Building a PC, give me your thoughts please

[fluidistic]

It's been years I don't have a desktop pc and now is the time I "need" one. My use will be numerical simulations, compiling general programs as well as latex documents, some lightweight/middleweight programming, using a virtual machine and use another OS's this way. I'll be using Linux only. My wish is that the computer is very responsive at all times, as well as being silent. A core i5 6500 or so would probably be more than enough but it's been so many years I'm dying to buy a monster pc that... below are my thoughts:
1) CPU : Intel core i7 7700. I know, a huge overkill but I really want it, and I don't mind spending the extra 150 bucks or so instead of picking a more reasonable core i5. This processor eats 65 W of electricity maximum.
2)A motherboard that's compatible with usb type C and that's able to deal with 2400 MHz RAM. Not sure which model at all.
3)A single SSD of 250 GB. After 2 years on my laptop, I have "only" used 41 GB of space, so 250 GB is more than enough. I do have an external usb 3.0 500 GB mechanical disk as a back up.
4)Either 8 or 16 GB of RAM. I've ran out of RAM (4 GB) quite a few times, either due to a bug in my python program or due to compiling programs.
5)Not sure about the PSU, a good quality one that provides 300 W?
6)No DVD/bluetooth/blueray/floppy disk devices. No GPU other than the one provided by the processor. I don't play to video games.

Do you think I'm doing something wrong? Thanks for your thoughts.

 

[phinds]

https://www.physicsforums.com/threads/building-a-pc.900739/#post-5670012

And one thing I should have added to that thread, having been reminded by your statement

1) CPU : Intel core i7 7700. I know, a huge overkill but I really want it, and I don't mind spending the extra 150 bucks or so instead of picking a more reasonable core i5. This processor eats 65 W of electricity maximum.

My whole system, i7, four hard drives, two DVD players, a couple of external USB-powered drives, a graphics card and sound card, ALL TOGETHER take a grand total of 80Watts so my 600W power supply turned out to be massive overkill. Oh, it DOES spike to almost 120Watts when the bigger of the two DVD players is active.

 

[Chronos]

300W PSU is a bit thin. The few extra bucks for a 600W PSU is like life insurance and makes more sense than using it, or other system components, as fuses. The I7 only makes sense if hyper threading is a benefit for your intended usage. Some software, especially games, actually performs worse with hyper threading than without it. I would be advise checking on this before burning the extra cash for it. On a 64 bit OS 4 gigs is not enough memory. 8 is usually the sweet spot although 16 may help with memory intensive software. A dedicated GPU is a plus for almost any video based software, mainly games, but, also stuff like CAD. Onboard V chips still have performance issues and can overwork your CPU. A high end GPU can cost more than the rest of your system, but, there are plenty of budget GPU's that will do exemplary work for most users. Not to mention that VRAM absolutely smokes DRAM on graphics.

 

[phinds]

300W PSU is a bit thin. The few extra bucks for a 600W PSU is like life insurance and makes more sense than using it, or other system components, as fuses.

I agree w/ that even thought my 600W is way overkill for my system.

The I7 only makes sense if hyper threading is a benefit for your intended usage. Some software, especially games, actually performs worse with hyper threading than without it. I would be advise checking on this before burning the extra cash for it.

I'm not so sure about that. I typically have a LOT of programs all up at the same time and with an i7 and 4 cores, the response time is amazingly fast. Of course a lot of that is due to the solid state boot drive

 

[john101]

Consider getting a Motherboard that'll take next gens cpu's and scaling back to the 6700 (negligible performance difference, but no in built 4k support) and include a GPU.

ps. you don't necessarily have to have on board usb-c support if you get a dongle instead.

 

[Chronos]

Agreed, an SSD is the single biggest speed boost I have ever seen on any of my builds.

 

[phinds]

Agreed, an SSD is the single biggest speed boost I have ever seen on any of my builds.

Yeah, I've really been delighted with it.

 

[Vanadium 50]

This is to be an SSD-only machine? What do you plan to do about swap space?

 

[phinds]

This is to be an SSD-only machine? What do you plan to do about swap space?

I don't see why that would be a problem. It's not like the SSD is volatile memory. I'm a power user with a LOT of installed applications and many of them frequently running at the same time, and my hibernation file AND my swap file together only add up to 30Gigs and he's only using 40 to 50 gigs of the 256 available. Where's the issue? MY swap and hibernation files are on my 256Gig SSD and I don't have any problem.

 

[rbelli1]

MY swap and hibernation files are on my 256Gig SSD and I don't have any problem.

Just make sure you get a quality drive. Windows (or probably any modern OS) will burn out cut rate drives in short order. I've had a bunch of drives that only lasted about a year of always on but low load usage. This was in a low ram (1GB) system. These were different models from the same "industrial" series. 4 out of 5 so far have gone south. I'm not sure why the one is still running as that is the highest usage one out of the bunch.

BoB

 

[Chronos]

Curious, SSD's are generally more reliable than HDD's and not really sensitive to usage. Workmanship is, however, important so brand choice can be an issue.

 

[Vanadium 50]

Where's the issue?

SSDs have a finite number of writes before they fail. Swap partitions have a large number of writes, and worse, writes to specific areas of the SSD.

 

[fluidistic]

Ok guys, thanks! I didn't think about the 300 W PSU being too little. Good idea, I'll go for a 600 W or so (maybe 500 W).

This is to be an SSD-only machine? What do you plan to do about swap space?

Very good question, I didn't think about it. If I have 16 GB of RAM, then I think I'll go without any swap partition. I know that many people don't have any swap partition, but I don't know the consequences. If the main consequence is that the whole system freezes for 30 minutes when the RAM is fully used, then it's not that big of a deal because I expect these situations to be of extremely rare occurrences. But if I hit the 16 GB of RAM limit often, I might add another 8 GB of RAM afterwards.
On my laptops the times I ran out of RAM, the whole system would freeze anyway (though for only 2 minutes until being responsive for 1 s and then frozen back again) even though it had just started to write a few kB into the swap partition while I had 4 GB of swap. I know Linux offers a swappiness command to select when the system should start to use swap, mine is set on 60 (the default).

 

[phinds]

SSDs have a finite number of writes before they fail. Swap partitions have a large number of writes, and worse, writes to specific areas of the SSD.

Huh. I didn't realize they stayed on specific areas. That is bad for an SSD. Can the swap file be moved to a non-boot drive?

EDIT: never mind. I found it on internet. Quite simple.

 

[Vanadium 50]

If the main consequence is that the whole system freezes for 30 minutes when the RAM is fully used

The main consequence is you get a kernel panic when RAM is fully used. That will leave your filesystem in a corrupted state. (The actual message is "kernel panic, not syncing, no killable processes). The slowdown is when RAM is almost fully used.

 

[Vanadium 50]

Can the swap file be moved to a non-boot drive?

Sure, no problem. For a while I even had a swap area on a RAID.

You can also, with some effort, move it around. You need to use a swap file and not a swap partition, and you need to destroy it and recreate it regularly. How well this works (and how well this works as a function of effort) will vary. I won't go as far as to say "never use an SSD for swap" but I would say that it's important to understand the implications and the tweaks. I would also argue that Linux benefits less from an SSD than Windows. This is more a statement about Windows than about Linux.

 

[phinds]

Sure, no problem. For a while I even had a swap area on a RAID.

You can also, with some effort, move it around. You need to use a swap file and not a swap partition, and you need to destroy it and recreate it regularly. How well this works (and how well this works as a function of effort) will vary. I won't go as far as to say "never use an SSD for swap" but I would say that it's important to understand the implications and the tweaks. I would also argue that Linux benefits less from an SSD than Windows. This is more a statement about Windows than about Linux.

I moved my swap file to a non-boot drive but can't get rid of the one one the boot drive since it's a system file and Windows refuses to allow its deletion. Any ideas? Presumably it isn't being USED, but I'd like to free up the space.

EDIT: Oh. All I had to do was restart the computer and it went away by itself.

 

[fluidistic]

The main consequence is you get a kernel panic when RAM is fully used. That will leave your filesystem in a corrupted state. (The actual message is "kernel panic, not syncing, no killable processes). The slowdown is when RAM is almost fully used.

I see. I've googled a bit and it seems that the kernel will try to kill some processes first, via OOM killer. One can force the kernel to always panic when no more RAM is available, but it's not clear to me why would OOM killer lead to a kernel panic if it does kill the processes that use the most RAM.

 

[fluidistic]

Ok, from what I've just been told on IRC, there used to be kernel panics before OOM killer was implemented. Now, people have the choice of what happens when one runs out of memory, either let OOM killer kill some process(es) based on a not so complicated algorithm which can be found on https://linux-mm.org/OOM_Killer or kernel panic (and maybe other choices that I forgot).
However on modern SSD, swapping shouldn't be bad because the data should not be written on the same areas. So while it is true that SSD flash blocks have a finite number of write before they die, they still outperform HDD in terms of longevity.

If this is true, then I'd set 16 GB of swap partition on my 250 GB SSD (once I get one, that is).

 

[phinds]

Swap partitions have a large number of writes, and worse, writes to specific areas of the SSD.

... However on modern SSD, swapping shouldn't be bad because the data should not be written on
the same areas.

One of you guys has it wrong and I'd like to know which. A brief internet search didn't turn up any answer. I can certainly see how it would be very advantageous for the O.S. to keep using the same space for the swap file once it is first assigned since that would allow it to rewrite small areas of the file (a registry section for example) without having to rewrite the whole thing, so that makes a lot of sense, but I can't find corroboration that it does so.

 

[fluidistic]

One of you guys has it wrong and I'd like to know which. A brief internet search didn't turn up any answer. I can certainly see how it would be very advantageous for the O.S. to keep using the same space for the swap file once it is first assigned since that would allow it to rewrite small areas of the file (a registry section for example) without having to rewrite the whole thing, so that makes a lot of sense, but I can't find corroboration that it does so.

From what I'm reading (thanks to google) people seem to say that swap on SSD is perfectly fine. SSD come with at least 3 years of warranty (vs 2 for HDD), are way more reliable than what they used to be a few years ago. They even say that letting swapiness to 60 (the default in linux) is fine.

 

[phinds]

From what I'm reading (thanks to google) people seem to say that swap on SSD is perfectly fine. SSD come with at least 3 years of warranty (vs 2 for HDD), are way more reliable than what they used to be a few years ago. They even say that letting swapiness to 60 (the default in linux) is fine.

Yes, that's also what I find, however the reliability of the SSD is not at issue. They are by their very nature limited as to the number of writes that they can sustain in a given spot. Now, that's a BIG number of writes but even so, if the swap file is always hitting the same spot that IS an issue and as I said in my previous post, it is perfectly logical to me that Vanadium's statement would be true since it makes a lot of sense for the O.S. to deal with the swap file that way. You are avoiding the question, not answering it directly, although I realize that you probably think you have answered it indirectly.

I'm still trying to find confirmation. @Vanadium 50 , can you point me at anything?

 

[Vanadium 50]

Swap file and swap partitions are not the same thing.

Normally Linux uses a swap partition. It is more difficult to wear-level a partition than a file - the file system can't help, because it's not a file system. (Look up TRIM for more details on wear-leveling)

Modern operating systems don't wait until they are dangerously short of memory before they think about swapping. Long before that, they start identifying pages that might be swapped and use idle time to copy them to the swap partition. If they need to swap later, great, it'll go faster since the work is mostly done. If not, well, it was filled during idle time so no harm no foul. This is probably not the optimal behavior for SSDs.

 

[phinds]

Swap file and swap partitions are not the same thing.

Normally Linux uses a swap partition. It is more difficult to wear-level a partition than a file - the file system can't help, because it's not a file system. (Look up TRIM for more details on wear-leveling)

Modern operating systems don't wait until they are dangerously short of memory before they think about swapping. Long before that, they start identifying pages that might be swapped and use idle time to copy them to the swap partition. If they need to swap later, great, it'll go faster since the work is mostly done. If not, well, it was filled during idle time so no harm no foul. This is probably not the optimal behavior for SSDs.

I"m sorry, I don't seem to be clear in my question. You stated earlier that the swap file writes to the same area of the SSD (or presumably whatever drive it is on) over and over, rather than re-writing the whole swap file to a new location. This makes good sense to me that the O.S. would do that, but I can't find any confirmation that it actually DOES do that and I'm asking if you can point me to documentation that shows that the swap file, once assigned to a disk area, does indeed keep hitting that exact same area.

I understand completely about partitions and that is irrelevant to my question.

 

[Vanadium 50]

I said

Swap partitions have a large number of writes, and worse, writes to specific areas of the SSD.

 

[phinds]

I said ... Swap partitions have a large number of writes, and worse, writes to specific areas of the SSD.

OK, my misunderstanding. Sorry, and thanks for setting me straight on what was said.

 

[rbelli1]

A modern SSD should include a wear leveling algorithm. If you write say the same 4k part of a file with 100GB on a 100GB SSD you should have effectively written everywhere once. The controller on the drive will do that automatically. So making a separate partition should not considerably impact the actual writing going on under the covers.

BoB

 

[phinds]

A modern SSD should include a wear leveling algorithm. If you write say the same 4k part of a file with 100GB on a 100GB SSD you should have effectively written everywhere once. The controller on the drive will do that automatically. So making a separate partition should not considerably impact the actual writing going on under the covers.

BoB

I don't follow you at all. Suppose you have an 8Gig swap file and you create an 8Gig partition for it. How is the SSD NOT going to write to the exact same place every time?

 

[john101]

That makes sense to me.

Perhaps: Windows can be set* to always write to a permanent swap file or handle swap itself in which case it writes to wherever it has space.

edit add *either on the windows partition or onto another partition (on the ssd or on another drive)

 

[phinds]

That makes sense to me.

WHAT makes sense to you? Please quote the post that you are agreeing with, else we don't know what you are talking about.

 

[john101]

Sorry, I meant your post: "Suppose you have an 8Gig swap file and you create an 8Gig partition for it. How is the SSD NOT going to write to the exact same place every time?"

edit add:given that the 8gig partition is the 'exact same place'. A 4k file using leveling is written to that 8gig but to different places on the 8gig.

edit add2:

options: turn off swap entirely. maximise RAM. put swap on another drive, hdd or ssd.

all of which, depending on what programs you run and how you use the computer, have their own issues. For me maxing ram and putting the swap (Linux) on a separate HDD is probably a way to go.

ATM I have max ram for the MB and the swap set by linux installation on the main SSD with a couple of, 1 and 2Tb, external mechanical drives for storage.

edit add3: also: maximise the swap space so that the levelling technology has more space to work on. I understand there are limitations to this. 16 g for linux. Don't know about windows.

 

[rbelli1]

How is the SSD NOT going to write to the exact same place every time?

https://en.wikipedia.org/wiki/Wear_leveling

BoB

 

[phinds]

https://en.wikipedia.org/wiki/Wear_leveling

BoB

You are not paying attention. Reread post #28

 

[Vanadium 50]

:given that the 8gig partition is the 'exact same place'. A 4k file using leveling is written to that 8gig but to different places on the 8gig.

Swap partitions are not swap files. When a swap partition is activated, the SSD has no idea that it's data is out of date. So it can't just start from scratch and remap the whole partition. Furthermore, typical segments swapped are smaller than the SSD cell size. That also runs the number of writes up. Basically, file systems are smarter than plain old partitions.

 

[john101]

Swap partitions are not swap files. When a swap partition is activated, the SSD has no idea that it's data is out of date. So it can't just start from scratch and remap the whole partition. Furthermore, typical segments swapped are smaller than the SSD cell size. That also runs the number of writes up. Basically, file systems are smarter than plain old partitions.

In order to try to understand what you just posted I'm reading/bookmarking this : https://www.linux.com/news/all-about-linux-swap-space .