VOGONS


Pascal GPU support in WinXP

Topic actions

Reply 100 of 119, by RetroGamer4Ever

User metadata
Rank Oldbie
Rank
Oldbie
Alexraptor wrote on 2026-10-28, 15:27:

Now this would definitely completely overkill, but I'm curious and have to ask anyway: Is it at all in the realm of possibility to backport SLI support for Pascal/Maxwell GPU's?

Pascal was the last Nvidia GPU chip to support SLI in hardware, so possibly, though I doubt it would be a worthwhile thing, unless it could be implemented and significantly improved for XP.

Reply 101 of 119, by kcajjones

User metadata
Rank Newbie
Rank
Newbie
Shagittarius wrote on 2026-10-28, 01:55:

Installed my 1080ti to my XP and Win10 dual boot today. Everything seems hunky dory so far.

I should have said in my post - I'm using PAE (physical address extension) patches to allow upto 4GB RAM. Maybe this is the source of my issues? I'll try a copy of Windows XP without this and report back.

Reply 102 of 119, by xtreger

User metadata
Rank Member
Rank
Member

A potentially silly question, but still something I'm curious about - is it similarly possible to backport one of the late 2000's card's drivers to Win98? Or even to fix some issues that unofficial NVidia 7xxx drivers give on Win98?

Reply 103 of 119, by SupraGSX

User metadata
Rank Newbie
Rank
Newbie
RedCharles wrote on 2026-10-26, 20:57:

No problem, I've been hammering it with Claude and ChatGPT working together. 19th try running right now. Claude says we've got a 50% chance of getting it working 🤣.

Worst case, I buy a quadro P4000, and carefully use an angle grinder to make it fit.

I'm on driver 31 now. Still going. Should know in a day or two if it will work.

Any luck?

Reply 104 of 119, by RedCharles

User metadata
Rank Newbie
Rank
Newbie

ChatGPT is building the 49th driver right now.

I had Claude generate a report for you.
Quadro P2200 on XP: Progress Report for SupraGSX
Summary
With three secure-boot fixes to your Forceware 382.69 miniport, the Quadro P2200 (DEV_1C31) completes NVIDIA's full GPU init on XP SP3. A hard boot freeze remains after that.
Fixes
• ACR version check (0x23): swapped in the 431.02 ACR.
• SEC2 dependency check (0x17): used 384.94's alternate SEC2, which needs only v1 PMU, FECS and GPCCS.
• WPR region field 0 reads 12, expects 15: accepted that pair only.
Freeze
With all three fixes, XP hangs at boot: black screen with signal, no event log, no dump. Test builds run to a checkpoint, then fail cleanly to Code 10 and log. Every checkpoint from Driver 28 to 48 boots, covering SEC2 readiness, init and load of all 42 engines, master init (returns TRUE) and the start of the HwFindAdapter tail.
Untested
The rest of HwFindAdapter (mostly registry reads), HwInitialize and the first mode set. We suspect display bring-up.
Questions
1. Did the 1080 Ti port hang at HwInitialize or the first mode set?
2. The P2200 is DisplayPort-only. Any known DP issues on the 368.81 base?
3. Does region field 0 reading 12 instead of 15 mean anything to you?

Reply 105 of 119, by SupraGSX

User metadata
Rank Newbie
Rank
Newbie
RedCharles wrote on 2026-10-31, 19:00:
ChatGPT is building the 49th driver right now. […]
Show full quote

ChatGPT is building the 49th driver right now.

I had Claude generate a report for you.
Quadro P2200 on XP: Progress Report for SupraGSX
Summary
With three secure-boot fixes to your Forceware 382.69 miniport, the Quadro P2200 (DEV_1C31) completes NVIDIA's full GPU init on XP SP3. A hard boot freeze remains after that.
Fixes
• ACR version check (0x23): swapped in the 431.02 ACR.
• SEC2 dependency check (0x17): used 384.94's alternate SEC2, which needs only v1 PMU, FECS and GPCCS.
• WPR region field 0 reads 12, expects 15: accepted that pair only.
Freeze
With all three fixes, XP hangs at boot: black screen with signal, no event log, no dump. Test builds run to a checkpoint, then fail cleanly to Code 10 and log. Every checkpoint from Driver 28 to 48 boots, covering SEC2 readiness, init and load of all 42 engines, master init (returns TRUE) and the start of the HwFindAdapter tail.
Untested
The rest of HwFindAdapter (mostly registry reads), HwInitialize and the first mode set. We suspect display bring-up.
Questions
1. Did the 1080 Ti port hang at HwInitialize or the first mode set?
2. The P2200 is DisplayPort-only. Any known DP issues on the 368.81 base?
3. Does region field 0 reading 12 instead of 15 mean anything to you?

1. 1080 Ti: did it hang at HwInitialize or the first mode set?
No my blockers were earlier: ACR version rejection, VPR/WPR state, then SEC2 readiness caused by a 56-byte XP boot descriptor versus an 84-byte descriptor expected by the newer bootloader. After correcting those, HwFindAdapter returned success and hardware Direct3D worked.

I also needed two nv4_disp.dll changes to select the existing Pascal display-object paths. Those are already included in Forceware; you should verify you retained the matching display DLL while replacing the miniport.

2. Known DisplayPort issues in 368.81?
Yes: link-rate selection, bandwidth/color-depth handling and scaling could produce blank or garbled output. I corrected several of these.

Start with one directly connected monitor at 1024×768/60 or standard 1080p/60. Trace entry and exit of:

- The remaining HwFindAdapter path.
- HwInitialize.
- Display-driver initialization and the first IOCTL_VIDEO_SET_CURRENT_MODE.

Your checkpoints establish substantial progress, but deliberately returning Code 10 before the remaining code runs cannot identify where the unrestricted boot hangs.

3. Does 12 instead of 15 mean anything?
Very much so, if “field 0” is the WPR read-permission mask. We observed this exact pattern:

Firmware policy Read mask Write mask

Older XP expectation 0xF 0xC
Newer ACR result 0xC 0xC

Supported Windows 7 comparison also used C/C. Thus, 12 can represent an intentional newer firmware policy rather than corruption. Accepting that pair only lets verification continue; it doesn’t change the hardware permissions or prove subsequent accesses are compatible.

If you can provide the exact comparison instruction and complete expected/actual region record, then compare that same state with a stock P2200-supported driver. Final 1080 Ti solution retained its original checks; changing F to C alone wasn’t the complete repair.

The strongest next step is locating the first operation that never returns, while checking that WPR compatibility change against the supported P2200 control. Display bring-up is a reasonable lead, but remains unproven.

Reply 106 of 119, by RayeR

User metadata
Rank Oldbie
Rank
Oldbie

Cool, so now full 4k (3840x2160 at 60 Hz) is possible on GTX 970? Both HDMI and DP?
I was hunting this for over 2 years after I found Yas patch that unlocked HDMI bandwidth to 600MHz (no DP) but I was still somehow limited to 1600p @89Hz in nvidia drivers (any custom resolution over 1600p was immediatelly rejected).
Also nice to see even newer GTX 10xx works...

Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 8GB DDR3, 128GB SSD, GTX970(GF7900GT), SB Audigy + YMF724F + DreamBlaster combo + LPC2ISA

Reply 107 of 119, by a_h_adl

User metadata
Rank Newbie
Rank
Newbie

Thanks SupraGSX for your effort making this possible. My Titan XP is one the way hoping to use it on my multi boot XP beast machine. So, reading this thread still not sure if it's fully functional yet with XP 32 and 64 bit?

Reply 108 of 119, by agent_x007

User metadata
Rank Oldbie
Rank
Oldbie
RayeR wrote on 2026-11-03, 07:36:

Cool, so now full 4k (3840x2160 at 60 Hz) is possible on GTX 970? Both HDMI and DP?
Also nice to see even newer GTX 10xx works...

Note : DP may require firmware upgrade (on both GTX 900 and 1000), for modern monitors to work properly.

Reply 109 of 119, by RayeR

User metadata
Rank Oldbie
Rank
Oldbie

OK, is there also 32b version of NVIDIA DisplayPort Firmware Updater.exe?
BTW what it exactly do? Does it flash something to video BIOS ROM? or somewhere else? Maybe the flashrom is divided to multiple sections separate for vbios and for DP FW?

Gigabyte GA-P67-DS3-B3, Core i7-2600K @4,5GHz, 8GB DDR3, 128GB SSD, GTX970(GF7900GT), SB Audigy + YMF724F + DreamBlaster combo + LPC2ISA

Reply 110 of 119, by RedCharles

User metadata
Rank Newbie
Rank
Newbie

I've ran 4k 60hz on a 750 Ti with these drivers.

Reply 111 of 119, by RedCharles

User metadata
Rank Newbie
Rank
Newbie

P2200 works. Took 69 Drivers. Edit: And just so you understand, the turn around on these was about an hour each. 200GB of drivers 🤣.

Just played Far Cry, Doom 3, CS 1.5 and FEAR. The P2200 draws 75w, board power only, and it's about 230% faster in FEAR than a 4gb 750 Ti; 56fps to 185fps average at 4k.

Still trying to get 4k 120hz to work. Once I have that worked out, I'll share.

Claude's summary of Codex's work.
1. Secure boot failed (Code 10). The card’s security chain rejected the driver’s firmware three times:

An old security loader: swapped for the one from NVIDIA 431.02.
A security-processor version mismatch: used an alternate SEC2 from 384.94, with its startup descriptor converted from 56 to 84 bytes.
A memory-permission value that was 12 instead of 15: newer firmware’s intended setting, so we accepted it.

2. XP froze at boot, leaving no logs. We built about 40 test drivers, each stopping cleanly at a later point, until the hang narrowed to one command-channel request in HwInitialize.

3. Skipping that request caused a blue screen instead. NVIDIA’s display driver hit a deliberate “fatal error” trap. We made later builds show the graphics firmware’s status on the blue screen: error 0x18. The firmware read a factory setting (“level 2”) it didn’t recognize.

4. The fix: graphics-engine firmware (FECS/GPCCS) from 431.02, which accepts level 2, with matching 84-byte descriptors. The error cleared, the 4K desktop appeared, and Far Cry ran with hardware 3D.

Last edited by RedCharles on 2026-11-05, 02:16. Edited 3 times in total.

Reply 112 of 119, by RedCharles

User metadata
Rank Newbie
Rank
Newbie

Update: I got 4k 120hz working on my monitor. Not sure how to share the driver or what rules there are about this sort of thing.

Driver 71 change log — Quadro P2200, Windows XP SP3 32-bit, ASUS XG27UQDMS, direct DisplayPort

**What changed from Driver 70**

Driver 71 adds one monitor-specific timing rule to `[nv_timingRestrictions_addreg]` in `Display.Driver\nv4_dispi.inf`. The NVIDIA miniport and display DLL are identical to Driver 70. Driver 69 is the rollback build.

```ini
HKR,,NV_R&T,%REG_SZ_APPEND%,"R&T0000=3840,2160,32,120,*,AUSAA34,ADD,OEM,106656,4000,48,32,2222,3,5,+-"
```

**Why the entry was needed**

The ASUS advertises 3840 × 2160 at 120 Hz using CTA video code 118. The part of the XP driver that builds the mode list accepts codes 1 through 34, so it skips 118. The working 4K60 mode comes from a separate detailed timing in the monitor’s EDID. That explains why XP listed 4K60 but omitted 4K120.

**How the timing fits**

The new entry keeps the porches, sync widths and totals from the monitor’s working 4K60 detailed timing. At 120 Hz, those totals give `4000 × 2222 × 120 = 1066.56 MHz`. The standard CTA 4K120 timing calls for 1188 MHz. The custom timing fits the driver’s tested 1074.63 MHz limit for RGB at 8 bits per channel over HBR3 ×4.

| Timing | Active | Front porch | Sync width | Back porch | Total | Polarity |
|---|---:|---:|---:|---:|---:|---|
| Horizontal | 3840 | 48 | 32 | 80 | 4000 | + |
| Vertical | 2160 | 3 | 5 | 54 | 2222 | − |

**Why selecting the saved mode matters**

In an offline replay of the XP driver, submitting 1066.56 MHz through the custom-mode validator triggered an EDID range rejection and a fallback to 60 Hz. Selecting the installed 120 Hz entry took a different path: the driver found the ASUS timing in its saved list and passed 1066.56 MHz to the DisplayPort output code. The bandwidth check remained active and selected 8 bits per channel on the link. XP’s desktop setting was 32 bits per pixel.

This is the timing change from an already working P2200 Driver 70 build. The single INF line does not supply the earlier P2200 support needed with a stock NVIDIA XP driver.

Reply 113 of 119, by SupraGSX

User metadata
Rank Newbie
Rank
Newbie

Did you use a 382.69 release as base or stock 368.81? I would ask your AI to compile a patch script that recreates the finished driver byte identical. That way anyone that wants to recreate it can both audit what's actually different (malware protection) and use its findings to recreate or modify the driver. All my releases have rebuild instructions for that purpose.

I'm glad you got it working!!

Reply 114 of 119, by agent_x007

User metadata
Rank Oldbie
Rank
Oldbie
RayeR wrote on Yesterday, 02:02:

OK, is there also 32b version of NVIDIA DisplayPort Firmware Updater.exe?
BTW what it exactly do? Does it flash something to video BIOS ROM? or somewhere else? Maybe the flashrom is divided to multiple sections separate for vbios and for DP FW?

You mean this :
Firmware 32-bit : LINK ?

What it does (my guess) : Updates card's vBIOS configuration for DP ports, to match 1.3/1.4 standard settings (since those weren't finalized when older cards were made).

Reply 115 of 119, by myne

User metadata
Rank l33t
Rank
l33t

So... I'm a little bit wondering if you might be approaching this a bit backwards.

You're modding an existing XP driver to suit new cards. 368.81

Have you considered modding a win10-32 driver back to xp?
391.35 | Windows 10 32-bit
https://www.nvidia.com/en-us/drivers/details/132839/

I built:
Convert old ASUS ASC boardviews to KICAD PCB!
Re: A comprehensive guide to install and play MechWarrior 2 on new versions on Windows.
Dos+Windows 3.11+tcp+vbe_svga auto-install iso template
Script to backup Win9x\ME drivers from a working install
Re: The thing no one asked for: KICAD 440bx reference schematic

Reply 116 of 119, by Trashbytes

User metadata
Rank l33t
Rank
l33t
myne wrote on Today, 07:55:
So... I'm a little bit wondering if you might be approaching this a bit backwards. […]
Show full quote

So... I'm a little bit wondering if you might be approaching this a bit backwards.

You're modding an existing XP driver to suit new cards. 368.81

Have you considered modding a win10-32 driver back to xp?
391.35 | Windows 10 32-bit
https://www.nvidia.com/en-us/drivers/details/132839/

I can think of a couple of reasons that would be a huge amount of work, there was a huge amount of driver changes between XP and Win10 mostly to do with how the driver model actually worked with the OS and hardware none of which XP has.

Not saying it couldn't be done but rather is that amount of work actually worth doing when you can take the final XP driver and mod that to support the newer cards, its far less work as you don't need to mess with the underlying driver model.

AI is inevitable, there is nothing you can do to stop it, no Ironman snap to save you from the Basilisk.

Reply 117 of 119, by myne

User metadata
Rank l33t
Rank
l33t

Funny, I figured the extremely well documented windows part would be simpler than the poorly documented, mostly reverse engineered Nvidia part.

I built:
Convert old ASUS ASC boardviews to KICAD PCB!
Re: A comprehensive guide to install and play MechWarrior 2 on new versions on Windows.
Dos+Windows 3.11+tcp+vbe_svga auto-install iso template
Script to backup Win9x\ME drivers from a working install
Re: The thing no one asked for: KICAD 440bx reference schematic

Reply 118 of 119, by SupraGSX

User metadata
Rank Newbie
Rank
Newbie
myne wrote on Today, 07:55:
So... I'm a little bit wondering if you might be approaching this a bit backwards. […]
Show full quote

So... I'm a little bit wondering if you might be approaching this a bit backwards.

You're modding an existing XP driver to suit new cards. 368.81

Have you considered modding a win10-32 driver back to xp?
391.35 | Windows 10 32-bit
https://www.nvidia.com/en-us/drivers/details/132839/

Yeah I considered it but you absolutely would not want to do that. wddm vs xddm is first obvious reason. Stability is next thing. Just porting the gpu you're not changing any of the os facing components, so all the tweaks and bug fixes Nvidia included through 15 years of xp driver development are in 368.81. With GPU only port it's more finite, there's clear things to get right and you can reference working drivers behavior. You also have to realize there's no source code to reference this is all binary patching.

Reply 119 of 119, by Trashbytes

User metadata
Rank l33t
Rank
l33t
myne wrote on Today, 09:46:

Funny, I figured the extremely well documented windows part would be simpler than the poorly documented, mostly reverse engineered Nvidia part.

True its well documented but XP doesn't have any of the driver model enhancements Win10 does nor does it have anyway of adding them via a driver patch, it would be an enormous amount of work and debugging of both the driver and the driver model along with all the core Windows system changes required to support that driver model.

If you want a taste of just how ...buggy that could be just go install Vista RTM unupdated on XP era hardware from 2000 - 2004 and watch how horribly Vista handles release drivers for it. Vista being the transition from the old driver model to the new driver model had to juggle both and was well know for being unstable in part because it had to work with both.

I think what Supra has done here is nothing short of black magic, getting Pascal hardware to work fully under XP is truly amazing, I for one am truly grateful for that alone.

Anything extra on top is just icing for me.

AI is inevitable, there is nothing you can do to stop it, no Ironman snap to save you from the Basilisk.