Long story short, my laptop burned out and I am taking that as a sign from the gods that it is time to build a new computer.
I am going the desktop PC route this time- hopefully it will be upgradeable in the future and easy to swap out parts if something should burn out again. I want it to be sort of future proof, so I am willing to shell out a little extra to get something that hopefully will last.
I was looking at CPUs. The fastest one seems to be the Intel Core i7-4790K Devil’s Canyon Quad-Core, clocking in at 4 ghz. But, not so fast! (see what I did there?) Talking to friends who know more about these things than me reveals:
Clock speed isn’t really the measure of a CPU anymore. Chip designers are making chips that do more per clock cycle, so a “slower” chip might actually do more work adding this consideration.
There isn’t much point to having 4 cores (as found in the i7). I want to build a nice gaming PC, one that will be able to handle the next iteration of VR (and perhaps some unknown future application that I will just Have to play with), and people tell me that 4 cores just don’t come into play for those applications, and that I might as well get a good i5 chip and save myself some $$.
So, are the i7s a waste of money? How do I judge which is the best chip, without getting something that I really won’t use?
The i7-4790K is already replaced by the i7-6700k Skylake CPU. It is arguably the fastest CPU that is available to consumers, but you could argue for the significantly more expensive Core i7-6950X is faster.
You’re right though, that spending more on a CPU isn’t a good investment. This is especially true on a gaming PC. An i5-6600k with a Nvidia 1070 is an excellent cost-conscious PC build these days. Treat yourself to an mITX system too. Nice and small without any real compromises these days.
While the i5s are very capable CPUs the i7s are not a waste of money. Its HTT works well and applications benefits from it.
I have an i3 CPU that has HTT and when I upgraded to an i5 I did not see much improvement. The i3 runs very well that I built a budget PC to keep the i3 CPU for backup.
There is a point to having 4 cores. Some games run better on 4. Any decent CPU will have at least 4 cores anyway.
If you get a k-series CPU, you may wish to have a water cooler or a good air cooler. Otherwise, a locked i5 14nm intel should be good.
I agree with the 1070 suggestion.
I wonder about the mITX. Is it common for people to be really all that bothered with an ATX or EATX case unless they have to lug it around?
I was thinking of holding out for an Nvidia 1080, since I am going to wait to bite on a VR system until the next iteration comes out. But that is not set in stone, I am not sure how much relative difference there is between that and the 1070.
So, 4 cores: yes then? Harrumpf, I am getting conflicting info.
Re: cost consciousness, I am more wanting to dodge diminishing returns than to get a budget machine. I will spend a few extra $100 for 50% better performance, 25% even. I want to avoid spending 50% more for a 5% improvement though.
Yeah, the i5 and 1070 is that point. You could spend less and get better performance-per-dollar, but an i5/1070 is basically the maximum performance before prices get insane. If you’re going to sit for a bit for a VR refresh, you may as well wait for a GPU refresh too. GPU refreshes determine the best time to buy a PC.
The support chip set is very important. The front side bus speed and if the memory can support that speed. The disk input output speed.
If you are going for wild speed. Get the fastest FSB ( front side bus ) and stuff it with memory that can match it. Use solid state disks. Make sure your disk input output system is up to that speed. ( it won’t be, but do the best ) 12 Gbps is the latest I think. But that is some expensive, RAID with SSD and SAS I/O.
I just learned all that stuff in the last couple of weeks, when I got saddled with work above my pay grade. But if you want bleeding edge, it is true.
Fast and lots of ram and fast disks. Solitary TByte solid state disks can do 520 Megabytes per second. Get SATA III at least on your motherboard.
I upgraded my I7 4 core ACER laptop, to 32 Gigs of ram and a solid state disk for the OS and another 1 TB SSD. It very noticeably increased the speed of processing 5 to 11 GB of 5 to 7 thousand separate files. Windows 7 boots in the blink of an eye. But oddly, it still takes a long time to shut down.
If you open up the resource monitor while you are running an application. You can usually see that system idle time is often the biggest percentage of your system time.
That is due to how much slower so many other things are, compared to your CPU processing capabilities.
Often memory is being used up to hold things until other transactions outside of the processor are being handled. Some things get bumped back and forth to even slower disk storage. The pipelines to those places are backed up. The processor is waiting to get room to put another thing in that pipeline. You have way more processor speed and power than resources speed to keep up with it.
Often, getting the latest, greatest, fastest processor, just means it is doing nothing, faster.
It is far more expensive and physically bigger to try and feed that processor all it can handle. If you can even assemble all the hardware in the right way.
I currently have tons of processor power and very fast disk in/out. But I can’t seem to get my network card to do what it is supposed to do. So it is all backed up.
I think for my purposes the Skylake i5 3.5 ghz is the best choice. The savings would be best put towards a good mobo and other hardware.
Which motherboard? I dunno. I am looking at this one, the ASUS ROG MAXIMUS VIII HERO, though this isn’t a Need Answer Fast kind of question. It supports DDR4 3600 and has an advanced audio chip which I may want to play with down the line.
I don’t think I will need 4 SSDs or 4 monitors and so on. I see some motherboards that support overclocked DDR4 up to 4000, but I don’t think I want to spend the money for everything it comes with. I need to do more research before I can decide.
It looks like USB 3.2 supports 16 mbps, but so far I do not see a single motherboard that supports that, they are all 6 mb/s. And, how much difference does it make to have 2 SSDs…?
Even if they didn’t, it’s nice to be able to have several CPU hogs running concurrently. There’s no law that says you have to use all four cores for the game you’re playing.
A lot of programs are run single threaded despite the number of cores. This is where the HTT shine.
For motherboards I like Gigabyte. Asus is another good brand and surprisingly MSI are popular despite being cheaper.
I built my i5 system in 2012 with a Gigabyte B75 motherboard which is very durable and an Ivy Bridge 3570 CPU. I’m not a gamer so this is more of a business grade with longevity in mind. I don’t like onboard video and audio so I have a Radeon HD 7770 video card and a SB Audigy sound card.
Every motherboard is going to be stable and perform adequately. You may get burned on a design flaw or manufacturing defect, but that’s as likely to happen on a $300 board as a $60 board. All you can do is check reviews to see if a bunch of people report the same specific issue.
A good starting point is to pick a budget-oriented 4-phase motherboard. More phases are usually better, but modern CPUs consume so little power that it doesn’t matter as much. It would if you were planning to heavily overvolt your CPU, but you shouldn’t do that. Such board are usually about $100. From that starting point, only move up motherboard tiers if they offer features that you want. I don’t mean that you think something sounds good; I mean you can actually articulate how it will help.
Useful features to me include: diagnostic LEDs to make boot troubleshooting a lot easier, m.2 slot for an SSD, SLI (maybe), PS/2 port for nkro, dual BIOS so you can flash updates with less fear, CPU socket clearance if you go mITX, onboard start/reset buttons, some other stuff.
Palooka has a good point. When you buy something look to the reviews and look for a run of a bad batch. This is the time period not to buy said product.
I’ve researched everything I’ve bought online and have yet to get a bad product. Reviews do help but read between the lines.
As you will have worked out - there is no simple answer. There isn’t really a simple sweet spot either. That moves with use case along with everything else.
Future proofing is hard, and IMHO largely a waste of time and money. It isn’t as bad as it was, there was a time when it simply wasn’t viable - your shiny new machine was already obsolete by the time you got it home.
You will find all sorts of evil issues that make things messy, and make it difficult to make comparisons. One thing is very clear - there is no one “fastest” CPU. Different computation jobs can have significantly different performance on apparently only slightly different CPUs.
Multiprocessing is only slowly improving in many codes - the reality is that for many jobs is is hard to make them parallel in a useful manner, and so they chug along single threaded. Gmes however should parallelise, and a reasonable expectation is that things will improve here. However there are hidden issues here. A typical die has say 4 CPUs, and three levels of cache, with at least the level 3 cache shared between the CPUs. Cache is really what makes your programs run fast. There are codes which simply get fasger the more cache you can give them. So much so that there are major top-500 supercomputers that are built out of i7 chips and have three of the four CPUs disabled on each die. The job works best if there is no competition for cache, and letting the other CPUs run actually slows the job down. Hyperthreading is a similar issue. For jobs loads that have lots of short actions, often punctuated by IO, and which have lots of parallelism (and thus lots of jobs ready to run) hyperthreading can work well. For compute bound jobs it is a net drain, and you get better performance if it is disabled. On a desktop PC with a mixed job load it might or might not be a net gain.
As alluded to above - the major part of future proofing can benot the CPU but the motherboard - and in particular the controller chip it uses. That determines the speed and amount of memory you can use, and determines the speed and number of peripheral busses. An SSD based machine is a no-brainer. Here I would ensure that motherboard supports M2 NVMe SSD, and not just M2 SATA. NVMe makes a big difference.
The point about spending 50% more to get 5% more performance is about right. It is easy to find oneself doing exactly that - especially buying the top of the line bleeding edge processor.
You can go with 1 SSD. I chose to have a half TB M1 SSD for the OS drive. That left me with two slots for normal SSD drives. Then a 1TB SSD for separate storage. When doing a certain processing job, I bring in up to 11GB in files and process them in a session. But there may be a lot of sessions. So I bring them into the OS drive and then process them out to the other drive.
