Radeon HD 2900 (R600) reviews thread

I'd offer this up to demonstrate how "unfair" and biased [H]ard have been to ATi in the past:

From [H]ard:

"If you want the best bang for your buck, the PowerColor Radeon X1950 PRO Extreme 256 MB is it."

http://www.hardocp.com/article.html?art=MTI2NSw5LCwxNTA=

"ATI has proven they are a leader and not a follower with the X1950 XTX. ATI has released the world’s first consumer 3D graphics card with GDDR4 memory clocked at the highest ever stock speed that chews through games when it comes to high definition gaming. Memory bandwidth looks to one again be the defining factor in 3D performance. With a re-designed heatsink/fan unit, faster memory, and lowered price, the ATI Radeon X1950 XTX and CrossFire Edition are both serious 3D gaming video cards for the [H]ardcore that offer some value over NVIDIA’s more expensive 7950 GX2. ATI’s CrossFire dual GPU gaming platform looks to have just grown up."

http://www.hardocp.com/article.html?art=MTE0NCwxMiwsNzA=

"If you want the fastest performing video cards right now that provide the highest level of image quality in games, it doesn’t get any better than the Radeon X1900 XT and XTX. You get High Quality AF, Adaptive AA, HDR with AA, and very fast shader performance. If you have been waiting to play F.E.A.R. at very high quality settings, your wait is over; the Radeon X1900 XT/XTX is the card for you."

http://www.hardocp.com/article.html?art=OTUzLDEzLCxoZW50aHVzaWFzdA==

The problem is with the product, not the reviewer.

 

This is certainly the largest problem with their testing methods. Even if the margin of error wasn't huge, is there any way to quantify it?

 

Unless we went through the tests ourselves, or have data about the tests, no real way. In game results as everyone knows can vary widely. And looking at it from the outside, way too many areas where things can shift.

 

The whole 'playable settings' result just doesn't sit well with me. It's no biggie though, I never even read [T] reviews, there are far better sources for controlled, explainable results.

 

Yay, 3 separate examples of pretty much the same product. Yes you've provided thorough proof

 

My only complaint with their reviews is we have to rely on what Brent or another editor deams "playable". That may change by a decent amount to your terms or my terms of what is playable. But that I guess is the same thing about not having a way to reproduce their testing...ect. Still I think their reviews add value so long as its taken into account with all the other information on a product....

 

Its one review thats done differently from all the rest, its better if reviews are done differently otherwise you end up with the same old same old. Now the margin of error, if someone was to do that for any of these benchmarks (time demo tests too), a review wouldn't be done in one week it would take one month. By then the review is useless, new drivers what not.

Actually the only other site that does actual game play tests that I can think of is B....somthing can't think of the name wierd.

Anyways I wouldn't be suprised if there were areas of games where the 2900 falls, looking at the synthetic results we see areas where the r600 is weak, is it a far stretch to think parts of some levels would push the r600 harder because of the relative weakness?

 

http://www.firingsquad.com/hardware/diamond_radeon_2900_xt_1gb/page18.asp

some of these also mimick [H]

 

Bit-tech.net?

 

yeah thats the one.

 

The 'same old' tells us what we want to know... The same old is essentially a baseline that provide a great way to visualise improvements with new parts. The same old just works. One reviewers' assumption of what's playable in a game really doesn't mean squat when you've got no way of quantifying what you're seeing. "Oh cool, 55fps in 'Dense Tree Area Outside Town' ", how do I compare this kind of result to that of an opposing part, without possibly reading a review on said part on the same site as this initial review? I can't...I can't in anyway justify taking said results seriously when there is absolutely no way of determining their validity. Now i'm all for abstract / unusual / (realistic?) testing methods, but in all honesty, I / We can't make use of them. There's no control in such cases.

One mans 'playable' is another mans 'who's playing this damn slidesh- oh hell no".

 

you can't have best of both worlds though, you can't get time demos do everything you do in game, time demos use triggers to set many things off, where in game that won't happen. So you end up loosing things for those controls. Thats why its best if you have both types of tests. Its always better to have more data then less data, or more of the same data is what I'm trying to say. I agree its impossible to assess [H]'s methodology without being there and seeing the tests run but we have to look at credibility and believablity for that.

 

I agree with jb. H uses an innovative method and that's interesting. They pioneered min & max framerates reporting and that's great. However, when they draw conclusions, they seem to draw some unusual conclusions -- ones that often lead me to wonder what their motivations are. They've several times taken products with performance within 5% of each other and deemed one "playable" and one not, they've often tested cards at significantly lower IQ settings from one another despite several other sites reporting respectable numbers with higher settings, and the actual conclusions pages themselves have a several times been in extreme exaggeration of their testing results. For the most part I've stopped reading their reviews, but, other than these flaws, sure, it's great that they try something different. Occasionally something there will get linked here that I'll read. My main concern is actually with their big brother attitude over there -- how many other hardware review sites proclaim their site and methodologies as perfect, and number one, and take time out of every review to point out how their unique (and unreliable) testing methodologies are so superior?

 

Absolutely not. Have you ever considered that the benchmark scenarios that Xbit chooses could be different from those of the websites that show severely different results? Ever consider that LOADS differ in different parts of the game, and some scenario are more efficient to render on one card versus another.

An example:

Right now we pretty much universally agree that Oblivion runs way better on R580 than G70/71. That's a fair statement. However there were a few publications at the time that were completely legitimate, but showed results where the nvidia part either met or beat r580. Yes, even in foliage. Those publications didn't lie, but got different results. Got it?

 

http://www.club3d.nl/index.php/products/graphics/item/270
Click on the high res image: Club3D HD2900XT seems to have 3 heat pipes.

 

You are talking about 1 game, I'm referring to a slew of games used by both site, with same hardware and drivers that show complete opposit results. If it was just 1 or 2 games, I could agree with you. But every single game tested by both sites? Come on man, even you should be able to see that.

 

I'm wondering how much of these past concerns are related to the discussed discrepancies and current review methodologies? The original context was cheating, but the gist of it applies to accurate game performance too.

 

I picked up a Powercolour 2900XT from OcUK and it also has the 3 heat pipes in the cooler.

 

It's not just one game. Different websites sometimes use different bencmark routines that run through different portions of the same game. It's not as easy as "this website sucks, this ones good" approach, but it does make a whole lot more sense. And it is definetely not every game that is skewed on Xbit's results, look again. Also, check the description of testing procedures, I believe those differ as well (AAA on i believe).

 

The early ones had two. I have a Sapphire that I got around launch, and it has two heatpipes. The Palit I got two days ago has three.