I am getting ready to build a new computer. One of the big issues, obviously, is processor. I've read multiple things regarding dual- vs quad-core. Seems like low end things, up to gaming, is better off on dual-core, whearas quad-core is good for multimedia applications. However, everything I read was coming from the Windows world.

How does Linux work with quad-core? I'm thinking that I read somewhere that programs without multi-threading can be "assigned" to a core when they start up so that all the cores are being used. I don't know if I'm remembering that right, though. If I were to buy a quad-core, would all four cores be used? I don't do heavy things. The most CPU-using programs I have are WoW, Eclipse, VirtualBox (building a Gentoo system in it so CPU gets used a bit) and Firefox (I really need to switch browers). I also often do multitasking, so I see my CPU jump up sometimes. If Linux is able to use all four cores, even with non-multithreading programs (via that assignment thing or whatever), it would be well worth it to go for quad-core. If not, though, then I might just get a dual-core.

Right now, I'm debating between the AMD Phenom 9850 BE 2.5GHz (I'd be overclocking it) and the AMD Athlon64 X2 3.2GHz. I only buy AMD . The money isn't the issue. I have no problem paying the extra $75 if all 4 cores are used. That is the deciding factor. Thanks!

Try this other thread, it's about the same kind of thing with answers to many of the same questions:
http://www.linuxquestions.org/questi...-e6850-633007/

You're mixing in some confusing factors that really aren't relevant to the question.

1) Linux: Linux neither limits your use of multiple cores nor promotes it. Whether four cores would be useful for you depends on the applications you run, not on Linux.

2) Assigning a program to a specific core does not increase the number of threads that will run at once. Most of the tasks will be idle most of the time (waiting for input or some other events). When multiple tasks need the CPU at once, then multiple cores will be used, regardless of whether those tasks are multi-threaded and regardless of whether programs are assigned to specific cores.

When at most one task needs the CPU at a time (which is typical for home computer use), multiple cores can be used only if that task is multi-threaded. The tasks you want to run that use a lot of CPU time probably aren't.

Very rarely all at the same time.

The literal answer to your question is yes, because as various minor tasks use small amounts of CPU time, that would be distributed over all four cores, just because they're there. Windows would distribute even a single compute intensive task over four cores, moving it on each time slice boundary. That gives the CPU chip slightly better thermal behavior, but gives the task worse cache behavior.

But you don't want a literal answer. You want to know if four cores will get more work done per second than two. I think they won't.

Dual core is best for WoW and for web browsing. The main activity uses only one core, but there is a significant amount of interrupt activity to service the network interface and keyboard/mouse and timers etc. A second core lets the main task proceed with fewer interruptions and better cache coherency. More than two cores would be wasted.

Get the dual core. I don't think you can safely overclock the Phenom much. The $75 won't buy you extra performance, it will buy you worse performance. For what you want to do, the difference between 3.2Ghz and 2.5Ghz is much more significant than the difference between 2 cores and 4.

I forget whether the 3.2Ghz is available in models with two different L2 sizes or whether all models have the same L2. If there are two different sizes, paying a moderate extra for the larger L2 makes good sense.

I didn't understand memory/motherboard issues myself well enough to make a good decision when I bought a similar system myself, but if you think you know how to overclock the CPU, you ought to understand such things. Within DDR2-800 (which I assume you're getting) the lower memory timing numbers (such as 4-4-4-12) can give significant performance improvement, but often cost more, require higher voltage (which not all motherboards can provide) and always run hotter, so you need a better case cooling system. If you understand such things and want to spend a little more for better performance, faster ram would likely be more useful than quad core. (Though if available, a bigger L2 would be more useful than faster ram).

I feel stupid asking this...but is DDR2-800 better than DDR2-1066? I've been around computers my whole life, but I'm just starting to study things. I could build computers, fix hardware issues, etc., but all I ever understood was the practice, not the theory. As I start putting this computer together, I am studying the theory more in depth to understand computers, electronics and the relationships of the various components so that I can build a better system and know what I'm building and so on. So, I apologize if this is one of those "duh" questions.

One other question...if DDR2-800 is better, will I lose anything by getting a motherboard rated for DDR2-1066?

No. DDR2-1066 is better.

DDR2-800 is the type of memory that 3.2GHz CPU is designed to work with. It will also work with faster or slower DDR2 memory, including DDR2-1066.

I don't know how much (if any) of the possible extra speed of DDR2-1066 memory that CPU can use. It probably depends on the motherboard and BIOS.

DDR2-1066 memory will cost more. You don't want to pay that unless you get the benefit. I don't know how to tell (before purchasing the motherboard and ram) whether you would get the benefit.

That main number for DDR2 memory is the maximum number of million cycles per second. The DDR2 memory gets a clock from the CPU and the DDR2 memory's cycle rate is double the rate of that clock from the CPU. So to get the 800 maximum rate for DDR2-800 the CPU must provide a 400MHz clock. To get the 1066 maximum rate for DDR2-1066, the CPU must provide a 533Mhz clock.

The CPU provides that clock as an integer divide rate from its own clock, so a 3.2Ghz CPU divides its clock by 8 to get 400Mhz. I think it can divide by 6 to get 533Mhz, so providing the clock for 1066 memory is not the problem.

The signal lines on the motherboard between the CPU and the memory are likely to be the problem. For this type of CPU, those signals lines aren't designed to handle faster than DDR2-800. The BIOS may have those limits programmed in and refuse to program the CPU to generate that 533MHz clock, because the motherboard designers didn't want you to run those signals at a speed that would make them unreliable.

If you really want to push it, a motherboard with only two DDR2 sockets has a better chance of being able to operate those sockets faster than a motherboard with more than two (The fewer points each signal connects to, the easier it is to lay out the board so the signal can transition faster).

I haven't checked prices, voltages and specs for DDR2 memory in a while. When I last did, there didn't seem to be any reasonable choice in DDR2-1066 that fit my needs. I was willing to pay a bit extra for faster memory, but there weren't good choices nor good documentation.

Pay attention to those other timing numbers (that 4-4-4-12 spec on the faster DDR2-800). Assuming the BIOS can configure the memory correctly, lower numbers there are almost as important for performance as a higher number in the main cycle rate, but those lower numbers don't represent extra demands on the quality of the physical connection between CPU and ram. So DDR2-800 rated 4-4-4-12 would be faster than DDR2-1066 rated 6-6-6-18 and that DDR2-800 would also be compatible with more motherboards.

Pay attention also to the voltage of the ram. 1.8 or 1.9 volt ram will work with almost any motherboard and with ordinary case cooling. Higher voltage ram may limit your choice of motherboard and require better cooling. That's the factor I couldn't figure out when I was shopping for ram (the higher speed ram usually requires higher voltage).

Ahem ... I have Conky configured to show all cores separately, I've to say most of CPU-intensive tasks are distributed equally on all four. AVC encoding, compiling. I'm running my box as distcc helper for other Gentoo boxes and even "incoming" jobs are run equally on all four cores. For instance, kernel compile takes around 5 minutes. I'd say I'm happy I got Q6600 instead of dual-core.
Buying AMD only? Dunno, I was big AMD fan myself, IMO Intel runs cooler right now and probably is less buggy. There are microcode updates available for Intel CPU-s, I did some simple tests and loading updated microcode definitely improves performance, plus bugfixes they provide. There is nothing like this available for AMD AFAIK.

Edit: Recent kernels do fairly good job handling multiple cores, including IRQ balancing. Just make sure you dedicate some time to configuring your kernel, it pays off.

Distributed equally on all four does not mean simultaneously using all four. Time slices are very short. Even Linux may move a single threaded task to a different core on a time slice boundary (though not as aggressively as Windows would).

That activity is very different from the typical home use (gaming, web browsing, etc.) discussed above.

If I wanted a computer to run a lot of compiling of big projects made up of large numbers of separate modules, I would certainly select an Intel Quad core. It isn't that hard to configure a make system to run multiple instances of the compiler in parallel. Even though each instance of the compiler makes no significant use of multiple cores, compiling a whole project makes great use of multiple cores (assuming you also have enough ram).

Even an AMD 2.5Ghz quad would be better than an AMD 3.2Ghz Dual for that use. But in my opinion, AMD quad core CPUs aren't really competitive with Intel. If your workload really justifies quad core, then I think it justifies selecting Intel (as you seem to have done). (Hopefully AMD will improve their quad core offerings and make that area competitive too).

I have an AMD dual core for my home computer and an Intel quad core at work. For all the above reasons, I think both those choices were correct.

Here is what I'm looking at in terms of CPU and RAM:

CPU:
http://www.newegg.com/product/produc...82E16819103228

RAM:
http://www.newegg.com/Product/Produc...82E16820148180

I was looking at this for a CPU at first:

http://www.newegg.com/Product/Produc...82E16819103194

It's about $70 cheaper and the reviews say that it easily overclocks to 3.2-3.4. But, as I kept looking around, I saw that my new choice has a 1MB cache. That, together with not having to overclock is making me lean towards that and think it's well worth the $70. I've never overclocked so...I kinda like the idea of not having to overclock, though it's the cache that is really doing it for me. Can I ask your opinion on these selections?

What about motherboard? What would be the biggest things to look for? Right now, I'm thinking 2600MHz HyperTransport, AM2/AM2+, DDR2-1066 with at least one PCI Express x16 and 8GB RAM supported. I figure with something like this, it gives me good upgrade possibilities (especially if AMD picks it up in the next year or two). Would an AMD chipset be the best?

Also, is ATI starting to do any better with Linux? I have my eyes on both an nVidia and ATI card. I plan on getting the motherboard/CPU/RAM in the next month and the video card in a couple months. I'm slowly building a new computer. I was going to buy it in one swoop...but I don't see myself doing that. This way, I get to have fun every couple weeks.

$150 seems a bit high for 2x2Gb. Also, do you know which AM2 motherboards support 2.2 volt DDR2 ram? I couldn't figure that out when I was making similar purchasing decisions.

That $70 overlooks the cooling device. The 3.2 CPU you chose includes a very good heat-sync/fan in that standard retail package you quoted. The cheaper CPU in a package for overclockers requires you to get a third party cooling device to cool it enough to overclock. There goes half of your $70 savings.

Good choice.

So far as I understand, the HyperTransport speed is almost meaningless on a single CPU (package) motherboard. The path between CPU and ram is what matters and that doesn't go through the HyperTransport. CPU to video and CPU to disk drive transfers do go through HyperTansport, but the HyperTransport speed is not a limiting factor for those.

HyperTransport speed is the critical speed in NUMA (non uniform memory access) multiprocessing (where each CPU has its own ram, to which it has direct access, but often needs indirect access to another CPU's ram). You're not doing that.

I don't think you're likely to throw out your AM2 CPU and replace it with an AM2+ CPU in the same motherboard. By the time your CPU is obsolete, your motherboard will also be obsolete. I'm less sure about DDR2-1066. I think it is a good idea to allow for growth to 8Gb in the same motherboard. I don't know if you could find an AM2 motherboard without a PCIx16 slot even if you wanted to.

Are you moving a video card from an older computer? Or are you getting a motherboard with built-in graphics?

Be careful when choosing a power supply. A good PCIx16 card for gaming places a significant extra load on the power supply. Many need a special power connector that most power supplies lack. Most draw surges of power that can burn out a power supply that would be perfectly acceptable without the gaming display card.

Since I don't do any gaming myself, I don't even know what runs in Linux. I expect most of those programs that need a high power PCIx16 card are Windows only. My Linux system has graphics built into the motherboard. For web browsing plus serious work, that's all you would want.

One of my sons burned out two power supplies before getting one that properly supported his gaming display card. Another got motherboard graphics but a power supply good enough to add PCIx16 graphics later. Then he discovered that the games he plays run as well with his motherboard graphics as with his friends' graphics cards (different games than the ones my first son found ran significantly better with the card than with his motherboard graphics). (None of my sons use Linux).

I recommend reading the following:
http://en.wikipedia.org/wiki/Memory_divider
http://en.wikipedia.org/wiki/Front_side_bus
(and make sure to look at the diagram on there)

It's a little complicated, I know, because it depends on a great number of things. Typically, higher numbers are better, but if the numbers don't match or compute between components, then it will do you no good.

Systems using the AM2 socket don't use a Front Side Bus, so much of what is shown in those web pages, especially that diagram, doesn't apply.

The memory controller is inside the CPU, not in the motherboard chipset.

The CPU uses HyperTransport to talk to everything other than the memory.

I was looking at some other RAM around the $80 price range. Most notably:

http://www.newegg.com/Product/Produc...82E16820211282

I overlooked the cooling device. I hadn't even noticed that. Thank you for that catch.

Thanks

I'm still learning the whole HyperTransport vs FSB. I had understood it that FSB can slow down a system because of the speed but the Intel/AMD benchmarks prove otherwise, as far as I can tell anyway. So, what is the story with the FSB vs HyperTransport since I'm obviously not getting it.

My thoughts exactly, regarding growth. I didn't do the same with my current system. It was only a 2GB max which is where I'm sitting at. I can't imagine using 8GB...but it wasn't that long ago when 1GB seemed like a lot. I don't really NEED 4GB (running Debian Lenny w/XFCE), but I like the idea of having it and knowing that I'm covered for a bit. I do like the idea of giving my VM a little more memory to play with too. I run openSuSE in VM (along with others, but that seems to be the one that needs the most RAM) because my mother uses openSuSE and when she has a SuSE-specific issue, it's easier to crank open the VM and figure out what's happening that way.

Yes. I currently have a nVidia GeForce 6600 256MB card. It works for the only game I play so I'm not in too much of a hurry to get a new one. I plan to, but it's working good for me now.

Right now, I have a 450W power supply. Since I have some ideas of how I want my system to end up, I'm thinking of upgrading to around 700W, probably. Better safe than sorry, I figure. As I understand it, it only draws as much as it needs, right?

My end system is going to include either 1 500GB hard drive or two 250s in a RAID setup(I currently have a 150GB external, a 120GB internal that's dying on me and a 40GB internal that is my main hard drive and I don't even use half and that includes all my music, backups, etc.), a DVD burner, a CD burner and the new graphics card I'm considering.

Thank you! I've been reading a lot lately from Wikipedia. I hadn't seen these yet, though. I'm going to go read up and get to learning.

The overall wattage may not mean much. The maximum current at each individual voltage may matter more. The two power supplies my son burned out had higher overall wattage than the one that works fine now.

For that CPU you chose, the motherboard will draw a lot on +12. I think the big power draw of a PCIx16 board is also on +12, but I'm not sure of that.

For my own computer and for my son that doesn't yet have a PCIx16 card, I chose this power supply from NewEgg
http://www.newegg.com/Product/ShowIm...W+Power+Supply
Notice it is just 430 Watts, but it allows up to 17A from either one of the two +12 rails (but limits the two +12 rails together to 360 Watts, which should be 30A). That is more current at +12 than most 450 Watt supplies can deliver.

I think the "80 PLUS Certified" is significant (I paid more to get that. I hope I wasn't fooled). Most power supplies have a minimum below which ideas like "it only draws as much as it needs" don't apply. So I don't think a 700W supply is a good idea without checking those specs. For that EA-430 I did check the manufacturer site to find the minimum below which the "80% Efficient" wouldn't apply and it seemed low enough for an idle computer (CPU clocked at 1Ghz instead of 3.2 and some other parts drawing less as well).

An idle CPU draws much less power than an active CPU even if the clock rate isn't changed, but it is almost linear with clock rate, so an idle CPU at 1Ghz draws less than a third of an idle CPU at 3.2Ghz. So most of that 125 Watts isn't being consumed most of the time. You wouldn't want a power supply that becomes inefficient when the CPU is idle and the draw on +12 may be under an amp.

I'm not 100% sure of the importance of two +12 rails. The CPU (especially that one) has a massively unstable power draw. The motherboard does voltage step down from +12 to the voltage the CPU actually uses and has lots of capacitors on the output side of that step down, which shield the CPU from instability in +12 and from seeing the noise from its own unstable power use. But those don't necessarily shield the +12 supply itself from the noise generated by the CPU's unstable power use. So I think the separate +12 rail is supposed to shield all the other devices that use +12 from the noise generated by the CPU's use of +12.

Anyway, make sure your 450 Watt supply provides enough on the CPU rail +12 for this CPU, or a lot more than enough if all the +12 is combined on one rail.

The requirement spec I saw for this CPU was 15A or more for the CPU rail +12 if they are split. That may be an excessive margin for safety, because it will actually only draw 11A running flat out. But I'm not an expert and that spec may be a real requirement, so I picked a supply that provides 17A to that rail.

The whole rest of my system (with motherboard graphics) probably doesn't use even 5A on +12, so the 30A total limit on +12 of that supply is major overkill.

Oh yeah, oops, I thought we were debating Intel processors and boards ... silly me, ignore what I said. But there is a link on the wiki on FSB about HyperTransport which is what AMD uses.