Turning GeForceFX to QuadroFX with SQ4 on 44.03. Important clarifications.

Unwinder

First, I'd like to thank all guys who reported about successful NV30/35->QuadroFX modification with Detonator 44.03 using SoftQuadro4 scripts. It was a surprise for me and due to their reports I investigated this issue deeper and I can confirm that _some_ boards (probably GFFX only) are really moddable with SQ4 using early DetonatorFX versions (starting from 44.03 up to the latest non-encrypted 44.65).
The explanation is pretty simple: NV's SQ4 protection system seems to be screwed for some boards in these drivers. Currently I've tried to use SQ4/NVStrap/NVStrap antiprotection for NV18 (the protection seems to work fine, OGL applications cause PC lockup during OGL context destruction) and NV30 (the protection doesn't work). I've also tried to update SQ4 for newer drivers and modded NV30 seems to work fine on 45.xx drivers too. In 5x.xx drivers modded NV30 behaves the same as GF4 - the OGL applications cause PC lockup. So the protection seems to be fixed.
I'm not pretty sure if the hole in protection system is NV3x specific or it applies to some NV17/18/25/28. So any feedback is welcome.

Bluefirexp · *Last edited by Bluefirexp; 10-16-2003 at 08:07.*

Yes, protection is there on my NV25 ( Leadtek A250 LE Ti4200 ) and detonator 44.03- lockups in Specviewperf.
I'm getting a Leadtek A350 TDH LX ( FX5900 ) tomorrow and will test it aswell

MikeMcNuke · 10-16-2003, 11:32 | posts: 17

Hi Unwinder,
just to give you some more verification: I sucessfully modded my GF FX 5900 Ultra with your scripts and the driver versions 44.03 and 44.65 as you said. WOW really great. Next week I will also try the rumoured hardware mod.
cya and many thx for SQ and RT
Mike McNuke

Arty · 10-16-2003, 12:46 | posts: 10

Thanks for the efforts everyone, and your great work untill today Unwinder..

I was about buy a new card either Radeon9800XT based one or FX 5900 Ultra based one, I was hoping to convert them to pro versions as I'm mostly working with 3D app rather than the games.

As the FX3000 beats GL X2 in pro apps, I guess I'll wait a bit more until that thread reaches somewhere.

I've read all the threads about FX to Quadro here. My question is, on the GF4 > Quadro mods there was hardware differences on chip and some features of the original quadro was not available after the mod.

Is it still the same way with current FX > quadro mods? or are the chips same again just like the old days and modding can enable perfect quadro.. ?

Thanks,
Arty

Bluefirexp · 10-17-2003, 07:52 | posts: 591 | Location: Stuttgart

interstingly Speviewperf 6.1.2 is causing a PC-crash already at the closing of the first window- so anytime a part of the benchmark is over and the window is going to close, the system hangs. But on Specviewperf 7.1 i can run about half of the benchmark- f.e. half of ugs-03- without any problems and only then there is a crash....
It seems to me, as if the protection is application specific

Bluefirexp · *Last edited by Bluefirexp; 10-17-2003 at 22:12.*

My A350 TDH LX is here, and with 44.65 it works now as Quadro 3000!

3dsmax 5 with maxtreme is superfast now, Autocad also. I will add screenshots and bench-results l8ter

EDIT: here my Specviewperf 7.1 results ( btw, ist there already a 7.1a?! ):
----------------------------------------------------------------------------------
softquadFX5900LX @ FX3000 @ 400/700

Athon XP 2.2ghz ( 11 x 200, 1024MB PC3200 timings: 11-2-2-2 )

3dsmax-02 Weighted Geometric Mean = 23.40
drv-09 Weighted Geometric Mean = 89.18
dx-08 Weighted Geometric Mean = 95.13
light-06 Weighted Geometric Mean = 27.77
proe-02 Weighted Geometric Mean = 20.80
ugs-03 Weighted Geometric Mean = 9.102
----------------------------------------------------------------------------------

It's obvious- the mod works!

Still, UGS-03 and PROE-02 are very slow, i wonder why FX5800 owners have so fast UGS and PROE.

...But i'm very happy with this overall result and the card is even not OCed and just @700mhz DDR.

10-17-2003, 17:32 | posts: n/a

Since I'm using a Geforce FX 5200 would you recommend the highest compatible driver is the 44.65 using SQ4/NVStrap/NVStrap antiprotection?

Also, when will there be an update for the Detonator FX Decoder? I tried using it with the 52.13 drivers but it wouldn't patch.

Bluefirexp · *Last edited by Bluefirexp; 10-17-2003 at 22:27.*

yes 44.65 is highest working until now.
btw, even a stupid FX 500 is faster in UGS and PROE than our softmodded FX5900@FX3000 cards, that's too strange...

. This incredible slowdown just can't be only hardware related, there must be some protection in FX3000 driver still...coz all other scores are as fast as real Quadros..
And the FX5800@FX2000 doesn't have this problem, as we see from posts in this forum..

10-18-2003, 20:39 | posts: n/a

Quote:

Originally posted by Bluefirexp
yes 44.65 is highest working until now.
btw, even a stupid FX 500 is faster in UGS and PROE than our softmodded FX5900@FX3000 cards, that's too strange... . This incredible slowdown just can't be only hardware related, there must be some protection in FX3000 driver still...coz all other scores are as fast as real Quadros..
And the FX5800@FX2000 doesn't have this problem, as we see from posts in this forum..

I think this is caused due the lak of Registercombiners, NV replaced them with FP-units due better DX9 Shader-Performance.

Samiam · 10-18-2003, 21:45 | posts: 144

anything above 42.51 locks up my 4200 using SQ. I tried 44.03 and got lock ups.

Bluefirexp · 10-19-2003, 00:09 | posts: 591 | Location: Stuttgart

Quote:

Originally posted by AlphaTier
I think this is caused due the lak of Registercombiners, NV replaced them with FP-units due better DX9 Shader-Performance.

Well, but why is then a REAL Quadro FX3000 faster than a FX2000 in PROE and UGS? There should be also the same difference as with the soft-modded cars i think...

10-19-2003, 05:43 | posts: n/a

Well, we just have to wait the REAL SoftQuadroFX then. Maybe there is some features can not be turn on by SQ4 script.
This is my latest test using Athlon XP2600+
Run All Summary

---------- SUM_RESULTS\3DSMAX\SUMMARY.TXT
3dsmax-02 Weighted Geometric Mean = 21.59

---------- SUM_RESULTS\DRV\SUMMARY.TXT
drv-09 Weighted Geometric Mean = 123.2

---------- SUM_RESULTS\DX\SUMMARY.TXT
dx-08 Weighted Geometric Mean = 107.5

---------- SUM_RESULTS\LIGHT\SUMMARY.TXT
light-06 Weighted Geometric Mean = 23.66

---------- SUM_RESULTS\PROE\SUMMARY.TXT
proe-02 Weighted Geometric Mean = 35.93

---------- SUM_RESULTS\UGS\SUMMARY.TXT
ugs-03 Weighted Geometric Mean = 44.45

Maya Benchmark 5 GFX score 2.55, Total score 2.25
3DMark2001 score 13433
3DMark2003 score 5432
(I don't understand why I got such a low scores on 3DMark tests)

10-19-2003, 21:20 | posts: n/a

Quote:

Originally posted by Bluefirexp
Well, but why is then a REAL Quadro FX3000 faster than a FX2000 in PROE and UGS? There should be also the same difference as with the soft-modded cars i think...

u have a proof for that? i, wanna see myself.

Bluefirexp · 10-20-2003, 00:05 | posts: 591 | Location: Stuttgart

yes- check out here

Unwinder · 10-20-2003, 06:51 | posts: 10,318 | Location: Taganrog, Russia

Very important info update

After investigating this situation a bit deeper I was able to find out the following:

1) There is no need to install SQ4. Professional capabilities on my NV30 are unlocked in case of installing only NVStrap/NVStrap antiprotection.
It seems like NV don't use GF4-styled PCI DevID override protection on NV3x, so the only differences between pro and gaming boards are PCI DeviceIDs again. The only conclusion I can made from it is that pro and gaming GPUs are equal again.
Considering the fact that NVStrap programs PCI DevID strapping bits that can be also hardwired via the configuration resistors on the PCB, I can say with 99.9% warranty that hardware mod is also possible.

2) Considering that fact that even in case of installing NVStrap/NVStrap antiprotection on 5x.xx I get lock ups in the OpenGL applications, I can assume that these drivers set introduce new type of protection against NVStrap (probably OGL driver checks hardwired ID). No promises on developing new generation of antiprotection scripts

P.S. I don't know if it applies to whole NV3x family or to NV30 only. So feedback from NV35 owners is welcome.

Bluefirexp · *Last edited by Bluefirexp; 10-20-2003 at 15:30.*

Yes, i had these assumptions also, i posted in this thead: http://www.guru3d.com/forum/showthre...threadid=66749

The only strange thing was the lowe performance in UGS03 and PROE02 of a softmodded FX59000>FX3000 compared to a FX2000.

And now i tested, applied ONLY Nvstrap-antiprotection to the driver and installed NVstrap: my NV35 card doesn't work as Quadro without SQ4.

Sandra reports Quadro FX300 OpenGL driver, the card is recognized as Quadro FX3000- BUT performance is low, 3dsmax doesn't find a quadro card also and the OPenGL-properties-tab doesn't contain the specific quadro options.

So without SQ4 the Quadro-mod doesn't work on my FX5900 ( NV35 )

Regarding the UGS and PROE performance weakness of my card now even with SQ4, i think, that the FX5900 seems to be hardware-different from a real Quadro FX3000 GPU.

Bluefirexp · 10-20-2003, 14:16 | posts: 591 | Location: Stuttgart

So you lucky NV30 users, but i don't need UGS performance luckily

10-20-2003, 19:10 | posts: n/a

Well, it still nice to have no hardware-different from QuadroFX, just like my FX2000 (even faster with 500/1000), it seems that no tests can tell the different between a REAL QuadroFX2000 to a SQFX2000.

10-20-2003, 21:14 | posts: n/a

Quote:

Originally posted by Bluefirexp
Yes, i had these assumptions also, i posted in this thead: http://www.guru3d.com/forum/showthre...threadid=66749

The only strange thing was the lowe performance in UGS03 and PROE02 of a softmodded FX59000>FX3000 compared to a FX2000.

And now i tested, applied ONLY Nvstrap-antiprotection to the driver and installed NVstrap: my NV35 card doesn't work as Quadro without SQ4.
Sandra reports Quadro FX300 OpenGL driver, the card is recognized as Quadro FX3000- BUT performance is low, 3dsmax doesn't find a quadro card also and the OPenGL-properties-tab doesn't contain the specific quadro options.

So without SQ4 the Quadro-mod doesn't work on my FX5900 ( NV35 )

Regarding the UGS and PROE performance weakness of my card now even with SQ4, i think, that the FX5900 seems to be hardware-different from a real Quadro FX3000 GPU.

Perhaps, it does not work, cause of we used the 44.03 driver, that has no Q3000 support. Perhaps they integrated the Special features in later Drivers. Tanks to UDA the Cards worked 4 us, but not fully accelerated. Mhm, but how 2 proof that??

10-20-2003, 21:19 | posts: n/a

they used 4523 Detos, http://www.spec.org/gpc/opc.data/vp7...00/config.html

it's possible to AntiNVStrap this Driver?

Bluefirexp · 10-20-2003, 22:18 | posts: 591 | Location: Stuttgart

@Alphatier:

i used 44.65 driver also and this one has FX3000 in it's nv4disp.inf:

NVIDIA_NV35.DEV_0338.1 = "NVIDIA Quadro FX 3000"

still, SQ4 is needed for getting a working Quadro...

Atari · 10-21-2003, 02:20 | posts: 64

Ok.. HAs anyone installed the 44.65's that actually have the QFX 3000 listed as supported devices in the driver set and saw if the UGS test went up?

10-21-2003, 03:12 | posts: n/a

Quote:

Originally posted by Atari
Ok.. HAs anyone installed the 44.65's that actually have the QFX 3000 listed as supported devices in the driver set and saw if the UGS test went up?

I tested it, the Driver still needs SQ4 to work properly as a Quadro Card. UGS Don't went up. All other scores went a little bit down than 4465. (drv9: GF(43) and Qu4403(150) Qu4465(100))

With only NVstrap Antiprotection, the Card will be recognized as a Quadro, but u don't have the QuOGL settings, and GF like scores.

10-21-2003, 03:15 | posts: n/a

Perhaps, the Qu300 is protected through BIOS, or other things changed (PCB and/or Chip)

Unwinder · 10-21-2003, 07:51 | posts: 10,318 | Location: Taganrog, Russia

Quote:

Originally posted by Bluefirexp
And now i tested, applied ONLY Nvstrap-antiprotection to the driver and installed NVstrap: my NV35 card doesn't work as Quadro without SQ4.

Sandra reports Quadro FX300 OpenGL driver, the card is recognized as Quadro FX3000- BUT performance is low, 3dsmax doesn't find a quadro card also and the OPenGL-properties-tab doesn't contain the specific quadro options.

So without SQ4 the Quadro-mod doesn't work on my FX5900 ( NV35 )

Regarding the UGS and PROE performance weakness of my card now even with SQ4, i think, that the FX5900 seems to be hardware-different from a real Quadro FX3000 GPU.

Unfortunately I cannot verify your information myself because I don't have NV35 sample, I’ve only NV30 and NV34 in my test labs. But if it is correct, it can only mean that NV35/NV35GL cores are really a bit different. In this case, the situation is quite similar to GF4 moddability: if you can remember, in our investigation (www.nvworld.ru\docs\sq4.html) we found that NV25 and NV25GL boards contain some differences besides PCI DeviceID (one bit in NV_PBUS_DEBUG_1 register used for Quadro identification). SoftQuadro4 script allowed to override the driver's access to this register and to unlock pro caps regardless of its state. If you can remember, we draw the conclusion that this register indicated some real physical differences in NV25/NV25GL silicon, because even after installing SQ4 there were some task (AA lines and two-sided lighting) in which real Quadro4 drastically outperformed modded GF4 (BTW both UGS and PROE and oriented on AA lines performance).