Anandtech has been exploring the Out of Memory issue with a couple of articles. We've been talking about this (a couple of threads down) but this is getting more interesting. Anandtech's first article is here: http://www.anandtech.com/gadgets/showdoc.aspx?i=3034They describe that Vista constantly runs a forecast (an EsTiMate) of the amount of memory a program will need in the future. This estimate does not consider how much ram is installed in the system or is currently in use. For instance during a flight, Vista and FSX will engage in a running dialog about how much memory they believe FSX might require in the future. If this estimate goes beyond 2 GB, Vista will abruptly shut down the program with an Out of Memory message. This is the infamous 2 GB barrier. The main takeaway of this first article is: 1) It does not matter how much physical memory is actually installed in the system (ram or page file) or how much of this memory is actually in use (ram or page file). 2) Each program has an individual, 2 GB "estimate" limit. To stretch an example: With 50 programs running, Vista might estimate that a total of < 100GB of total physical memory might be required. All programs will continue to run as long as no single program's ram - estimated requirement - goes above 2GB. 3) If any program exceeds its individual 2GB estimate limit, Vista will shut down that single program with that cryptic "Out of Memory" message. Bamm.As you consider this, it might seem that Vista has a severe case of paranoia. Staying ahead of the game with a good estimate of memory usage might be a good idea, but woah there big dog. Look at the big picture. What's my total available memory. Do I still have real resources available? If so, use it. Don't just shut me down because of your wildly pessimistic estimate. There on the other hand, this 2 GB limit and Vista's method of making these estimates are (were?) kernel-level operating system design decisions. There's gotta be some rational considerations in play. But can't I just turn it off and take my chances? Yes, we can turn the auto-shutdown point up to 3GB with the "3GB switch," but be careful here. A 32 bit operating system has a hard limit at 4 GB and really, really wants to reserve that second 2GB for the operating system (OS) and other InVitED guests. Pete Taylor's blog cautions: "Note this (3GB) switch needs to be used with care, since PCI-Express maps the entire address space of the graphics card into the OS address space."http://blogs.msdn.com/ptaylor/default.aspxEver wonder why the Vcard mfgs are not just ramping up Vcard memories? ATI's (proposed) 2900 Vcard with 1GB would only leave 1GB for the operating system. Now we want to burn up some more by giving another big chunk to the programs? A 1 gig Vcard and a fully implemented 3 gig switch leaves ZERO GB for the operating system.Anandtech's second article compares XP to Vista. (Note: In these articles, Anandtech calls these memory estimates "Address Space.") http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3044XP has the same system of estimating future memory requirements, but is not near as paranoid. To a lay person, it seems that XP uses a much more rational method to estimate projected memory requirements. These estimates are MuCh lower. Therefore, it seems that XP may allow programs to run with a (much?) higher - actual - memory load. With this information, it's hard to argue that XP is NoT a superior gaming platform. Anandtech describes that MS is aware of these Vista specific issues. Maybe we'll get a tweak in SP1.But there is no "Fix." We are now at the very edge of a 32bit operating system's capability. Who remembers the difference between Extended and Expanded memory. I was hoping to have (permanently!) forgotten. The article's title "A Messy Transition" starts to make sense. We are into that transition phase now.

>Vista specific issues. Maybe we'll get a tweak in SP1.>>But there is no "Fix." >>We are now at the very edge of a 32bit operating system's>capability. Who remembers the difference between Extended and>Expanded memory. I was hoping to have (permanently!)>forgotten. The article's title "A Messy Transition" starts to>make sense. We are into that transition phase now. >> SamHi Sam, I enjoy reading your posts and thanks for sharing your research and insight.I had to chuckle a bit when you asked about Extended/Expanded memory. Brought back some painful memories. I recall spending countless hours hacking Dr. DOS and 286 assembly code. Thankfully these days are over right? Wrong I guess! The more things change, the more they stay the same.Here is my take FWIW. I am 50 years young. I have retired from writing software after some 20 years of fortran, pascal, C and C++. This experience has included embedded systems, hardware drivers, video processing and higher level apps for Unix. Not to mention the countless "years" I spent hacking the platform environments for the development tools.We are on the edge of a fascinating period in the PC where performance increases will be astounding and reflective of the past decade. We will have a parallel processing capability with multiple cores and dedicated GPUs (perhaps multi-core GPUs one day). We will break the 32 bit

It's interesting to watch them push the envelope: http://www.theinquirer.net/default.aspx?article=41256AMD is showing off their new Barcelona/chipset combo. Their new chipset supports 4 PCIe Vcards . . . but notice they only used three. The 2900XTs are 512MB cards. Three of them (used here) might just be burning 1.5 gigs of the operating system's 2 GB reserve. The 4th would have used it all and left nothing for the this 32 bit operating system. With three 512MB cards running, will the operating system really be happy with just 500MB left for itself? Forget about that 3 gig switch! We'll see how it goes.