I believe NV34 has additional circuitry that among other things allows for more memory compression optimizations, effectively giving the chip a higher memory bandwidth.

NV34 (FX5200) is on an older 150nm process, has less transistors and lacks color compression and perhaps Z- compression, slower RAMDAC 350MHz (vs.400 on NV31) and only 4xAA (vs. up to 6x on NV31).
It was a replacement for GF4MX, the NV31 was more of a replacement for GF4Ti I think.

Definitely NOT the same 😉
NV34 @ 250/400 (effective) = 9863 (driver 71.84)

NV31 @ 250/400 (effective) = 9091 (driver 93.71)

^Sorry about different driver 😒
Pretty sure it does weird things to performance in older benchmarks.
NV34 @ 250/400 (effective) = 8055 (driver 71.84)

NV31 @ 250/400 (effective) = 6442 (driver 93.71)

^This is more reasonable, but still driver messing things up...

From practical point of view :
OC'ing 5200 to 5600 level is NOT easy, because it requires 5200 to have reasonable VRAM quality (as capable as 5600), which only earlier ones do.

I did a better 1:1 with Ultra cards (downclocking 5600U and overclocking 5200U) :
5200U @ 350/700 (driver 71.84) : 12191

5600U @ 350/700 (driver 71.84) : 12745

This time with the same driver for both 😉

Note : Later than 5x.yz series driver will limit mutlitexture rate on tests with 5200 cards, here's example of that :
5200U @ 350/700 (driver 56.72) : 11445

Biggest difference is with Codecreatures benchmark :
5200U @ 350/700 (driver 71.84) : 1455

5600U @ 350/700 (driver 71.84) : 2013

Comparing the 5200 with the 5600 is not entirely accurate.
Despite the fact that the 5200 and 5500 have identical GPUs, some of the functions are disabled in the BIOS of the 5200, such as IntelliSample and Z-buffer compression.
There are BIOSes for flashing 5200 to 5500, and most likely, they can be found on forums from those years.
https://www.laneros.com/temas/fx-5200-a-fx-55 … ing-bios.37904/
https://www.clubedohardware.com.br/forums/top … x5200-p-fx5500/

In general, given the identical GPU configuration of the 5200/500/600, there will be no significant difference, only proportional to their operating frequency.

shevalier wrote on 2025-11-28, 07:01:

Comparing the 5200 with the 5600 is not entirely accurate. Despite the fact that the 5200 and 5500 have identical GPUs, some of […]
Show full quote

Comparing the 5200 with the 5600 is not entirely accurate.
Despite the fact that the 5200 and 5500 have identical GPUs, some of the functions are disabled in the BIOS of the 5200, such as IntelliSample and Z-buffer compression.
There are BIOSes for flashing 5200 to 5500, and most likely, they can be found on forums from those years.
https://www.laneros.com/temas/fx-5200-a-fx-55 … ing-bios.37904/
https://www.clubedohardware.com.br/forums/top … x5200-p-fx5500/

In general, given the identical GPU configuration of the 5200/500/600, there will be no significant difference, only proportional to their operating frequency.

If the difference between 5200 and 5500 are bandwidth saving technologies it would be nice to confirm it by measuring the difference between cards with different bandwidth.

Just because the configuration is identical in the most basic view does not mean they are actually identical.

FX 5200 will most likely suffer higher penalty from MSAA in 1:1 clock comparison tests too.

In short, nothing worked out.
The theory has been completely destroyed.
With identical RAM timings, 5200 and the same one reflashed to 5500 show completely identical results.
Only the numbers in the name of the video card are more pleasant.

yes afaik 5200-5500 are 100% the same, the only difference should be the name, no extra features unlocked, I don't think the NV34 had features disabled, it was a a different chip lacking some stuff all the other NV3x had.

also be careful because I have seen 5600 series cards with 64bit memory bus, they are rarer but something to keep in mind also.

SPBHM wrote on 2025-11-28, 15:27:

yes afaik 5200-5500 are 100% the same, the only difference should be the name, no extra features unlocked, I don't think the NV34 had features disabled, it was a a different chip lacking some stuff all the other NV3x had.

also be careful because I have seen 5600 series cards with 64bit memory bus, they are rarer but something to keep in mind also.

Perhaps nVidia changed its mind after the preliminary release.
Intellisample and Z-buffer compression are mentioned in many places.
But, in fact, both functions (do not) work on the 5200 and 5500 in exactly the same way.

@agent_x007

WOW how does it possible fx5200 1500 points higher than fx5600 in 3dmark 2001 ?According to old reviews fx 5600 should be faster than fx 5200 even with same clocks. Fx 5600 has 80M transistors , 5200 only 45M. Those extra transistors were used for additional cashes , blocks etc. Something deffinately not right here.

ciornyi wrote on 2025-11-28, 17:23:

@agent_x007

WOW how does it possible fx5200 1500 points higher than fx5600 in 3dmark 2001 ?According to old reviews fx 5600 should be faster than fx 5200 even with same clocks. Fx 5600 has 80M transistors , 5200 only 45M. Those extra transistors were used for additional cashes , blocks etc. Something deffinately not right here.

Newer drivers are trash on GeForce FX.

ciornyi wrote on 2025-11-28, 17:23:

@agent_x007

WOW how does it possible fx5200 1500 points higher than fx5600 in 3dmark 2001 ?According to old reviews fx 5600 should be faster than fx 5200 even with same clocks. Fx 5600 has 80M transistors , 5200 only 45M. Those extra transistors were used for additional cashes , blocks etc. Something deffinately not right here.

Perhaps because 3DMark 2001SE uses DirectX 8.1.
In the 2003 version, which uses DirectX 9, everything may be completely different.

ciornyi wrote on 2025-11-28, 17:23:

@agent_x007

WOW how does it possible fx5200 1500 points higher than fx5600 in 3dmark 2001 ?According to old reviews fx 5600 should be faster than fx 5200 even with same clocks. Fx 5600 has 80M transistors , 5200 only 45M. Those extra transistors were used for additional cashes , blocks etc. Something deffinately not right here.

Drivers, 71 and earlier are just faster for DX6/DX7 stuff.
Look at Ultra scores for best accuracy.

The FX series could really use some comprehensive performance and image quality analysis. If someone wanted to put that kind of effort in. Each chip has peculiarities and the various drivers all have different image quality cheats. It got really dirty when Shader Model 2 games first came in. Radeon R300 is such a hugely superior architecture for SM2, anti aliasing and texture filtering.

shevalier wrote on 2025-11-28, 14:57:

In short, nothing worked out. The theory has been completely destroyed. With identical RAM timings, 5200 and the same one reflas […]
Show full quote

In short, nothing worked out.
The theory has been completely destroyed.
With identical RAM timings, 5200 and the same one reflashed to 5500 show completely identical results.
Only the numbers in the name of the video card are more pleasant.

I found a bit more difference for '03 for 5600 vs 5200 :
NV34U (350/700 = 2215):

NV31U (350/700 = 2847) :

Well if its just driver related issue thats sux. As fx 5600 should be at ti 4200 level . And as i have quite a few of those its scores 11k to 15k.
here is based on pentium4 3ghz

There are plenty of GF 5600s that are nowhere near 4200 Ti level of performance (due to crappy VRAM they use or insufficient core clock) - you have to watch out what you buy.

Good news is, you usually need a fast platform to see the difference :
4200 Ti (result 15k) :

5600U (result 14.4K) :

EDIT : Here's an example of "garbage" 5600 (result 11k) :

agent_x007 wrote on 2025-11-28, 20:46:

There are plenty of GF 5600s that are nowhere near 4200 Ti level of performance (due to crappy VRAM they use or insufficient cor […]
Show full quote

There are plenty of GF 5600s that are nowhere near 4200 Ti level of performance (due to crappy VRAM they use or insufficient core clock) - you have to watch out what you buy.

Good news is, you usually need a fast platform to see the difference :
4200 Ti (result 15k) :

The attachment 3DMark 01SE.PNG is no longer available

5600U (result 14.4K) :

The attachment 3DMark 01SE.PNG is no longer available

EDIT : Here's an example of "garbage" 5600 (result 11k) :

The attachment 3DMark 01SE.PNG is no longer available

Yeah but very first result you have showed with fx 5600 was 6500 points so thats why i was shocked , 11000 is not that bad .
PS . Fun fact geforce 5600U has 2 revisions first one has 350/350 second 400/400.