Five Overclockable 32 GB DDR3 Kits, Reviewed

[mapesdhs]

The only task I've come across so far which benefits a lot from higher clocked RAM is After Effects,
probably because it's often very bandwidth intensive. A friend did a test for me: reducing the RAM
speed from 2133 to 1866 increased render times by 10% (this is for a 3930K @ 4.5 with 64GB
GSkill TridentX).

Ian.

 

[faye__kane]

WyomingKnott asked the same question I wanted to ask:

> If I read the chart on page 9 correctly, most memory runs higher bandwidths at the stock settings than overclocked. Am I misreading something?

He asked that on the 22nd. I'm very surprised that it wasn't asked in the article, and extra surprised that nobody seems to be interested in the question, seeing as overclock performance is the point of the whole article.

I imagine that stock settings work better because the lower latency more than compensates for the slower clock. But it's like a company selling its candy bars 1/3 cheaper, then you discover that they're half as large.

-faye kane ? girl brain

 

[JackNaylorPE]

Hey Look, a RAM speed article where the results of the tests show virtually no perceptible performance changes in actual usage. Fancy that.

It's too bad RAM is so boring these days. It would be nice if the advice for RAM purchases wasn't "Just buy whatever is on sale as long as you have enough RAM for what you are doing, spend the money on your CPU/Video"

Ya mean like the 11% performance difference on F1 ?

Of course that one just stands out but overall 2-3% performance differences are not unusual ..... that might be considered not worth bothering about but when a 2.5% performance increase comes at a corresponding increase in system costs of 1 - 1.5%. For last 2 weeks GSkill 2133's were cheaper than 1600 or 1866. 2400 CAS 10 Mushikins weres $30 more.... at that kinda cost (1.5%) on a $2k box, I can't argue against that kind of ROI.

 

[Crashman]

WyomingKnott asked the same question I wanted to ask:

> If I read the chart on page 9 correctly, most memory runs higher bandwidths at the stock settings than overclocked. Am I misreading something?

He asked that on the 22nd. I'm very surprised that it wasn't asked in the article, and extra surprised that nobody seems to be interested in the question, seeing as overclock performance is the point of the whole article.

I imagine that stock settings work better because the lower latency more than compensates for the slower clock. But it's like a company selling its candy bars 1/3 cheaper, then you discover that they're half as large.

-faye kane ? girl brain

It was answered in the article, a couple times, and also re-answered in the response thread about four responses up

 

[Traciatim]

Hey Look, a RAM speed article where the results of the tests show virtually no perceptible performance changes in actual usage. Fancy that.

It's too bad RAM is so boring these days. It would be nice if the advice for RAM purchases wasn't "Just buy whatever is on sale as long as you have enough RAM for what you are doing, spend the money on your CPU/Video"

Ya mean like the 11% performance difference on F1 ?

Of course that one just stands out but overall 2-3% performance differences are not unusual ..... that might be considered not worth bothering about but when a 2.5% performance increase comes at a corresponding increase in system costs of 1 - 1.5%. For last 2 weeks GSkill 2133's were cheaper than 1600 or 1866. 2400 CAS 10 Mushikins weres $30 more.... at that kinda cost (1.5%) on a $2k box, I can't argue against that kind of ROI.

Yeah, but F1 is notorious for scaling really well with memory performance for some reason, which is probably why it's used in the article just to show that there can sometimes be a difference. The vast majority of the time the results end up like the metro chart where all the results are within 1FPS of each other. Which is why the advice on purchase ends up being to buy whatever is on sale for the best price at the size that you want. Generally it doesn't matter if you just get 1600. Like you pointed out though, sometimes you can get 2133 for a better price anyway, so you may as well, since it's not like it would hurt performance . . . but this isn't a performance decision, but a cost based one.

I just find it sad these days that with RAM and somewhat processors there isn't really all that much going on in the usable performance department. People with decently spec'd sandy bridge and above don't generally have anything to buy that would be deserving of an upgrade for the cash.

You also calculate the ROI on the entire machine, when you are going to be buying all of the other parts anyway the comparison should by to sinking the cost difference in to upgrading other parts and what overall system performance difference you would see. For example you can grab some G-Skill Ares CAS9 1600 2x4GB sticks for 70 bucks. The G-Skill Trident CAS9 2400 2x4GB kit is 135... nearly twice the price. But you could go from a GTX760 to a GTX770 for 100 bucks, which is only a 35 dollar difference if you save the RAM cost (since you were willing to spend the extra on the RAM anyway). So which of these are going to get you better overall more performance? For 35 bucks difference in cost the 770 is something like 20% faster, vs saving 35 bucks an having your performance go up by 11% in one case, but generally not even measurable in most other cases.

 

[JackNaylorPE]

You also calculate the ROI on the entire machine, when you are going to be buying all of the other parts anyway the comparison should by to sinking the cost difference in to upgrading other parts and what overall system performance difference you would see. For example you can grab some G-Skill Ares CAS9 1600 2x4GB sticks for 70 bucks. The G-Skill Trident CAS9 2400 2x4GB kit is 135... nearly twice the price. But you could go from a GTX760 to a GTX770 for 100 bucks, which is only a 35 dollar difference if you save the RAM cost (since you were willing to spend the extra on the RAM anyway). So which of these are going to get you better overall more performance? For 35 bucks difference in cost the 770 is something like 20% faster, vs saving 35 bucks an having your performance go up by 11% in one case, but generally not even measurable in most other cases.

That can be significant in a budget limited instance but for example.... anytime you get a % increase in performance for a smaller % in cost..... there's no other way to look at it but as having a positive return on investment. Yes, you may get better returns on investing money elsewhere but that doesn't change the fact that your return is bigger than your investment. I can certainly more easily justify the extra $30 I spent on the Mushkin 2400s than the $100 extra I spent on the 4770k.

But I wasn't budget limited..... so when you are not in a position to have to choose it becomes a decision in and of itself and it's hard to argue against getting 3% for 1.5%. The other thing is memory bandwidth has historically had a much greater effect on minimum frame rated than on average so I wish reviewers looked more at minimum frame rates, lag and SLI / CF impacts