Computer died, and tells me "System Failed CPU Test"

[Lazybones]

OK here are the MS links to these issues were it is spelled out

Without the /3GB flag windows applications are limited to 2GB of RAM no matter how much you have installed, the /3GB flag opens up an extra 1GB to applications for at total of 3 GB, IF they where COMPILED to look for it.
http://www.microsoft.com/whdc/system/platform/server/PAE/PAEmem.mspx

That remaining 1GB not included for apps can only be used for a limited number of things by the system running in kernel mode. Part of this space will be used up to track the extra application memory made available.
http://msdn2.microsoft.com/en-us/library/aa366778.aspx

If for some reason you need to run a MASSIVE amount of system process the kernel memory space does make a difference, however this limit is normaly only seen in large systems such as terminal server deployments
http://www.msterminalservices.org/articles/Windows-Terminal-Services-64-Bit-Environment-Part1.html

[Cova]

Neither Vista nor XP give "4GB [of] address space [to] each app".

The very first sentence of the first link Lazybones posted above...

Operating systems based on Microsoft Windows NT technologies have always provided applications with a flat 32-bit virtual address space that describes 4 gigabytes (GB) of virtual memory.

Once again..., 32-bit system = 32-bit pointers.  32-bit pointer can point at 2^32 or 4 294 967 296 bytes of RAM, which everyone just rounds to 4GB.  Big difference between address space and RAM though - you can (and almost always do) have a lot more address space than RAM - and those addresses can point at things other than RAM (eg. under Vista the video-cards onboard RAM is mapped into that address space).  Virtually every OS that supports virtual memory (eg. everything newer than DOS pretty much) works this way.  And if you program in good old C you can make a void-pointer point at any address you want and read that memory (you can write to it too - but that'll prolly result in bad things) - you might end up reading system RAM, video RAM, or something else, depending on how your address space is mapped.

[Melbosa]

Thank you for taking the time to compare the FS system to a custom-built, Lazy.

Awe not even an honorable mention for my quicky ... yeah Lazy's was closer to price with components than mine.  Mine was more the system specs, but with decent parts.

In theory though 32bit is upper limit 4GB.
But whether there is 3GB or 4GB the adapter cards memory, and kernel will still eat up a portion or hide a portion for itself

Take my machine at home.  2x1GB HD 2900 in crossfire.  4GB of RAM.  XP OS.  Really I can only address 3.5GB to apps, windows, etc if I'm lucky, but I have pushed 3.2GB of used system RAM outside my HD 2900s.  So for the cost and the redundancy, I paid the extra.

So if XP can only address 4GB including the video card's RAM and the motherboard's RAM (and PCI devices and other devices on your system), and you have 2GB on your video cards and 4GB in your machine, doesn't that mean that your computer has 2GB of *completely* unaddressable and unusable RAM sitting in it?

Doesn't quite work that way under XP or Vista.  Cova explains this pretty well in his posts; and the different between XP and Vista is how it treats Video RAM on a video card.  This also doesn't mean that the Windows Executive can't utilize the space for its own tasks, and not having to impede the application ram should it require it... very interesting stuff.

[Lazybones]

So what you guys are saying is that no one application can utilize a total of more than 2GB or 3GB (with the porper switch).

With the added note that even with the 3GB switch an application needs to be COMPILED to see more than 2GB.

But last time I checked you can run more than one application at a time on Windows... so by that means, you could still utilize 4GB of RAM, no?  This is how I understood it.

No matter how many applications you run, the sum of your applications could only use 3GB MAX. The other 1GB will be reserved for kernel memory use only, and the kernel is unlikely to use more than half of that 1GB, unless you are spawning hundreds of processes such as when you are running a terminal server and have hundreds of desktops and applications active at once.


64Bit applications running under 64bit windows get past these limitations.

[Melbosa]

LOL you caught my post before I removed it.  I got confused with the difference between virtual and physical.  My bad.  Hence why I removed it.  I am sorry.

I still think it is valuable to have that 1GB for Windows Executive and 3GB of physical mapped to the virtual application pool, versus only 3GB with 2GB mapped?  No?

I have cleared up some of my mis-understandings of the whole, but I still remain that 32bit Windows is still better off with 4GB than 3GB of RAM in the system.

[Thorin]

And here's an interesting article discussing Supreme Commander crashing when it hits the 2GB barrier (also sums up the different terms - virtual memory, virtual address space, physical memory, etc):

http://www.anandtech.com/gadgets/showdoc.aspx?i=3034

With a follow-up article:

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3044

that has an excellent comment:

You can paper over this by adding a few more gigs of ram

Actually, the whole point is that adding more RAM doesn't help. Due to the nature of 32-bit OS's, virtual memory is capped to 4GB and most OS's limit it to 2GB for user applications. This limit doesn't change whether you have 256MB of physical memory or 4GB.

Now are XP and Vista smart enough to use 4GB of physical RAM to store various virtual address spaces for various applications, or are they not able to properly fill and access the full 4GB of physical RAM (regardless of whether the virtual address space for each application is 2GB or 3GB)?

[Cova]

Now are XP and Vista smart enough to use 4GB of physical RAM to store various virtual address spaces for various applications, or are they not able to properly fill and access the full 4GB of physical RAM (regardless of whether the virtual address space for each application is 2GB or 3GB)?

On a system with 4GB of physical RAM, and not using the /3GB boot switch, there will be 2GB of physical RAM that is kept reserved for the kernel (if using /3GB, only 1GB is reserved for kernel).  On a desktop machine / gaming box, that is way more than enough, and a significant part of it is likely not able to be used.  I really don't know if the kernel can use its extra reserved memory as part of the filesystem cache - if it can it's not a complete waste as it'll reduce the IO load on your disks, though it probably could be used better elsewhere.

But on a server box its actually quite easy to run into problems with the kernel running out of reserved memory area.  Especially when using the /3GB switch.  Exchange 2000/2003 is actually REALLY bad for causing the kernel to use up its reserved space, which is why for 2007 they only support production on 64-bit OSs.

And I really don't know if Vista improved much on making use of un-used but reserved kernel memory.  It's gotten worse in so many ways that I wouldn't consider running it anyways, and could care less if it did manage to make better use of a couple hundred MB of RAM.

[Thorin]

Neither Vista nor XP give "4GB [of] address space [to] each app".

The very first sentence of the first link Lazybones posted above...

Yes, I missed a word in that sentence to spell out unconditionally what I meant.  That word is "accessible".  As Lazy's link points out, the application is given 4GB of virtual memory, but 2GB of that virtual memory are not accessible by the application as they're reserved for "Windows executive software".  Which, really, means the application can only use 2GB.

Now are XP and Vista smart enough to use 4GB of physical RAM to store various virtual address spaces for various applications, or are they not able to properly fill and access the full 4GB of physical RAM (regardless of whether the virtual address space for each application is 2GB or 3GB)?

On a system with 4GB of physical RAM, and not using the /3GB boot switch, there will be 2GB of physical RAM that is kept reserved for the kernel (if using /3GB, only 1GB is reserved for kernel)

As Lazy's link points out, each application is given 4GB of virtual memory (aka virtual address space).  There is not 2GB of physical RAM reserved for the application, nor for the kernel.  I'm being semantic here because we're blending together virtual address space and physical address space.  They are *not* the same and in fact only relate to each other through a complex virtual address-to-physical address mapping system, where the physical addresses can be either RAM or HDD locations (and in fact, the HDD locations are virtual addresses again, mapped through the HDD controller software).

[Cova]

Neither Vista nor XP give "4GB [of] address space [to] each app".

The very first sentence of the first link Lazybones posted above...

Yes, I missed a word in that sentence to spell out unconditionally what I meant.  That word is "accessible".  As Lazy's link points out, the application is given 4GB of virtual memory, but 2GB of that virtual memory are not accessible by the application as they're reserved for "Windows executive software".  Which, really, means the application can only use 2GB.

No...  The application is given 4GB of virtual address space (not virtual memory) - up to 2GB of that address space may point at physical RAM - the rest will point at either virtual RAM (eg. pagefile) or other application-addressable memory area's (eg. video RAM).  None of an applications 4GB of address-space is reserved for the kernel - 2GB of the physical address space is.  Which once again brings us back to the fact that a single application can only get 2GB max physical memory (3GB with /3gb switch if app was compiled for it), I think pretty much everyone agree's with that point.

Now are XP and Vista smart enough to use 4GB of physical RAM to store various virtual address spaces for various applications, or are they not able to properly fill and access the full 4GB of physical RAM (regardless of whether the virtual address space for each application is 2GB or 3GB)?

On a system with 4GB of physical RAM, and not using the /3GB boot switch, there will be 2GB of physical RAM that is kept reserved for the kernel (if using /3GB, only 1GB is reserved for kernel)

As Lazy's link points out, each application is given 4GB of virtual memory (aka virtual address space).  There is not 2GB of physical RAM reserved for the application, nor for the kernel.  I'm being semantic here because we're blending together virtual address space and physical address space.  They are *not* the same and in fact only relate to each other through a complex virtual address-to-physical address mapping system, where the physical addresses can be either RAM or HDD locations (and in fact, the HDD locations are virtual addresses again, mapped through the HDD controller software).

I'm very well aware of how page tables work - coincidentally I've been reading up on them recently while researching hardware-accelerated nested page-tables for virtualization software.  And if you read the quote you were replying to a little more carefully, you'll notice I began it with "On a system with 4GB of physical RAM" - I'm reasonably sure that on such a system 2GB of physical RAM really is reserved for the kernel, mostly because 2GB of the physical address space is reserved, and with a full 4GB of RAM the physical address space and physical RAM should be identical.

[Thorin]

Would you buy this?  Why should I not buy this?

http://www.futureshop.ca/catalog/proddetail.asp?logon=&langid=EN&sku_id=BDL10002503&catid=

Thank you to those who responded to this question.

Melbosa, you indicated that I shouldn't buy it because the parts may be (or probably are) cheaper than what you'd normally buy and how the same specs could be had with better (more expensive) parts.

Lazy, you indicated that the price appears to be good when comparing against a build-your-own system

Darren, that quote about stuff owning you?  To me that came across as a kick-him-when-he's-down comment.  I doubt you meant it that way, but I feel the need to explain to you why I replied rather aggressively.

And then the thread went off-topic in a giant let-me-prove-my-intellectual-superiority cockfight that did absolutely @%&#ing *nothing* to improve the quality, usefulness, or joviality of this thread.  Of course, none of us including myself got it totally right.  So instead of nitpicking the nitpicked nitpicking, I've started a new thread regarding Windows memory management.

Let's leave *this* thread to discussing possible alternatives to the Futureshop quad-core Acer computer package linked in the third post.

[Tom]

All I have to say is cheap is bad. You will be sorry, unless you only intend to spend a few minutes a day on the machine, and only keep it for a year. I've learned this by trial and error. Mostly error. Cheap hardware, even if made by reputable OEM/ODMs, is going to be bad.

[Thorin]

"ODM"?  Original Design Manufacturer, as opposed to Orange Democratic Movement?  Sorry, haven't seen that term before.

As far as cheap being bad, I assume you're implying that cheap price = cheap parts used, and that you're implying the inverse, expensive price = expensive parts used.  I would like to point out that there are plenty of manufacturers out there that use cheap parts but charge expensive prices based on their reputation (SONY YOU BARSTARD!!) and manufacturers that use expensive parts but charge cheap prices to win market share (GM used to sell Toyota Corollas rebadged as Chevrolet Prizms for less than the Corollas cost).

Thus, the question becomes whether there are cheap parts in this computer, or whether Futureshop is trying to maintain market share in a highly-competitive market.

Also, I'd like to point out that the computer that failed is one with enthusiast-decided hand-picked researched parts, while my package deal Dell from eight years ago is letting me type out this message.

[Tom]

"ODM"?  Original Design Manufacturer, as opposed to Orange Democratic Movement?  Sorry, haven't seen that term before.

Yeah, these days lots of hardware is made/designed by one ODM, and OEMs and retail fronts just rebrand the ODM hardware.

As far as cheap being bad, I assume you're implying that cheap price = cheap parts used, and that you're implying the inverse, expensive price = expensive parts used.  I would like to point out that there are plenty of manufacturers out there that use cheap parts but charge expensive prices based on their reputation (SONY YOU BARSTARD!!) and manufacturers that use expensive parts but charge cheap prices to win market share (GM used to sell Toyota Corollas rebadged as Chevrolet Prizms for less than the Corollas cost).

You are right, while cheap == bad, expensive does not necessarily guarantee quality. You have to be carefull, but _usually_ if you spend an extra couple hundred dollars on PC hardware, specifically on the PSU and Motherboard, theres a very good chance that you'll end out on top.

Thus, the question becomes whether there are cheap parts in this computer, or whether Futureshop is trying to maintain market share in a highly-competitive market.

Cheap + BigBox store == bad. Unless it purely just a big sale, which I doubt. I don't even like the mid range stuff at Futureshop. It ends up being the same stuff places like Memory express sells as low end hardware.

Also, I'd like to point out that the computer that failed is one with enthusiast-decided hand-picked researched parts, while my package deal Dell from eight years ago is letting me type out this message.

It all depends on useage, and if one of them happened to get those bad caps that hit the entire consumer electronics market several years ago. Of course there are other parameters (did the maker skimp a couple cents on capacitors or other part?)

[Melbosa]

Glad to see someone decided to break this RAM topic off.  Was going to do so when I got into work this am, but you beat me too it (damn traffic).

Melbosa, you indicated that I shouldn't buy it because the parts may be (or probably are) cheaper than what you'd normally buy and how the same specs could be had with better (more expensive) parts.

Well, I also stated that I had experience with Acer products, and that based on those experiences I didn't feel Acer was much better than an eMachine.

As far as cheap being bad, I assume you're implying that cheap price = cheap parts used, and that you're implying the inverse, expensive price = expensive parts used.  I would like to point out that there are plenty of manufacturers out there that use cheap parts but charge expensive prices based on their reputation (SONY YOU BARSTARD!!) and manufacturers that use expensive parts but charge cheap prices to win market share (GM used to sell Toyota Corollas rebadged as Chevrolet Prizms for less than the Corollas cost).

Thus, the question becomes whether there are cheap parts in this computer, or whether Futureshop is trying to maintain market share in a highly-competitive market.

So based on those experiences, and the comparison to parts I know and trust, I assumed the Acer machine must have some sub par equipment (cheaper) inside.  Either that or they are clearing bulk stock.  I tried to find out what parts were in the machine, but neither your link, nor a google search for Acer AM5620-E5301A returned anything really helpful.  nVidia 8600GT is stated, but that is a given that it is nVidia - is it on board?  If not who makes it (eVGA, BFI, etc)?  What RAM is in the machine (Aper, Kingston, Cruicial)?  HD Manufaturer?  You get my point.

Also, I'd like to point out that the computer that failed is one with enthusiast-decided hand-picked researched parts, while my package deal Dell from eight years ago is letting me type out this message.

This is very subjective.  How many times did you bang, move or accidently kick either computer?  Spillage?  Environmental variables (which is closer to a heat vent)?  Technology differences (8 year old computer was not as tempermental towards heat changes - hell you could run a P1 machine without a heat sync for what seems like a year compared to CPUs from 2-3 years ago, and that is just one example of technology differences)?  Manufacturering differences (as Tom pointed out)?  So yeah, nice to compare, but not very fair IMO.

[Lazybones]

Options?

To me there is only two options most of the time, build it my self or get a Dell.

enthusiast-decided hand-picked researched parts

What was the enthusiasts goal when building the system? I have built many systems, both for my self and for clients. Sometimes the goal was speed sometimes it was price and sometimes it was just getting all the right features and leaving room for growth. The most risk comes when the goal is speed as it normally brings new untested parts into the system.

My current system has been running hardware fault free (after the initial build, had some DOA parts) for close two years, and in that time I have changed the HD, RAM, CPU and Video card with faster components as they became affordable.