Anand's 9800 XT review

[banksie]

I liked the new set of benchmarked games though, I hope ATI can improve homeworld 2 performance a bit.

I understand that the Shadow bug has been fixed in the 3.8 drivers due out in a week or so. Something slightly odd is going on with them as they do seem to chew an inordinate amount of GPU processing time when anti-aliasing is enabled.

On my XP1800+ with a Radeon 9700 card I can run the game nicely with 4xAA, 8xAF (quality) and every other graphics option in game set but as soon as I enable the shadows performance crawls like it is thrashing texture data across the AGP bus. I have tried leaving everything but the AA at application preference and even 2xAA causes this effect.

There also is some oddness with the Vertex Buffer Object handling in the 3.7 drivers that causes problems.

All very puzzling.

 

[Johnny Rotten]

This is the first installment of a multipart series ... The extensive benchmarking we’ve undertaken has forced us to split this into multiple parts, so expect to see more coverage on higher resolutions, image quality, anti-aliasing, CPU scaling and budget card comparisons in the coming weeks.

Perhaps reading before flaming would be a better strategy?

 

[Natoma]

This is the first installment of a multipart series ... The extensive benchmarking we’ve undertaken has forced us to split this into multiple parts, so expect to see more coverage on higher resolutions, image quality, anti-aliasing, CPU scaling and budget card comparisons in the coming weeks.

Perhaps reading before flaming would be a better strategy?

I would have rather gotten the review a week from now in its entirety with the performance #s AND the IQ in the same review, then come to one single conclusion based on the entire picture. As it stands, Anand came to his conclusion based solely on performance. He didn't even mention image quality in the conclusion.

That is not what I consider a good review, especially in light of the severe IQ problems with the 50.xx+ drivers unearthed by Lars at THG.

 

[Grall]

Dilettante:

First of all, if you can do so much better, why don't you start a review site, huh? Second, who forced you to waste those precious few minutes of your life? Third, there weren't any pics in HOCP's review either, apart from graphs that is. Though those may have been done in html, I didn't check that carefully. Just make sure you bash ALL pic-less reviews equally, okay?

Thank you.

*G*

 

[3dilettante]

Dilettante:

First of all, if you can do so much better, why don't you start a review site, huh? Second, who forced you to waste those precious few minutes of your life? Third, there weren't any pics in HOCP's review either, apart from graphs that is. Though those may have been done in html, I didn't check that carefully. Just make sure you bash ALL pic-less reviews equally, okay?

Thank you.

*G*

Me start a site? No, don't think I could do it as well as some sites, which makes it much worse in my opinion; since if even I can see bad information and methodology, you know something's wrong.

I spent those few minutes on that article because I had previously found Anandtech's articles to be reasonably well grounded, just not this time.

I'd talk about a [H] review in a thread not about anandtech's article.
Though if you are talking about the review of the 9800XT, there were screenshots, including some AA and AF comparison. Not exactly in depth, but at least he threw a few pics out.
At least Brent in his review put down more than 3 sentence fragments (with prescott in white text) describing his system setup, used only released drivers, and didn't try to throw an untested and unknown prescott performance variable into the mix just to show off the supposed size of his e-wang.

 

[bdmosky]

I would have rather gotten the review a week from now in its entirety with the performance #s AND the IQ in the same review, then come to one single conclusion based on the entire picture. As it stands, Anand came to his conclusion based solely on performance. He didn't even mention image quality in the conclusion.

That is not what I consider a good review, especially in light of the severe IQ problems with the 50.xx+ drivers unearthed by Lars at THG.

That's nice and all that you would have rather waited a week, but I wouldn't have, and I'm sure there are many more out there like me that wanted this information now. After all, this was a preview of how 9800XT cards in retail might perform, not a review of an actual shipping card. The difference is that while this is indicative of how the card may perform, shipping drivers will be 3.8 catalysts, and performance and image quality may change. Likewise, the NV38 card which was previewed will be indicative of how retail GFX 5950s will perform, but of course image quality and performance may change.

I like previews... if you'd rather wait for a full fledged review, don't read the previews!

 

[Ailuros]

Any reader has the God given right to voice his/hers opinion on any publicly available material, wether you like it or not.

As for the recommendations someone to start his own website if he wants to do a better job when it comes to reviews, how hard do you actually think it is to write a review after all? Either way any reviewer will face criticism in varying degrees at any time.

Since the actual NVIDIA Detonator driver v45.23 has issues in some applications (see: Aquamark3: Accurate Benchmarking for Old and New (DirectX9) Apps?), we also use the older driver version v44.03. The new Detonator 50 alias ForceWare driver was not yet available at the time of this review. NVIDIA asked us not to use the beta version v51.75 of this driver in reviews because it contains a couple of bugs. All ATI cards were tested with Catalyst drivers v3.7.

http://www.tomshardware.com/graphic/20030930/radeon_9800-13.html

I don't think I need any additional comments to the above.

 

[Pete]

Weary != wary. That is all.

 

[Velocity]

This is something somebody at the Rage3D forum found out while looking at the Anandtech review.

See how 43.8 should be LOWER than 44.3!!

How does adding 4xAA/8xAF raise the FPS of the NV38?!?!?

Same thing with the 5600. It goes from 15.5 w/o AA/AF to 25 w/ AA/AF

Some fishyness seems to be going on here!

 

[ZoinKs!]

How does adding 4xAA/8xAF raise the FPS of the NV38?!?!?

Same thing with the 5600. It goes from 15.5 w/o AA/AF to 25 w/ AA/AF

A difference of a couple % could be due to the benchmark method. Using fraps as a benchmark requires the user to manually press keys to start and stop the benching process. A slight difference could be due to pressing the button at a slightly different point.

That's plausible for the small nv38 difference, but not enough to explain the big difference for the 5600 ultra.
This needs begs for some investigation. Anybody out there got Homeworld 2, fraps, and a 5600 ultra?

 

[Natoma]

I would have rather gotten the review a week from now in its entirety with the performance #s AND the IQ in the same review, then come to one single conclusion based on the entire picture. As it stands, Anand came to his conclusion based solely on performance. He didn't even mention image quality in the conclusion.

That is not what I consider a good review, especially in light of the severe IQ problems with the 50.xx+ drivers unearthed by Lars at THG.

That's nice and all that you would have rather waited a week, but I wouldn't have, and I'm sure there are many more out there like me that wanted this information now. After all, this was a preview of how 9800XT cards in retail might perform, not a review of an actual shipping card. The difference is that while this is indicative of how the card may perform, shipping drivers will be 3.8 catalysts, and performance and image quality may change. Likewise, the NV38 card which was previewed will be indicative of how retail GFX 5950s will perform, but of course image quality and performance may change.

I like previews... if you'd rather wait for a full fledged review, don't read the previews!

First off, this wasn't a preview. From the anandtech frontpage:

--------------
Fall 2003 Video Card Roundup Part I - ATI's Radeon 9800 XT
Review | Oct 1st, 2003 3:02 AM
After spending countless hours of benchmarking we're finally able to bring you results from our new suite of 15 game tests on ATI's Radeon 9800 XT as well as NVIDIA's NV38.
--------------

Second, you'd rather have inaccurate and incomplete information than wait a little while and get accurate and complete picture?

No wonder sites can continue to perpetuate misinformation with impunity. People simply don't care and want it now Now NOW.

 

[3dilettante]

See how 43.8 should be LOWER than 44.3!!

How does adding 4xAA/8xAF raise the FPS of the NV38?!?!?

Same thing with the 5600. It goes from 15.5 w/o AA/AF to 25 w/ AA/AF

Some fishyness seems to be going on here!

That first disparity is only caused by the use of slightly different values for the x-axes, notice how the top graph goes to 50, while the bottom goes to 45. As long as the values are consistent within their graphs, there is no problem, it just hints at a rush job where they skipped over normalizing all their graphs.

As for the NV38, I think the first graph may have switched the NV38 and FX 5900 around, which leads to the mistaken impression that AA and AF increased the speed. I ascribe to the "rushed job, wrong values" school of thought.

As for the FX56000 having such an improvement, either the two values were switched between the graphs--once again because of the lack of due diligence in going back over the data--or Nvidia's drivers are even more aggressive in their optimizations than previously thought. If any of the graphs wasn't CPU bound at no AF/AA, it would probably be the comparatively weak FX5600, which hints that the values were switched.

However, one should take these graphs with a grain of salt, the lack of significant differences between the modes hints that the game is cpu limited. It would have been much smarter to go with a 3.2 Ghz P4 or P4EE, or Athlon FX, instead of a prescott at 2.8 Ghz (which seems to be at best as fast in some preliminary benchmarks as a 3.0 Ghz P4). Instead, they rushed out using a weaker CPU just for the wow factor.

Perhaps the reduced performance of the ATI cards hints at some software or driver fault that is offloading extra work onto the CPU, maybe.

If Anand hasn't ticked off Intel due to any broken NDAs, it's probably only because most of the tests really don't reveal any good data on the processor.

If Anand hasn't ticked of Nvidia due to any broken NDAs, it's probably only because most of the tests really don't reveal any good data on the NV38 either.

 

[jimbob0i0]

I have no problem with Anand wanting to split a large review into several parts but I do have a problem in the way he has done it. As many have said FPS results without IQ tests is meaningless and using an undocumented/untested/unreleased CPU is pure my ____ is bigger than yours mentality and gives us no indication of exactly how these results relate to anything we can test.

If Anand wanted to undertake a 3 part review he should have (IMHO):

a) Used a well tested, fast, released, known CPU on a known motherboard
b) Done the IQ tests first so there would be something to relate to in terms of perspective for when the FPS tests came about
c) NOT put a conclusion at the end of each part as without the remaining parts any conclusion would be bogus.. the conclusion should be at the end of all three parts
d) Use shipping/verifyable drivers... not something provided by an IHV for a benchmark

 

[3dilettante]

Yeah, multi-parters aren't bad, but poorly divided parts most definitely are.

Assuming there will be parts detailing IQ, cpu scaling, and DX8 vs DX9 behavior, each article will essentially undermine the first part, further emphasizing it was a waste of time.

If things get worse, each article will conflict with the other articles as well. Maybe the cpu scaling article will be fill rate bound half the time, the image quality article done with no reference to what the performance was, and game tests will be cpu bound.

Maybe they'll put up some kind of uber-article that takes all the factors into account, but that would only show what should have been done in the first place.

 

[bdmosky]

First off, this wasn't a preview. From the anandtech frontpage:

--------------
Fall 2003 Video Card Roundup Part I - ATI's Radeon 9800 XT
Review | Oct 1st, 2003 3:02 AM
After spending countless hours of benchmarking we're finally able to bring you results from our new suite of 15 game tests on ATI's Radeon 9800 XT as well as NVIDIA's NV38.
--------------

lol... from webster.com

Preview--
1 : to see beforehand; specifically : to view or to show in advance of public presentation


Review--
1 also /'rE-"vyü/ : to view or see again
2 : to examine or study again; especially : to reexamine judicially
3 : to look back on : take a retrospective view of

It's not my fault Anandtech doesn't know their vocabulary. I don't really care what they wanted to call it, it was still a preview of a non-retail card. It may be the same revision as the shipping card, but it's not the same packaging or the same drivers...

Second, you'd rather have inaccurate and incomplete information than wait a little while and get accurate and complete picture?

Last time I checked I was grown up enough to realize that I can read preliminary results without immediateliy forming an absolute opinion and rant and rave over them... I like early information because I find it interesting and if I can take the information in with reason, then why not? I just think it's kind of ridiculous that everyone immediately jumps on Anandtech's back for every little thing they do, just because you believe their entire audience is a bunch of lemmings. Perhaps some of them are, but I'm not and I don't appreciate being thought of as one.

 

[Natoma]

bdmosky,

You are not the vast majority of end users who see Anandtech and think of it as the end all be all wrt performance. Most people don't visit B3D, nor have even heard of it. But they go to sites like Anandtech because they're advertised and sponsored in major magazines like Maximum PC.

You may not form an opinion before seeing the entire results pool, but I'd wager very heavily that you're in the minority on that one.

 

[tEd]

This is something somebody at the Rage3D forum found out while looking at the Anandtech review.

See how 43.8 should be LOWER than 44.3!!

How does adding 4xAA/8xAF raise the FPS of the NV38?!?!?

Same thing with the 5600. It goes from 15.5 w/o AA/AF to 25 w/ AA/AF

Some fishyness seems to be going on here!

that's no surprise. In his 5900ultra review he had many such obvious wrong benchmark numbers. Does he even look ones at the screen when doing this benchmarks? I highly doubt it.

incompetent

 

[Natoma]

tEd,

Wasn't there something like this in the HL2 preview a few weeks ago? Where all the cards lost a nice chunk of framerate while doing AA when going from 1024 --> 1280, but the 9600 framerate was identical?

It's been shoddy for a while.

 

[bdmosky]

You are not the vast majority of end users who see Anandtech and think of it as the end all be all wrt performance. Most people don't visit B3D, nor have even heard of it. But they go to sites like Anandtech because they're advertised and sponsored in major magazines like Maximum PC.

You may not form an opinion before seeing the entire results pool, but I'd wager very heavily that you're in the minority on that one.

Once again, I realize this... from above:

Perhaps some of them are, but I'm not and I don't appreciate being thought of as one.

Now, I still don't see what all the fuss is about... the preview didn't picture the NV38 as the end all be all of cards... it actually came in second to the 9800 XT the majority of the time. I read the preview and thought that Nvidia really has their work cut out for them if they want to win this round... I think even the majority of Anandtech's readers probably came away with this conclusion. Now, was this preview perfect and thorough? No... Did it cause irreparable harm to ATI? Absolutely not! I actually think it showed ATI in positive light.

 

[demalion]

There is a problem with giving bad information, even if the bad information isn't as bad as it possibly could be. "Not the worst it could be" does not mean "not very bad"...what does not saying the NV38 is "the end all be all" have to do with this?
Analyzing against arbitrary examples of what something is not does not change criticisms about the problems in what something is, there has to be some relationship from what you've shown it isn't that prevents it from being something else. The relationship for your example is "not the worst possible" means "not very bad", and it simply isn't valid.

Anyone who has heard good things about nVidia and bad things about ATI can reasonably translate these results into the NV38 being a better choice for the games that might be misrepresented currently. It would appear to be true that, in many cases, this would not be a reasonable conclusion with accurate information. Therefore, it seems to me that information that allows this disparity due to its inaccuracy might clearly be called "bad", and can clearly be said to not show ATI in a "positive light" in relation to reality.

The missing image quality analysis, the lack of the ability for independent verification, the presentation of comparison with settings/workload that seem to be vastly divergent in actuality...all of these seem bad or very bad by themselves, and the combination even worse. Just because the comparison could have been distorted more doesn't make it any better than it is...all it establishes is that it "isn't the worst possible".

Where does that get you?

The choices aren't as simple as "Perfect and thorough" and "Irreparable harm to ATI", nor does representing those as the choices and saying "it shows them in a positive light because it didn't do irreparable harm to them" make any sense whatsoever. I'll try to refrain from the obvious but emotionally weighted comparisons to crime that come to mind, but I ask you to give it some thought if it helps illustrate a problem with your above statements for you.

More directly: How does a comparison that puts someone at a gross disadvantage to where they actually stand, in this case by representing their doing more work as valid to compare to another doing less, show that someone in a "positive light"?