If you have the money, get the 32gb. It's possibly much more than you'll use at an given moment, but why the heck not?
In the end, I'd go for the fastest speed with the lowest latency and then worry about 32gb vs. 16gb.
You don't likely need more than 16GB and 8GB is probably even plenty. Go for the fastest RAM that works for your motherboard. That means not only the overall timing (1866) but the CAS write latency as well. Like memory with timings 9-9-9-24 and CAS latency 9 will be faster than memory with timings 11-11-11-28 and CAS latency 11. Though in practical use you may not actually be able to tell.
In my own trials with the Intel Burn Test, there's about a 10% overall speed boost in moving from DDR3-1333 to DDR3-1600, same RAM, faster speed, same CAS latency. Presumably, there's about 5% to be had going to 1866.
There's a caveat here, though. RAM chips have been getting faster, but very, VERY slowly. They keep getting larger, but they are barely getting faster. Because actual speeds of the physical memory cells are barely moving at all, they're stacking memory chips in parallel. This means that total bandwidth increases, but latency stays about the same.
What many companies will do is specify that you can run, say, DDR3-1333 at CAS level 8, and DDR3-1600 at CAS level 10, and that ends up being almost exactly the same speed, overall. CAS is a delay measurement, and the bigger it is, the more latency the memory has. The huge caches on modern CPUs try to compensate for that very high latency level, so you end up with this weirdly unpredictable behavior; some algorithms and programs benefit mostly from latency, and others from bandwidth. The only way to know for sure what's best is to test it yourself, with your OS and specific programs, and tinker. But this takes forever, and people who aren't trying to make things run fast on a professional basis (ie, highly paid admins) shouldn't bother.
Because of all the marketing-induced complexity around RAM, it seems to me that you're best-served by looking for DDR3-1600 with good, tight timings. As long as it supports at least the X.M.P. profile, which was generated by Intel, you'll know it will work nicely at that speed. You can go faster, but whether that will actually matter is something that you can't easily tell ahead of time... XMP memory is a good compromise between the various parameters, and the Intel chips are well-optimized to use it.
If you have the vertical clearance, the Corsair Vengeance chips are usually very good. Most of them have tall heatsinks, so if you have an overhanging CPU cooler, that can be an issue. If you think that might be a problem, you can get low-profile Vengeance sticks, but I think they cost a few bucks more.
If you go 32 gigs now, you'll usually have to go with slower timings; the denser RAM chips don't run as quickly. 16 gigs in 4 chips will give you the best overall speed at the moment.
You're probably on the right track with memory, but out of curiosity, are you sure your CPU cooler is seated correctly? RAM can be fickle, but going 0-for-2 on memory I'd have to think is a rare occurrence if you're getting it from a reputable vendor. 3rd party cpu coolers can sometimes be tricky. Did you apply enough/too much cooling paste? Did you apply it correctly?
Air bubbles are exactly the reason you use thermal paste. But it goes a long way. You want metal to metal contact, the paste is there to fill in any imperfections and effectively mate the surfaces at the molecular level. If you have an air gap then that's an insulating barrier. You can see your die temps in BIOS or a tool like CPU-z though.
Unfortunately it's typically easiest to swap out parts and track down the issue through process of elimination. Neither fun nor easy for the typical home user.
Man, I feel for you on this.
Maybe I am jumping too late, but I had similar issues when I installed build a computer a few years ago and it was the CPU overheating. The bios had a temperature stat and it was hitting 100C (it shouldn't be much above 60, but on average should be in the 30s). Turned out my CPU wasn't properly seated and the CPU wasn't getting the cooling it needed.
Well, at this point, you're trying to narrow down which component is failing. You know something is, but you're still not sure what. The memory was the easy/quick troubleshooting step. That didn't work, so now you have to start a divide-and-conquer process to figure out where the issue is.
First, you've done a fresh install of the OS, since this is a new computer, so you've ruled out software. You know it's a hardware problem.
So now you need to characterize exactly how it's failing. Is it a true freeze, or is just a major slowdown? A slowdown could be the CPU cooler. A hard lock probably is not. Intel chips should never run hot enough to lock up; they throttle themselves before that happens. And Intel CPUs are some of the most reliable electronic devices ever made; you can be essentially certain that the CPU itself is working. (with the caveat that a bad motherboard or power supply can cook a CPU: when you took it out of the package, it worked, but once in a very great while, you can toast one via another bad component. And it's quite easy to wreck one with aggressive overclocking.)
If it is hard locking, there's two major areas of possible failure: CPU-based stuff, and GPU-based stuff. A good way to differentiate is by using the Intel Burn Test to test the CPU, memory, and motherboard. Run Coretemp (make sure to get the non-crapware version, it's an alternate download... and don't forget to kick in a few bucks in exchange for no crapware), and run the IBT on Maximum for several hours. If temps get to 90C, then you do have a cooling issue. If temps stay at 80C or less, and you get inconsistent results from the IBT (the passes are giving different answers), then you've most likely got either a bum motherboard, or a power supply that isn't putting out enough juice. Since that's a Seasonic supply, a motherboard replacement is a higher-percentage bet. Either thing could break, but motherboards in general should have a higher failure rate, so that's your priority for replacement.
If that passes, if it produces consistent results for a few hours (and, preferably, stays below 80C, definitely below 90C) then you can trust your CPU/motherboard/memory. Try running Furmark, which will stress the hell out of the GPU, without putting very much load on the CPU. If you get crashes there, then you know it's the GPU, or the power supply again.
If everything tests okay individually, then run Coretemp, IBT, and Furmark at the same time. If you then get crashes, that means your system is failing under heavy current draw, so it's probably the power supply.
Oh, and just in general, I prefer it when CPUs don't exceed 70C by too much. 80C gets worrisome, and 90C is the hard cutoff, where performance falls off a cliff as the chip self-throttles. It's doing its best to make sure it never hits 100C, at which point it will be immediately slagged.
I'm at the point in this whole thing where I think it's the mb.
That's what I started to say (swap the motherboard) but then I backed off a little -- that's a giant PITA. So, I tried to give him what he'd need to PROVE it was the motherboard, since testing is easy and free, and swapping motherboards sucks.
I don't think we've ruled out GPU and power supply yet.
Oh! If you narrow it down to CPU/RAM/motherboard, try pulling the mboard out of the case and running on a piece of cardboard or something. If you have anything shorted out against the case, that can cause the crashiness, and running it out on something non-conductive helps rule out installation error.
Is it a true freeze, or is just a major slowdown? A slowdown could be the CPU cooler. A hard lock probably is not. Intel chips should never run hot enough to lock up; they throttle themselves before that happens.
This is true if it isn't due to an improperly mounted heatsink and fan. When I had this issue, it did do a complete freeze becuase even when it throttled, it still was overheating and the CPU automatically shuts down to protect itself.
The absolute worst new-build system part defect I ever had was when the case's power button was randomly sending the "I was pressed" signal up the wire and causing the system to shutdown. Yeah, that one was just a big "F U" from computing to me.
This is true if it isn't due to an improperly mounted heatsink and fan. When I had this issue, it did do a complete freeze becuase even when it throttled, it still was overheating and the CPU automatically shuts down to protect itself.
On a recent Intel chip? How long ago was this?
Those temperatures seem fine. They are a little high but within normal operating values.
Bite the bullet and RMA the motherboard. Pain in the butt but you're at that point now.
Good. Hang in there, it'll be worth it. Hopefully.
Yep, motherboard's about it, at this point. Best of luck. What a PITA, eh?
Borrow a Win7 disc from a friend, or "acquire" an ISO from the Internet.
You get that 30 day trial period before having to activate. Run it for a week, and if it works, buy a key and use it.
That way, you don't buy it until you're sure that Win8 was the problem.
Man. This is crazy.
Borrow a Win7 disc from a friend, or "acquire" an ISO from the Internet.
Pages