200 with 4X AA playable with the 8600 GTS (and higher performance still).

So yes, our focus is very much on high levels of resolution and AA, but it does depend on the game. Some games look better if you raise the resolution and lower the AA setting, while others look better to keep the resolution down a level and raise AA instead.

If anyone ever has a question about our methods or testing or suggestions or comments or whatever don't hesitate to drop me a line.

 

That's a very good strong point. It remains to be seen if ATI was able to produce a cost-effective $200 DX10 part and achieve performance matching last generation's $200 DX9 products. If they can only match the GTS then we will all have to accept the fact that DX10 support comes at a cost.

When we're back on a level playing ground - old DX10 vs new DX10 - maybe we will return to the heydays of world beating performance in the midrange.

 

I have a comment, but I will post it here instead and endure the wrath Geo for possibly throwing this thread of kilter.

What I did when I used to review was generate a "baseline" using the graphics card in question. I did this by using a resolution that I thought would be common to users who would be looking at the card, and adjusted game settings from there until I got playable results. I'd then use those settings for all the graphics cards in the review.

I'd rely on timedemos for generating performance results, but I would verify that the results they generated matched actual gameplay performance (by using Fraps as I played through the game). That way I didn't have to waste time with playing every game with each card, which gave me extra time to put toward offering more results from more games (I just had to verify the validity of the timedemo once then use it for each review and card).

For most of my reviews I offered results with and without AA and AF over 4 or 5 resolutions in all the games for all the cards in the review (the choice of AA/AF/resolutions would be geared toward the graphics card).

So basically I was offering real world results using verified timedemos and benchmark procedures, a lot of games (a dozen or so) with a lot of results so the reader could determine for himself what "playable" was in the games he was playing, and in a way that was easy to read and understand.

I always thought that this method worked very well and gave reliable and honest real-world results. What do you think?

- Mark "Ratchet" Thorne, formerly of Rage3D.com

 

Clearly there's some sensitivity to bandwidth in G84:

http://www.elitebastards.com/cms/in...sk=view&id=379&Itemid=27&limit=1&limitstart=9

All the percentages are wrong, the correct figures are:

• memory overclocked by 11.4%
• core overclocked by 9%
• HL2:EP1 performance gain of 15.1%
• Prey performance gain of 15.6%

Jawed

 

No I wasn't talking about timing. Just positioning...the X1650XT was the bane of the 7600GT. And the 8600 series is technically the replacement for that class of cards. I'm not making excuses for the card here - it can barely beat the 7600GT in some cases - but that fact is ignored in the midst of all the focus on how it does against the 256-bit hand-me-downs from last generation.

 

Techreport chimes in - http://www.techreport.com/reviews/2007q2/geforce-8600/index.x?pg=1

Better showing there but like all the other sites they emphasize the sacrifice of DX9 performance by going with a 128-bit DX10 card over the more mature 256-bit DX9 competition. Don't know how many people are looking at the 8600GT but it arguably puts on a better show than the GTS.

 

I am pretty pleased with the power consumption for a DX10 mainstream part.

 

Hey Mark,

I think this is primarily where we differ in our methodologies. We do play every game with each card in each evaluation. It is much more time consuming, but we feel it is the proper way to evaluate the gameplay experience. There is no better way that I know of to get a feel for a game (both performance and IQ and who knows what other issues that might be present) with a certain video card other than playing the game with the video card.

If your timedemos match the real-world game performance that's great, but I think you are missing a huge part of the gaming experience not playing each game on each video card. Again, you are correct, it is much more time consuming, but I'd rather spend the time and do it right IMO.

Thanks for the feedback

 

I do (or did), of course, play through portions of the games on the graphics card being reviewed to get a basic feel for the card and check for anomalies and bugs and whatnot, so I think I had that covered as well (and I guess I did this for all the cards considering they were all reviewed at one point or another). I obviously didn't use those sessions to measure performance simply because I don't think manual Fraps runs are reliable enough for that. Anyway, with that experience and all the results I generated I think I got a pretty good grasp of what each card is capable of in each game before I wrote the review, and I think I conveyed that to my former readers relatively well.

So, I dunno. From my point of view I think I went about my benchmarking procedure pretty thoroughly; I offered a lot of real-world, reliable, and objective results so readers could pick what best matched their own preferences and feel pretty sure that what they saw was what they could expect to see on their own platform. How's that not the right way to do it?

 

IMHO, the 8600 GTS will start looking much more appealing once the price sinks under the 150 dollar psychological barrier.
That may not take long, depending on ATI's side, although the latest news bit doesn't seem too reassuring.

And, let's face it, the 8600 GTS might not win against a 7900 GTX, but it sure is much better than a 7600 GT, both in speed, IQ and video processing features, and that segment of the market cares more about value than pure speed at unrealistic settings for the typical buyer of a mainstream product anyway.

 

Here are some interesting findings on the video processing engine of the 8600:

http://www.driverheaven.net/reviews/msibfg8600/media playback.php

 

QFT. 47W at full load according to xbitlabs (8600GTS) is just impressive for a GPU on a 80nm with 298 million transistors.

 

Do these VP features actually improve IQ over software based solutions or is it just a performance thing? What kind of score would a software player get in HQV for example?

 

Software scores are here :

http://www.tomshardware.com/2007/01/09/avivo-vs-purevideo/page12.html

I'd like to see Nero Showtime 2 and 3 tested .

 

Thanks man! I guess you really do get a lot of benefit from these hardware solutions.

 

I found that very odd too. I'm not sure if it is a result of there unique approach to comparing cards or because of bias. I actually like HardOCP's approach, but it makes the review much more subjective and liable for bias.

I've read perhaps 6-8 reviews and HardOCP is the only site that infers the 8600 beats, let alone crushes the x1950Pro.

 

The new features in G84/G86 compared to G80 are not for IQ afaik, only performance. The G80 has some improvements in IQ compared to G7x though, as the shader core is now used even more extensively there.

As the Tom's Hardware link above indicates, the HQV score of software solutions is definitely much lower. What will be more interesting is the 720p/1080p IQ, imo... Last time anyone checked that (anandtech, iirc) the results were rather, err, unimpressive let us say.