HardOCP review of HD2900XT now posted

[Mr. Cornell]

http://enthusiast.hardocp.com/article.html?art=MTM0MSwxLCxoZW50aHVzaWFzdA==

And looks very, very bad for HD2900XT. Note that Kyle used drivers 8.37.4, not the 8.36.1 drivers on retail version CDs that lot of early NDA-breaking reviews used.

[Nandro]

I didn't read the whole article, but as i mentioned in one of the other r600 threads, as seen in these graphs at the inq http://www.theinquirer.net/default.aspx?article=39580 the 16 rop's to NV's 24 appears to be startving the shaders in many situations. i don't understand that move at all. It should have had 32 at least to keep up with the rest of the procesing power. It shines whn the tmu/rop's aren't in play and looks crippled in every other situation. Shame on you ATIMD!!!

[joesaiditstrue]

has anyone seen the chile hardware review? the 2900XT came out on top in almost all of their tests...

http://www.chilehardware.com/Revisiones/Tarjetas-de-Video/ATI-Radeon-HD2900XT-200705131808/4.html

[Nandro]

Also, think of the mid range if they furthur cropple the rop/tmu's? They need to scrap this 250Watt dud and chalk it up to experience like Nv did with the FX 5xxx series.

[ixq]

has anyone seen the chile hardware review? the 2900XT came out on top in almost all of their tests...

http://www.chilehardware.com/Revisiones/Tarjetas-de-Video/ATI-Radeon-HD2900XT-200705131808/4.html

it was only against Both GTS models. and barely won..
BUT the DX10 benches looks mutch better. it just ran over both GTS models in that.

[rainless]

I wouldn't mind getting one of the 2600 cards. They seem to consume much less power, the price is right, and I really do like the little flame decal. :)

Unfortunately I can only buy one card, and, at the moment, it seems like it'll have t be the 8800GTS.

[joesaiditstrue]

it was only against Both GTS models. and barely won..
BUT the DX10 benches looks mutch better. it just ran over both GTS models in that.

i'm not gonna go back and paste exact numbers from their benchmarks but at 1280x1024 and higher, the 2900XT seemed to pull away generously from the two GTS cards, which is the card that this is supposed to compete with (at least in terms of performance, price/power consumption is another issue)

[rainless]

i'm not gonna go back and paste exact numbers from their benchmarks but at 1280x1024 and higher, the 2900XT seemed to pull away generously from the two GTS cards, which is the card that this is supposed to compete with (at least in terms of performance, price/power consumption is another issue)

I think you're going to have to post numbers.

Here's 1600x1200 and 1900x1200:

http://enthusiast.hardocp.com/article.html?art=MTM0MSw4LCxoZW50aHVzaWFzdA==

Still being beaten by the GTS.

[The_Iceman]

???
The [H] review seems to have disappeared. Maybe they got slapped for using the new drivers.....? Dunno, it just aint there.

[jtjuska]

The DX10 bench from Chili hardware is the really interesting one. I am surprised how well the 2900XT performed when you factor in its performance in DX9. I am not sold on its DX10 performance yet, there need to be a few more game benches that are DX10 before I can safely say one way or the other on it but it looks very interesting.

~jtjuska

[Mr. Cornell]

DX10 performance in Lost Planet:

http://forum.beyond3d.com/showpost.php?p=1004646&postcount=9

Even more ouch. The HD2900XT certainly delivers in the ouch-per-buck performance segment.

[rainless]

H review is the only review were they show lower FPS for the HD 2900XT then in any other review on all games tested. Something is wrong here and it's apparent H's review be taken with a pinch of salt.

They seem consistent with Tweaktown's review...

[Mr. Cornell]

H review is the only review were they show lower FPS for the HD 2900XT then in any other review on all games tested. Something is wrong here and it's apparent H's review be taken with a pinch of salt.


You should go read all those other reviews more carefully. When the other reviews test the same high res with high AA and high anisotropic the way HardOCP did, the numbers are similar. It's great that HD2900XT wins at 1024x768 with no AA and no anisotropic. Do you play games that way? Does anyone?

[ECH]

You should go read all those other reviews more carefully. When the other reviews test the same high res with high AA and high anisotropic the way HardOCP did, the numbers are similar. It's great that HD2900XT wins at 1024x768 with no AA and no anisotropic. Do you play games that way? Does anyone?

I think you need to comprehend the scope of my post. Your answer to my question is not about HD 2900xt winning. The question I bring up is that the frame rates are lower on the HD 2900XT from what I've seen in other reviews. Similar doesn't take the cake here. If anything you need to look at the other reviews again to see what I am talking.

[Dan0512]

Chilehardware's review was on Vista, I'm pretty sure since they ran the DX10 demo on it. Haven't studied the article closely enough.

I think I'll lay my final judgment in a week or so once the new driver is released, and when we see a detailed XP vs Vista environment and so on.

dan

[ECH]

They seem consistent with Tweaktown's review...
Can you show me 2 of the same games tested between H and TT?

[joesaiditstrue]

I think you're going to have to post numbers.

Here's 1600x1200 and 1900x1200:

http://enthusiast.hardocp.com/article.html?art=MTM0MSw4LCxoZW50aHVzaWFzdA==

Still being beaten by the GTS.

Fear
1280x1024 4xFSAA 16xAF

2900XT : 85fps
GTS640 : 87fps
GTS320 : 82fps

1600x1200 4xFSAA 16xAF

2900XT : 63fps
GTS640 : 61fps
GTS320 : 44

Prey
1280x1024 4xFSAA 16xAF

2900XT : 106
GTS640 : 107
GTS320 : 106

1600x1200 4xFSAA 16xAF

2900XT : 85
GTS640 : 82
GTS320 : 80

Splinter Cell: Double Agent

1280x1024 4xFSAA 16xAF

2900XT : 63
GTS640 : 56
GTS320 : 56

1600x1200 4xFSAA 16xAF

2900 XT : 48
GTS640 : 44
GTS320 : 41

Oblivion performance #'s showed that the 2900 XT fell between the 320 & 640MB cards

Call of Juarez DX10 Benchmark

1280x1024 Pure Speed

2900 XT : 29
GTS640 : 20
GTS320 : 16

1600x1200 Pure Speed

2900 XT : 24
GTS640 : 17
GTS320 : 14

[rainless]

Fear
1280x1024 4xFSAA 16xAF

2900XT : 85fps
GTS640 : 87fps
GTS320 : 82fps

1600x1200 4xFSAA 16xAF

2900XT : 63fps
GTS640 : 61fps
GTS320 : 44

Prey
1280x1024 4xFSAA 16xAF

2900XT : 106
GTS640 : 107
GTS320 : 106

1600x1200 4xFSAA 16xAF

2900XT : 85
GTS640 : 82
GTS320 : 80

Splinter Cell: Double Agent

1280x1024 4xFSAA 16xAF

2900XT : 63
GTS640 : 56
GTS320 : 56

1600x1200 4xFSAA 16xAF

2900 XT : 48
GTS640 : 44
GTS320 : 41

Oblivion performance #'s showed that the 2900 XT fell between the 320 & 640MB cards

Call of Juarez DX10 Benchmark

1280x1024 Pure Speed

2900 XT : 29
GTS640 : 20
GTS320 : 16

1600x1200 Pure Speed

2900 XT : 24
GTS640 : 17
GTS320 : 14

Thanks. Numbers always help :)

@ the other guy:

A game that both Hardop and Tweaktown have in Common? What about Oblivion? I already gave a link for it.

I actually think the 2900XT will be an absolute MONSTER once they control the power consumption and get it down to 65nm. I'm expecting pretty big things from the 2900XTX now. Hell even the 2600 should be pretty hardcore operating at 45 watts (!!!) and not even needing a PCI-E connection. Overall heat of the card should be a bare minimum at that low wattage which should give us plenty of room for overclocking. I'm still torn between going big at an 8800GTS 640 or going low with a 8600 GTS and overclocking it to a gigahertz or so under water. We'll have to see once I get back to Chicago.

Any word on when the 2600 and 2400 cards going going to be released?

[Rattle]

go gts 640 man why buy a cheaper card then sink all that money into cooling?
you'll still be way slower than a 640

[joesaiditstrue]

imo the 2900 xt is limited by it's 16 Texture units and ROP's, as can be seen in the technical beyond3d review

[deathman20]

I wouldn't mind getting one of the 2600 cards. They seem to consume much less power, the price is right, and I really do like the little flame decal. :)

Unfortunately I can only buy one card, and, at the moment, it seems like it'll have t be the 8800GTS.

Ya the 2600's anywhere from 45-65W Thats amazing. And at least on paper the 2600's is ~ 1/2 of the 2900XT. Can't wait to see what the die shrink does for them as well.

[deathman20]

imo the 2900 xt is limited by it's 16 Texture units and ROP's, as can be seen in the technical beyond3d review

Have to totally agree on this. While nVidia and ATI do separate things in the Texture/ROP's where the X1800/X1900's where keeping pace with the 800/7900's, brute force of more texture/ROP's helps a lot. Doubt ATI will change there unit for more on the 65nm die. Most likely we'd have to wait for the R700's to see its true advantage.

[rainless]

go gts 640 man why buy a cheaper card then sink all that money into cooling?
you'll still be way slower than a 640

Good point. :beer:

[deathman20]

Bah don't listen to him, you can get 1Ghz on the Core ;) Talk about all the girls that will love you for that.

[rainless]

Bah don't listen to him, you can get 1Ghz on the Core ;) Talk about all the girls that will love you for that.


:)

[Rattle]

lol whatever, water blocks and hose dont get me frames

[joesaiditstrue]

the way the 2900xt does it's AA is also strange, and not desirable

from anand: "Forcing MSAA resolve to run on the shader hardware is less than desirable and degrades both pixel throughput and shader horsepower as opposed to implementing dedicated resolve hardware in the render back ends. Not being able to follow through with high end hardware will hurt in more than just in lost margins."

:< looks like most of the faults of this card won't be solved through driver updates, it's just how the card was designed

[jtjuska]

Its very interesting to me that the chili review in DX10 was completely different than the results of the beyond3d results. This is the area I think that the 2900XT could really shine. Only time will tell for sure though. It will take more DX10 stuff and some fair benches between the cards to figure out who is the winner.

~jtjuska

[deathman20]

http://www.xtremesystems.org/forums/showpost.php?p=2189809&postcount=1204

This was a good catch by Ownage. There is some serious driver issues that need to be worked out so there is a lot of potential.

[ECH]

imo the 2900 xt is limited by it's 16 Texture units and ROP's, as can be seen in the technical beyond3d review
My understanding is that ROPs 16x2 or something like that.

[Nandro]

I said that so long ago that I cant think of how far back I would have to look. I believe it was in response to something ViperJohn said. The amount of shaders to ROPS is just totally disproportionate. The whole goal of DX10 was to have unification for an overall faster flow, so no part was over or underwhelmed with work. What I cant understand is that the math for this amount must have led them to believe this was the right amount. I would like to see the logic behind the decision, as I would have loved for this card to kick NV in the testes, thereby lowering prices all around, but I think NV will do as well or better in DX10 also. They have had more time, and also are already ahead in the next release as far as tweaks and changes needed to make it better while ATIMD is still trying to make their 250W space heater work right.
On a side note, didnt Antec just make a case with dual power supply's just for this card in crossfire?

[Super Nade]

XBit's review seems to be less critical of the 2900:

http://xbitlabs.com/articles/video/display/r600-architecture.html

[deathman20]

XBit's review seems to be less critical of the 2900:

http://xbitlabs.com/articles/video/display/r600-architecture.html

Not a bad review of it. Defiantly better then some of the other ones.

[thideras]

Many of you that want to overclock this video card are going to expose some incredible amounts of performance. I am thinking that with a cool case temperature and 300 watts of power feeding the card, you should see 250MHz overclocks on the core. Reaching the 1GHz+ core clock mark should be doable with the stock air cooler….if and only if, you make sure the card is getting cool air at the heatsink fan intake. Once we see water coolers for the 2900 XT there is no telling what the ceiling will pan out to be.
250 on the stock cooler?!:eek::eek:

[deathman20]

250 on the stock cooler?!:eek::eek:

200Mhz is doable on the stock cooler someone was shown doing it feeding cool air right into the card fan. It could be possible but defiantly not a suggestion hehe. Really to have a nice cool card water cooling would be the way. Which hopefully I'll be able to test this weekend.

[Shell]

200Mhz is doable on the stock cooler someone was shown doing it feeding cool air right into the card fan. It could be possible but defiantly not a suggestion hehe. Really to have a nice cool card water cooling would be the way. Which hopefully I'll be able to test this weekend. If you lap the heatsink to perfection and use a really good TIM such as LiquidPro, how much of a temperature drop would you expect?
Is the stock TIM any good?

When I got my X1900 the first thing I did was slap on my Gallium brew and enjoy a quick little boost before my first cooler shipped. :bday:

[OC Noob]

I never understood why these companies try so hard to be the fastest when they could just take their best product, price it to match the mid to mid-high range card of the competator.

They'll not have a top range offering, but only dumb@sses like us buy those. Most of the sales are mid to low range cards so if you offer a product that outperforms all the competitors products and is similarly priced where 99% of the sales are going to be, why give a rats @ss about the ultra high range?

Will the cards cost a few cents more to make, sure but wouldn't that be offset by increased sales?