1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Build Advice A Newbie's Hardware Guide (Updated 6/4/2014)

Discussion in 'Hardware' started by Dae314, 13 Jul 2011.

  1. slaw

    slaw At Argos buying "gold"

    Joined:
    13 Feb 2003
    Posts:
    1,050
    Likes Received:
    40
    Wow this is very comprehensive. I will book mark it now
     
  2. Dae314

    Dae314 New Member

    Joined:
    3 Sep 2010
    Posts:
    988
    Likes Received:
    61
    Guide Updates (1/8/2013):
    1. Links updated
    2. End of life products removed or replaced
    3. NVIDIA 760 added to Section 5 recommendations
    4. AMD FX-9370 and AMD FX-9590 added to Section 2 recommendations
    5. Seasonic FL2 series PSUs added to Section 6 recommendations
    6. Removed Section 13 sound card recommendations due to lack of content
    7. Added UPS information to Section 13
    8. Edited information in Section 12 Keyboards
    9. Added detailed information to Section 12 Mice
    10. This guide is now licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License. If you possess an out of date copy of the guide, please update your copy to this version or later.
    11. Full disclaimer added in new Section 18

    There were some major non-technical updates this month. Foremost of which is the fact that I've decided the license the guide under a Creative Commons license. Full details are now in the guide for those who are curious.

    The upshot of the license is that you may make copies, edit, and share the guide as long as credit is given, it is not used for commercial purposes, and you make your edits available under a similar license.

    I also did a huge overhaul on Section 13 and completely removed the sound card section replacing it with a section on UPSes and I almost completely rewrote the mouse section to cover more details regarding the technology.
     
  3. Dae314

    Dae314 New Member

    Joined:
    3 Sep 2010
    Posts:
    988
    Likes Received:
    61
    Guide Updates (15/9/2013):
    1. Links updated
    2. End of life products removed or replaced

    After last month's excitement I felt that a smaller update would be good.
     
  4. Dae314

    Dae314 New Member

    Joined:
    3 Sep 2010
    Posts:
    988
    Likes Received:
    61
    Guide Updates (2/11/2013):
    1. Links updated
    2. End of life products removed or replaced
    3. General update and grammar check for all content sections
    4. AMD R9 and R7 graphics cards added to section 5 recommendations
    5. Section 14 builds updated with R9 and R7 graphics cards where appropriate

    Sorry I skipped last month's update, I was a bit distracted with some life events (and a little bit of gaming). The past two months saw some major changes in the graphics area that warranted a full of the content sections (if I'm doing one why not do them all right?).

    Enjoy!
     
  5. PocketDemon

    PocketDemon New Member

    Joined:
    3 Jul 2010
    Posts:
    2,107
    Likes Received:
    139
    Having never looked at this before, i've just had a quick check of the SSD section & there's a reasonable amount where either you've gotten the wrong end of the stick - & so are mis-explaining things...

    ...&/or what you've written is outdated - h/w &/or s/w, along with costs/consumer expectations, having moved on from the premise that you were writing under; though some of it was outdated in July 2011.

    (if you want, i can have a look at either making notes or drafting a rewrite at some point)


    i'm also not overly keen on the SSDs you've chosen being lumped together like that as they give vastly different performance & so the differing price points have no hint of an explanation.

    Then, whilst i don't think that you're recommending entirely the right drives overall -

    1. the M500 *really* shouldn't be in there at anything below a 480GB capacity (& even that's not spectacular), as they're really shonky for the money.

    This is esp relevant since you've made the (very outdated) suggestion earlier that "(the) high cost presents a problem for even the most affluent system builders, so it has become almost standard for SSD users to use a small (i.e. 50GB-120GB) SSD as a boot drive" - this 'may' be true for a low to mid-end build, but it isn't the case for a high end one.

    (just to note - i'm hardly rolling in cash, but was using 256GB (nand capacity before OP, etc), albeit split between 2 drives in R0, in 2009 // adding another 256GB in 2010 // &, naturally, replacing everything since then - currently having 1280GB of nand capacity in day-to-day use between 2 machines)

    2. the M4's EOL so that sensibly needs to go as well

    3. & the 840 (non-pro) has been completely outclassed by the Evo for a couple of months now &, as they're much the same price, you're recommending the wrong drive there.

    Overall, given that the prices have dropped considerably, whatever SSDs you choose sensibly need splitting between at least the mid range & high end builds based on capacity & performance.
     
  6. Dae314

    Dae314 New Member

    Joined:
    3 Sep 2010
    Posts:
    988
    Likes Received:
    61
    Honestly, I wondered a little myself whether the SSD section was completely up to date when I was going over everything. However, due to the utter lack of SSD news coming into my normal sources, I decided not to pursue the matter further (I had the rest of the guide to grammar check >.<)

    Regarding the lumping, that was done a while ago because the SSD recommendations section was getting a bit out of hand length wise. A good remedy for this I suppose would be to simply recommend fewer SSDs.

    Regarding cost, I still see that 500GB or greater SSDs are rather prohibitively priced. 500GB is usually the minimum a regular user expects out of their storage, and it seems like most people are still going for the 256GB or lower models. Most of the builders I've talked to have builds with a mixed SSD/HDD environment, so I think the cost statement still applies. When I see more mid-range builders doing a full SSD build, I'll take out/change that statement. It would be good to put in a statement about the long running trend of falling prices though.

    I believe the M4 is in there because it's still showing up in the search results. Newegg's SSD search isn't the greatest and the M4 was a good drive. I had to refine the graphics card searches since Newegg completely changed that too, so maybe that's what I'll need to do on the SSD side as well.

    Thanks for the info on the 840.

    ---------------------------------------------------------------------------------------------
    Regarding the content section, you seem to have some up to date knowledge that I will need. Do you mind if I dip into that a little? It will save me some research time.

    I assume the problem of write amplification hasn't been solved yet since I don't see drives getting off NAND, but I have a feeling TRIM support is no longer an issue. What are the newest technologies you know of to combat write amplification?

    What is the current controller landscape? Along with that, what companies are still competing strongly in the SSD arena? I know OCZ is going down (already removed them) but what about Intel (haven't heard from them in a while) along with the others?

    What aspect of SSD technology is receiving the most research right now? I know from several months ago that there was a lot of research into alternatives for NAND technology, but are there any NAND areas that are being researched or has that all stagnated until a replacement for NAND arrives?

    Thanks in advance. Also, the above questions are open to anyone who wants to answer them since answers will direct where I need to do research.
     
  7. PocketDemon

    PocketDemon New Member

    Joined:
    3 Jul 2010
    Posts:
    2,107
    Likes Received:
    139
    All the words in the world.

    i'm wasn't actually saying that you should be recommending 480/500/512GB SSDs at all - just pointing out that the M500 is pretty shonky for the money until you get to the 480GB.

    The problem is that a 60/64GB & probably a 120/128GB SSD would now be starting to come in on a budget build - either instead of or in addition to a HDD...

    ...a half decent budget 120/128GB or 240/250/256GB (ie the Evo) would now be mid range...

    ...whilst a high end build would either be looking at something between a half decent budget & a top notch 240/250/256GB or 2 of them in R0.


    As to this bit, the easiest way to try & demonstrate what i was meaning was to look at 'the 'problem of write amplification' - as it happens to be an example of where you've gotten the wrong end of the stick, & simply isn't what you wrote it as being as such...

    Whilst trying to keep things as simple as possible, this is far more technical & contains far more info than you'd sensibly need to have in the guide - but i need to explain why your statement's wrong.


    So, in the guide, you've written that -

    "One problem with SSDs is write speed decay (this may also be known as write amplification). Write speed decay occurs on SSDs because freely programmable blocks, which the SSD uses to store data, become used up and locked as the drive is used. If there are no freely programmable blocks to use, then the SSD must first clear data from used blocks before writing new data to it."

    - however, the actual problems are those of (1) OS deletes not directly translating into the erasing of cells & (2) the fact that SSDs write in pages, but can only erase in blocks (leading to block fragmentation) - with a slowing of write (or read speed, depending on the controller) & write amplification being symptoms.


    A. Problems.

    1. The first of these is where trim can help - since, assuming the OS, controller driver, controller & SSD(s) can all send/pass on/act on the trim command, this informs the SSD that an OS delete has occurred so that the SSD can more quickly pre-erase any blocks where there is no useful data...

    ...however, whilst it's generally less efficient (the SSDs using SandForce (SF) controllers being, to a reasonable extent, something of an exception), in a non-trim environment garbage collection (GC) is used to do the same thing.

    2. With the second, on the SSD level, this is down to GC to try to sort by combining pages from blocks with dirty pages in so that more blocks can be pre-erased - however there are s/w options which can improve things.

    On the prevention level, for a few years there's been the intelliwrite part of Diskeeper (DK) & ttbomk, PerfectDisk (PD) added in a similar function a year or so back - in both cases, these act by combining writes such that there is less chance of block fragmentation occurring in the first place.

    Whilst on the solution level, both DK (since Hyperfast came in in Dec 2008), PD (first with a scheduled option, but now acting automatically like DK) & the Windows 8 defragger itself (a scheduled option) can all enhance block combining.


    in both cases in how aggressive a SSD acts on the trim command & runs GC depends upon the algorithms used in the controller's f/w.

    [NB GC has been used in SSDs since at least ~Q3 2009 so significantly predates your guide - albeit that whatever the current intel drive at the time was (the G2?) didn't get either it or trim added via a f/w update, & my recollection is that the only other drive released without it from that point on was a shonky JMicron one... but my memory's vague on it.

    Either way, nothing that anyone's remotely likely to buy today wouldn't have GC.]​

    * * * * * * *

    B. Symptoms.

    Moving on to looking at the symptoms you've mentioned, firstly, (1) the slowing of write (or read) speeds & (2) write amplification are, to a huge degree, independent of each other... Okay, they're both symptoms, but...

    [NB it's only the need to block combining & that some of the solutions will help both that perhaps gives a real link between the two.]​

    1. The slowing of write/read speeds occurs when there are insufficient free blocks in the partitioned area to write to - this being because block combining has to occur 'on the fly' - adding extra read--modify-write-erase steps before a write can occur.

    [NB with most SSDs it's the writes that slow significantly, however with the SFs (esp the 2nd gen ones) then the write speeds are pretty much maintained, but the read speeds slow...

    This has been the case since the 1st gen SFs in Q2 2010, whilst the 2nd gens were Q2 2011 - both predating your guide.]​

    Now, there are several things that can help here -

    (a) Maintaining a decent amount of free space - the more free space, the more blocks that will be free after block combining, when done ahead of time. For an OS, etc drive, 20% of the partitioned space is a good min point - though much less if you're effectively statically storing data on it.

    (b) increasing the OP by under partitioning the SSD - primarily, since the OP area cannot contain any data, increasing this means that (i) there is more space for GC to do its block combining & (ii) impacting upon the in a situation where there is the need to block combine on the fly as, in both cases there are more free blocks available.

    The recommendation for an OS, etc drive is increase it to 25% of the *total nand* in the SSD (remembering that 120 & 128GB SSDs both have 128GB of nand, 240, 250 & 256GB SSDs all have 256GB of nand, etc) - ie you partition to 75% of the relevant amount...

    ...again, there's no need to do this if you're effectively statically storing data on it.

    (c) Buying a SSD of an appropriate size so that you can do (a) & (b).

    (d) Deliberately allowing idle time so that GC has a better chance to do its thing.

    (e) Using (an up to date version of) DK or PD to help prevent block fragmentation from occurring in the first place.

    (f) & using (an up to date version of) DK or PD (or, i guess, the Windows 8 defragger - though it's less useful d.t. having to be scheduled) to force block combining to happen.

    [NB naturally these are all options.]​


    2. As to write amplification, this is caused by -

    (i) The need to block combine.

    (ii) That GC also needs to wear level (ie move data around periodically) beyond that which can be done solely as a part of block combining - otherwise you'd be constantly writing to a very ltd no of nand cells which would lead to them failing far more quickly than would be sensible.

    (iii) & with SSDs which have a SLC (if they use MLC nand) or MLC (if they use TLC nand) write mode, the rewriting of data from a faster nand 'cache' - basically, writing the same data twice (though at less of a longevity cost d.t. using an alt mode).


    Now, the only realistic way to eliminate write amplification is if the SSD's controller compresses the writes - which is what the SFs have done from the get go & they have a normal write amplification of <<1x (typically no more than 0.5x with the 1st gens & pushing down further with the 2nd with average workloads - though entirely incompressible ones will be ~1x) - meaning that less data is written to the nand than is sent to be written to the nand.

    [NB naturally the SFs suffer from slower write speeds when writing very incompressible data, but it's a trade off.]​

    Every other SSD will have a write amplification >1...

    [NB i guess, unless (with a non-SLC/MLC write mode SSD) you were to either solely write sequential data & never leave data on there long enough for there to be a need to move any of it for wear levelling, or initially write some data & then unplug the SSD forever; which would be highly atypical usages.]​

    ...with the write amplification increasing with all SSDs as writes become more random.


    As mentioned earlier, some of the solutions for the drop in write/read speeds will also help in this case, in particular -

    (b) increasing the OP by under partitioning the SSD - this makes by far the largest difference.

    (c) Buying a SSD of an appropriate size so that you can do (a) & (b).

    (e) (&) using (an up to date version of) DK or PD to help prevent block fragmentation from occurring in the first place.


    Otherwise, the block combining options *can* help to some extent, but only in a situation where you've got bunches of temporary data on the SSD that's going to be deleted in short order &, without it, the SSD's having to block combine 'on the fly' in order to write new data...

    ...simply that, solely in this situation, the SSD's being forced to move data that's very temporary into new blocks - which, when it's deleted, will lead to more blocks with dirty cells & so more block combining will be needed.


    * * * * * * *


    Anyway, if you've made it this far then hopefully you've understood what i was meaning about things...

    So, quickly running through your guide -

    1. The high cost & limitations on capacity are completely over egged with the fall in prices.

    2. The defrag statement isn't correct as DK, PD & Windows 8 all have defrag options which are appropriate for SSDs.

    3. Everything in the section above + trim is by no means *the* solution to the problem as you've written.

    4. Raid arrays - the SFs have always been robust in non-trim &, with a sandy bridge (with a modded bios), ivy bridge (with an up-to-date bios) or haswell board + up-to-date drivers then trim is supported in R0 on the intel controller.

    5. Most, if not all, pcie SSDs do not have working trim/unmap (the scsi version of trim) support, d.t. a lack of driver support... Well, certainly OCZ's don't support it Windows, & both intel's 910 & Micron's P320h don't appear to support it at all.

    So, 'if' trim were *that* essential (as you've suggested), you then should be actively discouraging the use of them (at least in Windows) in the same way as you've discouraged raid.

    This is, however, 'a' reason why many pcie SSDs have used a SF controller...

    ...why the intel 910 has a huge amount of OP; the 800GB intel 910 has 1792GB of nand...

    ...& why the Micron P320h benefits so hugely from dramatically increasing the OP, as there's nowhere near enough nand on it & so it's 'settled in' write performance is far worse than it needs to be.

    6. & then limiting the SSDs to only the high end isn't appropriate anymore.

    Tbh, for most people, the sensible choice in the mid or high end builds are either a 840 Evo (ideally 250GB) or Pro (ideally 256GB) as they give the best performance for the respective budgets.

    Whilst i'd normally put the Evo at least in the mid range, it does need the caveat that, for situations where you're writing huge amounts of data in one go (>3GB for both the 120 & 250), the write speeds will temporarily (15-30 seconds after stopping writes) suffer d.t. exceeding the limit of the MLC mode...

    ...so, in that situation, another SSD should be chosen.


    Beyond them, i'd personally link to this & this as giving probably the best overview of relative performance for heavy & light usage that's out there...

    ...& letting people do their own pricing on the various options...

    ...though noting that 'if' they have an unusual usage, to look more closely at the various b/ms available; ie, for highly sequential usage (ie a/v editing or batch encoding) the AS-SSD sequential read & write b/ms would be the most important.


    As to upcoming tech, well, it's been stated by Samsung that they will be adding their 'rapid' memory cache mode (that's optional for the Evo) to the Pro with an upcoming Magician s/w update - making the Pro even better.

    [NB as the 'rapid' mode uses system memory, there's the potential to lose data that's not been written in the event of a power outage or whatever (though this can happen, albeit to a lesser extent) with the write cache enabled in Windows or with data in the DDR cache that all non-SFs have.

    it also does not currently work with more than 1 SSD - either as separate drives or in any type of raid array.]​


    But beyond that...

    - Well, sometime around the middle of next year there's SATA express (it appears to be with coming in with the Haswell refresh that's incompatible with the current Haswell stuff & Haswell-E), which will vastly improve speeds; if you can afford to upgrade most of everything to have a compatible system.

    - Charge Trap Flash nand already exists in the enterprise sector thanks to Samsung's V-nand & is likely to enter the consumer market next year... this increases nand endurance back to 35,000 cycles (vs 3-5,000 now) @ a 3Xnm process allowing for further die shrinkages into the future & reductions in SSD cost.

    (it's likely that other manufacturers will have versions of the same tech before too long)

    - There's been some talk of more pcie options (beyond the M form ones for notebooks & whatnot) from other manufacturers for a bit, but i don't recall anything concrete yet.

    - &, otherwise, the move to something like resistive ram has no timescale whatsoever.
     
    Last edited: 4 Nov 2013
  8. Dae314

    Dae314 New Member

    Joined:
    3 Sep 2010
    Posts:
    988
    Likes Received:
    61
    Unfortunately, I only just now thought to check the working copy of the guide I keep on my computer (this is where I do updates before pushing them to google). Due to my SSD failing, I wiped out the contents of my data drive and restored an OS image on top of it. Windows file restore was less than ideal for restoring the files I wiped out and the version of the guide it gave me did not contain most of the edits I'd been working on in the week leading up to the SSD failure.

    After skipping December to enjoy the holidays I had hoped to push out an early update in January but I doubt I'd be able to get anything out before mid-January. This combined with the fact that I'm going to wipe my data drive (and thus rely on Windows File Restore one last time) when the replacement SSD comes does not give me very much confidence in getting an update out this month.

    Please accept my apology for the inconvenience this may cause you. Most of the info sections should still be relevant however the SSD subsection is mostly out of date. This was the section I was making most of my edits in since I plan to peel it off and give it a section separate from hard drives. I'll be trusting my backups to Acronis in the future, so hopefully this won't happen again.

    Happy New Year everyone!
     
  9. Dae314

    Dae314 New Member

    Joined:
    3 Sep 2010
    Posts:
    988
    Likes Received:
    61
    Guide Updates (9/2/2014):
    1. Links updated
    2. End of life products removed or replaced
    3. SSD subsection split into new Section 8 with more detailed and updated information
    4. Section 12 updated with QD-LED information
    5. Kaveri APUs added to Section 2 recommendations
    6. Changed line spacing to 1.15

    The SSD subsection is brand new and probably contains some grammar, content, and/or wording errors. Please point these out to me. The same goes for the QD-LED section.

    The new R7-250X card is not in the guide yet because I wanted to get the update out tonight.

    I noticed this time around that some of the sections are getting a bit heavy on one brand. This happened inadvertently over time and is not meant to be fanboyism. I will try to address this next update.
     
  10. Dae314

    Dae314 New Member

    Joined:
    3 Sep 2010
    Posts:
    988
    Likes Received:
    61
    Guide Updates (6/4/2014):
    1. Links updated
    2. End of life products removed or replaced
    3. Intel 570 SSDs added to Section 8 recommendations
    4. NVIDIA GTX Titan Black added to Section 5 recommendations
    5. Antec High Current Platinum series PSUs added to Section 6 recommendations

    Sorry for dropping off the map there for a month. Lots of news of future changes happened while I was gone, but it appears that the hardware landscape stayed fairly stable. Keep that in mind if you're planning to build a new box.

    Unfortunately I didn't get to address the brand bias in the recommendations though :(.
     
  11. bartiszon

    bartiszon Active Member

    Joined:
    25 Aug 2010
    Posts:
    548
    Likes Received:
    50
    Very nice guide Dae.

    Imho the monitor section needs a small update. From 9 recommendations 8 are Dells, so it might be worth to put some light into their designation.
    S - stands for "Studio"
    U - stands for "UltraSharp"
    M - is aimed more at mainstream users and representing the use of a more standard colour depth (16.7m colours) with standard gamut.
    H - is now used by Dell to signify 16:9 aspect ratio screens.

    Taking the above into consideration I would say U2713HM should be in "Mid-Range Build", but for users who want/need a bigger screen, while U2713H definitely can stay in "High-Range Build".
    U2413HM ??? - there is no such model. U2413 with a wide colour gamut and 16:10 ratio is in my opinion aimed primarily at higher end users.
     
  12. Dae314

    Dae314 New Member

    Joined:
    3 Sep 2010
    Posts:
    988
    Likes Received:
    61
    Yes, the monitor section is one of those with some problems with brand bias -_-. I just haven't had the time to research more monitor manufacturers who produce reliable, decent screens.

    The tiering is based mostly on price as opposed to functionality or whatever the manufacturer aimed the product for. As in what would someone on a budget of $1000 to $1500 (mid-range) be able to buy? Although there are some pieces that are borderline.
     

Share This Page