Building a PCLet's Talk Processors

[Saladsamurai]

Hello all. I am looking to build a new PC. I really know very little about computers, but since my degree seems to be
leading me into the computational realm, that will need to change quickly. Instead of just *building* a PC and going
through the motions, I would really like to *learn* what exactly it is that I am doing. I want to learn about the components
themselves and how they are specced out. So I would like to first talk about processors, since I believe that that will be the
main component that I will need to choose and the rest will follow suit. Then I hope that this thread will evolve into a
discussion about the remaining components.

EDIT Added for clarity: I suppose I should say what the PC will be used for, since that will dictate many choices: I am not a gamer, but I will possibly use the PC to do some simulations (i.e., CFD or FEA). I am getting into programming quite a bit, but will probably never take it to a professional level. I use tools like Mathematica and MATLAB a lot.

So here is what I already (think that) I know: The higher the processor speed and the more 'cores' the faster. That's all I know. So maybe
we can use a specific processor as an example and talk about what each of the specs means:

Intel Core i5-750 Lynnfield 2.66GHz 8MB L3 Cache LGA 1156 95W Quad-Core Processor
http://www.newegg.com/Product/...6819115215

I have wikipedia'd these, but it's a little wordy; maybe someone could paraphrase for me:

L3 Cache
Manufacturing tech (they are all the same, so I guess I don't really care about this one)

What does the "LGA 1156 95W" mean?

What is the difference between the one I linked above and this one? http://www.newegg.com/Product/...6819115131

They are both Quad core 2.66GHz. What does the other stuff mean?

 

[Borek]

LGA 1156 95W

Socket LGA 1156 rated at 95W. That is, you won't be able to put the processor into mobo that doesn't have this socket type and is not able to supply 95W to it.

L3 cache - do you know what memory cache is? In short - that's a very fast memory sitting in processor and mirroring content of RAM, used by processor to speed up access to the code and data. L3 means there are three levels of cache, and L3 cache has 8 megabytes. In general - the more, the better.

 

[rcgldr]

If you go through the "products for work" section at the intel website you end up component specific info, such as a web page like this one:

http://www.intel.com/products/processor/corei7/specifications.htm

 

[dE_logics]

The number of cores is not always a direct measure of performance, here are a few factors --

1) The processor architecture
2) The cache
3) Core frequency
4) Features

Here 1, 2 and 3 are MAJOR factors.

So it might be that for e.g. a company gives a processor a very high core frequency, e.g 3.2 ghz (it's rated in GHZ for modern processors) and lots of cache memory...e.g 1 MB for L1 cache and 6 MB for L2; but the processor architecture is very bad. Then it might happen that a processor with core frequency 1.8 Ghz and cache sizes of 259 KB, 515 KB might beat the previous processor.

From all the 4 stated above, a high core frequency is the only one which has disadvantage if increased.

Your processor has lots of features also (if it's a good one)...many times to utilize these, your OS should not be the standard one...since you need to compile the source code in such a way so as to make utilization of such features.
So to use your features completely -
1) You have to quit all MS products (most probably the have the worst quality of precompiled binaries)
2) You need to start using a source based distribution which uses a rock solid compilers like GCC.

That's not all...the configuration part is remaining.

Many developers and advance users support AMD for a solid reasons, but right now Intel has the upper hand when it comes to top notch performance (i.e only higher end processors). It's said that AMD processors have better features, and technically better architecture, but aaa...we can't say anything about that right now.

 

[dE_logics]

And yes...if you're looking for pure 64 bit computing...you have no option but AMD.

Intel processors, at the current state of time are just compatible with 64 bit, not actually 64 bit...it's called EMT64.

 

[Topher925]

So I would like to first talk about processors, since I believe that that will be the main component that I will need to choose and the rest will follow suit.

I think you're starting off on the wrong foot here. The CPU is most definitely NOT the main component of a PC and is usually one of the last components I pick out when doing a build.

The very first thing you need to look at when building a PC is the chipset of the motherboard. The chipset is a collection of processors and different components and basically is your computer or can at least be considered the heart of it. Its the only portion of the computer which you are directly interacting with (except maybe the VGA card) and is also responsible for just about everything that goes on behind the scene. Its what controls the work the CPU does, handles data from RAM and harddisks, interfaces to the AGP, and controls all the peripherals, etc. In other words, it is by the far the most important thing to consider when building a PC.

There are a lot of different chipsets out there. Some by processor manufacturers (AMD, Intel) some by GPU manufacturers (Nvidia), and some by other parties (SiS). They all have their pros and cons such as stability, speed, and features (overclocking). Chipsets also will have a significant effect on the price of your PC. Motherboards with an expensive chipset with many features such as RAID drivers, voltage controllers, and timing features, can cost well over $200. While a motherboard with a cheaper more "barebones" chipsets, like the Intel G31 chipset I have in my PC can cost as little as $40.

You can read more about it here:
http://en.wikipedia.org/wiki/Chipset

 

[Saladsamurai]

Okay folks. This is great information; exactly what I am looking for. I suppose I should say what the PC will be used for, since that will dictate many choices: I am not a gamer, but I will possibly use the PC to do some simulations (i.e., CFD or FEA). I am getting into programming quite a bit, but will probably never take it to a professional level. I use tools like Mathematica and MATLAB a lot. (I am going to add this 'uses paragraph' to the OP for clarity.)

So let me address some things:

Socket LGA 1156 rated at 95W. That is, you won't be able to put the processor into mobo that doesn't have this socket type and is not able to supply 95W to it.

L3 cache - do you know what memory cache is? In short - that's a very fast memory sitting in processor and mirroring content of RAM, used by processor to speed up access to the code and data. L3 means there are three levels of cache, and L3 cache has 8 megabytes. In general - the more, the better.

This comes across crystal clear. Thanks Borek.

The number of cores is not always a direct measure of performance, here are a few factors --

1) The processor architecture
2) The cache
3) Core frequency
4) Features

Here 1, 2 and 3 are MAJOR factors. <snip>...
...
So to use your features completely -
1) You have to quit all MS products (most probably the have the worst quality of precompiled binaries)
2) You need to start using a source based distribution which uses a rock solid compilers like GCC.

Thank you de_logics! Great info. Ok, so you say that processor architecture is an imperative, but how do we distinguish different architectures? Is it simply either AMD or Intel? Or is there more to it?

With regard to "quitting all MS products," I think we might be getting a little too advanced for me. I want to build a fast machine, that is the design goal here. Now perhaps I will dual boot in Linux at some point. But since I have never actually, used Linux, that is down the road a little bit; this will mainly be a Window machine.

And yes...if you're looking for pure 64 bit computing...you have no option but AMD.

Intel processors, at the current state of time are just compatible with 64 bit, not actually 64 bit...it's called EMT64.

I am sorry, but I don't understand this You say that "Intel processors, at the current state of time are just compatible with 64 bit, not actually 64 bit."

What is the distinction here btween compatible with 64 bit and is 64-bit?

<snip>The CPU is most definitely NOT the main component of a PC and is usually one of the last components I pick out when doing a build.

The very first thing you need to look at when building a PC is the chipset of the motherboard.

Holy throw a wrench in my plans Batman! Well, then. If I know ahead of time that I am definitely going with an Intel processor, should I be going with an Intel MB? I am assuming the answer will be yes. So then, I guess I need to be looking MBs since I do know what distinguishes them from each other.

 

[Borek]

Holy throw a wrench in my plans Batman! Well, then. If I know ahead of time that I am definitely going with an Intel processor, should I be going with an Intel MB? I am assuming the answer will be yes. So then, I guess I need to be looking MBs since I do know what distinguishes them from each other.

It is not MB that matters, but chipset first, MB second.

In my experience faster chipset means faster MB, but not all MBs with the same chipset are equal - some will be faster, some will be slower. These differences are smaller than those from the difference between chipsets, and sometimes they are marginally small, but sometimes they can mean a lot.

Note, that ordering of the MBs depends on the benchmark used. There is no one fastest combination of MB/processor for every task.

You may visit tomshardware.com and check their tests.

 

[Saladsamurai]

It is not MB that matters, but chipset first, MB second.

In my experience faster chipset means faster MB, but not all MBs with the same chipset are equal - some will be faster, some will be slower. These differences are smaller than those from the difference between chipsets, and sometimes they are marginally small, but sometimes they can mean a lot.

Note, that ordering of the MBs depends on the benchmark used. There is no one fastest combination of MB/processor for every task.

You may visit tomshardware.com and check their tests.

OK then But I still do not know how to rate a chipset or how to pick one. So how do I choose? I read the Wiki on chipsets but I am still not sure what metric we use to distinguish a 'better' chipset from a 'worse' one.

I can't say that I am experienced enough of a computer user to even know that I would see a difference.

 

[Borek]

That's not easy, as you need to have access to different mobos - and that's why I have referred you to tomshardware. They do comparative tests like here and http://www.tomshardware.com/charts/2009-desktop-cpu-charts/benchmarks,60.html (while these two are not chipset comparisons, I think you should be able to dig these as well).

Note: there are other sites that do similar tests, I just happen to be Tom's fan. In fact many years ago he wanted me to work for him, but I was busy with other things at the time.

 

[rcgldr]

processor architecture

For intel most desktop cpu's will be "Nehalem":

http://en.wikipedia.org/wiki/_(microarchitecture)

You have to quit all MS products ... gcc

I don't see how the OS affects the choice of compilers, and Visual Studio supports features such as SSE2.

And yes...if you're looking for pure 64 bit computing...you have no option but AMD. Intel processors, at the current state of time are just compatible with 64 bit, not actually 64 bit...it's called EMT64.

I'm not aware of any cpu that emulates 64 bit mode. Almost all current Intel CPU's are 64 bit cpus with 32 bit backwards compatability. EMT64 is an old name for AMD's 64 bit extension to the x86 instruction set, x86-64, which was licenced and implemented by Intel years ago. Microsoft calls it x64, AMD calls it AMD64, Intel calls it Intel64. This is to diffentiate from the older IA-64 bit instruction set used on Intel's Itanium cpu's which is a completely different instruction set.

 

[Saladsamurai]

That's not easy, as you need to have access to different mobos - and that's why I have referred you to tomshardware. They do comparative tests like here and http://www.tomshardware.com/charts/2009-desktop-cpu-charts/benchmarks,60.html (while these two are not chipset comparisons, I think you should be able to dig these as well).

Note: there are other sites that do similar tests, I just happen to be Tom's fan. In fact many years ago he wanted me to work for him, but I was busy with other things at the time.

Yeah, I looked at that website, but I cannot find any charts for chipsets. But moreover, even if they did have charts for chipsets, what am I looking for? What makes Chipset_A better than Chipset_B ?

 

[dE_logics]

I don't see how the OS affects the choice of compilers, and Visual Studio supports features such as SSE2.

I'm talking about how the how the programs are compiled...you do not know that when using precompiled and 100% proprietary programs.

Also GCC's development is pretty fast...it might be that AMD and Intel too contribute to it; but with MS, its solely depends on them...their mood...so development might be slow and buggy.

I'm not aware of any cpu that emulates 64 bit mode. Almost all current Intel CPU's are 64 bit cpus with 32 bit backwards compatability. EMT64 is an old name for AMD's 64 bit extension to the x86 instruction set, x86-64, which was licenced and implemented by Intel years ago. Microsoft calls it x64, AMD calls it AMD64, Intel calls it Intel64. This is to diffentiate from the older IA-64 bit instruction set used on Intel's Itanium cpu's which is a completely different instruction set.

Yeah, I was wrong about that...confusion with the 64 wars.

but how do we distinguish different architectures? Is it simply either AMD or Intel? Or is there more to it?

Usually companies provides an architecture codename...you should check them out. Also the nomenclature of the processors is such that you can make a difference with the architecture...for e.g. Intel's core2duo is different from the Pentium dual core processors. If this name is common among certain set of processors (with difference of just certain numbers e.g. E7500 and E8400 with Core2duo processor models); is usually means same architecture...you can check their codenames for further analysis.

With regard to "quitting all MS products," I think we might be getting a little too advanced for me. I want to build a fast machine, that is the design goal here. Now perhaps I will dual boot in Linux at some point. But since I have never actually, used Linux, that is down the road a little bit; this will mainly be a Window machine.

If you're in this field, I think it will be mandatory to at least have hands on Linux since that really let's you play around to extreme depths.

Linux is easy actually...start with Ubuntu but for best processor optimization you need source based distributions like Gentoo and LFS (DO NOT jump to Linux with them, switch when you think you have gained enough experience with Ubuntu).

And also...prefer an Nvidia graphs chip since you'll be using it with Linux.

 

[mugaliens]

The very first thing you need to look at when building a PC is the chipset of the motherboard.

Precisely.

http://en.wikipedia.org/wiki/Intel_Nehalem_(microarchitecture)" family of processors (though it can handle the Core i3 and Core i5 processors, too).

LGA 1156: Lynnfield (Core i5-7xx, Core i7-8xx, Xeon X34xx). Quad-Core 45 nm Dual-Channel, PCIe - performance/mainstream desktop socket set.

LGA 1366: Bloomfield (Xeon 35xx and Core i7-9xx). Quad-Core 45 nm Triple-Channel - extreme/performance desktop socket set.

If you want to wait, the Gulftown (also LGA 1366) is a six-core, hyperthreaded processor to be released Q1 2010 in Core i7-980X.

 

[Topher925]

Yeah, I looked at that website, but I cannot find any charts for chipsets. But moreover, even if they did have charts for chipsets, what am I looking for? What makes Chipset_A better than Chipset_B ?

Its difficult to compare chipsets by a single metric as there are a lot of variables. Instead, you should focus more on the features you want your chipset to have. For example, you stated that you mostly will be doing FEA/CFD along with matlab. Software such as this runs best on a solid stable platform that won't do strange things when pushed to its limits. Something with a dual BIOS would probably be ideal. Computational software usually requires lots of RAM, the faster the better. So you will want something that supports the highest FSB you can get (with overclocking if you feel adventurous) and something that supports large amounts of RAM. These days, 8gb is typical for 775 socket based chipsets but you will probably find you may want 16gb if not more. Then you will also want to consider if you want to dual/triple channel your memory as well.

Theres lots to consider when looking at a chipset, a lot more to cover in just one post. I recommend you join a computer forum as lots of them have good articles that give in depth answers to the questions you are asking. I like www.overclockers.com, but there's lots of good ones out there.

Also, take benchmarks from Toms Hardware and all other sites for that matter with a little bit of skepticism. Often you will see benchmarks from different sites under the same conditions to be radically different from each other.

Another thing worth mentioning is that I recommend you look into AMD processors. While the focus of this thread has mostly been on Intel products, I highly recommend AMD CPUs as they, without contest, give the most performance for the dollar. The introduction of the i5 closed that gap quite a bit but its still more than possible to build a high end AMD machine for a few hundred less than an Intel with similar performance.

 

[Borek]

Also, take benchmarks from Toms Hardware and all other sites for that matter with a little bit of skepticism. Often you will see benchmarks from different sites under the same conditions to be radically different from each other.

It is also worth to remember that same processor/mobo combination can be better in one benchmark and worse in other; there is no one, universal benchmark that tells the only truth.

And getting back to chipset tests - you will probably not find just a chipset comparison, but comparison of mobos built on different chipsets. Sometimes these give a clear winner (like "all mobos with chipset x are substantially faster than all mobos with chipset Y"), sometimes results are inconclusive.

As Topher wrote - you should first decide what the computer will be mainly used for, then look for the hardware that best suits your needs.

 

[Saladsamurai]

Ok then I am going to back it up a little here. This has been a great discussion and a real eye-opener to me as this is my first build. It is fun, but there is a serious learning curve associated with it.

Topher: I see what you are saying. However, in the OP I said that I would possibly be using CFD/FEA. First off CFD/FEA licenses are so expensive that I would first have to learn how to use Linux so that I could mess around with any of the decent open source stuff. So that was my point; I would like to have the *option* of being able to really *tinker around* with some open source CFD/FEA.

I had initially started the OP with the intent of learning about the nomenclature associated with CPUs, which I have. Now I learning about mobos and their associated chipsets. I never included a price tag on the build, which upon reflection, I probably should have done. Since i7 processors and 16MB Ram are automatically excluded when I say that my build has a target price of $800 (USD).

i7's have a hefty price tag on them. And even the entry-level ones at $290 will incur additional expenses as I hear that their compatible mobos are pricier.

So I am onto read up about mobos now. yay.


edit I am reading up on mobos at Tom's. People keep saying that that I want a "stable" mobo. How do we quantify stability? What metrics should I be concerned with?

http://www.tomshardware.com/reviews/lga-1156-motherboard,2463-2.html" is the chart I am looking at. There is no "stability" category

 

[minger]

I'm not the most computer knowledgeable person, but I do dabble a lot in the numerical world. If you're going to be running simulations, it is important to first decide which codes you're running. For example, we use a in-house code that is specifically optimized for multiple cores. Additional code is required to utilize these extra cores, they don't just start solving automatically.

If you're going to be using a third-party software such as ANSYS/CFX, note that their HPC (high-performance computing) sucks. I have just done some benchmarking, and it's not impressive. So, in that sense, having a lot of cores might not benefit you that much.

One open-source CFD package, OpenFOAM says that they get great scaling (3x at 4 processors; which isn't great), but again, getting that functionality can be difficult.

So, I suppose the only semi-worthwhile advice that I could give is to have plenty of fast RAM. You don't want to be writing to disk 100,000 times, each consisting of potentially hundreds of thousands of grid points.

Good luck

 

[mugaliens]

Good points, topher925 and minger.

1. AMD's do provide more bang for the buck, while Intel's flagship chips provide more oomph, but at a premium.

2. I'd go for the quad-core for no other reason than most games will use it. :)

3. More ram helps only if it's needed. Since most of the Nehalam mobos max at 16 GB, I'd shoot for two 4 GB sticks to give you 8 GB, leaving you slots for more, if it proves necessary. If not, you've saved some bucks.

 

[dE_logics]

Ok then I am going to back it up a little here. This has been a great discussion and a real eye-opener to me as this is my first build. It is fun, but there is a serious learning curve associated with it.

Topher: I see what you are saying. However, in the OP I said that I would possibly be using CFD/FEA. First off CFD/FEA licenses are so expensive that I would first have to learn how to use Linux so that I could mess around with any of the decent open source stuff. So that was my point; I would like to have the *option* of being able to really *tinker around* with some open source CFD/FEA.

I had initially started the OP with the intent of learning about the nomenclature associated with CPUs, which I have. Now I learning about mobos and their associated chipsets. I never included a price tag on the build, which upon reflection, I probably should have done. Since i7 processors and 16MB Ram are automatically excluded when I say that my build has a target price of $800 (USD).

i7's have a hefty price tag on them. And even the entry-level ones at $290 will incur additional expenses as I hear that their compatible mobos are pricier.

So I am onto read up about mobos now. yay.


edit I am reading up on mobos at Tom's. People keep saying that that I want a "stable" mobo. How do we quantify stability? What metrics should I be concerned with?

http://www.tomshardware.com/reviews/lga-1156-motherboard,2463-2.html" is the chart I am looking at. There is no "stability" category

The most difficult thing that you will be doing for the computer is using software computational tools, that require a good processor, or optionally a graphs chard which can also be used in scientific computation. But we have special software for that (i.e software which makes use of that graphs technology to increase the computing power.) and most probably you won't find those software with ease...specially narrowing the category to computational tools, it's still rarer. But if you do get such card, it's good...it's not that expensive, and might come onboard.

I recommend you to spend mostly on the processor and the motherboard. If you're opting for Nvidia graphs then you should opt for an "integrated nvidia graphs controller" that should be cheep. If the integrated graphs support "CUDA" (very unlikely)...it's still better. For $800 you can get a lower end (or actually a bit above that) gaming PC also, so you might reduce the budget. However if you're willing to spend that much for non-gaming purposes, do it all on CUDA(2) and the processor(1).

Yeah, and also ensure that the graphs chip is HD ready...that's standard now a days. All 8xxx series and above chips have that (including the GT series) have it.

 

[Borek]

edit I am reading up on mobos at Tom's. People keep saying that that I want a "stable" mobo. How do we quantify stability? What metrics should I be concerned with?

Something like MTBF (Main Time Between Freezes) will do. But these are random things, so they are hard to measure. Basically when they say mobo was stable it means "we have not seen computer freeze during our tests".

 

[dE_logics]

Something like MTBF (Main Time Between Freezes) will do. But these are random things, so they are hard to measure. Basically when they say mobo was stable it means "we have not seen computer freeze during our tests".

MTBF is Mean time before failure.

I think such sort of stuff should be out of the question here...since the MTBF is really high for at least current generation chips. Although it's said that AMD uses better quality raw material (as a result they get 30% profits and Intel gets 50%), but I think that won't matter...they will at least for for 10 years continuously @ 60°.

 

[Borek]

MTBF is Mean time before failure.

Obviously now that I don't have Best Humor badge I will have to use smileys.

I think such sort of stuff should be out of the question here...since the MTBF is really high for at least current generation chips. Although it's said that AMD uses better quality raw material (as a result they get 30% profits and Intel gets 50%), but I think that won't matter...they will at least for for 10 years continuously @ 60°.

I think it was always more about overclocking - the better and more stable chipset, the more it is resistant to increased frequency/voltage.

 

[dE_logics]

Obviously now that I don't have Best Humor badge I will have to use smileys.



I think it was always more about overclocking - the better and more stable chipset, the more it is resistant to increased frequency/voltage.

Really, MTBF is mean time before failure.

http://en.wikipedia.org/wiki/Mean_time_between_failures

Yes, if you're planning to set of core frequency to 3.6 - 4 ghz without a cooler which is more expensive than your processor, then it does matter. May be if the processor runs at 90d it will fail in a year.

 

[dE_logics]

I read about MTBF...so it is actually a quality control sort of thing (which makes me think I was not right with this previously).

 

[Borek]

Really, MTBF is mean time before failure.

I really know F in MTBF stands for Failure and I really thought translating it in this context as Freeze would be funny. Seems I was wrong.

 

[Saladsamurai]

I really know F in MTBF stands for Failure and I really thought translating it in this context as Freeze would be funny. Seems I was wrong.

Oh! It's hilarious Borek! Don't you worry! :rofl: