Mark Pickavance tries to build a console-beating PC while celebrating the Pentium's 20th anniversary
Almost exactly a year ago, I built a SteamOS PC from an AMD APU processor and a H75 chipset motherboard. I recall thinking at the time that with the addition of a decent video card, the system might be comparable with the latest-generation games console, at least on paper. Indeed, if you ask most people why they bought a games console over a PC, it certainly wasn't because the games were cheaper!
There is a general view that the PC, if specified correctly, is actually a superior gaming experience to the console, but it's hampered by what a system like that might cost.
At the time of writing, the PS3 is £329 with a single controller, and the Xbox One without Kinect is exactly the same price also with a single controller.
Obviously, there are other deals around that bundle games and extra controllers, but £329 seems to be the magic number these days. However, when both these consoles launched, they were closer to £400, so I'm not going to beat myself up too much to get down to that level - not least because most buyers don't own any games before they own the console, whereas most PC owners are likely have titles already.
Also, to use the Xbox One, you'll need to fork out for a Live Gold code for the first year - another £39.99. The PSN isn't compulsory like Live, but it costs about the same. Factor in a second controller, which almost everyone buys, and you're back to sub-£400, unless you crave the Kinect, and that's another £130 on top.
After much consideration, I set my target at £400 for a working system that could generate the same level of detail or better at 1080p in a selection of games compared to the console versions. With most console games that equates to 60fps, below which things can get a little choppy.
With my £400 budget in mind, I set about building the very best games PC I could, in the hope that when it was built I'd have something that could at least give those consoles a run for their money. But to achieve this I needed a radical plan, one that didn't involve stealing the parts.
The Planning Stage
I'd be the first to accept that as a tech journalist I do get a rather skewed idea of what a PC build costs, on the basis that I had most of the bits for this one on the shelf.
Not having to pay for items definitely makes parts cheaper, even if it's slightly delusional. Therefore I decided early on that what I'd do was use a part that I had handy but price up an equivalent if I was forced to source a part for hard cash. This might seem like a cheat, but frankly I'd be happy to test the exact bill of material for this computer against what I actually put together, and as they both have the same CPU and video card they'd be amazingly close performers.
Because of this, I've ended up with probably a better CPU cooler and better memory than I really needed, but that was for my financial benefit more than it was to suggest you use the same.
That point made, all good projects start with a critical part, and in this computer there are two of them. I've remarked recently in the magazine that Intel seems content to keep the cost of its processors artificially high, as AMD doesn't currently represent much of a challenge to its desktop product range.
At around £150 for a good Core i5, that chip would consume 40% of my budget, so it's out of the question. What I needed was a processor that's much cheaper but which can be tweaked to get extra performance out of it. The obvious candidate for that is a curious processor that Intel launched to celebrate 20 years of the Pentium, the amazing Pentium G3258.
On paper this looks a bit of a dud. It's only dual core, doesn't have Hyper-Threading or much in the way of cache memory. Normally, that's a combination that would keep gamers heading in the opposite direction, even if it does have a 3.2GHz default clock.
Still, it has one feature that none of its Pentium brethren can match: it's completely unlocked for modification, and as one of the Haswell generation, it's also been fabricated using a 22nm process - the perfect ingredients for some fine old-fashioned tweakery.
That's one part of the recipe, but if I insisted on using the Intel HD Graphics GPU in the G3258, then this wouldn't be much of a gaming experience. Luckily, because the Pentium only cost £50, there was plenty left in the kitty for a decent video card too.
At various points in this project, it was going to be an Nvidia GTX 750Ti, and then I realised that most of the cheaper cards of that series don't have DisplayPort. As this is a technology that I'm slowing moving to, in the end I went with the even cheaper Radeon R7 260X. That has the ports and power to drive a game very fluidly at 1080p.
The case, PSU, cooler, memory and storage were almost secondary considerations to the CPU and GPU showboat, and the chosen motherboard was a LGA 1150 one that I had unused.
Because of that, I've built two shopping lists (see below): one that I actually used and another that I could have, had I started with absolutely nothing. As you can easily see, there's a pretty huge difference. And, you could make the revised list even less if you used a stock air cooler for the processor. I put £40 for the Ballistix RAM as an arbitrary amount, because the ones I used you can't buy any longer. The conclusion must be that if you exclude the cost of the OS and some peripherals, you can build a PC for console money, but will it perform like one?
Before I can answer that question, I need to build it and then configure it to get the very best performance out of the hardware.
The Building Phase
All my PC builds follow a very similar pattern these days, as I've worked out the most efficient approach to assembling the parts by now. The first job is always to strip the case down so I can access all the things I need for mounting the motherboard and running cables.
Once that's complete, I mount the CPU on the motherboard and mount that, because it's easier to get the processor in the socket away from the case. What I don't generally do is mount the cooler at this point, because it can make working around the motherboard difficult in respect to the power supply.
In this design I've decided to go with a closed water cooling system, not for any other reason than I'm going to overclock the processor, and it helps in that endeavour.
What I'll say about that choice is that with the G3258 it isn't a necessity, because it's possible to overclock this CPU by some way purely on air cooling.
Therefore, if you want to build this system even cheaper, you could remove this item and use the stock item, saving yourself at least £50 or more.
Equally, the case I used isn't the cheapest, and neither is the main drive or the PSU, so there's room to either reduce the costs even further or redistribute the expense to get a better GPU or a CPU that doesn't need overclocking.
The case I've chosen is the excellent Corsair Carbide Series 100R. I happen to have the 'Silent' model that costs £53, although it makes another version with a clear side panel that's only £44.50.
The motherboard was an ASRock Z97 Pro3, though anyone wanting to do this more cheaply should use a H81 platform.
Normally in these features I'd detail the exactly build method, but there isn't anything special about putting this system together other than the fun I had with the H80i GT water cooler.
Unless you've got a case with a specific place designed to take the radiator, then finding the perfect place for it can be something of a challenge. Corsair suggests you place it over the rear fan vent, but there just wasn't clearance over the CPU to make this work. Ideally a ventral position would have been perfect, but there is no vent or space on this case for that. In the end, I did something counter-intuitive and placed in between the internal and external bays, so it could suck cool air from the outside. The disadvantage of this is that it now pushes warmed air in towards the video card, although I retained the rear fan to expel it from the case.
If you don't use water cooling, then this is a very simple build that would take anyone with experience probably less than an hour to complete.
When it was all working, I installed Windows 10 preview edition, purely because it wouldn't keep asking me for a valid licence code while I worked on it.
Tweak Town
Once you've built a PC, the first thing you need to do is make sure it works as is, before you do anything else with it. This part is crucial, because there's nothing more infuriating that trying to overclock a system that has an inherent fault from the outset.
There are numerous free tools you can use to do the soak test, but I'd recommend Prime95 (mersenne.org) as a straightforward torture testing device. If the PC can run this straight for 24 hours, then you can be reasonably confident that it's a stable platform on which to build.
I'm old enough to remember some of the frankly insane things that you needed to do to get chips over their specification in the past, but these days it's all rather civilised. That's because instead of trying at every opportunity to a scupper you, the chip makers have realised that the enthusiast community are good customers, especially if they think they're getting something for free. Now, both AMD and Intel make processors that are 'unlocked', allowing some creative alterations to be made that can improve performance, sometimes by disturbing amounts.
Intel processers that are unlocked are denoted by their 'K' classification, and those are the ones that are most prized by enthusiasts. Usually they cost a bit extra, but the object of my intentions here, the G3258 is about £50.
These days, I've generally matured in my attitudes, and ragging processors till their pins glow red isn't something I commonly do. But hey, it's the 20th anniversary of the Pentium, so it's time to unleash the demonic hell-hounds of thermodynamics.
The G3258 is a 3.2GHz dual-core CPU without Hyper-Threading, which on the face of it doesn't seem a great choice. That said, what we're fighting here is heat, in that realistically the chip will work well until it reaches about 100°C. So nothing unfortunate happens, I've put an arbitrary cap of 90°C on proceedings, and we'll see how far the chip will go before it gets that toasty.
What I have on my side is numerous things, not least that with just two cores there isn't much surface area to generate heat, and plenty of chip surface to dissipate it. This is also a 22nm fabrication, and I've got the H80i GT water cooler sucking the heat away from the CPU.
On paper, this sounds like a perfect combination of circumstances, but how fast can I get the 20th Anniversary Pentium to go, and will it actually be worth the effort?
First port of call on this mystery tour is 4GHz.
4GHz Or Bust
On some chips, throwing them from 3.2GHz to 4GHz might be considered excessively optimistic, as it's 25% faster than stock speed. But my research showed just how adaptable the G3258 is, and 4GHz is just a short paddle in the greater ocean of tweakery beyond.
Pips Squeaking
Getting the G3258 to 4GHz isn't at all difficult, because it's patently well within it's true performance envelope. Even though I didn't leave the system at this speed setting for long, I did a few benchmarks to access what the real benefits were. This is a critical part of the exercise, because it’s not unheard of that speeding up a chip can actually make it slower in some tests.
I wasn't expecting this to be the case, and in general it wasn't. However, it's worth pointing out that when I ran 3DMark Fire Strike, the score only moved from 3361 to 3591 - less than 7%. Therefore, in that test, pure clock speed is only a portion of the whole story, and other factors are in play.
To reduce the impact of the video card in the equation, I ran that old classic Super PI 1.5 4M, which took 43.031 seconds versus the 58.643 seconds baseline 3.2GHz level. That's 27% more computing power, on a 37.5% higher clock. However, going higher than these speeds seemed destined to frustrate me.
The system posted fine at 4.5GHz with sufficient Vcore voltage (1,3V), but Windows was highly unstable, and most of the time would crash before it fully booted. The obvious choice was to hike the voltage to get stability back, but that proved not to be the magic bullet in this case. I was also getting close to 1,36V - about the most I really wanted to use.
Eventually, as suggested by someone else, I tried enabling Internal PLL Overvoltage, and amazingly that was the critical factor in getting my G3268 to 4.4GHz. I was actually a bit disappointed with this, because I've seen lots of reports of people getting to 4.7 and even 4.8GHz on air cooling. My chip was a little disappointing in this respect, and I can only conclude that this specific processor on this motherboard wasn't a perfect marriage for some reason. That said, a 1200MHz increase in clock speed (37.5%) isn't that bad, even if I'd have liked more.
What was slightly sobering, however, was that for the last 400MHz, a 10% clock boost, I managed to increase the Fire Strike benchmark from 3591 to just 3663. That's only a 2% increase in video performance, suggesting that the 260X was approaching its limits. Comparing a PCMark 8 Home test, which went from 4637 to 4878, a more than 5% increase, supports that view. But it also shows there's more to performance than clock speed in this instance.
Costs And Conclusions
When I actually put together the parts list and priced this computer, I was a little shocked, as it wasn't anywhere near the low cost I was looking for.
It's reasonable to expect that a £500 PC should easily be able to beat a £329 console. But my reliance on off-the-shelf items has actually made this computer a good bit more expensive than I'd intended.
Using the cheaper 100R case, an alternative motherboard, 4GB of RAM, a refurbished cooler, a slightly smaller PSU and a cheaper Kingston SSD, I knocked off a massive £150. If I'd dropped the water cooler altogether, then it would have been cheaper than either of the games consoles, and by a decent chunk. Probably enough to buy an OEM licence of Windows 7, which I didn't include. Many people could easily run SteamOS, which is free, or another flavour of Linux.
If you look at the performance levels that I was getting out of the R7 260X, this is a much more powerful system than either console and capable of higher detail levels and higher resolutions. This system could also use Eyefinity multi-display mode or one of the new variable frequency monitors supporting AMD Free Sync.
But there's another story here, and it's the G3258, an insane piece of silicon that can overclock to almost unprecedented levels, even if the one I own seemed a little shy.
If you think 4.4GHz is impressive, then you should search the internet for the lunatic who through the use of extreme cooling technology managed to get it to 6.7GHz! Obviously, that's not a practical exercise to run on a daily basis, but it shows how paring down the system to two cores and controlling the voltages can yield spectacular results, in theory.
These things said, it's easy to get carried away and forget that what I managed to do was use tweaks and water cooling to drive a cheap chip up to the level of a medium performer. For another £50, I could have used a low-end Core i5 at stock speeds and got more real performance.
As an aside to these experiments, and another reason I used Windows 10 preview, was that I ran some tests using the new DirectX 12 and also AMD's Mantle API. For those unaware of these technologies, they aim to get the GPU working to maximum potential by reorganising how the API drives the video card.
With this functionality in place, at 3.2GHz the 260X using DX12 performed over seven million draw calls per second, compared with just 700k on DX11 multi-threaded.
And at 4GH,z it shifted a stunning 9.2 million calls compared with 890k on DX11.
The hint, if you didn't get it, is that under Windows 10, the performance of all video cards is likely to improve massively, once game designers support the new API.
However you get there and even before DX12, it is possible to build a PC that can easily outperform a console for around the same money, but if you take the G3258 route, it might involve more work than just assembling the parts.
Warnings - Please Read!
In this feature, I'm exploring taking a chip and motherboard beyond their known specifications, and that comes with some penalties that need to outlined.
When you overclock a component, there isn't any guarantee that it will work correctly, you're probably shortening its life, and it might fail completely.
Doing this, I've broken equipment in the past, because unpredictable things can happen when a chip gets very hot or more voltage than it was intended to handle. Therefore, if you can't afford to replace your processor, motherboard and PSU, then simply don't do it.
Experimenting in this fashion can also damage your OS installation, so never do it on a system that you're using to work with. And in a worse-case scenario, you could start an electrical fire, with all the negative consequences that could follow that scenario.
Because of this, never experiment without some suitable (not water!) means of extinguishing an electrical fire handy, in a well ventilated space and ideally with someone else handy if things get really out of control.
Take your time, and never leave an overclocking system unattended when you're torture testing it.
We're not the Mythbusters, looking for an excuse to blow things up. We'd much rather that our and your experiments end without a YouTube disaster video. So please, be safe!