What would be a good upgrade for this PC?

[Chalenge]

Upgrading from an 8800GT to a 9800GTX is a waste of money.

[Ghastly]

Are you sure of this?  I'm running 32bit XP with a 1GB card on 3.25GB RAM (4GB installed) and all is available.  Nothing changes on my RAM between my old 256MB card and my 1GB card.

I think you may be thinking of video cards that use shared memory.

Pretty darn sure... that's been my experience, and that of others on the board to date.   With the advent of PCI-E, there is no longer a video memory aperture, it's direct mapped into the address space.

<S>

[Masherbrum]

Upgrading from an 8800GT to a 9800GTX is a waste of money.

Not according to someone who builds PC's for a living.   You go ahead and figure out who that "someone" is.   

[1701E]

[. . .]
Are you sure of this?  I'm running 32bit XP with a 1GB card on 3.25GB RAM (4GB installed) and all is available.  Nothing changes on my RAM between my old 256MB card and my 1GB card.

I think you may be thinking of video cards that use shared memory.

Windows 32-bit only can access ~3.25Gb of RAM total throughout the system.  So if you have 2Gb of System RAM and a GPU with 1Gb VRAM that's 3Gb of the RAM it can access.  However, Onboard GPUs with shared RAM actually "steal" System RAM to run, so in essence it'd be like having 1.5Gb of System RAM and a 512MB GPU.  If the System is accessing the 1GB GPU that's 2.25Gb (assuming there is 4GB) of System RAM it can use where it needs (background tasks, game, etc.) and any 1 program, I think, can only access 2Gb at a time.  So if the background tasks need 1Gb of System RAM that leaves 1.25Gb of RAM for the game to run off of.


Least that's what I've gathered from listening to Skuzz, TD, and the others around here who know way more then I ever will.
I'm sure I'm wrong somewhere (or all of it!), but hopefully Skuzz/TD or someone can correct it.

[Skuzzy]

I was trying to stay out of this thread.

32 bit Windows OS's can address 4GB of RAM.  However, all memory addressable buffers (sound card, video card, ethernet....) are mapped starting at 4GB and then it goes down from there.  If you have a 256MB video card, then instantly subtract 256MB from 4GB to find the highest amount of memory the OS will be able to use for applications.

If you have a 1GB video card, subtract 1GB from 4GB (regardless of the physical amount of RAM in the computer), which yeilds 3GB.  This means 3GB will be the most system RAM available to the OS for applications.

The reason?  When all 32bits of a binary number are set, it is equal to 4GB (4294967296-1).  This is the largest directly addressable number for a 32 bit system.  However, if Microsoft would properly implement PAE (physical address extensions), then a 32 bit OS could address 64GB of RAM.  UNIX OS's have been doing it for years.

[1701E]

Knew someone smart would correct me soon enough.

[Tigger29]

I was trying to stay out of this thread.

32 bit Windows OS's can address 4GB of RAM.  However, all memory addressable buffers (sound card, video card, ethernet....) are mapped starting at 4GB and then it goes down from there.  If you have a 256MB video card, then instantly subtract 256MB from 4GB to find the highest amount of memory the OS will be able to use for applications.

If you have a 1GB video card, subtract 1GB from 4GB (regardless of the physical amount of RAM in the computer), which yeilds 3GB.  This means 3GB will be the most system RAM available to the OS for applications.

The reason?  When all 32bits of a binary number are set, it is equal to 4GB (4294967296-1).  This is the largest directly addressable number for a 32 bit system.  However, if Microsoft would properly implement PAE (physical address extensions), then a 32 bit OS could address 64GB of RAM.  UNIX OS's have been doing it for years.

See Rule #2 

[Ghastly]

...{snip}...  However, if Microsoft would properly implement ...{snip}... UNIX OS's have been doing it for years.

Skuzzy, if you start in on the things that Unix OS's were doing correctly before the PC even existed that Windows still doesn't do correctly today, you'd need more than 32 bits just to assign each bullet a number...

<S>

[BaldEagl]

Ripley's right on the RAM speed.  It's a waste of money so you can say you've got "fast" ram.  With his CPU stock the most he'll get out of his RAM is 667 regardless what speed the RAM is capable of.  Even if he OC's the CPU it's doubtful he'd exceed 800Mhz RAM.

Simple math:  dual core double pumped CPU on a 1333 FSB = 1333/4 = 333.333 clock cycle.
DDR2 RAM processing requests from the CPU = 667 (333.333 clock cycle x 2)

[Ghastly]

Ripley's right on the RAM speed.  It's a waste of money so you can say you've got "fast" ram.  With his CPU stock the most he'll get out of his RAM is 667 regardless what speed the RAM is capable of.  Even if he OC's the CPU it's doubtful he'd exceed 800Mhz RAM.

Simple math:  dual core double pumped CPU on a 1333 FSB = 1333/4 = 333.333 clock cycle.
DDR2 RAM processing requests from the CPU = 667 (333.333 clock cycle x 2)

{small voice}

What BE meant to say was quad pumped CPU (# of cores doesn't matter - or at least, I've never read so)  on a 1333 FSB = 1333/4 = 333.333 clock cycle

Yes, I know,there's one in every crowd ...

P.S. I just realized that my post could be taken as though I'm supporting the argument against faster RAM.  I meant it neither for or against, simply a clarification of the calculation.

PPS. OK, I looked at the thread in depth, and now I'm really confused,  I just shouldn't be looking at this stuff at 3 in the morning when I can't sleep.  I have to be missing something stupid, because I don't get the arguments against (potentially) upgrading RAM. His current memory is limited to 6.4 GB/s max.  With no processor overclock the system clock is set to 333, and the memory clock for DDR2 800 is 200, so the motherboard is using a 6:5 divider.  So how wouldn't DDR2 1066 at 8.5 GB/s max be a better solution, regardless of the 8:5 divider (if I looked it up right)?

Assuming reasonable timings on the DDR2 1066, why are you guys saying that faster RAM would be "worthless"?  It might not be cost effective, but I don't see where the argument that

"DDR2 however has only a multiplier of two (2) so sync ram speed is actually 666Mhz".

or 

With his CPU stock the most he'll get out of his RAM is 667 regardless what speed the RAM is capable of

apply.

As far as I know, assuming no OC on either processor or RAM, that's not going to happen until he gets memory that has to be clocked faster than the processor clock (which is 333 MHZ without any OC, and would correspond to something like DDR2 1333 (actually, something beyond that since DDR2 1333 - which AFAIK doesn't exist - would be where you'd get to a 1:1 divider, or "synced" memory ).  DDR2 1066 is still only clocked at 266 Mhz.  And would be as much as 1/3 faster on memory intensive operations.

<S>

[Chalenge]

Not according to someone who builds PC's for a living.   You go ahead and figure out who that "someone" is.   

It doesnt matter he is wrong. The 9800 GTX has a texture fill rate of 43.2 and the 8800 GT of 39.2 and that difference (10%) isnt worth an upgrade. The GTX 280s are pushing 48.2 and the 285s 51.8 which is much better for the money (25% performance increase).

I knew this of course because I own a ton of Geforce cards and yes I have compared them. The 8800s are all you need for AH. The 9800s are good if you need a card and dont want to spend a lot of money but they are virtually the same as 8800s. In a game like ROF or WOP the 280s will get you to top performance and the only time anyone would need the latest 400 series is for something like FSX. 285s texture fill rate actually beats the 480s on paper but 480s have four times the raster engines effectively quadrupling the power.

[MrRiplEy[H]]

{small voice}

What BE meant to say was quad pumped CPU (# of cores doesn't matter - or at least, I've never read so)  on a 1333 FSB = 1333/4 = 333.333 clock cycle

Yes, I know,there's one in every crowd ...

P.S. I just realized that my post could be taken as though I'm supporting the argument against faster RAM.  I meant it neither for or against, simply a clarification of the calculation.

PPS. OK, I looked at the thread in depth, and now I'm really confused,  I just shouldn't be looking at this stuff at 3 in the morning when I can't sleep.  I have to be missing something stupid, because I don't get the arguments against (potentially) upgrading RAM. His current memory is limited to 6.4 GB/s max.  With no processor overclock the system clock is set to 333, and the memory clock for DDR2 800 is 200, so the motherboard is using a 6:5 divider.  So how wouldn't DDR2 1066 at 8.5 GB/s max be a better solution, regardless of the 8:5 divider (if I looked it up right)?

Assuming reasonable timings on the DDR2 1066, why are you guys saying that faster RAM would be "worthless"?  It might not be cost effective, but I don't see where the argument that  or   apply.

As far as I know, assuming no OC on either processor or RAM, that's not going to happen until he gets memory that has to be clocked faster than the processor clock (which is 333 MHZ without any OC, and would correspond to something like DDR2 1333 (actually, something beyond that since DDR2 1333 - which AFAIK doesn't exist - would be where you'd get to a 1:1 divider, or "synced" memory ).  DDR2 1066 is still only clocked at 266 Mhz.  And would be as much as 1/3 faster on memory intensive operations.

<S>

Actually most Intel chipset/bios combos automatically adjust the memory multiplier to 2.4 etc. utilizing the 'faster' ram. However benchmarking has proven over and over that running the memory asynch actually carries a performance penalty that eats away increments of the speed boost. That combined with the fact that the CPU can only utilize the bandwith it's FSB asks for make investing to a 'faster' ram an act of futility. Very, very poor way to spend your money and downright wrong to suggest this to someone who is looking for an _upgrade_. That 80 bucks combined with selling the old card on ebay would constitute the price of a next-gen GPU and true speed increase for example.

In car terms 'faster ram' is equivalent to changing your air filter to a free-flow model for the price of a ECU chip or a supercharger. Gee, I wonder which one I would take. Filter or a supercharger?

[Ghastly]

OMG, you are absolutely right...

http://www.tomshardware.com/reviews/ram-speed-tests,1807-14.html (Note that even the synthetic memory tests shows what's really only a marginal difference between fastest and slowest in the testing.)

I've still got some sort of block thing going though, because I still don't ken the how and why.  I'm missing a piece of the picture or getting senile.   And I *thought* I knew this... I need to dig out the timing charts (the ones with the clock cycle diagrams) because I'm obviously making an erronous assumption somewhere.

<S>

[MrRiplEy[H]]

OMG, you are absolutely right...

http://www.tomshardware.com/reviews/ram-speed-tests,1807-14.html (Note that even the synthetic memory tests shows what's really only a marginal difference between fastest and slowest in the testing.)

I've still got some sort of block thing going though, because I still don't ken the how and why.  I'm missing a piece of the picture or getting senile.   And I *thought* I knew this... I need to dig out the timing charts (the ones with the clock cycle diagrams) because I'm obviously making an erronous assumption somewhere.

<S>

You're not TOTALLY in the woods though because if you raise both CPU FSB and memory speed you get considerable speed gains. But that falls into overclocking territory again.

<S>

[SectorNine50]

Heh first of all you're so off base it's not even funny. You're confusing the Intel "quad pumped" 333mhz fsb rating with your DDR2 rating.

Yes, the E8400 actually runs at 333Mhz bus speed. After multiplier of 4 it becomes 1333. DDR2 however has only a multiplier of two (2) so sync ram speed is actually 666Mhz.

I would never in a million years pay the price of 4gb ram sticks to get 200mhz bus speed change as a result. That's just retarded folks. He will get perhaps 1 fps gain (IF he's lucky) and pay 80 bucks or more.

You totally misunderstood what I was saying.

I'm saying, that right now, he's at 1333/4=333.25Mhz FSB, and his RAM is at 800/2=400Mhz.  He's already A-Sync, to sync he'd have to downclock his RAM to 666.4 (333.25 pre-multiplier).  I'm very familiar with bus and RAM speeds, there is no reasons to belittle me.  I wasn't aware of the tests on Tom's Hardware, but I had always personally gone for sync'd speeds just because I thought it made sense.

Basically what I was saying was that because he's already A-Sync, going to higher speed RAM would be a technical upgrade.  I never once told him he should do it.