Late, noisy, HUGE!!! - makes 17k Mark...

Status
Not open for further replies.
T

T2k

Veteran
17,000 3DMarks for a GeForce FX at CES

http://www.3dchipset.com/index.php#1031

Not that much w/ az Athlon 2400+... :p

BTW, the original article at http://www.nordichardware.se/nyhete...19&PHPSESSID=0a076dd4e0e69b967ef5a5e98d37ebdd links this:

GFFX.jpg


MAN... WHAT A HUGE CARD IS THIS??? :oops: :oops: :oops:

More images here: http://www.nordichardware.se/artiklar/Grafik/2003/GeForce_FX/
 
I'm not up to date of 3Dmark scores. Is this a good score, or not really that much higher than an R9700 Pro (stock obviously)?
 
Is it just me, or does there look to be a glaring design flaw in the FX cooling system?

I've never seen it from that angle before, but when looking at the side it's just staring me in the face.

:?
 
I dont see a design flaw (other than it being large), but I do think 17K 3DMarks is a pretty pathetic score given all the hype for how much faster shaders/gpu instructions are supposed to be. 9700 Pros easily hit just short of 16K.

What would be of more interest with 3dmark would be the individual test scores, mainly those that do not attribute to the overall score (high poly/high light source and advanced pixel shader tests). These might tell a different story, but Id expect a card of this market segment to have a much better score pushing through Nature and the high detail tests (especially Dragothic).
 
I suspect that the default settings where used. The resolution though was the standard 1024x768x32bit though.
Click to expand...

There is the problem right there...the default settings are outdated..If I lay down serious cash (which I'm not :D..come on 9500 get here )..but If I did would be looking for how fast it is with eye candy.

Cards of the 9700/Nv30 Caliber should be defaulting to 4X FSAA at 1024 x 768 as that is what the selling feature is of these high end cards..eye candy.

Most overclocked 9700's are hitting high 16's...I await for a real review with none other that REAL GAMES ..go figure :p
 
Well, from an engineer's point of view having the finned area remote from the GPU core is, well... hell, I can't imagine why they did that. There are other ways of achieving the airflow routing they desired without a squirrel cage fan mounted in that particular position.

The "heat spreader" on the GPU core is connected to the fins by way of a "thermal bus" for lack of a better term. A few hand calculations would show that whatever gain they picked up with increased air flow could have easily been offset with a more efficient heatsink.

Wow. That just doesn't look too smart at all.
 
Looks like a standard heat pipe config just layed out differently the finned area is connected to the heat pipe which is a vacuum tight copper tube. Inside that tube is a wick and a fluid (not sure what it is..maybe something like Glycol or Anitfreeze.)

When the heat pipe conected to the front plate gets warmer this turns the liquid into a high pressure vapor that moves toward the cooler finned area, as the vapor reaches the cooler plate, it condenses and releases the heat into the fins. The wick then helps move the condensed liquid back to the other end to start all over.
 
Big, hot, ugly and late. 'Tis very sad :(
Wasn't this thing supposed to be +30% faster across the board? For the price they're asking it had better be. 10% just aint cutting it.

At least we now have R350 and NV35 to look forward to
 
Sharkfood said:
I dont see a design flaw (other than it being large), but I do think 17K 3DMarks is a pretty pathetic score given all the hype for how much faster shaders/gpu instructions are supposed to be. 9700 Pros easily hit just short of 16K.
Click to expand...
I don't think so. +10% on default resolution, on a 2400+ system, seems to me quite a big advantage.

For exemple a R300 does 14k on a PIV 2.53
http://www.anandtech.com/showdoc.html?i=1683&p=13
 
Doomtrooper said:
1024 x 768 NO aa on a high end card you are going to be CPU limited, there is no way around it....
Click to expand...
*nod*

Hence, these 3DMark scores are neither disappointing nor elating. Just rather pointless.

ta,
-Sascha.rb
 
Well, this may indicate one of two things, if we assume that this benchmark is CPU-limited:

1. nVidia put 3DMark2001 optimizations at a high priority for early optimizations (wouldn't be surprising).

2. nVidia's current GeForce FX drivers are less CPU-dependent than ATI's Radeon 9700 drivers. Given nVidia's unified driver architecture, this also wouldn't surprise me, given that the 9700 is pretty significantly CPU-limited.
 
Given the poor nature scores that were leaked a while ago, and considering what kind of boost a good score in that test gives to the overall result, could it be that the drivers are now optimised to perform better with alpha textures. Also would the use of colour compression without FSAA make any difference to the performance in those tests?

I still think 17k out of the box on a stock 2400+ is quite nice. A good step up from a GF4 or 8500. Unfortuanately it looks like we'll be paying £1 per 3DMark :D
 
Chalnoth said:
2. nVidia's current GeForce FX drivers are less CPU-dependent than ATI's Radeon 9700 drivers. Given nVidia's unified driver architecture, this also wouldn't surprise me, given that the 9700 is pretty significantly CPU-limited.
Click to expand...

IIRC both the Gf3 and Gf4 seem to push out more fps over the 8500 and 9700 respectively on slower machines eg ~ 1gig machines.
 
Sharkfood said:
I dont see a design flaw (other than it being large), but I do think 17K 3DMarks is a pretty pathetic score given all the hype for how much faster shaders/gpu instructions are supposed to be. 9700 Pros easily hit just short of 16K.
Click to expand...

Nah, with a Ahtlon XP 2400+ that shold be fairly decent. Since 3dmark depend a lot on the CPU/FSB bandwidth in a number of the benchmarks, I wouldn't be too surprised by a mark like that.

Sharkfood said:
What would be of more interest with 3dmark would be the individual test scores, mainly those that do not attribute to the overall score (high poly/high light source and advanced pixel shader tests). These might tell a different story, but Id expect a card of this market segment to have a much better score pushing through Nature and the high detail tests (especially Dragothic).
Click to expand...

Indeed to the first part, but in regard to the second part I can't see what advantage the NV30 should off over the R300 besides the higher clockspeed. But then again we might not know everything about the architecture goodies in the NV30.
 
my AIW 9700 Pro on a 2400+ XP 512MB , nForce2 and 120GB WD 7200 RPM HDD gets almost exactly (13,980) 14,000 3DMarks.
 
EvilDeus-
I don't think so. +10% on default resolution, on a 2400+ system, seems to me quite a big advantage.
Click to expand...

It's interesting that Anand's score that you posted is barely able to muster 13,900 3DMarks when every other source I've seen generally is in the mid 14,000's for the same P4 system. Non-crippled systems here, of over 20+ systems, I run a standard test myself here at 15,180 as the "litmus" test on retail 9700 Pro's on the same system. But all other sources are easily in the 14,000's.

ExtremeTech/P4-2.53gz - 14,539 3DMarks
http://www.extremetech.com/article2/0,3973,475971,00.asp

HardOCP/P4-2.53ghz - 14,641 3DMarks
http://www.hardocp.com/image.html?image=MTAyOTczNDY5M0dMRUNjQkJVRmtfMl8yX2wuZ2lm


Sharkeys/P4-2.53ghz - 14,845 3DMarks
http://www.sharkyextreme.com/hardware/videocards/article.php/3211_1449721__5

Toms/P4-2.53ghz - 14,292 3DMarks
http://www17.tomshardware.com/graphic/20020819/radeon9700-13.html

Doomtrooper-
1024 x 768 NO aa on a high end card you are going to be CPU limited, there is no way around it....
Click to expand...

Actually, this is partially incorrect.

The final score is an accumulation of seven (7) benchmark tests, from which three (3) of these tests are CPU bound (Game 1,2,3 low detail) and four (4) tests are GPU bound (Game 1,2,3 High Detail + Nature).

This is why an average delta of >1-1.5K can be achieved through simple GPU overclocking without modifying the CPU speed one bit. A simple clock bump of 15-30mhz on an R300 core verifies this trend.

Obviously, due to the the three (3) CPU bound tests, isolating a CPU increase also raises the total score as well, which is why the individual scores for the tests would be of more interest.
 
Sharkfood said:
EvilDeus-
It's interesting that Anand's score that you posted is barely able to muster 13,900 3DMarks when every other source I've seen generally is in the mid 14,000's for the same P4 system. Non-crippled systems here, of over 20+ systems, I run a standard test myself here at 15,180 as the "litmus" test on retail 9700 Pro's on the same system. But all other sources are easily in the 14,000's
Click to expand...


Well i put the reference afterward. But 14500 to 17k it's 15% more on this bench it's quite a lot in fact, if it's true :)
 
It might just be me, but looking at the picture above the card appears to be covering the SECOND PCI slot as well :oops:
 
Status
Not open for further replies.
Back
Top