X2900XT Review

[alexk]

First?
http://www.vr-zone.com/?i=4946

Edit: HardOCP's review is up!
http://enthusiast.hardocp.com/article.html?art=MTM0MSwxLCxoZW50aHVzaWFzdA==

[PaulTa]

Yup. It's been all over the overclocking forums.

I think that some of the problem might be driver related, but other than that I'm a bit disappointed in the card as far as DX9 performance goes. I play STALKER avidly, so if the 8800 series does better with STALKER I'll be buying an 8800GTS 640mb.

For those interested you can buy the HIS 2900XT from Case-mod.com for 414$. That pretty decent considering HIS usually has an enthusiast cooling solution on their cards, and they apparently are also going to bundle a voucher for a good game in there (I've heard that it's either Crysis or Half Life 2: Episode 2).

[alexk]

so if the 8800 series does better with STALKER
I'm sure it does. STALKER was heavily optimized for Nvidia cards from the beginning - I remember reading a very old interview with STALKER's team where they specifically mentioned Nvidia and all the optimizations they were doing for their cards (unfortunately I can't find a live link anymore), plus there were numerous posts from ATi users all over the official forums (especially Russian ones) complaining about horrible performance with "full dynamic lighting" enabled.

[jonnyGURU]

It's driver and currently available benchmarks.

We can't give the card a fair shake until we see it on 3DMark'07 running Vista. :D

[SKYMTL]

Well, looks like it takes only 17W more than the 8800GTX....

That makes it consume what? 165W at the most? I told you so....

[alexk]

Well, looks like it takes only 17W more than the 8800GTX....

That makes it consume what? 165W at the most? I told you so....

First of all, ATi themselves (just read the review more carefully... it not that hard to do :p ) claim that it consumes more than 200w, second of all, do you have the exact data of the 8800GTX's power consumption? Because from what I've seen there's no official specifications anywhere on Nvidia's official page and a different websites claim totally different power consumption for it, up to 200w at full load:
http://www.pcworld.in/review/article.jsp/artId=4685885/secId=3342337

[SKYMTL]

200W at full load???? Are you kidding me?

Hilbert over at Guru3D has shown time and again that they take sufficiently LESS than that. Take this for example:

http://www.guru3d.com/article/Videocards/408/4/

Two of the Water Cooled editions PLUS THE REST OF THE SYSTEM under load draw 471W. So are you telling me his X6800, Raptor HD, 2GB ram, fans +++ draw ONLY 71W? In addition to that, the reading at the outlet is done AFTER efficiency is factored in. So, even if the PSU being used was working at 80% efficiency the components would only be drawing about 377W. 200W per 8800GTX....I don't think so

Don't be so naive.

In addition, the reviewer at VR Zone mentioned what the ATI mouthpiece SAID but if you read further you will see that they actually TESTED it under load conditions with a clamp meter:

http://www.vr-zone.com/?i=4946&s=9

I see 17W more than a GTX under load.

[PaulTa]

So, what's the general Jonnyguru concensus here... the X2900XT a success at the 400$ price point?

I know it's hard to pass judgment since (as Jonny stated) we don't have proper benchmarks for it, but I'm personally disappointed with the DX9 performance. I'd say that as a direct competitor for the 8800GTS 640mb, it's a failure.

[CTman]

Here's another review (a very short one though) of the HD 2900XT:
http://www.tbreak.com/reviews/article.php?cat=grfx&id=511&pagenumber=1

[PaulTa]

Damn, now that review puts the 2900XT above the GTS... What a sad, confused little hardware enthusiast I am.

I need to buy a card tomorrow, so here's hoping that something more definitive comes out!

[alexk]

lots of stuff which can be summarized in single sentence
So in other words, you don't have an exact power consumption of a single 8800GTX card? Just like I thought :p

In a more serious note, all these "show a full system power consumption" benchmarks are pretty useless (in my opinion) when it comes to trying to determine the power consumption of the video cards - first of all, I don't really know HOW exactly these total power loads were measured (did they use the "min/max" function on multimeter (if it even has one) or did they simply stared at its LCD once in a while and recorded the highest wattage they managed to see? They don't exactly say that in any of the reviews you've linked and I wasn't physically present at these reviews, so why should I assume they did it the "right way"?). Second of all, ALL of the games/benchmarks they've used for testing might simply NOT provide maximum load for these cards - after all they are OLD, DX9-based games/demos which do not use any of the DX10-specific hardware functions that are present on the G80/R600 GPU's. Last of all, I'd rather trust ATi than some random review site since ATi themselves have ALL the proper equipment (and software, including possibly the custom-made DX10 benchmarks) to test the absolute maximum power consumption of each reference card (or maybe even GPU) itself, without trying to do some random approximations based on some specific system's total power consumption :p

[madmat]

Most sites use a power angel (or kill-a-watt) and read the wattage display while running a benchmark. After that they divide the AC wattage used by the efficiency of their particular PSU to arrive at the DC wattage used. Simple no?

[SKYMTL]

Or they take a card they KNOW the wattage requirement of and then measure the full system load with that card installed. Then they put the new card in the same system, take the reading and subtract the difference to see how much more or less the new card uses.

[SKYMTL]

So in other words, you don't have an exact power consumption of a single 8800GTX card? Just like I thought :p

We do have it:

http://www.xbitlabs.com/articles/video/display/gf8800gtx-roundup_6.html

Xbitlabs does VERY thorough tests when it comes to power consumption.

[alexk]

We do have it:

http://www.xbitlabs.com/articles/video/display/gf8800gtx-roundup_6.html

Xbitlabs does VERY thorough tests when it comes to power consumption.

I can see that they did put some effort to try to measure it objectively (such as actually trying to measure the power delivered through PCI-E slot by adding some shunts into the power lines of the slot and connecting them to multimeter), unfortunately their results still do not correlate with any of the various results I've found on internet (claiming anywhere from 145w (some Beyond3D article) to 177w (Wikipedia table) and all the way up to 200w (PC World India article)) plus they only done tests with 3DMarks, which is, as I said before, highly unlikely to provide a full load to such GPU (no DX10 features), so I can't really consider them as valid results :p

[alexk]

B.t.w, whoever was interested in results of STALKER benchmarks:
http://enthusiast.hardocp.com/article.html?art=MTM0MSwxNCwsaGVudGh1c2lhc3Q=
"Min/Max/Avg framerates for the ATI Radeon HD 2900 XT were 4/40/21.3. The GeForce 8800 GTS 640 MB had 24/82/38.9 framerates in this game"

Like I said before, STALKER is a very Nvidia-optimized game (http://blogs.nvidia.com/developers/2007/03/stalker_intervi.html), so these results are not really surprising and I highly doubt that they will change with any further driver releases :p

[CTman]

Here's the giant review (26 pages!) of the HD 2900XT by GURU3D:
http://www.guru3d.com/article/Videocards/431/1/
(Dynamic lighting is still disabled in the S.T.A.L.K.E.R: Shadow of Chernobyl test).

Power consumption HD 2900XT (vs. Geforce 8800):
http://www.guru3d.com/article/Videocards/431/13/

[Spectre]

B.t.w, whoever was interested in results of STALKER benchmarks:
http://enthusiast.hardocp.com/article.html?art=MTM0MSwxNCwsaGVudGh1c2lhc3Q=
"Min/Max/Avg framerates for the ATI Radeon HD 2900 XT were 4/40/21.3. The GeForce 8800 GTS 640 MB had 24/82/38.9 framerates in this game"


Seems crossfire didn't work so well at 55c.......

[jonnyGURU]

We do have it:

http://www.xbitlabs.com/articles/video/display/gf8800gtx-roundup_6.html

Xbitlabs does VERY thorough tests when it comes to power consumption.

They do. But their figures need to be used in context. That's why they always group them together.

You can't assume that the numbers that Xbit puts up are "maximum" or even "average." All we can gather from the numbers is that a 7950 GX2 draws more power running the same application as an 8800GTX. That's all.

From Xbit:


We loaded the GPU by launching the first SM3.0/HDR graphics test from 3DMark06 and running it in a loop at 1600x1200 resolution and with enabled 4x full-screen antialiasing and 16x anisotropic filtering. The Peak 2D load was created by means of the 2D Transparent Windows test from Futuremark’s PCMark05 benchmarking suite. We did not test the XFX GeForce 8800 GTX XXX Edition separately. Its shader processors, which are the biggest consumer in the GeForce 8800, are clocked at the same frequency as the reference card’s, while the growth of the frequencies of the main domain and memory is not as big as to influence the card’s power draw much.


So there's no way that's the most power the cards draw.

And FYI: If you're using 3DMark'06 to benchmark power consumption, the scene with the Zeppelin flying around the sea dragon (called "Canyon Fight") uses more power than anything else in 3DMark'06.

[SKYMTL]

And running CoH uses more power than pretty much anything I've tried...period.

Even running ATItool and Orthos together takes less power than CoH.

[jonnyGURU]

And running CoH uses more power than pretty much anything I've tried...period.

Even running ATItool and Orthos together takes less power than CoH.

Correct.

[PaulTa]

Here's my take on it-

Hello all,

I, like most of you, have been eagerly anticipating the arrival of the latest and greatest ATI offering, the HD 2900XT. Now that we have it in stock at the local e-tailers, a flood of reviews have come out. Everything from performance at insanely high resolutions to discussion about the weight of the cooler has been mentioned, but what's the verdict? With over 20 reviews out and about right now, it's hard for someone to make a snap decision. Well here's my take on it, from the perspective of someone looking for the best graphics solution at the 400$ price point.

First off, let me say that I'm very picky about the reviews that I accept. I like to see the test system used, detailed explinations of the settings, and graphs that are easy to read. My first and last resource for some of the best reviews on the internet is [H]Enthusiast. Why? They have all of the above in an easy to read format, and enough definitive tests to convince an agnostic.
(Here they are- http://enthusiast.hardocp.com/article.html?art=MTM0MSwxLCxoZW50aHVzaWFzdA==)

So is the latest and greatest ATI offering worth the money spent? To be brief, I'd say no. Now before you get your panties in a wad, let me explain.

Price Point-
At 400+ United States Dollars, the HD 2900XT doesn't come cheap. It sits nestled between the 8800GTS 640mb and the 8800GTX, both of which I consider superior solutions for the money. The 8800GTS 640mb can be had for as low as 330$ on newegg AR, and the GTX is closing in on the 500$ price point as well. Sitting pretty between the two means that the HD 2900XT needs to offer some considerable performance boosts compared to the 8800GTS 640mb, and that is not something it has been able to do.

Performance-
6 months of delays has netted ATI a card that can't currently compare to it's direct price competitor (only considering MSRP, not actual retail), the 8800GTS 640mb. Anywhere from 10% to 30% under the Nvidia solution is where the HD 2900XT consistently sits, and that's just not enough for 400$. At 300$, this would be an excellent choice.

Power/Heat-
Even on an 80nm manufacturing process the HD 2900XT expells more heat than it's direct competitor, and draws nearly 50w more than the GTX. This is frankly unacceptable, especially considering that most GPUs sit right under a very heat sensitive component, the CPU (and in some cases the NB as well).


So what do we have? A component that you pay more for, and in return get less performance, higher power draw, and higher heat output. I can understand a life-long ATI enthusiast toughing it out with this card, but for anyone else to make the purchase right now is absurd. At the very least, wait for DX10 to see if the HD 2900XT is a complete failure or not.

The results clearly show this huh? In what, artificial benchmarks like 3dmark?

Check the review I linked you. I personally consider it one of the most thorough and complete review methodologies on the face of the planet for graphics card testing. It's not as if their results are way from left field either. Many reviews show the HD 2900XT playing second fiddle to the 8800GTS 640mb. Considering that the 8800GTS 640mb comes in at about 50-70$ cheaper, requires less power, and puts off less heat, I'm inclined to say that Nvidia has the winner right now.

Like I said, DX10 may yield different results. Even if it does, I wouldn't consider "hoping for the best" the right thing to do with 400 hard earned USD. Hell, it could out-perform the GTS 640mb by 10% consistently with better drivers, and I still would say that the performance increase doesn't justify an extra 50-70$, more heat, and more power.

[SKYMTL]

The HardOCP review is skewered oddly considering the current crop of ATI drivers seem to be having issues with AA implementation. So, using their example makes your argument moot at best.

Personally, I hold no love for HardOCP's testing methodology with graphics cards but they are great with everything else. In addition, there seems to be some VERY odd discrepancies when it comes to power consumption.

[PaulTa]

If all other reviews were indicating that the HD 2900XT was better than the 8800GTS 640mb, I would have been suspicious of this review. Unfortunately that isn't the case.

[alexk]

And running CoH uses more power than pretty much anything I've tried...period.

Even running ATItool and Orthos together takes less power than CoH.

In my system I get the highest power consumption when I run the Oblivion (and yes, I do have CoH too and I did test it using an actual game replay rather than built-in benchmark)... But then again, I only have a lowly 7900GTO and I use a tweaked oblivion.ini file (to enable multithreading for certain things as well as for general performance improvements... all of them are documented at www.tweakguides.com), so my results can't be really compared with the other systems (which have different cards and more likely use an unmodified version of Oblivion)...

[Super Nade]

I think it is only appropriate that folks wait for a mature driver-set to roll out before calling the card a flop.

[Spectre]

I think it is only appropriate that folks wait for a mature driver-set to roll out before calling the card a flop.

I think manufacturers should ship products that have drivers that work decently ;)

[alexk]

I think it is only appropriate that folks wait for a mature driver-set to roll out before calling the card a flop.

A universal (and absolutely redundant) excuse which can theoretically apply to ANY GPU manufacturer (both ATi and Nvidia) at any given time :p :D

Edit: Here's a nice example of what I'm talking about, taken from today's Guru3D review:
"So we started testing an 8800 GTX with this software [Call of Juarez demo], and yes ... stumbled into bad performance and weird performance issues.....Later that evening NVIDIA noticed that the press obtained the demo and one day later issued a driver which was showing excellent performance and no weird stutters or anything anymore. The driver is the 158.42 driver which be will be released to the public this week."

[Super Nade]

I think manufacturers should ship products that have drivers that work decently ;)

Ideally, yes. But this happens everywhere (sadly). Motherboards with crap BIOS'es etc...AMD handled this badly though. Crap drivers + delayed product is not a very good thing. :)

From a technical standpoint, I'm just curious to see if there is any merit to AMD's architecture.

[nicolasb]

Dissappointing, dissappointing, dissappointing. :(

You may recall I posted a few months ago saying I was considering the possibility of putting together an R600 crossfire system, subject to favourable benchmarks on release (and I received a tirade of vitriolic abuse as a consequence). Now that the benchmarks have come out... well, I am no longer considering this possibility.

(sigh)

What's particularly disheartening is that all of the areas where R600 should be particularly strong have actually turned out to be its most serious weaknesses.

It ought to be at its best in shader-heavy games, but it isn't.

The massive memory bandwidth it gets from the 512-bit bus ought to make it excel at very high AA and AF settings, leading to excellent image quality (traditionally ATI's strength); but the reality is that R600 suffers more of a performance hit with increasing AA than G80 does (possibly because the MSAA hardware is actually broken and it has to use the shaders to do AA instead of the ROPs).

It also takes a much larger hit with high AF settings than G80 does (possibly because of a woeful lack of texturing hardware).

Its AF is worse than that of G80 in terms of the image quality.

The new AA modes are superficially interesting but (judging by preliminary reviews) cause too much blurring of the overall image to use in practice.

The HDMI-audio and hardware MPEG4 and VC-1 decoding ought to make it ideal for an HTPC setup, but (according to the reviews) the noise level that its cooling system generates is so high that it's useless for anything AV-related.

I've been waiting since January for ATI to release the bloody thing, because my new build really must combine high-end gaming performance with hardware decoding of high-definition MPEG4; but, even now R600 is actually out, there is still no suitable product on the market!

:wall:

I suppose I'm just going to have to buckle down and wait another three or four months until R650 appears and hope that it's the product R600 should have been. :mad:

[alexk]

my new build really must combine high-end gaming performance with hardware decoding of high-definition MPEG4; but, even now R600 is actually out, there is still no suitable product on the market!


So you think that Nvidia's cards (such as 8800GTS or GTX) can't provide these 2 things together :confused:

[nicolasb]

So you think that Nvidia's cards (such as 8800GTS or GTX) can't provide these 2 things together :confused:Well, no, they can't. Otherwise I wouldn't have a problem, would I? :rolleyes:

The newer mid-range G8x chips do offer hardware MPEG4-decode, but the original G80 ones (8800GTS and 8800GTX) don't. The 8600 and below don't offer the level of gaming performance I'm after.

[burebista]

X-Bit: ATI Radeon HD 2000 Architecture Review (http://www.xbitlabs.com/articles/video/display/r600-architecture.html).
Legit Reviews: DirectX 10 Benchmarking - Lost Planet: Extreme Condition (http://www.legitreviews.com/article/505/1/).

[alexk]

Well, no, they can't. Otherwise I wouldn't have a problem, would I? :rolleyes:

The newer mid-range G8x chips do offer hardware MPEG4-decode, but the original G80 ones (8800GTS and 8800GTX) don't.

WTF are you talking about :confused: ALL of the Nvidia's cards starting from certain GF68xx models have a MPEG-4 (for all codec implementation, such as h.264, VC-1 and MPEG-2, regardless of the bitstream or profiles) hardware acceleration:
http://www.nvidia.com/page/purevideo_support.html

Same goes for ALL of the 8800 series of cards:
http://www.nvidia.com/page/8800_features.html

Even my 7900GTO does the hardware acceleration/offloading when I play 1080p h.264 movies (from Apple's website) using WMP11 (which supports DXVA and, therefore, Nvidia's PureVideo) - I get less than 45% CPU utilization with my AMD 3800+ CPU :p

The only significant thing that newer cards (8600 and 8500) add is a better offloading in terms of CPU utilization (up to 100% offloading), nothing more.

[nicolasb]

Even my 7900GTO does the hardware acceleration/offloading when I play 1080p h.264 movies (from Apple's website) using WMP11 (which supports DXVA and, therefore, Nvidia's PureVideo) - I get less than 45% CPU utilization with my AMD 3800+ CPU :pYes, but I'm not interested only in downloaded material, I'm also interested in HD-DVD and BluRay discs. If we check some figures from Anandtech you'll see that CPU utilisation during BluRay playback can hit 99% using a Core 2 E6300 and an 8800GTX card. That's much too high for me to be comfortable with.

http://www.anandtech.com/video/showdoc.aspx?i=2886&p=4 (third graph)

You are correct that the 8800 cards offer some MPEG4 decoding facilities, but they don't offer enough. If they did as much as G84/G86 do I would be happy. But they don't.

[alexk]

You shouldn't trust these single outdated benchmarks so blindly :D First of all, their measured "maximum CPU utilization" could be a single momentary spike which caused NO frame drops and NO playback slowdowns (even if it was 100%). Second of all, that test is outdated - here's a newer benchmark, using latest Nvidia drivers and latest version of PowerDVD:
http://www.anandtech.com/video/showdoc.aspx?i=2977&p=5
http://www.anandtech.com/video/showdoc.aspx?i=2977&p=4
Notice the (meaningless) "maximum" graphs - they don't reach even the 90% for a 8800GTX card. And that is with h.264 (the most demanding out of all MPEG-4 codecs).
Anyway, if you have a motherboard with dual 8x/16x PCI-E slots, you can always buy a cheap (they are about $90 right now) 8500-series card and use it as a secondary card just for a MPEG-4 movie playback... Or use these $$$ for a faster CPU. You really don't have to depend on a single company (ATi) for all of your demands :p

[madmat]

nicoslab, are you overclocking that 6300 at all? If you notice, core speeds do play a part in the maximum CPU utilization on that chart. While it may look grim for a stock clocked E6300 the outcome may well be a lot better for someone running theirs with the chip balls to the wall.

[nicolasb]

nicoslab, are you overclocking that 6300 at all? It's a hypothetical 6300, which could well turn into a hypothetical 6600 when no one is looking. :) I'll probably be overclocking when gaming, but not necessarily all the time. Noise is more of an issue for video playback, and the system is going to be noisy enough with a high-end video card in there without running the CPU cooler flat-out as well. If anything, it would be nice to have the freedom to underclock when desired.
http://www.anandtech.com/video/showdoc.aspx?i=2977&p=4
Notice the (meaningless) "maximum" graphs - they don't reach even the 90% for a 8800GTX card.Personally I think the fact that it does reach 89% is rather more telling. :) That's just too close to the point where it's dropping frames for me to be comfortable with. Even if it only drops them occasionally, once is enough. Compare that with the figure for the 8600GTS on the same graph - just 33% CPU. Big difference.

It isn't actually just CPU offload that I'm interested in anyway, although that's the most important thing. The HDMI implementation on R600 is much better than it is on G80. Proper audio via HDMI, an implementation of HDCP that isn't broken in the way that it is in G80 (ask anyone who has tried to play back HDCP-protected video on a 30" Dell monitor!) and so on. Individually these are not make-or-break features, but they add up.

[alexk]

Compare that with the figure for the 8600GTS on the same graph - just 33% CPU. Big difference.


Yea, the CPU usage difference is big, but I still do not see how it translates into stuff like dropped frames (or slowed down playback). Nobody has shown that, for example, even with 90% CPU utilization you will get at least one dropped frame :p


It isn't actually just CPU offload that I'm interested in anyway, although that's the most important thing. The HDMI implementation on R600 is much better than it is on G80. Proper audio via HDMI, an implementation of HDCP that isn't broken in the way that it is in G80 (ask anyone who has tried to play back HDCP-protected video on a 30" Dell monitor!) and so on. Individually these are not make-or-break features, but they add up.

Well, if I would be playing the HD movies on my PC, I would just use a certain software (Wikipedia mentions it) which would allow me to playback ANY Blu-Ray or HD-DVD on my PC without HDCP getting into way (regardless of the display and video output/input type) ;) I see nothing wrong with doing that since I want to watch my (legally bought) movies the way I want it and all this "copy protection" BS only benefits the companies who invent these things, nobody else (not movie studios, not movie actors/artists) :p About sound over HDMI - I guess you want to hear something other than plain old DD/DTS sound? If yes and your receiver already supports stuff like DTS-HD and Dolby TrueHD decoding, then yea, I can see the usefulness for having a proper HDMI audio output.
Anyway, personally I would (and probably will) rather buy a stand-alone HD-DVD and/or Blu-Ray (or a combo player) - in my opinion, it is much more convenient to use: you get a proper remote control, you're basically guaranteed to have a proper performance (no dropped frames ever), you get less noise (these things do have fans but they are much, MUCH more quiet than any average PC, especially any PC with this ATi's new abomination (2900XT)), they eat significantly less power and they provide a plenty of video/audio outputs for just about any type of display (even for the ones with a component input only - I don't think that there's any "copy protection" implemented for it yet, and probably will never be). These things aren't cheap, but neither are the similar PC drives :p Of course, I am speaking from a position of a person with a proper entertainment system (with no silly things such as PC's being used for it and with plenty of room to put all the stand-alone players near it) :p