Radeon HD 2900 (R600) reviews thread

One user told me, that his new HD2900XT (built by Sapphire) has not the same cooler, as other cards. Most HD2900XTs are equipped with dual heat-pipe cooler, but his new card carries slightly different heat-sink with 3 heat-pipes:



(mirror link)

(most other XT's from Sapphire use dual heat-pipe design)

Did ATi change the design of XT's reference cooler? That would be quite good (lower temperatures, noise...)

I checked ATi's pictures and almost all of them features dual heat-pipe version, only one (two XTs in CrossFire) shows the better cooler...

 

Yes its different, nice find.
Here is the original naked:


Deja-vu

Looks like sapphire (or amd) made the 3 heatpipe version fit to the reference cooler case

 

Check this out! New Xbit review with Cat 7.5's. This has AAA enabled on both platforms as a fyi. The x2900xt kicks fuckin ASS, beating the 8800ultra in several cases WITH AA and AF enabled. Its not until several pages in that the x2900xt starts to show its usual crappy + strange performance. The first couple pages make u actually think that the extra memory bandwith and shader power do something!

 

Only see an Improvement in Farcry myself and the Battlefield/Call of Juarez at the lower resolutions. Several Cases sounds like 7 games, its infact 4 "cases" over three games out of eighteen where it was faster then the Ultra. Also got kicked to the curb when resolution went up with the exception of farcry again. Not really anything great from that driver in my opinion . Still broken performance in STALKER, FEAR, SupCom etc etc or performing way too close to X1950 performance.

 

In those first couple pages it comes close or beats the gtx in no less than 7 benchmarks in that review and 5 games in that review. Looks closely. Funny you chose 7 as the number you thought it sounded like. And several means at least 2 fyi Ill list the games for you and resolutions.

Battlefield 2142: All res, beats gtx at 1600 and 1920

COJ: All res, beats gtx at all res, ultra at 1280.

Far Cry, Pier: Beats gtx at 1600 and 1920, ultra at 1920
Far Cry Research: Beats gtx at all res, ultra at 1920

Far Cry Research HDR: Debateable because at 1920, gtx is 13% faster, but who would really notice the diff. Also, Min fps is higher at ALL resolutions in this benchmark than Ultra. Which would make the gameplay experience better on the 2900xt.

Fear Extraction point: Comes within 3 frames of the GTX at 1920, though Min fps is less than half of the gtx. However, considering that the 1950xtx has higher min framerate, Id bet that will change.

Splinter Cell DA HDR: Comes within 2.5 fps of the GTX at 1600, betas GTX at 120.

X3 Reunion: Comes within 2.4 fps of GTX at 1280, within 4.6 fps of GTX at 1600. Gets killed at 1920 for some reason. though still much faster than GTS.

Command in Coquer 3: Is equal with GTX and Ultra. Min frame beats GTX at 1920. Of course all cards still average 30fps so this is a "who cares"

Company of Heroes: Comes pretty dang close at all resolutions, including within 5.1 fps at 1920 (80.3 vs 75.2).


So it is actually 8 games and 10 benchmarks in that review that the 2900xt does very well against the GTX. It does get killed in a few games, including NWN2, Supreme Commander. However, in the majority of the games in that review it does very well, and although in some of them, like HL2 it only pulls up to the GTS that is not a clear LOSS IMO, considering the price point. There is clearly some room for improvement in the drivers, as there are cases where the 1950XTX beats it, or is even with it.

I'm not a fan of this card because I, like many others, was expecting much better performance than it delivered, but it is encouragin to see driver improvements. If it can keep this up, at 350-400 it would be an extremely tempting buy over the GTX.

As you said, in Stalker, SC performance is definetely broken, and fear could certainly improve, but if you look at it, thats 3 games (2 really since fear performance scales really well on 2900xt to 1920 vs gtx) out of 11. NWN2 is also extremely shitty, but who really cares, that game sucks .

 

A review from one of 2 sites that manage to get results 180 degrees from each other. [H] being the other one. As far as I'm concerned, both sites should be removed from the good review list this round as IT is clear they both have done something to configure the results the way they wanted them to be. [H] shows total dominance of the 8800s while XBIT shows the opisite. And with the same exact hardware and drivers. I'm sorry, but both are worthless this time. DH, FS, or even B3D would be better than those 2.

 

I personally can't stand H reviews. Their testing methodology reeks. I've always thought Xbit had some of the best reviews around. Care to back that up? I think Techreport ranks up there as well, though they don't have a review with 7.5/8.38 drivers.

 

Look at all of XBIT initial scores with the 2900, then look at [H]. You'll see they are 180 of each other even thou they both used the same drivers, cpus, mobos, ram amounts(not sure on type) and video cards. And both sites go against what every other review has said. The XT is a good card, loses some, equal in others and when it wins, it competes with the GTX. So if 1 review shows all score skewed one way and another shows them skewed the other with same hardware, then it has to be faulty benching or they puposely skewed it to make the card look good/bad depending on how they thought the out come should be. ANd yes, I agree XBIT usually is one of the better sites. But for some reason, I find their result very suspect, [H] as well mind you, when every other tom dick and harry site that reviewed the cards got results that placed it anywhere from xt<gts<gtx to xt/gts<gtx to gts<xt<gtx.

 

aha but you said

 

and it did in quite a few

edit: eh ok ok, so my first post was a bit exaggerated. I'm just pretty excited at how well its doing. I thought that AA just totally murdered this card, but I guess it can hold its own in many situations. Hopefully later drivers will boost perf. like they did on the X1800, and work out the kinks in games like Supreme Commander. Peace

 

Sorry for being lazy, but with the reviews sumed up and looking exclusively at AA/AF numbers, is HD 2900XT faster than 8800GTS 640MB?

 

According to Brent and the gang over at [H], I certainly don't think they'd say so....(using Cat 7.6)

*Note: Be sure to read more than just this to be sure I'm not taking anything out of context....there's plenty of info in that review...

 

Picked up the new GFW issue and Jason Cross' testing showed the 2900 XT to be 20-30% faster than a GTS. That [H] review and its scores are really something of an anomaly.

 

Hmmm... [H]'s testing is anomalous and shows NVIDIA with unexpectedly large leads. That certainly doesn't happen in every generation of cards...

 

Having been an editor over at [H] in the past, I can say with confidence that Brent is a class-act editor and I highly doubt he'd skew anything for vendor A or vendor B....Regardless, we can all have our own opinions....moving on...

I'm not 110% sure, but isn't [H] the only review using the latest Cat 7.6 drivers? If so, what's the use in debating the results from an older driver in another review?

I guess we'll all have to wait for 7.7's to made a decision...

 

Tim can speak for himself, but I don't think either of us were trying to allude to intentional skewing of test results. Just pointing out that those scores aren't exactly congruent with most other results now available for comparison. And the 7.6s didn't see that much of a sea change, at all, to say those [H] scores stand completely aloof.

 

For whatever reason, there does seem to be a pattern of [H]'s results usually showing larger discrepancies in favor of NVIDIA cards. For example, I'm seeing the 2900 XT do very well in Crossfire in TR's initial R600 review, including some games where [H] shows it getting demolished. If someone decides to ask about this, [H] will reply with "oh we do real-world testing and this is what you would see" or something like that. It's maddening, because on the off chance they have found something nobody knows how to reproduce it or explain it.

I've also wondered about the margin of error in [H]'s approach; I can't help but think that it's astronomical compared to other benchmarking schemes.

 

Understood...

 

Well I for one smell something fishy. [H] is the only site that show Nvidia winning by good margins overall while xbit shows the exact optisite using exact same hardware and drivers. So either one or both sites are wrong.

 

Or they have totally separate benchmarking methodologies, and the areas that they test are stressing different components of the cards, etc. I'm a long, long way from attributing this to malice.