New FX Extremetech Review
- Thread starter Doomtrooper
- Start date
Doomtrooper
Veteran
Works fine here...
Some key notes:
Fan was measured at 58db at full load...
Wins most fill rate tests (we knew that would happen)
Some key notes:
Fan was measured at 58db at full load...
Wins most fill rate tests (we knew that would happen)
Click the individual links, that works. Many images are broken though, specifically those showing performance graphs it seems...
Not sure if it's intentional or not, as a pic of the monster cooler in action is also broken... I haven't actually checked all pages yet so maybe some graphs do work.
*G*
Not sure if it's intentional or not, as a pic of the monster cooler in action is also broken... I haven't actually checked all pages yet so maybe some graphs do work.
*G*
simplicity
Newcomer
The ExtremeTech home page currently says "Flash: First nVidia GeForceFX Test Results will be posted at midnight pacific time! "
Not what I expected
After seeing the benchmarks, I can see why nvidia was waited so long before allowing reviews. Their drivers are still pretty raw. Usually they are the technology leaders so people don't notice how raw the drivers are, but this time around they are certainly behind the times. Reminds me of the 8500 launch.
After seeing the benchmarks, I can see why nvidia was waited so long before allowing reviews. Their drivers are still pretty raw. Usually they are the technology leaders so people don't notice how raw the drivers are, but this time around they are certainly behind the times. Reminds me of the 8500 launch.
Working now, but flaky. Here's the whole thing on one page:
http://www.extremetech.com/print_article/0,3998,a=36238,00.asp
http://www.extremetech.com/print_article/0,3998,a=36238,00.asp
Hellbinder
Banned
Well, I have been able to get all the pages to work now. If these numbers hold true the GFFX is DOA for any sane thinking person imo.
On average it gets beatten in every game tested with 4x FSAA and 8x AF by 30%. The coupple exeption games SS:SE and JK2 have always (imo) offerd really bizzare results favoring Nvidia, so thats nothing new.
I am now eagerly awaiting the rest of the reviews.
On average it gets beatten in every game tested with 4x FSAA and 8x AF by 30%. The coupple exeption games SS:SE and JK2 have always (imo) offerd really bizzare results favoring Nvidia, so thats nothing new.
I am now eagerly awaiting the rest of the reviews.
Some of the synthetic tests confirms the theories some have had (the excellent performance in the 8-light geometry test for example and the low performance in the single texturing fillrate test also), however the gaming benches looks decidedly BAD for Nvidia at this point (especially at high res/AA+AF settings), however I don't know how conclusive these numbers really are. We need more tests from more sources.
I'm not going to declare the GFFX a lame duck on basis of one short (p)review...though I'm tempted to at this stage, lol.
*G*
I'm not going to declare the GFFX a lame duck on basis of one short (p)review...though I'm tempted to at this stage, lol.
*G*
Typedef Enum
Regular
You can now pretty much say that the FX is, perhaps, the biggest 3D flop since the Glaze3D hype from a few years ago.
We've heard such things as "most important contribution in 3D since the founding of the company" and other marketing related crap from the past several months.
When you look @ the numbers from ExtremeTech's review, they look downright pathetic! I mean, how on earth can a part that comes 6 months post R300, with a 200 MHz. core clock advantage, get creamed so badly in some of the tests? Epseically when it comes to shaders!
In the Advanced Pixel Shader test, the FX gets totally slaughtered...
In the Vertex Shader test, the R300 kills the FX...
So, the one thing that seemed...if nothing else...the one "clear" advantage nVidia would have over the R300...doesn't seem to be an advantage at all.
But the bottom line is what has become very apparant thus far...Sheer Bandwidth still makes a difference, and the FX part is simply too deficient in this department. They need to get much more bandwidth out of their architecture than is presently available. The article seems to hint that a 256-bit version is in the pipeline, but I'm not sure if it was just supposition or not.
Anyhow, I strongly suspect that nVidia is going to end up losing a heck of a lot of credibility with this product, and it's only going to get worse once ATI nails em' with the knockout punch...R350.
About the only thing you can really say, at this point, is that the marketing team @ nVidia is going to end up earning their paychecks for the next couple of months...They're going to have to _really_ spin some serious wool to make this situation turn out in their favor.
We've heard such things as "most important contribution in 3D since the founding of the company" and other marketing related crap from the past several months.
When you look @ the numbers from ExtremeTech's review, they look downright pathetic! I mean, how on earth can a part that comes 6 months post R300, with a 200 MHz. core clock advantage, get creamed so badly in some of the tests? Epseically when it comes to shaders!
In the Advanced Pixel Shader test, the FX gets totally slaughtered...
In the Vertex Shader test, the R300 kills the FX...
So, the one thing that seemed...if nothing else...the one "clear" advantage nVidia would have over the R300...doesn't seem to be an advantage at all.
But the bottom line is what has become very apparant thus far...Sheer Bandwidth still makes a difference, and the FX part is simply too deficient in this department. They need to get much more bandwidth out of their architecture than is presently available. The article seems to hint that a 256-bit version is in the pipeline, but I'm not sure if it was just supposition or not.
Anyhow, I strongly suspect that nVidia is going to end up losing a heck of a lot of credibility with this product, and it's only going to get worse once ATI nails em' with the knockout punch...R350.
About the only thing you can really say, at this point, is that the marketing team @ nVidia is going to end up earning their paychecks for the next couple of months...They're going to have to _really_ spin some serious wool to make this situation turn out in their favor.
This is a major disappointment. Even with half-done drivers I can't think of a way to spin this in the favor of the GeForce FX.
I really thought that nVidia had something up their sleves to support the massive fillrate and thus shader ops power. They are gonna take some serious heat for lacking in memory bandwidth tricks (although the card in itself is very good).
I really thought that nVidia had something up their sleves to support the massive fillrate and thus shader ops power. They are gonna take some serious heat for lacking in memory bandwidth tricks (although the card in itself is very good).
Interesting tidbits from ET's review:
Re: anisotropic filering--
3dMark theoretical subtests:
Guessing ATI fans won't be complaining as much about APS anymore... In any case, the geometry and pixel shading tests at least are victories for GFFX. But the fillrate test tells the story, IMO. Only in the multitexturing test (which is very unlike real multitexturing situations) does GFFX get even close to its theoretical fillrate. Perhaps the only realistic game situation it might get close is when doing the z-pass in the Doom 3 engine (no textures -> low bandwidth requirements).
And just to be fair, a couple tests where GFFX wins some:
Wonder what's going on with the AF/AA scores on Serious Sam?? And BTW: bad Doomtrooper for not posting these too!
EDIT: added bit about 2nd PCI slot being open
EDIT2: replaced images with text description of benchmark results
So you *can* put something in the 2nd PCI slot, although e.g. a PCI video card with DRAM on the back probably won't fit.
So under the current drivers, apparently running OS X 10.2 or Longhorn *would* trigger the loud fan.
Mentioned several times. Extremetech thinks there's still room for improvement in the drivers, although whether they made this up themselves or whether Nvidia indicated this was the case, who knows.
Re: anisotropic filering--
What about the new "adaptive" aniso algorithm? Did Extremetech get this wrong?? Of course, it would help explain the nasty AA/AF scores (although so would the AA bandwidth hit).
3dMark theoretical subtests:
Code:
3dMark Fillrate tests-
single-texture multitexture
GFFX 5800 U 1587.5 3557.8
16x12
9700 Pro 1819.6 2564.1
16x12
GFFX 5800 U 879.2 3055.1
16x12 4xAA 8xAF
9700 Pro 1168.8 2523.5
16x12 4xAA 8xAF
3dMark Triangle rate tests-
1 light 8 lights
GFFX 5800 U 83.8 28.9
16x12
9700 Pro 65.0 14.7
16x12
GFFX 5800 U 50.1 23.1
16x12 4xAA 8xAF
9700 Pro 44.1 13.5
16x12 4xAA 8xAF
3dMark Shader Tests-
Pixel Shader Vertex Shader
GFFX 5800 U 141.0 120.5
16x12
9700 Pro 101.7 150.0
16x12
GFFX 5800 U 36.0 45.6
16x12 4xAA 8xAF
9700 Pro 84.4 57.6
16x12 4xAA 8xAF
3dMark Shader Game Tests-
Advanced Pixel Shader Nature
GFFX 5800 U 67.9 54.6
16x12
9700 Pro 99.2 50.5
16x12
GFFX 5800 U 32.5 21.3
16x12 4xAA 8xAF
9700 Pro 60.9 27.2
16x12 4xAA 8xAFGuessing ATI fans won't be complaining as much about APS anymore... In any case, the geometry and pixel shading tests at least are victories for GFFX. But the fillrate test tells the story, IMO. Only in the multitexturing test (which is very unlike real multitexturing situations) does GFFX get even close to its theoretical fillrate. Perhaps the only realistic game situation it might get close is when doing the z-pass in the Doom 3 engine (no textures -> low bandwidth requirements).
And just to be fair, a couple tests where GFFX wins some:
Code:
Jedi Knight II and Serious Sam SE-
Serious Sam SE Jedi Knight II
GFFX 5800 U 69.3 133.4
16x12
9700 Pro 61.5 129.9
16x12
GFFX 5800 U 41.9 100.8
16x12 4xAA 8xAF
9700 Pro 32.6 99.7
16x12 4xAA 8xAFWonder what's going on with the AF/AA scores on Serious Sam?? And BTW: bad Doomtrooper for not posting these too!
EDIT: added bit about 2nd PCI slot being open
EDIT2: replaced images with text description of benchmark results
