Saturday 10 December 2016

Building A PC For The Future

Building A PC For The Future

Mark Pickavance talks about things you might well want to consider if you're considering building a PC in the next six months

There was a time that choosing the parts for a new PC was remarkably simple, because the choice was so limited. When Intel launched the i486, it only made two versions initially: 25MHz and 33MHz models.

These days, the opposite is true, and you're the confronted with a relatively huge range of processors, memory, GPU and motherboards that are widely available.

With so many options, it's difficult to know where to start. I created this small feature to show people preparing to build a new PC what things they might want to consider including and the things they could probably leave out. My suggestions won't fit every scenario, but they should help anyone building something new to strike a decent balance between what’s useful now and what could be ideal further down the road.

But first, let's talk about processors and the best choices for those who want longevity for their system.


The CPU Game


There's a problem with buying processors that stems from the very limited brand options the PC currently has. Specifically, it's Intel or AMD, as most processors made by other companies aren't X86 architecture compatible, and therefore not really a PC.

For a long time, AMD has been trying to compete with Intel, a company that on employees alone is about 12 times larger, and on revenue is about 30 times smaller.

While chips like the Opteron really demonstrated how far down the wrong path Intel went, since then, it's been all Intel if you want raw performance on a desktop system.

The problem with this dominance is that there's no pressure on Intel to reduce prices or advance performance levels significantly, resulting in a degree of stagnation. These days, we're lucky if we see a 5-10% performance bump for a new generation, and frequently Intel messes with the socket so you can't realistically use new CPUs on other boards and vice versa.

Because of how often it does this, I wouldn't think about a mid-term upgrade on the processor for any new PC I was building. Instead, I'd accept that the CPU placed in it at the outset will probably be the one it will end up with forever.

That does seem to fly in the face of having a socketed processor, but my experience is that when you want to upgrade, you'll find you can't, because Intel has done something to make it either impossible or impractical.

What's particularly irritating is that occasionally it won't change the physical socket, so its old chips will actually go on a new chipset, but they won't work for numerous reasons. So be aware of that.

Because of this. I'd really recommend you look at what Intel has in its Core i5 range when you come to build, because these generally offer the best combination of price and performance.

Unless you have modest requirements, i3 is underpowered, and i7 is not worth the price for the nominal improvement. You can occasionally get deals that might make the i7 affordable, but most of the time you just don't need all those cores and threads unless you're into heavy processing.

Buying today. I'd probably go with a Skylake-cored processor, mostly because compatible LGA 1151 motherboards are relatively cheap, and the Core i5 6500 can be found for less than £200. I'd avoid 'K' class overclockable processors for no other reason than if you want a PC to work for a long time, then just don't overclock it. I'd also avoid Haswell chips for no other reason than they're older technology that costs nearly as much as Skylake for the same core/dock.

When you buy a CPU, you need to think hard about the motherboard to go with it, because the two will only work well if they're considered in parallel.

Some readers might be wondering why I haven't mentioned the AMD Zen processor, which we'll see soon, allegedly. These might be wonderful, but so many times before, AMD promised plenty and delivered much less. It would be good for the PC industry if Zen was a chip that would fire back up the old Intel versus AMD antagonism, but more it's realistic to accept it probably won't. I so hope I'm wrong on this point, because Intel badly needs a big kick up the pants at this time, as much as AMD needs a little success.

Motherboards


While the brand options might be limited, the motherboard choices for the currently available chips are huge. Intel makes at least six platforms for its current LGA 1151 processors, some being more affordable than others. At the cheap end are the H110 and В150 boards, and at the top of that food chain is the Z170, which can easily be twice the price.

My advice is to buy the Z170, Q170 or H170 and not any of the others for one critical reason: PCIe lanes. In the cheaper boards, the chipset has fewer PCIe lanes, so the system is reliant on those that come with the processor.

That's fine until you use a discrete video card that needs x16 PCIe 3.0 lanes, and the system is left with virtually no lanes available for anything else. The importance of this can't be overstressed, because if you want the very best storage options (covered elsewhere), then you'll need those spare lanes.

If you wondered why many H110 motherboards don't have many slots, then the answer is that there just aren't the PCIe lanes to support them, and in some situations using a feature that uses lanes may disable some or all of the slots that they have.

Avoid this problem from the outset and pay another £50 for a motherboard with lots of lanes, like the excellent Z170, and don't get stuck with a system that has upgrade problems down the road.

The temptation, if you just got a windfall, might be to go for an X99 LGA 2011-V3 motherboard. Personally, I wouldn't. Few regular users of desktop computers need that sort of power, and everything to do with these systems is overpriced. Yes, they are the quickest systems, and they have the most PCIe lanes, but paying between £200 and £500 for a motherboard, and then at least another £400 to £1,600 for the processor is a mug's game.

Storage


They might be cheap, but hard drives are slow and use too much power. Any system you're building now should have an SSD, and only a hard drive if you want cheap extra storage. If your budget is tight, a 256GB SATA SSD is relatively inexpensive and can easily hold the Windows OS and a few important apps.

The current version of the Serial ATA interface is SATA-3. And the fastest drives on that interface can read at 550MB/S and write at 520MB/S, which is reasonably quick.

But it isn't the fastest by any standard, because that award goes to NVMe drives that connect directly to the PCIe bus. NVMe drives come either as a self-contained PCI Express card that needs a PCIe x4 slot or as an M.2 NVMe card.

To confuse matters, M.2 can also be used for SATA, where the same performance limitations as SATA-3, so be careful not to confuse the two. Motherboards can come with SATA-3, M.2 NVMe, M.2 SATA or even a new version of SATA called SATA Express.

The one that is getting the most traction is M.2 NVMe, and that requires for lanes of PCIe to provide the bandwidth needed for a single high-performance drive. There seems little evidence that the SATA Express connector that uses two lanes of PCIe is becoming popular or even getting support from drive makers.

Therefore even if you can't afford M.2 NVMe right now, getting either a motherboard with the M.2 slot on it or one with enough spare PCIe lanes so you can drop one in with a cheap M.2 card is the right choice.

The performance these drives deliver is up to six times, better than the highest-end SATA SSD you can buy, and they're also very low power consumers compared to conventional SATA drives.

Video Cards


This is a difficult area to give advice on, because each time AMD or Nvidia comes along with a new generation, they make most of their previous designs obsolete. I myself spent a decent amount on a GTX 960, only to have the better GTX 1050 Ti come along for less within six months.

Because of those pitfalls, I'd avoid spending any more than £200 on a video card of any flavour, because in less than four years the performance that costs you £500 now will be £150 then. And unless you're a 4K gamer, you won't notice much difference.

I also wouldn't consider using multiple GPUs as a strategic plan, because in this writer's experience if you don't deploy them from the outset, you'll never buy the second card at a later stage. When you come to the point where you want to add that power, you realise that you could put that money towards a new more powerful single card and sell your old one to soften the hit, and get general speed improvements.

It's also worth noting that the latest Nvidia cards only really work in two card combinations and that using multi-GPU mode doesn't always make a game or app faster by default.

Another consideration is that PCI Express, currently at version 3.0 will get a hike in bandwidth in 2017 with the introduction of PCIe 4.0. This jumps from a maximum of 8GTps to 16GTps (GTps = Gigatransfer), and it also has massive power reduction benefits that should help portable systems.

To support this, both Intel and AMD will be launching new chipsets, new processors, and also you'll need new video cards to work with it. But, it will be backwardly compatible with PCIe 3.0, so existing video cards should work well in PCIe 4.0 slots.

I'd moan that things change too often, but PCIe has been around seven years. With that information, those wanting to build the ultimate gaming system might need to wait a little longer, if they want something that exploits these new enhancements.

Useful To Have Along


When I considered writing this feature I thought about exactly what I'd put in a new PC if I built one today. After much chin scratching, one feature that I really would include is USB type-C, because in about five years I predict it will have taken over USB ports entirely.

You can get it on some motherboards already, but it's something you can also easily add with a card or even with a front-panel header plate.

Not that I would dispense entirely with conventional USB this minute, but this change will be necessary going forward. For this reason, I’d make sure any motherboard you buy has a proper 20-pin USB 3.0 header on it, for the front-panel ports.

Those We Leave Behind


If you have an ATA drive still, with some converter to SATA or even an IDE card, ditch it now, please! The performance of any IDE drive compared with SATA is utterly abysmal, and they use lots of power to do things slowly too. The same is also true of SATA-1 and SATA-2 hard drives, where the amount of bandwidth is very limited.

For exactly the same reasons, I'd dump any IDE optical drives, and on my latest system, I've actually dumped SATA optical too. I don't use discs often, and when I do, I have a USB 3.0 optical drive for the job. Having hardware hanging around the PC on the off-chance that I'll use it seems ludicrous.

USB 2.0 is also not something that you'd want to encourage in a new system, especially when USB 3.0 is backwardly compatible with its older incarnation. Those with a USB 2.0 card can leave it in their old PC, because a cheap hub on a single USB 3.0 port on their new system would be much more effective.

Examining many systems over the years, I've seen some crazy stuff still on a modern PC, like a modem. What's important to remember is that systems generally work better the less complicated they are. It's also much easier to identify what might be wrong if you do have a problem.

If you're transferring to a new flagship, it's a good time to perhaps leave as much of the legacy junk on that system as you can and truly get a fresh start. Unless you butcher it entirely, you can always fire it up to use that SCSI scanner or whatever, should you ever need that facility again.

Final Thoughts


The concept of future proofing is one we've explored in numerous articles in Micro Mart, and it comes with a substantial caveat. When you construct a new computer, you've no idea what revolutionary thing someone will come up with next that makes what you've got obsolete. And if the last 40 years of computing guarantees anything, it's that those developments will happen, and they will change everything.

Obsolescence is part of the personal computer ride, and we've all watched things we spent plenty on to be the best be relegated to less than scrap value in short order.

The fine irony is that at some point, probably not that far down the road, the PC as a concept will probably hit the buffers entirely, and the ideas that have been central to it, like the X86 architecture and the Windows operating system, will go to the knacker's yard too. But that’s fine, because if it were otherwise, I'd be typing this on Wordstar 1.0 on a CP/M running Z80 based machine.

If a PC can give you five good years use before you even think about moving on, then you've had some decent use out of it, and those that get ten out their hardware, I salute you.

If you have no intention of upgrading your PC ever, then you should really entertain the idea of a laptop, unless you're a gamer. Those who do like to tinker can stick to the desktop form factor, with a decent nod to those changes they're likely to make over its life. Having slots, bays and a case big enough to handle those changes just makes sense. Or it will when you come to make those alterations.

Being realistic, you can never cover all the possible directions that technology will go in, but you can avoid heading down the most obvious cul-de-sacs with a little research.