StarDestroyer.Net BBS

Hardware

Overclocking

Simple Version:: Don't do it. Just don't. Don't Do it in any form!

Complex Version:: Other than the obvious degradation of the computer's life due to increased thermal output, you greatly increase the probability of data errors creeping in and compromising the reliability of your data.

MSDN Article on this

Intel vs AMD

Simple Version: Right now, Intel has the performance crown, while AMD has the value crown.

Power Supplies

Simple Version: Get a power supply from a known big name manufacturer like Asus, Corsair, etc.

Avoid no name "clone" manufacturers like the plague.

Add in a little bit extra power than what you need -- e.g. if your computer needs 400W, put in a 500W power supply.

Your computer does not need a 800 to 1,000 W PSU. Period.

Complex Version: The power supply is the one component in your computer, if it fails, can take down just about everything, from the motherboard to your hard drives.

So don't skimp on it.

The reason for adding a little extra power capacity (100W~) over what you need are:

1.) To provide a buffer for possible future upgrades -- you might want to add more internal hard drives in the future, or upgrade to a video card that draws more power than your older one.

2.) To provide an extra margin of durability -- PSUs designed for 500W when they pull only 400W, run cooler, and because they're built for a hotter thermal environment, last longer. This ties into the "if this fails it can kill everything else." bit about PSUs.

Motherboards

Simple Version: Buy from quality brands like:

Asus
Gigabyte
MSI
Intel
Supermicro (Good for workstations/servers)
Tyan (Good for workstations/servers)

Avoid deals that are "too good to be true".

Additionally, check the specifications of the motherboard carefully.

A specific model of motherboard with two SATA connectors may be fine for a simple media center (with a single optical drive and hard drive); but would not be acceptable for someone who uses more than one hard drive for storage.

How much memory can the motherboard handle? You want to be a little ahead of the curve in regards to total memory capacity, so you don't have to replace your board that often.

Complex Version: If the power supply is the heart of the computer, the motherboard is the brain. If you buy from a quality manufacturer (see previous list), you should have no problems, provided you do not overclock.

The reason for paying close attention to the boards' specifications in your initial buy is while there are Expansion Cards which can add extra SATA connectors/controllers to the computer; they generally are not as reliable as a built in motherboard connector/controller.

Memory

Simple Version: Buy from known major manufacturers like Kingston, Corsair, or Crucial. Buy memory within the same spec as the original memory; e.g. PC3200, if you are adding more memory to an existing installation.

Performance RAM isn't worth it, period.

Complex Version: Not much to add here really. Oh, and don't overclock.

Graphics Cards

Simple Version: Pick a card with a chipset from ATI or NVIDIA; and from a known manufacturer like BFG, etc.

Complex Version: Picking a card from a major chipset designer and then from a major manufacturer ensues that driver support and overall quality will be decent.

This is especially important if the hardware is an integrated graphics system -- avoid no name manufacturers or chipsets like the plague. Intel, ATI and NVIDIA are the ones to get in the integrated market.

Cases: Get a case from a major manufacturer like Antec. They tend to be put together a lot more sturdily than the cheap clone cases; and are much, much easier to assemble; tend to come with special thumb screws which sure beat using a screwdriver!


Example of Thumb Screw

Computer Assembly

If you're doing major computer assembly work; e.g putting together an entire system from the boxes they came in; wear an ESD Wrist Strap.

While computer parts are surprisingly durable (for example, I've had cards which have sat unprotected against other cards for nearly a decade boot up and run with no problems; but those are exceptions to the rule); you want to be sure that you haven't blown your investment in computer parts by shorting out a key component.

While you can RMA those components; it's annoying to wait for a new part to come in.

Minor assembly work; e.g. adding a new hard drive, or changing the memory in an existing system can be done safely by regularly touching unpainted metal parts of the case to ground yourself and dissipate any possible charge.

Do assembly in a relatively clean room or area. Do not do it in carpeted areas; and if you have any pets in the house, lock them out of the assembly area until the computer is assembled and sealed up.

Example: Cat comes up and rubs against you, while meowing; transferring a static charge from her fur to you.

Software

Picking the right OS

Traditionally, the rule of thumb has been to wait six months or at least one service pack release before buying a Microsoft OS. This was born in the very bad old days of Windows 95; and then reinforced by the bad release of Vista into an environment where a lot of hardware drivers weren't ready for it.

Nowdays, Windows 7 is actually quite nice, stable, and very compatible.

So the problem becomes one of picking the right version.

Essentially, Windows 7 Home Premium is all you really need if you're the average and slightly above average user.

If you handle some rather obscure software which needs virtualization via Virtual PC; Windows 7 Professional works quite nicely. As a bonus, it can handle 192 GB of memory; which is quite a bit more than Home Premium's 16 GB; but I think by the time we need more than 16 GB, we'll be near Windows 8's release.

Ultimate is just a waste of your money.

Advantages of going from x86 to x64 binaries

Simple Version:: If your programs need more than 2GB per process or 4GB total, 64-bit is good for you --phong

Complex Version:: Generally, the advantages from going from 32-bit to 64-bit are very small for most applications (5-10% performance improvement); and in many cases, there is actually a decrease in performance.

A good rule of thumb would be to look at the kind of datasets that are handled by the application/game.

An emulator for the NES for example, would not benefit much (if any) from an upgrade to 64-bitness due to the fact that there's only 2 to 32 kb of memory in the system being emulated.

By contrast, a graphics editing application that routinely handles very large images -- a 5000 x 5000 pixel bitmap at 24 bit consumes about 72 megabytes of information would benefit from 64-bitness, due to the fact that many graphics applications save "undo" steps so you can go back to a previous version of the image, and if you have several large images open at once, this causes huge memory requirements.

----------------

This will do as a start; some other topics I have in mind are "you only need one antivirus program"; "personal firewalls are unneccessary, windows firewall works", and how to prevent browser hyjacks via the "map the whole thing as a bitmap so that no matter what button you push, the spyware installs."; and why to watch Windows Update; because sometimes they like to try and push automatic software installations along with security updates (I am looking at you IE8 and probably IE9 a few years from now).

Stark can probably add some more pointers from his time as IT.

I think cases with clip assembly are even better than thumbscrews; they can be a pain in the ass to remove anything, but you don't have loose bits. Sadly they're Miami Vice colour coded, generally.

Shep, I think it'd be worth editing in some kind of statement around 'don't fill your computer with useless crap like 5 spyware scanners or Norton', since people around here seem to be paranoid and feel that more apps = safer. It'd save someone posting 'the Windows firewall is all most users need' every time, too.

Let's see, there's also some tips to avoid Email drivebys; by suggesting users stay away from Outlook. I honestly Have not been following the email application wars; has Microsoft finally fixed Outlook's crippling vunerabilityies to just about everything, or should we still use Eudora or Thunderbird?

Well it's a good rule of thumb to not open, click links within or forward unknown, unsolicited or spam emails to anyone, regardless of what client you use.

MKSheppard wrote:Other than the obvious degradation of the computer's life due to increased thermal output, you greatly increase the probability of data errors creeping in and compromising the reliability of your data.

Overclocking is harmless when a processor has been de-rated from its actual bin for sales (market segmentation) purposes, rather than being sold at its actual max capacity (including tolerance). However, determining when this is the case is difficult. Highly overclocked systems can be as reliable as a normal system, if they have monster cooling systems - but validating this requires proper testing (e.g. a solid week of multi-threaded prime number generation with no data errors). Processor life will still be shortened if the voltage is increased, but most hobbyists change computers far too often for this to be an issue.

Of course you should not be overclocking if you need a 'cheat sheet' on basic PC facts.

Your computer does not need a 800 to 1,000 W PSU. Period.

A high-end factory-overclocked GPU will consume between 200 and 300 watts. If you have two of those plus an overclocked CPU and a bank of RAID drives, you can exceed 800 watts power draw. Again though, the few people insane enough to do that shouldn't need a cheat sheet.

Complex Version: If the power supply is the heart of the computer, the motherboard is the brain.

No, that would be the CPU. The motherboard is more like the spinal cord.

Performance RAM isn't worth it, period. Complex Version: Not much to add here really. Oh, and don't overclock.

The whole point of performance RAM is so you can overclock it. This can actually work fine if you know what you're doing; some RAM and motherboard manufacturers do fully validate, just for specs not endorsed by JEDEC (e.g. DDR-500 was never an official standard, but it was supported by several high-end vendors).

If you're doing major computer assembly work; e.g putting together an entire system from the boxes they came in; wear an ESD Wrist Strap.

I would say that you shouldn't be doing 'major computer assembly' work if that list contains anything you didn't already know (or know better than). If you can't be bothered taking a course on computer assembly or at least spending a month or two educating yourself from Internet articles (from reputable sites), just pay the extra $30 or whatever to have the local computer shop build it for you.

Complex Version:: Generally, the advantages from going from 32-bit to 64-bit are very small for most applications (5-10% performance improvement); and in many cases, there is actually a decrease in performance.

You should do it anyway. 32-bit is a dead end; it means there's a much larger chance of having to purchase a new OS during the lifetime of the PC.

An emulator for the NES for example, would not benefit much (if any) from an upgrade to 64-bitness due to the fact that there's only 2 to 32 kb of memory in the system being emulated.

64-bit is not just address space, it is also double the number of registers, and that will accelerate quite a lot of programs. It also allows you to do 64-bit integer operations without spanning two registers and messing about with carries, but very few programs make heavy use of 64-bit integer math (although some of the programs I run do). The NES emulator example is idiotic since flawless full-speed NES emulation has been possible since about 1992.

MKSheppard wrote: Add in a little bit extra power than what you need -- e.g. if your computer needs 400W, put in a 500W power supply.

Your computer does not need a 800 to 1,000 W PSU. Period.

Christ, it feels like it was only yesterday that if you got up to a 500W it was a big deal. Hell, my old computer was cruising on a 400 despite running four hard drives and an 8800. Big surprise to find out one day when I went to replace the 350 it'd been running on and find you couldn't get them anymore.

350W power supplies are still made by reputable manufacturers.

What is the value of extended warranties, in-store replacement programs, et al?

I think that depends on their terms and how much the individual values the support. I generally encourage people buying laptops to buy quality support programs, just because they're generally used for work and I don't want to provide free tech support. I wouldn't waste my time with a retailer warranty.

So, just manufacturer's warranties then. What about length of warranties? Most of what i've heard is that if a component needs to be RMA'd, it usually happens within the first few months.

Starglider wrote:I would say that you shouldn't be doing 'major computer assembly' work if that list contains anything you didn't already know (or know better than). If you can't be bothered taking a course on computer assembly or at least spending a month or two educating yourself from Internet articles (from reputable sites), just pay the extra $30 or whatever to have the local computer shop build it for you.

Who the hell needs a course on computer assembly or months of education? Swapping parts like GPU/RAM/harddisc or assembling a computer isn't a foreign language. If you don't feel comfortable to assemble a computer from scratch, why even waste money/time on courses/self-education instead of paying for assembly?

MKSheppard wrote:Let's see, there's also some tips to avoid Email drivebys; by suggesting users stay away from Outlook. I honestly Have not been following the email application wars; has Microsoft finally fixed Outlook's crippling vulnerabilities to just about everything, or should we still use Eudora or Thunderbird?

Outlook is fine.

Starglider wrote:Overclocking is harmless when a processor has been de-rated from its actual bin for sales (market segmentation) purposes, rather than being sold at its actual max capacity (including tolerance). However, determining when this is the case is difficult. Highly overclocked systems can be as reliable as a normal system, if they have monster cooling systems - but validating this requires proper testing (e.g. a solid week of multi-threaded prime number generation with no data errors). Processor life will still be shortened if the voltage is increased, but most hobbyists change computers far too often for this to be an issue.

Intel Burn Test is the current "gold standard" tool (it's essentially LINPACK and tuned to stress all aspects of a CPU hard) but even passing that is no guarantee.

64-bit is not just address space, it is also double the number of registers, and that will accelerate quite a lot of programs. It also allows you to do 64-bit integer operations without spanning two registers and messing about with carries, but very few programs make heavy use of 64-bit integer math (although some of the programs I run do). The NES emulator example is idiotic since flawless full-speed NES emulation has been possible since about 1992.

64-bit architecture also increases cache pressure since pointers are twice as large (and depending on the compiler, so might data words) so the extra register advantage might not hold.

Stark wrote:I think that depends on their terms and how much the individual values the support. I generally encourage people buying laptops to buy quality support programs, just because they're generally used for work and I don't want to provide free tech support. I wouldn't waste my time with a retailer warranty.

This is pretty good advice; I usually suggest an accidental-damage warranty for laptops since they're jostled around quite a bit. Most laptops also lack a proper load-bearing frame which adds further stress to key components inside as the shell gets twisted a bit.

Artemas wrote:So, just manufacturer's warranties then. What about length of warranties? Most of what i've heard is that if a component needs to be RMA'd, it usually happens within the first few months.

For desktops the bathtub curve tends to hold well but laptops are a crapshoot due to a much harder operating environment.

Intel Burn Test is the current "gold standard" tool (it's essentially LINPACK and tuned to stress all aspects of a CPU hard) but even passing that is no guarantee.

No consumer grade hardware offers a 'guarantee' of data integrity. If you want that, you will have to use ECC, RAID and other more esoteric enterprise technologies.

64-bit architecture also increases cache pressure since pointers are twice as large (and depending on the compiler, so might data words)

Not always. For example the Sun Java VM has a 'packed pointer' mode, which exploits the fact that all object pointers are 8-byte aligned. Pointers are stored in 32-bit fields and shifted up 3 bits when deferenced, giving an address space of 32 gigabytes. The cost of doing the shift is normally negligible compared to the rest of the memory access, so this is a good choice for everything except massive server applications. There are lots of other pointer packing tricks you can do manually in C (or Java if you are insane like me), though of course it only makes sense if the application spends a lot of time in tight loops chasing pointers.

so the extra register advantage might not hold.

The register advantage is independent of the cache pressure issue. More registers can dramatically speed up function calls at or near the leaf of the call tree, because x86 is so register starved it wastes a significant amount of time thrashing stack (at least, if you program the way the OO/pattern weenies want you to, with loads of tiny functions).

Starglider wrote:

Intel Burn Test is the current "gold standard" tool (it's essentially LINPACK and tuned to stress all aspects of a CPU hard) but even passing that is no guarantee.

No consumer grade hardware offers a 'guarantee' of data integrity. If you want that, you will have to use ECC, RAID and other more esoteric enterprise technologies.

There is some reasonable expectation that consumer technology won't spontaneously do crazy things, though. I know that the more guarantees you want the more it'll cost (ECC, disk checksum, chipkill, etc.).

Not always. For example the Sun Java VM has a 'packed pointer' mode, which exploits the fact that all object pointers are 8-byte aligned. Pointers are stored in 32-bit fields and shifted up 3 bits when deferenced, giving an address space of 32 gigabytes. The cost of doing the shift is normally negligible compared to the rest of the memory access, so this is a good choice for everything except massive server applications. There are lots of other pointer packing tricks you can do manually in C (or Java if you are insane like me), though of course it only makes sense if the application spends a lot of time in tight loops chasing pointers.

I know there are all sorts of interesting tricks done but how often are they done, especially in code that doesn't live in some sort of VM?

The register advantage is independent of the cache pressure issue. More registers can dramatically speed up function calls at or near the leaf of the call tree, because x86 is so register starved it wastes a significant amount of time thrashing stack (at least, if you program the way the OO/pattern weenies want you to, with loads of tiny functions).

I know the advantage is independent (and long overdue) but the cache issue still comes to play when looking at overall system performance.

charlemagne wrote:

Starglider wrote:I would say that you shouldn't be doing 'major computer assembly' work if that list contains anything you didn't already know (or know better than). If you can't be bothered taking a course on computer assembly or at least spending a month or two educating yourself from Internet articles (from reputable sites), just pay the extra $30 or whatever to have the local computer shop build it for you.

Who the hell needs a course on computer assembly or months of education? Swapping parts like GPU/RAM/harddisc or assembling a computer isn't a foreign language. If you don't feel comfortable to assemble a computer from scratch, why even waste money/time on courses/self-education instead of paying for assembly?

Seriously. One of my earliest jobs involved hand assembling motherboards, and the amount of training they gave us was a couple of days. It's not that hard to learn how to put together a system.

As far as avoiding outlook, I'm going to wager that nobody offering that suggestion has worked in an environment where using it was mandatory.

Graphic card rule of thumb: ATI is for value, NVidia is for performance.... although lately that dynamic has been changing to more like ATI for sheer performance and value, NVidia for less heat and power usage and more driver reliability and support.

Personally, I used to use strictly NVidia cards until my latest computer, which I went ATI, and I don't think I'll be looking back.

Shinova wrote:Graphic card rule of thumb: ATI is for value, NVidia is for performance.... although lately that dynamic has been changing to more like ATI for sheer performance and value, NVidia for less heat and power usage and more driver reliability and support.

Those rules of thumb are not accurate; ATI's latest GPU runs cooler than nVidia's and ATI's drivers do not cause Microsoft to have to limit maximum memory size in 32-bit Windows.

ATI has been consistently better in its compliance with standards in the DirectX 10+ generations of hardware. From having Vista drivers that were stable, DX10 supporting (if slower then XP drivers) at Vista's release while nVidia took months after release to properly support DX10, to supporting 10.1 which nVidia passed over. It has a definite lead with DX11 cards do to that, since DX11 is a further evolution of DX10.1. nVidia still hasn't released a DX11 part and won't until next spring, and who knows what the maturity of their drivers will be - their history in the DX10 era certainly doesn't inspire confidence.

All in all, AMD/ATi is generally the smart choice to buy if you aren't stuck on XP.

I still remember the bad old days where ATI simply had absolute horrible driver support back in 2000ish or so. It's sure taken them a while to erase that legacy.

MKSheppard wrote:I still remember the bad old days where ATI simply had absolute horrible driver support back in 2000ish or so. It's sure taken them a while to erase that legacy.

Yep. In the 1990s ATI was predominantly an OEM manufacturer and their drivers were properly tested and optimized for 2D compatibility and speed only until the first generation (DX7) Radeons. Even after that it took them couple of years to reach NVidia. The DX9 chipsets were the first ones which had pretty good drivers from early on. At least they finally got them working, unlike S3 which spent a large sum of money on trying to catch up with Nvidia, but never had properly working drivers for the Savage chipsets and was ultimately sold to VIA. If you didn't know, in the mid-1990s S3 was the largest graphics chip company in the world.

About overclocking: there's little reason to do it nowadays, since processors are so cheap. Back in the day they were expensive and what is still probably the most common overclocked processor ever, the Celeron 300A to 450 MHz, gave you a huge performance improvement and saved you a lot of money compared to a Pentium 2/III 450. In some cases it was faster, too, thanks to the on-chip L2 cache (all 128 KiB of it, inclusive...) running at core frequency. In those days you could also immediately notice such performance increases (and even much smaller ones) in everything you did with your PC.

My first overclock was a Pentium 90 to 133. It worked just fine, no noticeably instability1 for more than 9 months under heavy loads and then died without a warning. Of course back then hardware temperature monitoring was a virtually unknown concept on desktop computers. I bought a real Pentium 133 to replace it and after that I did not overclock my own systems. Still, I did have many friends who used the overclocked Celeron 300A and none of them ever had any problems with it.

1 Other than what was normal for Windows 95, and it was stable in Linux and OS/2 as well.

Generally, on overclocking, it can be summed up as "No cost saving rational reason to use it, but if you know what you're doing it can be quite fun."

That's how I treat most of my OC efforts, which are usually on older PCs that do not touch anything important.
But I'd also not count myself as anyone who needs "Computer rules of the thumb"

I just want to say that I find this thread both useful and handy. If I knew more about power supplies, I could have avoided a lot of pain.

Zixinus wrote:I just want to say that I find this thread both useful and handy. If I knew more about power supplies, I could have avoided a lot of pain.

The power supply situation is kind of funny and the current situation resembles more the world of fashion than technology. A few years back many people found out that they had insufficient power supplies (mostly because they were using older ATX12V 1.x and even ATX 2.x spec PSUs with motherboards that draw processor current from the 12 V rail), which created the myth that you absolutely need a 600+ Watts PSU, if you want to do anything else than surf the net. In reality most people could manage with a 400-500 W PSU just fine and for a non-gaming system even a 300-400 W PSU is usually more than sufficient.

The industry responded also with a big branding effort, which in fact is mostly just a re-badging effort. In reality there are only a few PSU manufacturers and most of the brand name PSUs are either straightforward re-branded OEM units from those big manufacturers, or in some cases built to ordered specs. The main players are SeaSonic and FSP (Fortron/Source). Enhance seems to be popular with many of the (re-)branders. I usually buy a Fortron branded PSU if I can find one; they are very reliable and have an excellent value. Unfortunately the branding effort has lead to a situation where they are a bit hard to find sometimes, and it's often difficult to know who actually made all the re-badged units. SeaSonics are more widely available, but I have less experience with them.

Marcus Aurelius wrote:The industry responded also with a big branding effort, which in fact is mostly just a re-badging effort. In reality there are only a few PSU manufacturers and most of the brand name PSUs are either straightforward re-branded OEM units from those big manufacturers, or in some cases built to ordered specs. The main players are SeaSonic and FSP (Fortron/Source).

Is Asus one of the rebranders?

Marcus Aurelius wrote:In reality there are only a few PSU manufacturers and most of the brand name PSUs are either straightforward re-branded OEM units from those big manufacturers, or in some cases built to ordered specs. The main players are SeaSonic and FSP (Fortron/Source).

A few brands do add substantial value, e.g. the Zalman heatpipe cooled PSUs (of which I have two) are based on Fortron boards, but have a substantially lower noise level and operating temperature (for any given power output) than conventional PSUs based on the same hardware.