backtop


Print E-mail del.icio.us 333 comment(s) - last by t5chris.. on May 13 at 7:25 PM


The 12 inch long OEM ATI Radeon HD 2900 XTX

Sizing up the difference between the 12 inch and 9.5 inch cards
AMD's flagship ATI Radeon HD 2900 XTX fails to usurp the GeForce 8800 GTX's performance crown

AMD is close to unveiling its long-awaited R600-based ATI Radeon HD 2900 XT. DailyTech previously posted benchmarks comparing the Radeon HD 2900 XT and GeForce 8800 GTS 640MB.

Up until March of 2007, the spearhead of the ATI Radeon HD 2900-family was the upcoming ATI Radeon HD 2900 XTX. This model is the big daddy of AMD’s DirectX 10 lineup, poised to take on NVIDIA’s GeForce 8800 GTX and upcoming 8800 Ultra.

ATI Radeon HD 2900 XTX video cards feature 1GB of GDDR4 video memory, besting the GeForce 8800 GTX’s 768MB of GDDR3 video memory.  GPU manufacturers often wait to set the exact clock frequencies of bins until just weeks before launch.  Our Radeon HD 2900 XTX sample was released to board partners in the second week of April, features memory clocked at 2.02 GHz and a core clock of 750 MHz. 

Physically, the Radeon HD 2900 XTX core is identical to the Radeon HD 2900 XT core.  Both feature 320 stream processors, but the XTX differs by bringing GDDR4 to the package.

Despite the reference clock speed differences, DailyTech managed to push the ATI Radeon HD 2900 XT up to 845 MHz core and 1.99 GHz memory.

Unlike the ATI Radeon HD 2900 XT benchmarked yesterday, the HD 2900 XTX is a 12-inch card geared specifically towards OEMs and system integrators. Expect retail box ATI Radeon HD 2900 XTX graphics cards to use the shorter 9.5-inch design, as with the HD 2900 XT.

The test system specifications are as follows:
  • AMD ATI Radeon HD 2900 XT (745 MHz core, 800 MHz GDDR3)
  • AMD ATI Radeon HD 2900 XTX (750 MHz core, 1010 MHz GDDR4)
  • AMD ATI Catalyst v8.361 (drivers slated for retail)
  • NVIDIA GeForce 8800 GTX (650 MHz core, 1000 MHz GDDR3)
  • ASUS P5N32-E SLI (nForce 680i)
  • Intel Core 2 Extreme QX6800
  • Corsair XMS2 PC2-8500 (800 MHz, 5-5-5-18, 1T)
  • Acer X241W
It's important to note that the GeForce 8800 GTX is a vendor overclocked board that comes shipped with the 650 MHz core clock.

Frames per second 1280x1024
Game
Radeon HD
2900 XTX
Radeon HD
2900 XT
GeForce
8800 GTX
Company of Heroes
97.1 N/A
128.6
F.E.A.R.84
79
125
Half Life 2: Episode 1
117.9
118.9
157.1
Elder Scrolls IV: Oblivion
100.3
101.4
110.5

Frames per second 1600x1200
Game
Radeon HD
2900 XTX
Radeon HD
2900 XT
GeForce
8800 GTX
Company of Heroes
73.7 N/A
94.5
F.E.A.R.58
54
90
Half Life 2: Episode 1
91.5
90.8
134.2
Elder Scrolls IV: Oblivion
87.9
86.2
102.9

Frames per second 1920x1200
Game
Radeon HD
2900 XTX
Radeon HD
2900 XT
GeForce
8800 GTX
Company of Heroes
53.2 N/A
80
F.E.A.R.53.7
52
81.7
Half Life 2: Episode 1
68.2
67.8
100.2
Elder Scrolls IV: Oblivion
75.1
73.4
98.4

The benchmarks DailyTech performed yesterday utilized release candidate drivers. Today's tests used retail drivers ATI released to its board partners.

The less than stellar performance benchmarks are no surprise to board partners and AMD insiders.  Two independent ATI board builders told DailyTech that Radeon HD 2900 XTX will not be a part of their initial portfolios.  Given the additional memory cost and bulky card footprint, it seems unlikely the XTX will ever see the light of day.


Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Very strange...
By tenks on 4/26/2007 2:13:30 AM , Rating: 4
I am really surprised with how crappy the performance is for the XTX....This seems really odd as the XT beat the GTS right? Also I noticed the clocks for the 8800GTX are higher then the 575/900mhz clocks of the original. Are you using an OCed 8800?




RE: Very strange...
By KristopherKubicki (blog) on 4/26/2007 2:15:20 AM , Rating: 3
It's a vendor OC (its best if we dont' reveal the exact card), but you are correct it's not the reference clocks.


RE: Very strange...
By mkruer on 4/26/2007 3:12:27 AM , Rating: 1
I seriously question Anandtechs bench marking ethics. Either do Stock to Stock OR Best over clocked to Best over clocked. This Stock to Overclocked is crap. The only time that Stock to Overclocked is legit is when you are comparing the same make an model to set up a metric for how the part scales with speed.

Anandtech is becoming or perhaps has become another THG.


RE: Very strange...
By oddity21 on 4/26/2007 3:17:54 AM , Rating: 2
Since they had a limited time window to benchmark the cards, you shouldn't expect a thorough review.

Plus, it doesn't really matter if the XTX's performance is this poor.


RE: Very strange...
By mkruer on 4/26/2007 3:31:25 AM , Rating: 3
That makes very little sense. Limited time window yes, but if they have a test system that is fully reproducible, then why can they not come back put in a stock nvidia card and generate the results for the stock nvidia comparison? Or are you saying that the test is not reproducible? If that’s the case, then you might as well chuck out this entire article, because it’s pointless.


RE: Very strange...
By oddity21 on 4/26/2007 3:34:49 AM , Rating: 2
I agree that they could've (and should've) done that. Then again, this is better than nothing, IMO.

Let's just wait for more information. I'm as shocked as everyone else by this.


RE: Very strange...
By mkruer on 4/26/2007 3:54:49 AM , Rating: 2
They still can. If Kris is still reading this it would take all but a minute to add the note.


RE: Very strange...
By DTAllTheBest on 4/28/2007 5:27:20 PM , Rating: 1
Yeah! Everybody lets wait. I hope the GTX will not beaten up by this card.


RE: Very strange...
By Ard on 4/26/2007 4:55:39 AM , Rating: 1
Not quite sure why anyone is complaining about the use of an OCed 8800. That's pretty much all you can buy. Who, in this ridiculously competitive, uses stock clocks anymore?


RE: Very strange...
By theapparition on 4/26/2007 8:00:26 AM , Rating: 5
Agreed, but would like to point out the 8800 is stock. Stock is what you buy, and if the vendors are factory overclocking with full warrantee support, then so be it.


RE: Very strange...
By Dactyl on 4/26/2007 2:02:12 PM , Rating: 2
quote:
Agreed, but would like to point out the 8800 is stock. Stock is what you buy
What's stock?
Stock is when you clock, no less than 30 up
Stock is when your chips, ain't safe because of heat


RE: Very strange...
By coldpower27 on 4/26/07, Rating: 0
RE: Very strange...
By Sunrise089 on 4/26/2007 3:18:47 AM , Rating: 4
Good point, but this is Dailytech, not AT. Though we all know most of the staff here are AT vets, they have to be different companies for NDA purposes.


RE: Very strange...
By mkruer on 4/26/2007 3:53:10 AM , Rating: 2
Yeah that didn't occur to me until after I posted. But still with such a close association I would assume they would be using similar if not identical ethics clauses.


RE: Very strange...
By mkruer on 4/26/2007 3:20:59 AM , Rating: 2
I didn't mean to should pissy, but at the very least when anyone does one of these "comparisons" the should note it on the graphs listed above. Place a asterisk notifying that the results list from a card are from a non-stock card. Is that too much to ask?


RE: Very strange...
By JSK on 4/26/2007 3:24:38 AM , Rating: 4
quote:
It's important to note that the GeForce 8800 GTX is a vendor overclocked board that comes shipped with the 650 MHz core clock.


Maybe if you just read the article they wouldnt need asterisks?


RE: Very strange...
By mkruer on 4/26/2007 3:39:31 AM , Rating: 2
I did see that, that’s why I made the comment. Apparently you have never written a proper thesis paper; forgetting to reference your notes usually lead to a failing grade on the paper.


RE: Very strange...
By TomZ on 4/26/2007 9:04:15 AM , Rating: 2
Good think this is a news site then. :o)


RE: Very strange...
By behemothzero on 4/26/07, Rating: 0
RE: Very strange...
By analex on 4/29/2007 5:10:38 PM , Rating: 2
quote:
I did see that, that’s why I made the comment. Apparently you have never written a proper thesis paper; forgetting to reference your notes usually lead to a failing grade on the paper.

Maybe you're the one who has never written a proper thesis paper. That bolded part is bad sentence construction.


According to you, what would be correct?

Changing the punctuation, like so: "I did see that -- that’s why I made the comment." Or reconstructing the sentence completely? Like so: "I made the comment because I saw that?"

Perhaps, I, too, am wrong?


RE: Very strange...
By deeznuts on 4/26/2007 1:14:27 PM , Rating: 4
quote:
I did see that, that’s why I made the comment. Apparently you have never written a proper thesis paper; forgetting to reference your notes usually lead to a failing grade on the paper.
If you saw it then why did you even complain about ethics? Questionable ethics is when you compare stock vs. overclocked and dont' even mention it. He made full disclosure here, get over it.


RE: Very strange...
By Crank the Planet on 4/27/2007 1:23:27 PM , Rating: 3
Also agreed that he discloses the fact that he is testing a stock card and an OC'd card. It doesn't matter if you can buy it OC'd and it has a full warranty- OC'd is OC'd. That means it has been MODIFIED from how the manufacturer created it. The fact that you can buy a Saleen and know you are getting a high performance car is the same thing- it's no longer just a Mustang.

Another note- even though he puts in his disclaimer the title is what's misleading. It has been worded to create the shock that everyone is experiencing so more people will read it. That's called journalism- lol


RE: Very strange...
By Ard on 4/27/2007 2:39:19 PM , Rating: 2
Get off the overclocked vs. stock comparison already. If you buy a factory overclocked card, that's stock, out of the box performance. End of story. That's what the XT/XTX are going to be competing against or do you expect consumers to just look at stock cards because it wouldn't be a fair comparison? The point is, the XTX would get slapped around by a stock GTX as well. Hell, ATI admitted as much. Leave it alone, it's done.


RE: Very strange...
By defter on 4/26/07, Rating: 0
RE: Very strange...
By mkruer on 4/26/07, Rating: 0
RE: Very strange...
By JSK on 4/26/2007 4:02:58 AM , Rating: 3
The 8800Ultra clocks were posted over at Dell's site earlier this week...


RE: Very strange...
By cheetah2k on 4/26/2007 4:10:38 AM , Rating: 2
Post a link dude


RE: Very strange...
By JSK on 4/26/2007 4:15:35 AM , Rating: 4
RE: Very strange...
By cheetah2k on 4/26/2007 5:44:45 AM , Rating: 2
cheers!


RE: Very strange...
By enlil242 on 4/26/2007 8:50:55 AM , Rating: 5
My first thought was like yours, stock to stock or nothing. But then I thought about it and came to the conlcusion that I have waited for a DX10 card until ATI's XTX was released and expected it to stomped on the 8800GTX, not catch up to it.

I would have thought, if anything else, the XTX would be on par with an OC'd 8800. To wait this long for ATI to meet the 8800's performance is a failure in my opinion. Let's hope further benchmarks reveal a bit more.


RE: Very strange...
By CBone on 4/26/2007 11:08:48 AM , Rating: 3
If it comes from the factory at those clocks, it is stock. Would you prefer they underclock it to reference clocks? What would that prove?


RE: Very strange...
By coldpower27 on 4/26/2007 1:02:53 PM , Rating: 2
It's legitimate if your comparing a retail available OC 650MHZ card, I know it's not the Nvidia reference design one, but this card is available to be bought, then it's a retail sample vs retail sample.

They will probably do a reference vs reference later.


RE: Very strange...
By ninjit on 4/26/2007 6:23:31 PM , Rating: 2
it appears that some people still need to be reminded
dailytech is NOT anandtech, they spun-off and formed their own website for news.

Anandtech now uses dailytech as their daily news feed (the headlines section).


RE: Very strange...
By retrospooty on 5/2/2007 11:32:45 AM , Rating: 2
"I seriously question Anandtechs bench marking ethics."

YOu know, it IS mentioned in the article that the 8800 is factory OC'd. Its not like they are trying to SKU the data. The point of the article is that the XTX seems not too fast.


RE: Very strange...
By Chadder007 on 4/26/2007 8:43:04 AM , Rating: 2
I wonder if there is a major flaw in the drivers perhaps?? At least I hope so anyway.


RE: Very strange...
By Anh Huynh on 4/26/2007 2:22:06 AM , Rating: 5
Also note the HD 2900 XT and XTX share the same exact GPU, whereas the GeForce 8800 GTS and GTX differ in shader processing capabilities. The GTS has 96 stream processors and the GTX has 128, which makes a larger performance difference than slightly different clock speeds.


RE: Very strange...
By redbone75 on 4/26/2007 2:30:02 AM , Rating: 2
That seemed odd to me as well: the HD 2900 XT bested the 8800GTS, but the 2900 XTX not only failed but failed miserably against the GTX? Would like to wait for retail cards and a more thorough test; however, that would really suck if there were no retail availability of a XTX.


RE: Very strange...
By redbone75 on 4/26/2007 2:36:13 AM , Rating: 5
Also, looking at the numbers, there would seem to be no reason whatsoever for someone to purchase a XTX seeing that it only provides an order of 5-6 fps over the XT. What?! It was even slower in Half Life 2: Episode One to the XT by a frame! There's got to be something seriously wrong with either the card, the drivers, or something. That, or the XTX simply sucks, but I find that hard to believe considering the specs of the card and the stellar showing the XT had against the GTS.


RE: Very strange...
By sol on 4/26/2007 2:38:21 AM , Rating: 4
The difference between 8800 GTS and GTX is huge though. The GTS is actually too crippled if you ask me. :-)


RE: Very strange...
By defter on 4/26/2007 3:40:45 AM , Rating: 5
Yes, since GTS has some units disabled.

Compared to GTS, GTX has:
- 50% more shader power ((1350x128)/(1200x96)=1.5)
- 38% more ROP power ((575x6)/(500x5) = 1.38)
- 53% more TMU power ((575x128)/(500x96) = 1.53)
- 35% more memory bandwidth (86.4/64 = 1.35)


RE: Very strange...
By oddity21 on 4/26/2007 2:47:00 AM , Rating: 2
Architectural efficiency problems, maybe? And isn't XTX just an XT with 1GB memory, since the clock difference is so minute?


RE: Very strange...
By KristopherKubicki (blog) on 4/26/2007 2:50:48 AM , Rating: 2
It's my thinking that the XTX could be set with higher clocks (remember the final clocks *still* aren't set in stone), but the big thing is the memory doesn't really seem to show better performance. I'm pretty convinced though that OEMs aren't picking this card up.


RE: Very strange...
By oddity21 on 4/26/2007 3:11:01 AM , Rating: 3
Is it possible for you guys to run another benchmark using seriously video memory-intensive games like, say, Oblivion with Qarl's Texture Pack 3 plus 16xAF? That thing eats video memory like mad.

Doubters should take note of the last paragraph. If the 2900XTX is as awesome as the hype, board builders wouldn't have dropped it from the launch lineup.

This is bad for the competition, IMO.


RE: Very strange...
By cheetah2k on 4/26/07, Rating: 0
RE: Very strange...
By oddity21 on 4/26/2007 4:21:47 AM , Rating: 5
quote:
Imagine an XT with GDDR4 mem and increased clocks? Nice!


2900XT + GDDR4 + increased clocks = 2900XTX

Is it not? XD


RE: Very strange...
By JSK on 4/26/2007 4:23:38 AM , Rating: 2
Id mod you up if I could.

If they are going to delay the XTX till 65nm, that may make sense, but it certainly isnt ready now.


RE: Very strange...
By cheetah2k on 4/26/2007 5:44:27 AM , Rating: 2
Yes, it would be an XTX, but it seems that the XT base is more optimised. I wonder what the board revision is on the XTX tested in this part-review, and compare that to the XT tested...

Oddity21, I know you wernt being facetious


RE: Very strange...
By oddity21 on 4/26/2007 8:40:25 AM , Rating: 2
While I wouldn't consider optimization to be a factor in these matters in other circumstances, this benchmark forces it into my mind. It's shocking how more memory and greater clock speed result in a performance dip.

I still hope that everything turns out differently on May 14th, for the sake of the competition. Doesn't look like it though. *sigh*

Still, it isn't like this is going to be ATI/AMD's end. The 2900XT is set to comfortably thrash the 8800GTS, while the real battle will be fought between the 8600GTS/8800GS and the 2600 series cards anyway.


RE: Very strange...
By defter on 4/26/2007 3:36:02 AM , Rating: 4
That's not suprising, afterall the core clock is almost the same and only difference is in memory clock and size:
- extra 512MB will not make any difference in today games
- XT already has ~100GB/s of memory bandwidth, that's quite plenty. Increasing memory bandwidth by e.g. 20% will not bring significant performance increase without corresponding core clock increase.

Thus, it's not suprising that the fate of XTX is still unknown, it has roughly the same performance with significantly higher production costs (more expensive memory, twice as much memory), this naturally would lead to lower margins for AMD and their board partners.


RE: Very strange...
By ogreslayer on 4/26/2007 7:27:29 AM , Rating: 2
Why would you think its strange the clock difference over an XT isn't there. Unlike the GTS and GTX cards relationship the XT and XTX are the same core with all the same features enabled.

This is exactly where I thought the XTX would be unless it had a massive increase in speed. Although, the drop in speed at 1920x1200 to the GTX seriously surprised me. It seems that the 1GB of GDDR4 is not doing what they thought it would. I'm sure if they get it to 1200 or so; it will sort itself out. But they'd also need to move the core clock into the 800 range.

And its not like an XT hoses a GTS. Beats it sure, but it is not smacking it up and down the street like it stole the XTs money. This was always gonna be the problem with R600. and is probably why they are looking to compete in the price area rather than the performance area. The card is not that powerful and now way to late for its own good.

Even at stock the cards would perform similar, I'd assume the XTX and the GTX would swap positions in some tests. The real point is that the Ultra is supposed to be nothing more than a cherry picked OC'd GTX. And its not like a good chuck of the companies don't sell OC cards.

I'd assume for the clocks however that the card was a certain 8800GTX, one with water-cooling. That's just a little on the unfair side, even if you are trying to point out the foolishness of the XTX's clocks when an Ultra or even more powerful products are gonna be on the way.


RE: Very strange...
By aguilpa1 on 4/26/2007 8:33:34 AM , Rating: 2
well, no not necessarily, you don't need H2O cooling to match the speed of 650Mhz GPU 1000Mhz Mem on the GTX, I have two running SLI and they can both run at that speed with no issues air cooled. They actually run set to 650Mhz GPU and 1024Mhz Mem. The GTX is just that good and that stable.


RE: Very strange...
By coldpower27 on 4/26/2007 1:07:49 PM , Rating: 2
There is no difference between the XT and XTX in terms of shader processors so the only increase in shader power comes from increase core rate, with the GTS to GTX 8800 line there is a 33% increase from the increase in units alone, and more plus the clockspeed hike.


RE: Very strange...
By wolf on 4/26/07, Rating: -1
RE: Very strange...
By KristopherKubicki (blog) on 4/26/2007 5:59:15 PM , Rating: 5
If we took legal advice from everybody who posted a forum comment we would have already been out of business.


RE: Very strange...
By JumpingJack on 4/26/2007 10:53:52 PM , Rating: 2
:) :) :) He did not sound very lawyer-like in his admonishment did he??

Isn't if funny to see people react in such a bizarre fashion when a review/preview/news article produces data they do not like to see ;) ....

Ostriches and sand.... ostriches and sand...


RE: Very strange...
By Ard on 4/26/2007 6:23:43 PM , Rating: 3
Ahh, laymen's who think they know the law. Common sense alone should tell you that legal agreements and contracts are only binding on those who sign them.


RE: Very strange...
By wolf on 4/26/07, Rating: -1
RE: Very strange...
By IntelGirl on 4/27/07, Rating: -1
RE: Very strange...
By scrapsma54 on 4/27/07, Rating: 0
RE: Very strange...
By Ard on 4/27/2007 3:06:13 PM , Rating: 3
Wow, 24xAA, huh? Nothing more than a pissing contest considering you're hard pressed to see a difference btw 8xAA and 16xAA. Next.


RE: Very strange...
By falc0ne on 4/26/2007 6:30:46 PM , Rating: 2
I was expecting sincerely a lot more performance from ATI/AMD...and by the numbers I see, their "ace" is on par in performance with 8800GTS or worse(let alone 8800GTS vendor overclocked,ie e-vga). I have no idea what's on their mind. How can you come up with this and expect to compete...If they don't cut the prices they don't stand a chance


RE: Very strange...
By HurleyBird on 4/27/07, Rating: -1
Numbers are FAKE People
By NextGenGamer2005 on 4/26/2007 4:36:05 AM , Rating: 5
Just let me show you what ATI's current Radeon X1950 XTX scores in these same games, with the same settings:

Company of Heroes - 1600x1200 - 69.2fps
Company of Heroes - 1920x1200 - 53.2fps

F.E.A.R. - 1280x1024 - 70fps
F.E.A.R. - 1600x1200 - 53fps
F.E.A.R. - 1920x1200 - 47fps

Half-Life 2: Episode One - 1280x1024 - 107fps
Half-Life 2: Episode One - 1600x1200 - 82fps
Half-Life 2: Episode One - 1920x1200 - 69fps

I would show you the Elder Scrolls IV: Oblivion scores as well, except that in this case the DailyTech numbers for the GeForce 8800 GTX are completely out of whack. I'm not sure how they got Oblivion running on all-high settings at 1920x1200 at 98fps, but in EVERY other place on the Internet, an XFX GeForce 8800 GTX XXX can only manage 69fps on an INDOOR scene at that resolution...let alone an outdoor one.

Now, look closely at those Radeon X1950 XTX framerates (they all came from X-bit Labs past reviews). Do you guys really think ATI would spend all these months and 700+ million transistors for a card that never performs more then 5% better, and in some cases, actually WORSE then their current champ? Yeah, I didn't think so either...

I am extremely dissapointed in DailyTech right now...




RE: Numbers are FAKE People
By JSK on 4/26/07, Rating: 0
RE: Numbers are FAKE People
By JSK on 4/26/2007 4:41:02 AM , Rating: 2
If thats a testament to the drivers, early hardware, GDDR4 issues, who knows.


RE: Numbers are FAKE People
By Sunrise089 on 4/26/2007 6:00:33 AM , Rating: 1
Nice research, and you're right - AMD wouldn't design a card to perform in that manner, but the consensus seems to be DT's benches are legit. So we fall back to either the design is (badly) flawed, or something is wrong with the drivers.


RE: Numbers are FAKE People
By hadifa on 4/26/2007 6:18:21 AM , Rating: 5
I checked Tomshardware VGA chart for oblivion score. It reports that
Oblivion outdoor, 1920*1200*32, no AA, 8xAF, Max quality, HDRR manages 28 FPS

From Anh description I understand that the test was an outdoor.
While there are different outdoor areas where the performance will be different still I am not sure if the 98.4 FPS for the 8800 GTX is correct.

Looking at the numbers for 2900XTX, they seem to be very close to the 1950XTX which is difficult to believe.

For example:

Company of heroes 1280*1024:
1950 XTX 99 (tweaktown) E6600
2900 XTX 97 (dailytech) QX6800

Company of heroes 1600*1200:
1950 XTX 70 (tweaktown) E6600
2900 XTX 73 (dailytech) QX6800

http://www.tweaktown.com/articles/1084/6/page_6_be...

FEAR 1280*1024:
1950 XTX 80 (tomshardware) No softshadow 4AA 8AF , X6800
2900 XTX 84 (dailytech) with softshadow 4AA 16AF , QX6800

FEAR 1600*1200:
1950 XTX 57 (tomshardware) No softshadow 4AA 8AF , X6800
2900 XTX 58 (dailytech) with softshadow 4AA 16AF , QX6800

http://www.tomshardware.com/2007/02/12/the_amd_squ...

In Fear the softshadows can make a big difference but in company of heroes it seems the 2900 XTX has no advantage over 1950 XTX.


RE: Numbers are FAKE People
By DingieM on 4/26/2007 7:18:25 AM , Rating: 2
Looking at your post now I believe the XTX with correct revision and mature drivers will utterly crush the 1950XTX as it should be.


RE: Numbers are FAKE People
By hadifa on 4/26/2007 7:40:41 AM , Rating: 2
Looking at Sven numbers, the difference between the 2900XTX and the 8800GTX is more like the difference of previous and the current generation rather than same generation cards.

Just a look at 2900XTX guarantees that the weak results can't be because of limited bandwidth or insufficient shader units or the like. It is a flaw and one can only hope it is a simple problem that can be easily fixed.


RE: Numbers are FAKE People
By coldpower27 on 4/26/2007 1:21:20 PM , Rating: 2
These results are actually in line with what we expect, there is no difference in terms of shader units and given the core rates 745 to 750 this is within margin or error mode, add to the fact that GDDR4 is typically higher latency as well as having 1GB is sometimes slower.

Not to mention the fact that more memory bandwidth would only come into play if the card is starved for bandwidth in the first place.

At over 100GB/s for the XT, it already has plenty to work with. So 20% more isn't going to really help, when the limitation is shader power.


RE: Numbers are FAKE People
By Regs on 4/26/2007 10:32:59 PM , Rating: 2
I would agree. The XTX just doesn't offer anything more to play with. AMD's line ends with the XT. End of story. They have nothing to compete with on the higher end.

Which would explain why AMD wanted to wait for the mid-range to launch with the high end because they have no high end.

So lets hope when 65nm comes out they are able to pump some life into the high end, or else they'll be locked into another price war with Nvidia. Nvidia could crush AMD right now the same way Intel is. Offer a higher performing and higher quality product, but at a competitive price.

Loyal as I am to AMD, though after 5 long years with nothing to show, they literally should just knock on my door and piss on my feet. That's how angry, disappointed, and terribly violated I feel to be any such loyal customer to them.


RE: Numbers are FAKE People
By elmikethemike on 4/26/07, Rating: 0
RE: Numbers are FAKE People
By coldpower27 on 4/26/2007 1:23:09 PM , Rating: 2
The numbers are accurate given the core rates and resolutions covered, it seems that the XTX need a core clock bump to over 800MHZ to becomes competitive with the GTX.


RE: Numbers are FAKE People
By tungtung on 4/26/07, Rating: 0
RE: Numbers are FAKE People
By slacker57 on 4/26/2007 1:36:38 PM , Rating: 5
quote:
Just look how flashy this site has been reporting the iPhone, and didn't even bother to mention that LG has actually a product similar to it and its out in the market now (the LG Prada).


http://www.dailytech.com/article.aspx?newsid=5758

Ooh, irrelevant AND incorrect. A double Whammy!


RE: Numbers are FAKE People
By coldpower27 on 4/26/2007 1:55:12 PM , Rating: 4
Worthless unless the testing suite is standardized and done by the same reviewer as you then introduce a huge margin of error.

Company of Heroes done on Xbitlabs normalizing to the X1950 XTX values.

8800 GTX
1600x1200 88.7
1920x1200 71.6

Half Life 2: Episode One It looks like your running into serious CPU limitations, as the decrease is only likely coming from overhead rather then shader limitation,

Performance if your shader limited should be close to these values:

Normalized to 1920x1200 as 100%
1600x1200 = 83.3%
1280x1024 = 56.9%

The figures you have provided show that the X1950 XTX is shader limited in Half Life 2: Episode 1 beyond 1280x1024.

1280x1024 = 107
1600x1200 = 82
1920x1200 = 69

Now let's see the 8800 GTX figures in those benchmarks

1280x1024 = 108
1600x1200 = 104
1920x1200 = 98

1600x1200 is 83% the workload of the 1920x1200 resolution, but the 8800 GTX is still providing 94% the performance of the 19x12 res at this resolution.

It likely going by this analysis the 8800 GTX is still CPU limited at 19x12 and like memory bandwidth limited, as the bottleneck isn't shader power. It's not surprising that the results here are much better as the Athlon FX-60 was a CPU bottleneck at Xbit, given they are using Core 2 QX6800 here with 2.93GHZ and the 20% advantage per clock it gives, give or take.

In F.E.A.R, without knowing what benchmarks were run, it's hard to say.

The article you also reference was quite old only 1 month or so after the 8800 launch, the drivers have matured since then, so Dailytech could be using a new set giving the 8800 GTX a considerable boost, as I heard the FW 15x are quite good.


RE: Numbers are FAKE People
By JumpingJack on 4/26/2007 10:56:24 PM , Rating: 3
quote:
Do you guys really think ATI would spend all these months and 700+ million transistors for a card that never performs more then 5% better, and in some cases, actually WORSE then their current champ?


That is precisely what AMD did when they released Brisbane...


RE: Numbers are FAKE People
By Azured on 4/27/2007 12:32:47 PM , Rating: 3
Brisbane is simply a process shrink. It may not do much for performance (in fact it should'nt do anything), but it does make the core a bit smaller, and thus cheaper to manufacture.


RE: Numbers are FAKE People
By JumpingJack on 4/28/2007 1:07:54 AM , Rating: 3
A shrink that took a hit in performance due to latency increase in L2 cache.... not much, but it did underperform at equivalent clocks....


RE: Numbers are FAKE People
By scrapsma54 on 4/27/2007 12:39:35 PM , Rating: 1
Good call. I also stated in one of my posts how they failed to mention what os they ran under. I believe Amd and Microsoft have been under close collaboration in order to design their graphics from the ground up for Vista, and not backwards compatible for Dx9, therefore, these numbers are shams until dx10 games are released, then we can see who really has their spine pressed to the back of the throne.


RE: Numbers are FAKE People
By Hawkido on 4/30/2007 6:13:00 PM , Rating: 2
I can't say the numbers are fake, and I really don't like entertaining the thought.

My thoughts are:

Remember when there was a DirectX change (8 or 9 I can't remember) and the cards made for the newer DirectX actually performed slower on the older directX games then the last generation cards on the older DirectX games.

The 2900 IS a DirectX10 card. So is the 8800. While you can compare the DirectX9 capabilities and say which you will buy now. I will not buy one till I see DirectX10 benchies.

How often do you see Feraris running Rally? Its fine and interesting to see how they perform in DX9, but that is not what they are designed for. And if you can't wait for a new card then get the one that performs best in DX9, because we really don't know how long it will be before DX10 sees the light of day. We really won't know which design will perform better until DX10 games are out and the crappo games out first are brushed aside.

I think the 8800 has more brute strength, but the AMD/ATI option has more Finesse. The ATI, with 360 individual stream processors, will be able to better adapt to a changing and varied environment. From what I have read about the 8800 the Shaders are programmed by the environment and cannot be changed just because you are facing a different direction (which can lead to lower minimum FPS, and exceedingly high Maximum FPS), whereas the ATI can change the programming of each shader with every rendering of every frame and yeilding a far more consistant FPS. Thus making far better use of its many simple shaders, than the the Brutish 8800. If the games are all programmed the way the 8800 is designed then the ATI 2900 will be destroyed, because it will not be able to flex because the program will not allow it, and the 8800 will power its way through each frame, while the 2900 will struggle to keep up.

Remember brute strength can be beaten by finesse. Just compare a v6 race car to a v8. The v8 will usually be faster in the straight stretches (where it is easy), but can't handle the corners as fast.

Do you really need 50-500 FPS? Wouldn't a consistently smooth 60 fps be far more preferable? Guess when the 500 FPS is going to drop to 50 FPS... Right when you need it the most, because all the explosions are happening, debris is flying, guns a blazing, planes crashing, you get the picture. Always when you need it the most is when your card is going to chug. The card that can on the fly reprogram its shaders to render what is needed (be it pixels, vertices, or physics) will best suit you in the end by not dropping the quality and framerate on you.


Horrible
By JackPack on 4/26/2007 2:03:41 AM , Rating: 1
And the XTX draws how many watts?

ATI might have a better mid-range product, but nVidia having a halo product like 8800 GTX/Ultra - that's priceless.




RE: Horrible
By PrinceGaz on 4/26/2007 4:07:57 AM , Rating: 4
Priceless is a rather ironic way of describing it- having a better top-end product like the 8800GTX/Ultra is indeed priceless if you mean that they make very little money out of it due to the relatively tiny number of them sold.

The mid-range is where the money is at, and having the best performing high-end part isn't going to make many people buy lackluster 8600 series cards. As for the 8500GT, on paper it looks like it will perform absolutely dreadfully even compared to the 8600GT, and be even poorer value despite its lower price.


RE: Horrible
By Mudvillager on 4/26/2007 4:42:25 AM , Rating: 2
Actually they've sold incredibly well...


RE: Horrible
By JackPack on 4/26/2007 4:45:53 AM , Rating: 2
quote:
The mid-range is where the money is at, and having the best performing high-end part isn't going to make many people buy lackluster 8600 series cards.


Of course it will. That's why it's called a halo effect. Most uninformed consumers simply ask whether nV or ATI is better. They don't spend the effort analyzing the GT/GTS benchmarks. The rationale is that if nV has the best card, then the rest of their products can't be that bad.

If vendors like Dell recommend the 8800 GTX with their gaming systems, people automatically assume any card from nV is the way to go. It's like E6300 vs. 5600+ at the same price. Sure, the 5600+ performs better, but people are still buying the E6300 because Intel claims Core 2 Duo is the "world's best processor" and everyone seems to agree.


RE: Horrible
By StarOrbiter on 4/26/2007 6:10:14 AM , Rating: 2
My AMD fanboy friend. If I'm not mistaken, that is after AMD Price cuts. When put on par with the recent Intel price cut, the E6420, which is priced where the E6300 was is better than the 5600+


RE: Horrible
By Martimus on 4/26/2007 9:43:34 AM , Rating: 2
Wow, that came out of left field. What is it with Fan Boy attacks that have very little to do with the actual post. I am just getting tired of reading these unfounded attacks that people write to each other that I actually created an account just to point out how stupid they sound.