Howdy, Stranger!
It looks like you're new here. If you want to get involved, click one of these buttons!
Quick Links
New low budget recommendation? New unlocked Pentium Dual Core
AMD's 4/6-core FX CPUs have long been a popular low budget option for gaming rigs. THey have the right combination of "Enough Power" and "Low Price", with the added bonus that the motherboards also tend to be a little bit less expensive.
http://www.legitreviews.com/intel-pentium-g3258-dual-core-processor-gaming-performance_145873
The Pentium G3258 is a dual core chip, with no hyperthreading, so it's just capable of 2 threads period. That could be a deal-breaker depending on what software you are running.
The main attraction: This is an unlocked CPU (like the Intel K-series, and the AMD Black editions), and it's MSRP is under $75 US.
That being said, when the software doesn't need more than 2 cores and depends more on IPC, this chip performs rather well. Particularly when overclocked, and preliminary indications show that it overclocks rather well (4.8G in this review).
The benchmark results in the link above show some interesting data. Games that leverage 3+ cores definitely show a disadvantage, but there are a lot of games where that doesn't matter at all, and in those games this CPU is matching, and at times beating, Core i7 4770K's.
The main question is, regardless of the competitive IPC and the attractive price, is dual-core good enough for most gamers today, or do we really need quad+ cores? There is an ever-increasing number of software titles out there where having more than 2 cores does matter. I don't know that we can say that about 4 cores yet, but there will be a time in the not-too-distant future where that will come to pass as well.
Comments
I agree, but this is for people on very tight budgets. Obivously you get the best you can afford, but the question is - is dual core still an option today, or is it the floor below which is no longer even worth considering regardless of price?
You make a good point about the Lightforge engine.
The thing is, as I see it, by the time those games actually release more and more people should be able to afford quad core rigs as the price for quad core processors drop.
Very few games today utilize 4 cores.
I'm going to say no, or at least not for gaming. Let's suppose that you're looking to overclock it and these are the contenders:
http://www.newegg.com/Product/Product.aspx?Item=N82E16819117374
http://www.newegg.com/Product/Product.aspx?Item=N82E16819113286
You might intuitively say, hey, the FX-6300 costs $45 more. That's not a fair fight. That's what Intel is hoping that you'll do: compare a $75 Intel CPU to a $75 AMD CPU.
But you don't use a CPU in isolation. If you're looking to overclock the Pentium G3258, you'll need either a Z87 or Z97 motherboard. There are 183 such motherboards on New Egg. The cheapest is $85 with shipping. The cheapest Socket AM3+ motherboard is $40. Now granted, the latter is a piece of junk. But you can get a decent 970 chipset motherboard for $70 that will suit an FX-6300 just fine.
The third cheapest Z87 or Z97 motherboard on New Egg is $100. Even there, do you really want to trust a severe budget motherboard with a hefty overclock? With an FX-6300, you are, after all, putting a 95 W CPU in a motherboard built for 125 W CPUs, so there's plenty of headroom there. The Intel motherboards being more expensive isn't some fluke or motherboard manufacturer conspiracy; Intel simply charges more for Z87 and Z97 chipsets. So let's say that by going with the FX-6300, you conservatively save $30 on the motherboard.
Now let's look at CPU cooling. AMD ships decent enough CPU coolers with their FX processors, so on a budget build, you use the stock cooler. Again, a 125 W cooler for a 95 W CPU means you've got plenty of safe headroom. Intel's stock coolers are notoriously junk, of course. Yes, it will be fine at stock speeds. But with a huge overclock to justify the chip? Not so much. So let's say that you get a $30 aftermarket cooler for the Pentium G3258. Now AMD just saved you $30 on a CPU cooler, too.
The net result of all of this is that once you finish the build, the overclocked Pentium G3258 conservatively costs you $15 more than an FX-6300. So basically, you're paying $15 more for a dual core than you could have had a capable six-core processor for--and without the risks of overclocking damaging anything.
-----
Overclocking the Pentium G3258 to 4.8 GHz isn't realistic for consumer use. It's not just that huge overclocks cause electromigration. It's that buying a $350 motherboard, a $200 power supply, and a $100 CPU cooler so that you can more heavily overclock a $75 W CPU is doing it wrong. Now, I understand using top notch components so that you can see what a CPU can really do if you build around it. It's kind of like I understand liquid nitrogen overclocking for demonstration purposes. But that's not relevant to everyday use.
-----
Right now, most of a game engine either uses such a small fraction of one core as to be inconsequential or else is pretty easy to scale to as many cores as you care to. The notable exceptions are reading from a hard drive and using a graphics API to communicate with the video card. Let's ignore the former, as when that's a bottleneck, faster single-threaded CPU performance won't help it anyway.
So what about single-threaded rendering? API improvements are supposed to ease that greatly. Nvidia recently claimed that if you really build around it, you can reduce the CPU overhead from passing stuff to the video card by as much as 90%. I think that 90% is unrealistic for real-world usage, but reducing the CPU overhead of rendering by 2/3 might well be realistic. Basically, if you need to draw things using a given program with 100 different textures, it's unrealistic to expect that all textures will be exactly the same resolution. But having only 10 or 15 different resolutions among the 100 different textures is a lot more plausible. So what happens if games do that and overnight suddenly scale to three times as many CPU cores as before? Still happy with a dual core?
Nor is this off in the distant future, either. Mantle already has it. DirectX 12 will have it and is coming next year. OpenGL 4.3 added one API command to handle part of this two years ago, and OpenGL 4.4 another last year. The big one is texture switching, which is still only available as an extension, but will probably be made part of the core API soon--possibly with a release of OpenGL 4.5 at SIGGRAPH in a month.
Once this is available, what stops existing games from adding it to the graphics engine? I haven't tried it myself, but it looks like it might well be practical to tack onto an existing game just by adding a few hundred lines of code. To really do it properly, you'd want to only have a handful of different texture resolutions. Going back and altering all of a game's textures is not going to happen, of course; this may be the difference between reducing rendering overhead by 90% like Nvidia says is possible and "only" reducing it by 2/3.
But a few hundred lines of code to use three new extensions without altering a single shader or anything outside of the rendering thread? I'd expect more than a few games that are already out to do that if it's as easy as I think it will be. New versions of graphics APIs just add more options to previous versions. You could make an OpenGL 4.4 version of a game by using a single extension and then otherwise using purely OpenGL 3.3 code. Or 2.1, at least if you're not using the stuff that was deprecated in version 3.1. Which you shouldn't, as that's really archaic. (Color Index Mode, for monitors with few enough colors that it's easier to call them color 1, color 2, etc. than to have red, green, blue values!) I don't know for an absolute fact that you can do that with DirectX, but I'd be surprised if you couldn't.
And it doesn't look to me like it's going to be especially hard to code, either. It's not like tessellation, where there's little point in trying if you haven't studied manifolds--something that virtually no computer science majors will see. You don't have to alter shaders at all, and that's the hard part of OpenGL/Direct3D.
There's an enormous difference between "can make use of four cores" and "needs four cores".
The project I'm working on will get hyperthreading to kick in at times on my quad core--meaning, there are moments where it would use 8 cores simultaneously if you had them. I'm pretty sure that it would use 16 cores if you had them, though I haven't had the chance to test that. For that matter, it would try to use 64 cores at once if they were offered, though I suspect that it wouldn't scale well to multiple sockets.
But it still runs passably on my laptop, an AMD E-350, which is a 1.6 GHz dual core--and with poor performance per clock cycle by modern standards, at that, so it's "slow" even compared to other 1.6 GHz dual core CPUs without hyperthreading.