backtop


Print 53 comment(s) - last by gazzhazz.. on Aug 23 at 9:27 PM


Benchmarks provided to vendors for the Radeon X1950XTX CrossFire versus GeForce 7950GX2 Quad SLI - Click to enlarge
Radeon X1950 CrossFire puts Quad SLI in its place

DailyTech has received early benchmarks of ATI’s upcoming Radeon X1950XTX and X1950 CrossFire graphics cards. ATI’s upcoming Radeon X1950XTX and X1950 CrossFire is expected to make its debut on August 23rd. Specifications for the Radeon X1950XTX and X1950 CrossFire are finalized with a 650 MHz core clock and 2 GHz effective memory clock. The core clock is unchanged from the previous Radeon X1900XTX while memory clock receives a hefty 450 MHz boost. This time around the Radeon X1950XTX and X1950 CrossFire are equipped with 90nm Samsung GDDR4 memory.

The benchmarks compare ATI’s Radeon X1950XT in CrossFire against NVIDIA’s Quad SLI. The test setup used for NVIDIA’s Quad SLI is a Dell XPS 700 system equipped with an Intel Core 2 Extreme X6800, 1GB of DDR2-800 memory and two GeForce 7950GX2 graphics cards for four total GPUs. The ATI CrossFire test system is identical with the Quad SLI system except the nForce 590 SLI Intel Edition motherboard was swapped out for an early Radeon Xpress 3200 (RD600) motherboard and two ATI Radeon X1950XTX/CrossFire graphics cards.

These early benchmarks are favorable to ATI. Call of Duty 2, Half-Life 2: Episode 1 and Serious Sam II heavily favor ATI, most likely due to better multi-GPU scaling on ATI’s side. FarCry, Quake 4 and Doom 3 performance shows the Quad SLI system creeping up to the performance of ATI’s X1950XTX/CrossFire system. However, the Quad SLI system still falls behind, close but no cigar. F.E.A.R. is the only game that can take advantage of Quad SLI and shows the Quad SLI system beating out the similarly configured X1950XTX/CrossFire system; though at 2560x1600 with 4xAA and 8xAF the X1950XTX/CrossFire takes the lead once again. With Elder Scrolls IV: Oblivion the X1950XTX/CrossFire system manages to take a nice lead over Quad SLI at 1600x1200 with 8xAF. The lead narrows when the resolution is raised to 2560x1600 with 8xAF.

While these early benchmarks show ATI’s X1950XTX/CrossFire beating out NVIDIA’s Quad SLI there’s more to the story. Since the benchmarks are only comparing performance with 4x anti-aliasing and 8x anisotropic filtering at most, it doesn’t show the true performance of NVIDIA’s Quad SLI. The true performance of Quad SLI being its capability to render high levels of anti-aliasing without taking a heavy performance hit. Ryan Shrout at PC Perspective has written an excellent article covering early Quad SLI performance.


Comments     Threshold


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

The 7950 GX2 numbers seem a little odd
By s12033722 on 8/9/2006 1:37:30 PM , Rating: 2
For instance, Anandtech's number for a single 7950GX2 in Half Life 2 Ep 1 at 1600x1200 no AA/AF is 135.8 FPS. http://www.anandtech.com/video/showdoc.aspx?i=2769...

It seems unlikely that two 7950GX2's in SLI would have only 92.4 FPS as shown in this graph. Also, where does it state that this is Quad SLI vs Crossfire? The image title seems to indicate single card vs. single card.

In any case, the x1950 looks like a very nice card!




RE: The 7950 GX2 numbers seem a little odd
By TheDoc9 on 8/9/2006 1:50:48 PM , Rating: 2
exactly. It seems strange how 4 gpu's can't beat 2.


RE: The 7950 GX2 numbers seem a little odd
By Anh Huynh on 8/9/2006 2:00:23 PM , Rating: 2
Quad SLI has very poor scaling. The only game that takes advantage of the four GPUs is F.E.A.R. PC Perspective outlines the performance. Most of the time Quad SLI has troubles keeping up with SLI. Quad SLI only shows its true performance with SLI8XAA and SLI16XAA where it takes no performance hit.


RE: The 7950 GX2 numbers seem a little odd
By Some1ne on 8/9/2006 4:08:21 PM , Rating: 2
Are you sure that that is accurate? The 7950GX2 appears to the system as a single graphics card, hence its ability to work even on platforms that do not support SLI of any sort. Given that, although running two of these cards in SLI has been branded "Quad-SLI" to indicate that there are really 4 GPU's, to the rest of the system it should appear to be just a plain SLI setup, and should scale at least as well as a plain SLI setup would. I don't see why a game would have to do anything special to support quad SLI beyond what is done to support plain SLI...all the algorithm does is divide the data in two either by frame or by individual parts of a single frame, and then ship each half off to the GPU's, and if the data can be split in half, each half can be split again without doing anything special.

I agree with the general skepticism being expressed by others here. I think that the graph shows either a single X1950 against a single 7950GX2, or possibly even two X1950's in crossfire against a single 7950GX2. Keep in mind that a single X1900XTX is demonstrably slower than a 7590GX2, and that the X1950's are really just X1900's with GDDR4 instead of GDDR4, so it really seems like a stretch to believe that a simple memory upgrade has boosted the card's performance to the point where it absolutely rrounces the 7950 as that chart suggests. In fact, when you think about it, the results on that chart look suspiciously like what one would expect if you were to bench two X1900's in crossfire against a single 7950GX2.


By Anh Huynh on 8/9/2006 8:44:05 PM , Rating: 2
7950GX2 shows up as two GPUs and will run SLI. You can disable one GPU and just run single GPU mode.

http://www.anandtech.com/showdoc.aspx?i=2769&p=4

The X1950XTX has a major memory bandwidth advancement. With those limited AA and AF modes Quad SLI isn't able to shine. Quad SLI is designed for higher AA/AF without taking a performance hit. Check the Pcperspective article linked towards the end of this news post.

I double checked to make sure that it is indeed CF and Quad SLI.


By OrSin on 8/10/2006 9:49:43 AM , Rating: 2
7950 Gx2 will not work in a non-SLI systems unless you disable one of the cores. Also the QUad rives sucks and and many sites should that that 2 cars in sli is getting these quad systems. I dont thing the numbers are that far off.


By hstewarth on 8/9/2006 2:44:12 PM , Rating: 2
Its very likely that it was tested with early. I think until recently quad sli supports was not available completely in the drives. In this case the driver would treat the configuration as 2 boards.


By Loc13 on 8/10/2006 10:57:13 AM , Rating: 2
http://www.tomshardware.com/2006/08/07/get_quad_sl...

At the time this article was written, this Monday in fact, the quad sli driver was still in beta. Quite frankly, the results suck looking at those benchmarks. But as of today, it's not beta anymore, so the performance might be better. Who knows.


By gazzhazz on 8/23/2006 9:27:54 PM , Rating: 2
Just to set things straigt, In that benchmark things have been setup so the ATI automaticaly gets the advantage.

For example the new X1950 is designed with all that insane ram to take advantage of crossfire, in a single card setup that isnt there. Also the quad sli drivers are pritty awful at the moment and thats a fact, this benchmark simply draws off the x1950 strength in crossfire and the gx2 weakness in quadsli.

So in all, I'd ignore the benchmark, its unfair like most manufactures benchmarks.

As an upgrade that takes up 2 slots on your motherboad id have the gx2 anyday, its only £20-50 more, it uses less power, u dont need an sli motherboard or compliant psu to run it and it will last alot longer than the x1950.

And ive seen the single card benchmarks, in Prey at various resolutions with 4AA and 16x aniso the fps is;

x1950 1600x1200 62.2, 1280x1024 80.6
gx2 1600x1200 86.2, 1280x1024 108.1

I rest my case lol


RE: The 7950 GX2 numbers seem a little odd
By mb on 8/9/2006 1:52:07 PM , Rating: 2
Good catch, that does seem to be quite a difference (135.8 FPS from the AT article vs 92.4 FPS from the graph above) especially considering the above tests were done with a much faster CPU (X6800 vs FX-57).

Something isn't right.. someone is cooking the numbers somewhere.. and my bet is the source for the graph above as I trust AT a little more than some random graph.


By theprodigalrebel on 8/9/2006 2:06:24 PM , Rating: 2
The reported FPS means nothing unless you know EXACTLY what map/timedemo was tested. Most games these days have levels where you can run 100+ FPS one moment and drop to half later.


By therealnickdanger on 8/9/2006 2:50:08 PM , Rating: 2
Doesn't Anandtech do their demos without sound? That can make a big difference, but I would agree that 40fps is a HUGE difference. That AT review also compared a X1900XT to the GX2, not an XTX - which I found strange.

In the end, I think we can all agree that the X1950XT are impressive and that AT will give the final word on performance very soon. There's no denying that faster memory, smaller die size, and other enhancements should make an impact against the typically unimpressive Quad-SLI, but I'll wait for AT to make the call...


By PLaYaHaTeD on 8/10/2006 3:49:59 PM , Rating: 2
Apples to oranges, no aa & af compared to 4xaa 8xaf. Isn't that a perfectly acceptable reason for the drop in framerates? How come everyone is generating their own conspiracy theory?


By ElFenix on 8/16/2006 12:35:48 AM , Rating: 2
anandtech (and almost all other review sites, for that matter) test with the default graphics driver settings. for nvidia that is only 'quality' and not 'high quality'. ATi's default graphics driver settings are 'high quality'. if the vendor, when benching, turned the nvidia settings to where they should be, the nvidia scores would drop in comparison to anandtech's misleading scores.

plus, nvidia takes a bigger hit than ati when enabling AA and AF, and have for the past several years.


“We do believe we have a moral responsibility to keep porn off the iPhone.” -- Steve Jobs

Related Articles
Major Radeon X1900 Price Drops
August 8, 2006, 1:44 AM
ATI Radeon X1950 Announced
July 21, 2006, 5:58 PM
Here Comes "Conroe"
July 13, 2006, 12:47 PM
Samsung Shipping Production GDDR4
July 5, 2006, 10:00 AM













botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki