Initial thoughts about HD2900XT

[Blackstone]

I installed the HD2900XT today. I am using the 8.37.4.3 drivers (the file itself is called 47097) that I downloaded from the ATI website today.

My system is
E6600 @ 3.4 Ghz
2 Gigs Ocz Ram @ Cas 3, 1:1
74 gig raptor
PCP&C Silencer 750 quad PSU

I can sort of confirm one aspect of the [H] review. With HDR and 4x aa forced at the driver level, Oblivion runs great, 1680 x 1050 4xaa, HDR, everything maxed, except for areas that include extremely high levels of trees. I don't have any problems with grass distance, per se, it is only when I get into really dense forest areas that things slow down quite a bit. This is not because of the grass, however, it is because the complexity of the trees. This part of the card's performance is quite disappointing. It is playable at these settings however, and the results are for the most part, stunning. For the most part there is no lag at all. 2xaa runs much better with these settings in those really dense areas. It is the kind of thing that makes you cringe when it happens, however. Does anyone have a rational explanation as to where the deficiency might be with respect to the ATi hardware?

It is sort of late so that is all I will say right now, I'll post more tomorrow. I don't plan on giving objective numbers because I don't have that kind of time, but I will give some subjective analysis. All I can say right now is I really hope that future drivers improve performance in Oblivion. This is still a big upgrade from my X1800XT, however.

Also note as a preliminary matter that there is very little subjective improvement in BF2. Increasing aa to 8x did not yield a significant improvement in IQ. The difference between 4x and 8x aa is not that great. If BF2 is your main game you don't really need this card yet.

Also, I will add, no thermal problems or excessive noise. Of course, my system is not quiet, and this is not a quiet card, but the drivers are handling the fan very well. It spins up when it should and when I alt-tab out of a game it spins back down to acceptable levels very quickly. Under heavy load, this card is noisy, but not as noisy as my Tuniq Tower 120 on high.

Also note it did not increase my case temps or cause any instability. It really blows that hot air out pretty hard, which is good. It seems to be doing a good job of managing the heat, which was something I was really worried about. I do have a very well ventilated case however--Silverstone TJ07 with top exhaust fans, rear intake fans for the cpu, separate hard drive cages with fans, a 120mm front intake fan for the video card and mobo, and an expansion slot cooler that I could probably do without now.

 

[Blackstone]

 

[trick_m0nkey]

I honestly hate 3d mark scores. Keep playing games and keep us updated.

 

[Blackstone]

I honestly hate 3d mark scores. Keep playing games and keep us updated.

I don't really believe in using 3dmark scores to compare competing graphics cards of the same generation because I think actual games are more useful for that. However, we all know how demanding these tests are and how they run on our current hardware, so the scores--especially the framerates of the video card tests, are very useful in that respect.

 

[buffbiff21]

3dmk scores are in no way the universal way of determining the strength of a card.. it is only a way for manufacturers (nvidia, ati) to get a general idea of the power.

other things like driver optimization and certain game's programming come into play.

id take real world benchies anyday.

 

[Silus]

I installed the HD2900XT today. I am using the 8.37.4.3 drivers (the file itself is called 47097) that I downloaded from the ATI website today.

My system is
E6600 @ 3.4 Ghz
2 Gigs Ocz Ram @ Cas 3, 1:1
74 gig raptor
PCP&C Silencer 750 quad PSU

I can sort of confirm one aspect of the [H] review. With HDR and 4x aa forced at the driver level, Oblivion runs great, 1680 x 1050 4xaa, HDR, everything maxed, except for areas that include extremely high levels of trees. I don't have any problems with grass distance, per se, it is only when I get into really dense forest areas that things slow down quite a bit. This is not because of the grass, however, it is because the complexity of the trees. This part of the card's performance is quite disappointing. It is playable at these settings however, and the results are for the most part, stunning. For the most part there is no lag at all. 2xaa runs much better with these settings in those really dense areas. It is the kind of thing that makes you cringe when it happens, however. Does anyone have a rational explanation as to where the deficiency might be with respect to the ATi hardware?

It is sort of late so that is all I will say right now, I'll post more tomorrow. I don't plan on giving objective numbers because I don't have that kind of time, but I will give some subjective analysis. All I can say right now is I really hope that future drivers improve performance in Oblivion. This is still a big upgrade from my X1800XT, however.

Also note as a preliminary matter that there is very little subjective improvement in BF2. Increasing aa to 8x did not yield a significant improvement in IQ. The difference between 4x and 8x aa is not that great. If BF2 is your main game you don't really need this card yet.

Also, I will add, no thermal problems or excessive noise. Of course, my system is not quiet, and this is not a quiet card, but the drivers are handling the fan very well. It spins up when it should and when I alt-tab out of a game it spins back down to acceptable levels very quickly. Under heavy load, this card is noisy, but not as noisy as my Tuniq Tower 120 on high.

Also note it did not increase my case temps or cause any instability. It really blows that hot air out pretty hard, which is good. It seems to be doing a good job of managing the heat, which was something I was really worried about. I do have a very well ventilated case however--Silverstone TJ07 with top exhaust fans, rear intake fans for the cpu, separate hard drive cages with fans, a 120mm front intake fan for the video card and mobo, and an expansion slot cooler that I could probably do without now.

My take on why the HD 2900 XT performs poorly, especially when taking into account its specs, is efficiency.

G80 has an amazing scheduler (or thread processor) that distributes the workload (which is a mix of vertex, geometry and pixel shaders) through all its available stream processors, very effectively. This means the stream processors are working 100% to achieve the common goal, of rendering game's frames (for example), This is the main purpose of an unified architecture.

R600 however, even despite the fact that it has more stream processors - 320 SPs vs 256 SPs of the 8800 GTX (256, because if we count NVIDIA's stream processors the same way ATI counts theirs, a 8800 GTX would have 256 SPs) - is obviously not being as efficient in distributing the same workload, through all its stream processors. R600 seems to have it all in terms of specs, but it's clearly not taking advantage of them.
The fact that it consumes so much power, indicates that it's indeed trying to use all its stream processors, but since performance is so poor, my guess is that it really is not doing so, in an efficient manner.

 

[Decibel]

Lol, you guys are getting a little extreme with this anti 3dmark thing.

I know that a lot of us when we get a new bit of kit run it through a lot of different benches. Sure it's useless when it comes to games, but it's still fun to see what kind of numbers it pumps out.

I always download a bunch of tech demos too.

Thanks for the info Blackie. Lots of us are interested.

 

[Silus]

Lol, you guys are getting a little extreme with this anti 3dmark thing.

I know that a lot of us when we get a new bit of kit run it through a lot of different benches. Sure it's useless when it comes to games, but it's still fun to see what kind of numbers it pumps out.

I always download a bunch of tech demos too.

Thanks for the info Blackie. Lots of us are interested.

The HD 2900 XT actually shows how worthless the use of 3DMark is. The XT performs amazingly in 3DMark, yet in games, where it matters, it performs poorly.
So yeah, you might like to see what numbers it provides, but it's no real indication of performance in the real tests - games.

 

[Decibel]

Are you arguing or agreeing, I can't tell... In no way was I saying that 3dmark matters when it comes to opening my wallet.

You:
...it's no real indication of performance in the real tests - games.

Me:
...it's useless when it comes to games...

I think it's pretty stupid that Blackstone is catching any grief for running 3dmark. People get new gear, they're going to test it.

 

[Silus]

Are you arguing or agreeing, I can't tell... In no way was I saying that 3dmark matters when it comes to opening my wallet.

I think it's pretty stupid that Blackstone is catching any grief for running 3dmark. People get new gear, they're going to test it.

Arguing ? People are really touchy these days...

I was just sharing my opinion on the subject and used the XT to back up the claim that 3DMark is really worthless,

 

[debello_64]

I find 3dmark is usefull when you are comparing two identical/similar machines. It has its place in this world, just not in the NV vs AMD debate.

The 2900XT is not 8800 series kill, I have no delusions of grandeur on that fact. But me being a AMD hommer I bought it (the thought of no nvidia or intel inside brings me joy), not a fanboy I am able to admit when mine doesnt perform on the same level. Although I know the 2900xt will get better over time, it may never beat the GTS or the GTX, but close enough is good for me.

Besides it has flames on it, how can you go wrong with that.

 

[the gamer]

I am getting a 2900XT in July, yes I am a AMD/ATi fanboy, though I was thinking of getting a GTS couple months ago, but bought other stuff instead.

 

[Silus]

I am getting a 2900XT in July, yes I am a AMD/ATi fanboy, though I was thinking of getting a GTS couple months ago, but bought other stuff instead.

In the end, it's your call of course.
It's just that some of us (me included) don't understand this fanboy thing. Why would anyone spend money on a product, that performs more or less the same as the competition, while costing more and sucking more power, producing more heat and noise. The Black Box doesn't even make a difference anymore, since Valve in all its wisdom cancelled it.

 

[debello_64]

The Black Box voucher will still work, the general agreement reached on Valves forum is that the Black Box cancellation is for the retail version only. Mainly because B&M stores where saling EP 1 for far less then you could buy it on Steam. If ATI had to endorse the Half Life 2 vouchers we got with the 9800, I am sure these will be no different.

 

[Dethman]

It seems to me, everyone worshiped 3dmark scores, then the 2900 comes out, and everyone says they are BS... I think given time, with drivers, or the next revision, ati will be back on track

 

[mascaras]

The Black Box voucher will still work, the general agreement reached on Valves forum is that the Black Box cancellation is for the retail version only. Mainly because B&M stores where saling EP 1 for far less then you could buy it on Steam. If ATI had to endorse the Half Life 2 vouchers we got with the 9800, I am sure these will be no different.

Valve Black Box coupons still valid for R600 buyers

The recent news of the retail black box cancellation has raised the question if buyers of the HD 2900 XT, who also received a coupon for a free "Black Box" will receive an upgrade to the orange variant. Valve has now confirmed that only the retail offering has been axed and anyone with a coupon will still receive the same games as found in the Valve Black Box. Sorry guys - no free Episode One and Half-Life 2 for y'all.

Source: TheINQ

http://forums.techpowerup.com/showthread.php?t=31697

regards

 

[Silus]

It seems to me, everyone worshiped 3dmark scores, then the 2900 comes out, and everyone says they are BS... I think given time, with drivers, or the next revision, ati will be back on track

What does "everyone" mean to you ?
Because that "everyone" certainly does not include the [H] crowd. We don't really like to use 3DMark, to show what a graphics card can do, for a while now.

 

[Mat3]

R600 however, even despite the fact that it has more stream processors - 320 SPs vs 256 SPs of the 8800 GTX (256, because if we count NVIDIA's stream processors the same way ATI counts theirs, a 8800 GTX would have 256 SPs) -

I'm pretty sure that's not quite accurate either. For the R600 shaders, in the 5-processor block, the fat one can do most of the normal stuff in addition to the extra functionality, while that extra unit in the G80 is quite limited in what it can do. So a perfect comparison of the quantity of shaders is impossible.

 

[LawGiver]

i still laugh at those morons who said the 2900 is so hot it will limit your system wide overclocking. They were so desperate to criticize the 2900 they would say anything no matter how ridiculous it sounded.

 

[Silus]

I'm pretty sure that's not quite accurate either. For the R600 shaders, in the 5-processor block, the fat one can do most of the normal stuff in addition to the extra functionality, while that extra unit in the G80 is quite limited in what it can do. So a perfect comparison of the quantity of shaders is impossible.

You are confusing two things. What stream processors can do and how can we count them. I'm talking about the latter and if we count NVIDIA's SPs the same way ATI counts theirs, then a 8800 GTX has 256 SPs.
And you are talking about what stream processors can do, in both cards. That equates to the peak FLOPS capacity of the graphics card, not how many stream processors it has.

 

[anonuser]

If the only game you play is 3dmark, the hd2900xt is a good choice money wise over the gtx

 

[Ranari]

Besides it has flames on it, how can you go wrong with that.

I put stickers on my car and gained 10hp! (Sorry, couldn't resist, lol)

And yes, I'd probably buy it for the flames too.

It's just that some of us (me included) don't understand this fanboy thing. Why would anyone spend money on a product, that performs more or less the same as the competition, while costing more and sucking more power, producing more heat and noise.

I don't know, but I'll take a stab at it: For some people, they just have money to burn and are curious. I used to do this when I was younger and had no bills to pay (this is not implying that the people doing this are young and/or have no bills to pay), and would go out and buy both a Geforce2 GTS and a Voodoo5 5500 just for the sake of my own personal curiosity. Some, like me, might also realize that regardless of what you're upgrading from, both a 2900XT and a GTS/GTX are going to provide wicked fast gaming experiences. Even despite its drawbacks, the 2900XT is no slouch. It's not like it's the same as comparing the Geforce 5800 to the Radeon 9700. The 2900XT at least keeps up with the GTS and shows promise of only getting better. Lastly, I think some of us realize that competition in the long run is greater than a temporary short coming. ATI has proven to provide great hardware to the community. They just either underestimate or flat out got lazy this time around, but that doesn't mean they're incapable (like 3dfx was). If Nvidia has no competition, then the consumer will pay for it.

And, lastly, maybe they're emo and just want to be different!

 

[Cyrilix]

I put stickers on my car and gained 10hp! (Sorry, couldn't resist, lol)

And yes, I'd probably buy it for the flames too.



I don't know, but I'll take a stab at it: For some people, they just have money to burn and are curious. I used to do this when I was younger and had no bills to pay (this is not implying that the people doing this are young and/or have no bills to pay), and would go out and buy both a Geforce2 GTS and a Voodoo5 5500 just for the sake of my own personal curiosity. Some, like me, might also realize that regardless of what you're upgrading from, both a 2900XT and a GTS/GTX are going to provide wicked fast gaming experiences. Even despite its drawbacks, the 2900XT is no slouch. It's not like it's the same as comparing the Geforce 5800 to the Radeon 9700. The 2900XT at least keeps up with the GTS and shows promise of only getting better. Lastly, I think some of us realize that competition in the long run is greater than a temporary short coming. ATI has proven to provide great hardware to the community. They just either underestimate or flat out got lazy this time around, but that doesn't mean they're incapable (like 3dfx was). If Nvidia has no competition, then the consumer will pay for it.

And, lastly, maybe they're emo and just want to be different!

I sort of agree with this. Even though I think the 8800 series is currently better, I don't think the 2900 series is a disappointment. In the end, a consumer would likely be happy with both. Also, there is no market that works in the absolute sense that the one is better will take 100% of the market. That would be disastrous. Say, one generation, nVidia is better and 100% of discrete cards sold came from nVidia, and the next generation, ATi is better, and the opposite happens -- things just don't work that way. People will be more inclined to buy one or the other due to its advantages and disadvantages, but you'd have to be god himself (disclaimer: I don't believe in god) in order to enforce that kind of market policy.

So as it is, I would not buy an 2900 XT unless things were to shape up in a major way, but I would not be quick to discredit it either (which literally half of the nV camp has been doing on this ATi board). I feel some people need to take a softer stance instead of foisting their opinions and hatred on the other camp.

 

[Blackstone]

Well, the reason why I went ATi is because I have a Crossfire chipset (Asus P5W DH Deluxe) and I plan on eventually going dual graphics. I think Nvidia's offerings are more attractive at this point, I don't have any particular attachment to ATi, but going Nvidia this round would have eliminated dual graphics as an option so I'm sticking with ATi. If I had an Nivida chipset I would have bought a single GTX.

 

[Blackstone]

This card is not a dissapointment--yet. I think the drivers need to mature a bit and we also need to know how it does with Crysis. I'd like to play Crysis at 1680 x 1050 with like 4xaa so if it gets close to that I'll be happy. I'll be less happy if the GTS can do that but this card can't. If the GTX can do it and the 2900 can't, I won't mind as much because that is a more expensive card.

 

[Anarchist4000]

Did you just quote and respond to yourself?

 

[Blackstone]

Did you just quote and respond to yourself?

No I was just supplementing my previous post, should have just edited. lol

 

[phide]

I'd like to play Crysis at 1680 x 1050 with like 4xaa so if it gets close to that I'll be happy. I'll be less happy if the GTS can do that but this card can't.

That's probably a reasonable expectation. I'd assume that both NVIDIA and AMD intend to work with Crytek, to one degree or another, and that will give both time to optimize. R600 doesn't seem to be as broadly well-performing as G80 -- it requires much more attention to compiler optimization.

Given what 3DMark shows in the way of R600's shader performance, the capability of surpassing GTS performance seems to be there.

 

[Blackstone]

That's probably a reasonable expectation. I'd assume that both NVIDIA and AMD intend to work with Crytek, to one degree or another, and that will give both time to optimize. R600 doesn't seem to be as broadly well-performing as G80 -- it requires much more attention to compiler optimization.

Given what 3DMark shows in the way of R600's shader performance, the capability of surpassing GTS performance seems to be there.

I'm trying the 8.38 RC3 drivers and they didn't really make a difference.

Also, 8xaa in Battlefield2 is just not possible it is a slideshow--20-25 fps. Does the GTS do 8xaa in BF2?

 

[MelonSplitter]

What does "everyone" mean to you ?
Because that "everyone" certainly does not include the [H] crowd. We don't really like to use 3DMark, to show what a graphics card can do, for a while now.

Hey Silus, what does "we" mean to you, because I've been a member of [H] for a while and I disagree with you. I will always use 3dMark.

 

[Brent_Justice]

I'm trying the 8.38 RC3 drivers and they didn't really make a difference.

Also, 8xaa in Battlefield2 is just not possible it is a slideshow--20-25 fps. Does the GTS do 8xaa in BF2?

The 8800 GTS will do 16X TR SSAA in BF2

 

[Blackstone]

The 8800 GTS will do 16X TR SSAA in BF2

Well how could the GTS do 16X in BF2 and the R600 can't handle 8X? I mean, BF2 is an old game already, it should chew it up and spit it out. Does ATi have an explanation for this?

 

[Sniper_Merc]

Well how could the GTS do 16X in BF2 and the R600 can't handle 8X? I mean, BF2 is an old game already, it should chew it up and spit it out. Does ATi have an explanation for this?

That whole conversion math from canuck bucks caused them to price it too high, and converting from mm to inch meant they wound up accidently using 130nm process....

 

[Brent_Justice]

Well how could the GTS do 16X in BF2 and the R600 can't handle 8X? I mean, BF2 is an old game already, it should chew it up and spit it out. Does ATi have an explanation for this?

cauuuuse the 8800 gts is faster?

did you read our review?

 

[Silus]

Hey Silus, what does "we" mean to you, because I've been a member of [H] for a while and I disagree with you. I will always use 3dMark.

"We" means the ones that like the [H] testing methodolgy better than any other and do not use synthetic benchmark tools, for the purpose of testing new cards. You're obvisouly not part of that group.

 

[Pepsiennis]

"... Some, like me, might also realize that regardless of what you're upgrading from, both a 2900XT and a GTS/GTX are going to provide wicked fast gaming experiences."

Thank you for saying that! Ferrari, Lamborghini... at this level of card, if one is splitting hairs over trivialities like power consumption, one should probably slip down a rung or two to a more prosaic level of performance. The people who are eyeing an HD 2900XT or an 8800 series card are going to get what they expected when they bought the card, even if it means chomping at the bit as drivers develop (ATI w/ 2900XT, Nvidia w/ Vista). I hope to hear more from Blackstone as to his latest ensemble (Get it... Mr. Blackstone, worst-dressed fashion list? Is this mic on? Hmph, tough crowd.) Anyway, thanks B!

 

[Blackstone]

Thank you for saying that! Ferrari, Lamborghini... at this level of card, if one is splitting hairs over trivialities like power consumption, one should probably slip down a rung or two to a more prosaic level of performance. The people who are eyeing an HD 2900XT or an 8800 series card are going to get what they expected when they bought the card, even if it means chomping at the bit as drivers develop (ATI w/ 2900XT, Nvidia w/ Vista). I hope to hear more from Blackstone as to his latest ensemble (Get it... Mr. Blackstone, worst-dressed fashion list? Is this mic on? Hmph, tough crowd.) Anyway, thanks B!

Here is the link to my latest post.
http://www.hardforum.com/showthread.php?t=1195344

 

[Deusfaux]

The HD 2900 XT actually shows how worthless the use of 3DMark is. The XT performs amazingly in 3DMark, yet in games, where it matters, it performs poorly.
So yeah, you might like to see what numbers it provides, but it's no real indication of performance in the real tests - games.

So you say it's worthless. Then you admit people might have legit uses for it.

Then you finally clarify what it's not good at doing.



3DMark has worth and value, yes we know it's not the best determinant of real-world performance, but that is only ONE particular use of the program and so it isnt completely garbage in the end.

 

[Pepsiennis]

...3DMark has worth and value, yes we know it's not the best determinant of real-world performance, but that is only ONE particular use of the program and so it isnt completely garbage in the end.

It appears to me, although I'm sure the piling on will commence shortly, that 3DMark is useful in that it is a neutral field (i.e., not optimized for a particular brand) on which to test a video card, and the system it resides in, gauging standardized performance formats. But that's more of an 'apples to apples' thing, and so won't be useful for those who prefer fruit salad.

 

[snaggletooth]

1. build your rig
2. bench your rig with 3dmark
3. compare your results to systems which contain the same components as yours
4. tweak your system until you're happy where you are in the delta

--how to use 3dmark effectively.


when used in conjunction with other games and testing software, 3dmark isn't 'useless' for the end user; no matter how many times the parrots squawk with objection.

it is useless when supplied as 'reliable' data in a published review. yes, we all understand that. thank you for pointing it out, again.




OP, keep that subjectivity comin! great thread, aside from all the life lessons we're getting about things which are completely obvious to everyone in this and all [H] discussions.