April Computer Build Thread

[Murgos]

Thanks babe, you're the best.

[murdoc]

Darkness filled the room, shadows dancing across surfaces as the PS2's screen saver whirled in rhythmic circles.  On the couch slept a nubile young woman, a too large nightgown drapped over her curves.  As the controller slipped from her hand and bounced off the floor her lids bolted open.  Her eyes focused intently and she screamed, "Fuck off!  Write your own wank material!"  Then she drifted back to sleep, content.

Fin.

Holy HELL that was HOT.

[Lantyssa]

Thanks babe, you're the best.

[Engels]

The lack of a bundled front fan does not mean the case doesn't have front fan mounts. Read the specs. All those Antec cases have front fan mounts (some have 2).

Its not listed on the newegg specs, but the 'product tour' mentions it has two 92mm front fan mounts. Not quite sold on the smaller fans, but at least its an option.

[Engels]

Update!

I don't think I need a new case after all. I did a little experiment. Firstly, I swapped out the rear Antec 120 fan I had for a Silverstone variable speed fan I had rotting in the computer parts bin. Then I turned off the front 120 fan altogether, and the internal temperature dropped, the CPU temp dropped and the vid card temp dropped, all about an average of 7 degrees. That is with the rear fan set on a low setting. If I crank it up to be just a mild hum, my vid card keeps acceptable temperatures. There's no rattling, either, so I'm basically fixed.

What I think happened with the front fan as an intake is that the air was blowing over the 4 hard drives, heating up, which was being shunted to just below the 8800 card, which is so long that it was essentially creating a warm air bath that couldn't rise very easily. With the front fan off, the air from the hard drives rises by convection alone and then gets flushed out the back fan. It makes ense now, but I wouldn't have figured it out without the happy accident.

[Trippy]

Early GTX280 benchmarks:

http://www.theinquirer.net/gb/inquirer/news/2008/06/14/complete-gtx280-scores-here

[Engels]

Looks like a good improvement at high resolutions, moderate improvement at medium resolutions. They say too big and too hot and too slow, but I don't see any figures addressing that. Maybe I'm missing something.

[El Gallo]

OK, I built my last computer several years ago. It's time for an upgrade. But I'm old. Spending a little more money is worth saving the aggrivation of putting shit together and the risk of diagnosing and repairing things on my own.

Can anyone recommend a reliable computer retailer where I can select from a big range of good parts for my rig and have them put it together and repair at my house it if it breaks?

I'll build it myself again if the cost difference is huge or if my only choices are places where I won't have much choice about what goes into it. But the peace of mind is worth a few hundred bucks.

[Engels]

Dell gaming rigs are pretty decent, and of course you can get a bells and whistles warranty with them, but it ain't cheap. Monarch computers used to be good for highly customizable builds, but they went belly up, I think.

Falcon NW seems to have a slightly cheaper build than Dell:

http://www.falcon-nw.com/config/build.asp 2k for a system with a E8400, 8800GT, 4 gig of 800mhz ram and a 250 gig drive. Its not a bad system, although of course, a home build of the same would probably hit ~1.2k

[schild]

$1.2k? What are you talking about? $800-$900 easy.

[Engels]

$1.2k? What are you talking about? $800-$900 easy.

Eh, probably right, but I wasn't in the mood to look up all the independent parts.

[Cyrrex]

I just bought a pre-built machine that eats AoC like a tasty snack, and it only cost 700 dollars.  I could have done it up to a hundred bucks cheaper.  Will the whole thing melt within a year?  Possibly. 

The caveat is that I don't care much for bells and whistles (including lots of extra peripherals, drives, etc.).  I just want the raw horsepower.  Also, people are probably spending (wasting?) more money than they need to on the quad processors and absolute top-of-the-line vid cards.  The price differences can be huge, but the gains are often marginal.

[rattran]

Yeah, because all the video editing / photoshop / folding or whatever people do with their machines is a waste. I don't use my computer just for gaming.

Claiming 'raw horsepower' then slagging off quadcores and high end video is kinda 

[Cyrrex]

Sorry, didn't mean to slag off quad cores per se...what I meant is that, depending on how much multi-tasking you are really doing, a powerful dual core may be nearly as good as a more pricey quad.  For many things, the speed of the cores is more important than the sheer number of cores.

But certainly, it depends on what you use your machine for, and how many things are being done in parallel.  For me, it is primarily just gaming.

[Salamok]

For many things, the speed of the cores is more important than the sheer number of cores.

This is becoming less true with each new cpu release.  Intel is getting much, much better at getting their multicore procs to work in unison and present themselves to software/os as a single core cpu. 

Your statement was has a larger degree of truth with the Pentium D than it does with the core 2 duo and beyond.

[Cyrrex]

I agree.  The point is, you can still get some pretty fast dual core processors that are good enough for most uses, without paying a premium.  While I'd love a shiny new Q9700 (or whatever the latest is), fact is that one could get by nicely with far less.

[Trippy]

This is becoming less true with each new cpu release.  Intel is getting much, much better at getting their multicore procs to work in unison and present themselves to software/os as a single core cpu. 

Your statement was has a larger degree of truth with the Pentium D than it does with the core 2 duo and beyond.

Huh? No. There's no "getting their multicore procs to work in unison and present themselves to software/os as a single core cpu" going on here.

[Engels]

And uhm, Cyrrex, since it was asked, what company made your pre-built machine?

[Salamok]

This is becoming less true with each new cpu release.  Intel is getting much, much better at getting their multicore procs to work in unison and present themselves to software/os as a single core cpu. 

Your statement was has a larger degree of truth with the Pentium D than it does with the core 2 duo and beyond.

Huh? No. There's no "getting their multicore procs to work in unison and present themselves to software/os as a single core cpu" going on here.

lets rephrase then, the os is getting better at bouncing tasks between multi cores + having shared on die cache that can be allocated dynamically to each proc will allow a quad core to outperform the equivalent single or dual core chip on any given windows application that remotely taxes the cpu (even if it isn't specifically written for multiple processors).

My original statement was based on a hazy recollection of an article (Maximum PC maybe?) when the core 2 duo 1st came out and was benchmarked against the FX62.  For some reason I have a fuzzy memory of them wondering how intels core 2 duo would compete against the FX62 when the FX62 handled single threaded apps in a way that utterly spanked the pentium D's handling of the same apps.  The answer (obvious now) is that the core 2 duo handles it better.

Anyhoo just being able to offload windows' bloated services onto another core and present your app with a fully untaxed core (or more) for it's own use is going to improve performance.

[Trippy]

This is becoming less true with each new cpu release.  Intel is getting much, much better at getting their multicore procs to work in unison and present themselves to software/os as a single core cpu. 

Your statement was has a larger degree of truth with the Pentium D than it does with the core 2 duo and beyond.

Huh? No. There's no "getting their multicore procs to work in unison and present themselves to software/os as a single core cpu" going on here.

lets rephrase then, the os is getting better at bouncing tasks between multi cores + having shared on die cache that can be allocated dynamically to each proc will allow a quad core to outperform the equivalent single or dual core chip on any given windows application that remotely taxes the cpu (even if it isn't specifically written for multiple processors).

Still wrong.

[Salamok]

This is becoming less true with each new cpu release.  Intel is getting much, much better at getting their multicore procs to work in unison and present themselves to software/os as a single core cpu. 

Your statement was has a larger degree of truth with the Pentium D than it does with the core 2 duo and beyond.

Huh? No. There's no "getting their multicore procs to work in unison and present themselves to software/os as a single core cpu" going on here.

lets rephrase then, the os is getting better at bouncing tasks between multi cores + having shared on die cache that can be allocated dynamically to each proc will allow a quad core to outperform the equivalent single or dual core chip on any given windows application that remotely taxes the cpu (even if it isn't specifically written for multiple processors).

Still wrong.

show me a benchmark

[Miasma]

I just popped in to ask if anyone had heard what the new graphics cards are going to be like but saw that it was already asked/answered.  Holy crap that 200 series is a monstrosity, my case is big but I'm pretty sure it would still hit the drive bays, I'll have to open it up to remember.

With one gig of ram it would seriously cut into the 4 gigs my 32bit os can handle.  I'd basically have to commit to a 64bit os to make the graphic card worthwhile, crazy.  I won't be upgrading any time soon I think.

[Engels]

Er, I'm pretty sure that the ram on a graphics card is evaluated independently of your mainboard's ram and is used regardless of wether you've used up the 3~ gig the 32 OS can read. Trippy or someone else may correct me if I'm wrong.

[Grand Design]

Its the memory addresses that are important.  Regardless of how much RAM you have, your devices (GPU etc.) will claim their addresses before the OS loads and if you've hit the memory ceiling (meaning, run out of addresses) then that RAM is unavailable simply because the hardware cannot physically talk to it.

[Engels]

So, say I have a 1 gig vid card, and 3 gigs of ram, for a total of 4 gigs worth of 'address space'. We know that a 32 bit OS will only see ~3 gig, so what gets seen? I ask because in your explanation and using my theoretical example, either the OS would not use any dedicated vid ram, or only 2 gig of system ram and the 1 gig of vid ram. That doesn't seem to be true to my experience.

Now here's what I don't get, GD, you say that devices will claim address space BEFORE the OS loads, but it is the OS itself that's limiting the adresses it can use. Explain.

[Miasma]

A 32 bit os can see about four gigs total.  I have four gigs of system memory and a card with 640 megs, if I bring up task manager windows thinks I only have about 3.3 gigs of system RAM due to the video card and I imagine other memory.

[Engels]

Aha, turns out I was off mark, and you guys were pretty much right, with the caveat that Vista locks itself to a total of 3.12 ram, aparently.

http://support.microsoft.com/kb/929605

For example, if you have a video card that has 256 MB of onboard memory, that memory must be mapped within the first 4 GB of address space. If 4 GB of system memory is already installed, part of that address space must be reserved by the graphics memory mapping. Graphics memory mapping overwrites a part of the system memory. These conditions reduce the total amount of system memory that is available to the operating system.

The reduction in available system memory depends on the devices that are installed in the computer. However, to avoid potential driver compatibility issues, the 32-bit versions of Windows Vista limit the total available memory to 3.12 GB. See the "More information" section for information about potential driver compatibility issues.

If a computer has many installed devices, the available memory may be reduced to 3 GB or less. However, the maximum memory available in 32-bit versions of Windows Vista is typically 3.12 GB.

[Grand Design]

Now here's what I don't get, GD, you say that devices will claim address space BEFORE the OS loads, but it is the OS itself that's limiting the adresses it can use. Explain.

This is the common misconception.  Its not the OS that limits the amount of addressable memory, it is the physical limitations of 32 bit architecture.  The motherboard and chip are designed to be 32 bit, and therefore the OS is designed to 'talk' to those 32 bit components.  You cannot simply drop in a 64 bit OS and expect it to work - it will try to talk in its 64 bit words to components that can only understand 32 bit words.  So, Windows takes the blame, but Windows was designed to be 32 bit in order to talk to 32 bit hardware.

Each piece of memory must be addressable in order to be used.  2³² = 4,294,967,296  or 4 gigabytes, means that 4 gigabytes is all that can ever be 'seen' by hardware that can only talk in 32 bits.  Now, if you have a video card with 512 megabytes, each byte of that memory needs an address from the hardware in order to be used.  You are trading out the addresses that would be available to your system RAM in order to use the closer and therefore faster dedicated graphics card RAM. 

But you've just lost 512 MB!  Its not fair!  Gates screws us again!  Chances are you were not using that RAM anyway.  XP limits applications to a 2 gigabyte pool unless you tinker with its default settings. 

So, say I have a 1 gig vid card, and 3 gigs of ram, for a total of 4 gigs worth of 'address space'. We know that a 32 bit OS will only see ~3 gig, so what gets seen? I ask because in your explanation and using my theoretical example, either the OS would not use any dedicated vid ram, or only 2 gig of system ram and the 1 gig of vid ram. That doesn't seem to be true to my experience.

If you have 3 gigs of RAM and 1 gig on a video card, your OS hardware will be able to address every location in all of that RAM (theoretically, without getting into holes.)  Do not confuse memory with addresses.  Memory is a physical thing, addresses are pointers to their physical locations.  I guess an example would be your hands.  You are physically limited to pointing at ten things at one time.  You may have a hundred things to point at, but you only have the fingers to point at ten at a time.  Its a physical limitation with modern computers as well.

[rattran]

Or you simply install an x64 os,. drop in 8gb of memory and move on.

The GTX280 seems somewhat... weak at this point. About the same as a 9800gx2, without as much heat/power, for a $200 premium. Still, pushing the 9800gtx/+ to ~$200 is a good thing. We'll see if later drivers can improve on it.

I'm hoping ATI gets their shit together and actually has a decent competing product in the 4870.

[Engels]

Now here's what I don't get, GD, you say that devices will claim address space BEFORE the OS loads, but it is the OS itself that's limiting the adresses it can use. Explain.

This is the common misconception.  Its not the OS that limits the amount of addressable memory, it is the physical limitations of 32 bit architecture.  The motherboard and chip are designed to be 32 bit, and therefore the OS is designed to 'talk' to those 32 bit components.  You cannot simply drop in a 64 bit OS and expect it to work - it will try to talk in its 64 bit words to components that can only understand 32 bit words.  So, Windows takes the blame, but Windows was designed to be 32 bit in order to talk to 32 bit hardware.

Except these days most CPUs are 32 AND 64, so what you end up with is determined by your OS, even though the hardware can go either way, as Rattan points out.

[Grand Design]

You can see my knowledge of hardware is dated.  Looking at my chip, it can definitely run 64 and I'm in a 32 bit environment. 

[Trippy]

This is becoming less true with each new cpu release.  Intel is getting much, much better at getting their multicore procs to work in unison and present themselves to software/os as a single core cpu. 

Your statement was has a larger degree of truth with the Pentium D than it does with the core 2 duo and beyond.

Huh? No. There's no "getting their multicore procs to work in unison and present themselves to software/os as a single core cpu" going on here.

lets rephrase then, the os is getting better at bouncing tasks between multi cores + having shared on die cache that can be allocated dynamically to each proc will allow a quad core to outperform the equivalent single or dual core chip on any given windows application that remotely taxes the cpu (even if it isn't specifically written for multiple processors).

Still wrong.

show me a benchmark

You are the one making the claim, show me yours first. It's like you are claiming Santa Claus really exists and asking me to disprove you.

[Trippy]

Actually I'm not doing anything important at the moment so I'll be kind and show you where you are wrong rather than making your figure it out for yourself.

lets rephrase then, the os is getting better at bouncing tasks between multi cores + having shared on die cache that can be allocated dynamically to each proc will allow a quad core to outperform the equivalent single or dual core chip on any given windows application that remotely taxes the cpu (even if it isn't specifically written for multiple processors).

Though this is stated really badly your claim is that a Core 2 Quad will have a larger shared L2 cache compared to a single core processor or a Core 2 Duo and therefore for a single application it's possible they will have a more L2 cache to work will and therefore be faster (given the same clock speeds I'm assuming).

First, by definition these processors are no longer "equivalent" since one of them has more total L2 cache (in theory).

Second, and this is my main point, a Core 2 Quad is two Core 2 Duos smushed together and actually have two separate L2 caches, each one shared by 2 cores. E.g. a Q9550 with "12 MB L2 cache" is actually two separate 6 MB L2 caches. I.e. it's essentially two E8300s packaged together. A Q9300 (the low end 45nm C2Q) actually only has two 3 MB L2 caches and is in effect gimpier in terms of L2 cache than many of the older Core 2 Duos which have a (shared) 4 MB L2 cache.

But let's assume at some point in the future a Core 2 Quad actually has a real shared global L2 cache that all 4 cores can share. Does that mean you'll get better performance in non-synthetic benchmarks? Maybe. There are very few real-world apps these days that can fit entirely inside a L2 cache. The more memory an app needs beyond what can fit inside the L2 cache the less benefit a larger L2 cache is. Your typical high-end 3D MMORPG will take ~1 GB RAM these days. Though how that 1 GB of RAM is accessed is not uniform, boosting your L2 from, say 6 MB (a current Core 2 Duo) to 12 MB (a theoretical C2Q with a single shared L2 cache) is not going to make a big difference in your L2 cache miss rate as the extra 6 MB of L2 cache represents 0.6% of the total RAM needed.

If a programmer wanted to he or she could spend a lot of time optimizing the code to take advantage of that extra cache (assuming the OS doesn't yank it away from the program) which could give a nice speed boost but that's something each game would have to be specifically tailored for and for the PC is not worth the effort since there are so many different processors that the game can be run on. For consoles I'm sure the AAA titles spend a lot of effort on optimizing this sort of thing since the hardware is fixed.

If you are running some synthetic benchmarks and with 12 MB of L2 cache you can fit the entire thing in there while with only, say, 6 MB of L2 you can only fit half the benchmark, yes you'll see a huge improvement in the benchmark numbers. But that sort of thing doesn't really reflect how real-world apps perform.

Edit: that first point wasn't necessary

[Murgos]

If a programmer wanted to he or she could spend a lot of time optimizing the code to take advantage of that extra cache (assuming the OS doesn't yank it away from the program)

God that would suck.  And also be highly processor dependent.  Just figuring out the scheme would take quite a long time.  You would basically have to divide up your program into 16 or 24 or whatever (based on the set associativity of the cache) discrete memory locations populated in a very specific pattern.  You're pretty much talking about hand placing several million instructions.

Anyway, there have been lots of studies that show there is a diminishing return on cache sizes and associativity so even if you did have some MONSTER cache it's not really going to end up that much faster than a slightly more limited cache.

[Dtrain]

ATi Radeon 4850 is coming out next week @ $200, which is prompting NVIDIA to drop the price of the 9800 GTX to $200. The two cards side by side are more or less even in terms of performance, with a slight nod to the 4850. NVIDIA still retains overall performance advantage with multiple cards, due to uneven performance scaling results from Crossfire. However, compared to the 8800 GT in SLI, the 9800 GTX SLI does not have significant performance gains in most games.

The NVIDIA 9800 GTX+ is slated for a mid-July delivery @ a $230 price point for ~ a 9% performance increase.

Cliff notes: If you're buying 1 card in the $200 range, buy a 4850. If you're buying 2 (or more) $200 cards go with 8800 GT's for $150 each.

Hard mode: Wait for the 9800 GTX+ to come out to see how it effects the pricing on the rest of NVIDIA's product line.