The Quadro FX 2500M 2010 Win7 driver worked OK on my latest M90 Win10 Pro x64, except for the same issue I had with a Go 7900 GS. Memory leaks -- considered normal for Win7 and earlier -- became a serious issue because of Win10's CONSTANT compression / decompression of memory (a technique they obviously "borrowed" from the Android system). Specifically, after an hour of heavy use, the display would lock up and take a minute to go blank and then fix itself. Once it happens, you have to reboot, else the lockup recurs every 2 to 5 minutes. To make a long story short, after a few days trying a dozen different drivers, by sheer LUCK I stumbled on what I believe is the ONLY solution. This driver is not on Nvidia's site, nor on any manufacturer site, and does not come up on MS Update catalog with a regular search. I can't find any reference to it on on this site, nor anywhere on the internet, except a few sites prior to 2016 that give dead links to what were probably incomplete versions. The correct links are: http://download.windowsupdate.com/m..._c9188d8cd4c54a0f0ed64a9e51791fc15350fa05.cab for Windows Vista / 7 / 8 / 8.1 / 10 (32 bit) and: http://download.windowsupdate.com/m..._a849e18a177b21e2d348609936c2a96ffc9bcb80.cab for Windows Vista / 7 / 8 / 8.1 / 10 (64 bit) You can bring up this driver, dated 12/7/2012, by inputting the version #: 9.18.13.768 in the Microsoft Update Catalog search box. You can also get it by inputting: nvidia preview. So it is a "preview" not a regular driver update, and you can see some other distinguishing characteristics in the description. E.g. Only English is supported as a language, and no supported hardware is listed. Clearly, Microsoft has no desire to make this widely available. I think they took Nvidia drivers from several sources and recompiled them to fix the memory leaks, and that would create obvious licensing issues. So when you download, you will find a nv_whql.inf file with no less than 566 adapters listed, which may well cover every adapter made by Nvidia from April of 2004 to the driver date. The original GeForce 6800 Ultra is there as well as the NVS 110 and several oooold nForce models. It works perfectly with my FX 2500M as well as my Go 7900GS (and fixes a few minor issues I had with my mod of the no-Go 7900 GS desktop version). I honestly believe all these legacy cards could work under Windows 10 with this driver! And for anything later than 12/7/2012 we have the search tool on Nvidia's site. All bases are covered. Note that I force fed this driver with the "Have a disk" method, but I think the included setup.exe program should work. I have the Nvidia control panel, which might be from remnants of other drivers I tried, though it didn't work right before this install. It does have two NvCplSetup*.exe programs as well as a 3DVision_307.68.exe program. (Note this isn't the same 307.68 driver posted in the archives, here, which only supports 31 Dell computers.) I didn't see the PhysX program in the driver, or in Program Files and Features, which is fine with me, since my card is too old to support PhysX anyhow. You might have to get PhysX from another Nvidia download. It is WHQL signed, so no need to disable driver signature. I'm downloading a few copies of this driver and storing in safe places, in case MicroSloth decides to pull the plug. Regards, Ol' Geezer
A bit of skepticism: From several sources - like what? And you do not fix memory leaks by recompiling the same source files, you have to find the leaking places and fix them.
You're not going to say the sources are illegitimate, are you? It's right there in the Microsoft Update Catalog. The sources are all 566 Nvidia drivers -- just look at the .INF file, they're all in there. And it works for me with a Go 7900 GS as well as an FX 2500M upgrading those to work with Win10, where nothing else will -- and I looked literally for YEARS with the Go 7900 GS, and a couple weeks with the FX 2500M. It was common knowledge, back in XP days, that C, C++ compilers produced code that leaked. They eventually got their act together and wrote leak-proof compilers, actually advertised as such. I don't know for sure -- can't prove it -- but my best guess is Microsoft got hold of the source code for every Nvidia driver, and recompiled to get leak-proof binaries. It was a proof of concept thing. They wanted Win10 to be the OS for EVERY machine, but it constantly compresses / decompresses memory to avoid the sloooow disk VM. Memory leaks that only require a reboot once a week under Win7 would require it once an hour under Win10. They needed to show the old cards could theoretically work under Win10. They have this driver in their catalog, but it does not turn up with a regular search. You have to know the driver number. Hardware names or IDs don't cut it. For example, if you input FX 2500M into Microsoft's Update Catalog, the latest driver you get is from 3/5/2009, numbered 7.15.11.7967. This "preview" driver is dated 7/18/2013, number (not shown in the Version column) 9.18.13.768 . Yet it clearly supports the FX 2500M, installs and reports itself as an FX 2500M driver, and works BETTER than the latest version available anywhere else. Situation is similar with the Go 7900 GS. These facts alone make it worth further investigation, IMHO. OG
I was developing with C/C++ from 1991 till XP days, and I have never heard about compilers to produce code with memory leaks. But even if it is true it would be easier for Microsoft to ask NVIDIA to recompile their own sources. PS Anyway the background story is completely irrelevant because contribution would be in drivers themselves. The scepticism was not about the drivers, it was namely about the story.
I agree, especially in more classic C/C++ where the developer is solely responsible for memory management, and not the language or the compiler. This is what makes C/C++ both strong and hard, since you actually have to pay attention. The compiler really isn't involved in managing memory.
you know you can disable the memory compression in windows 10, right? https://forums.anandtech.com/thread...-am-i-reading-task-manager-correctly.2509117/ start powershell as admin and use these commands Disable-MMAgent -mc <-- disables Enable-MMAgent -mc <-- enables
Read OP, he attributed memory issues to windows 10 memory compression: "became a serious issue because of Win10's CONSTANT compression / decompression of memory (a technique they obviously "borrowed" from the Android system)."
He attributed memory leaks to bad build of drivers. Win10 he mention only worsened the issue. But of course, the link to OP is there.
I think we're arguing semantics. Nearly all programmers know that C and C++ don't have automatic garbage collection, while nearly all other languages do. If a programmer is inattentive (or possibly hell bent on getting the last ounce of speed for some graphic-intensive game), and writes leaky code, then compiles it with C, the compiler will produce leaky code. It eventually became much easier to write tight code, with built in tools in C++ as well as many open source add-ons to check for this. All those leaky XP programs didn't just vanish because programmers got a conscience! I never meant to imply MS recompiled and did nothing else to tighten the code. But with modern tools the process is perfunctory. I'm sure it could be largely automated. As far as MS asking Nvidia to make this driver, I don't think Nvidia considers itself an employee of Microsoft. I don't think they'd do it without a profit motive. And why would they write this superior driver, and then never offer it on their site?? There are hundreds of references to various .INF files on Nvidia's site, but exactly three references to a nv_whql.inf file. It's in the forums, and points back to MS' Update Catalog. That one is from 2010, and is well known. AFAICT, the 2013 driver linked here is practically UNknown. Best I can tell, if you have a nv_whql.inf file, it originated from the MS Update Catalog. Does that PROVE they compiled the driver? No. But these drivers all span several different manufacturers. t would be strange for a company like Dell to write it and include support for a dozen competitors. BTW, there are TWO references to nv_whql.inf on Dell's site, both pointing to that same 2010 driver on MS' site. The problems I had with the Go 7900 GS and FX 2500M had all the earmarks of memory leaks. The binaries that come with this 2013 nv_whql.inf file seem to have solved those problems completely. Time will tell for sure, but I it's probably a better solution than turning off Win10's compression. Just to test, I've had 10 or more pages open at a time, with heavy graphics and streaming videos (I have no use for such a heavy load normally). Under XP this would slow the computer to a crawl with the HD sputtering away. With Win10 and this driver, everything is smooth as glass. It's an amazing thing to see! There are several other drivers returned by searching KB2836964 in the update catalog. (Articles on this KB update were either deleted from MS' site, or inhouse and unavailable to moi.) I don't have the hardware to test them, but it looks like they're similar in providing support for legacy Intel and ATI cards.
Well, I turned off memory compression (along with page combining) long ago, just in case. I have enough RAM and I don`t need CPU spending time on these features. Also I am glad drivers for my 1070 are completely leaks-free. PS And I am super glad that I do not use C/C++ at work.
See, I'm a big fan of the M90, so I'm limited to 3.25 GB RAM The compression / decompression is like a MIRACLE. Been getting into VB.NET myself. We're close to the day when programmers set up the interface, and maybe point to an equation or two. The language does all the grunt work.
I think the reason this is the only driver that works for me might be my 1440x900 resolution displays combined with these old cards and Windows 10. Older drivers seem to work OK with these cards in most cases, but the 1440x900 res seems to pose special problems for Win10.