You Should Have Bought Faster RAM: practical evidence that your once "average" DDR3 kit bottlenecks your new GPU way more than your old CPU does

[Guest]

I've long believed that 100% of all conclusions made by various tech news outlets (including LTT, IIRC) between roughly 2012 - 2014 or so that "RAM doesn't matter for gaming" or "you should buy the cheapest 16GB kit you can find" were deeply misguided in that they failed to recognize that in fact, they saw "no meaningful difference" solely because the graphics cards available at the time were not themselves actually fast enough to be bottlenecked by system memory.

 

Today, I was bored, and so decided to test my theory. Pictured below are the CapFrameX charts of two 10-minute 128-tick CS:GO bot deathmatches on the Overpass map. Both runs use the same hardware, consisting of an i7-4790K @ 4.6GHz all-core paired with a GTX 1660 Ti running a +150MHz core clock / +1000MHz memory clock OC. The only difference is that the first chart has the 16GB (2x8) G.SKILL TridentX RAM kit running at 1600MHz / 10-12-12-31, while the second chart has it running the intended 2400MHz / 10-12-12-31 XMP profile.

 

As far as in-game settings and resolution, both matches were played on a 1080p / 144Hz monitor with all graphical options cranked to their maximum (except for motion blur, which was disabled) and no v-sync.

 

TLDR: in this case, DDR3-1600 CL10 proved to be a whopping 56 average FPS slower than DDR3-2400 CL10, paired with an i7-4790K and a GTX 1660 Ti.

 

[straight_stewie]

Can you share your specific testing methodology? There are many factors that could cause significant differences in results, especially if you only ran the test once each.

[WkdPaul]

* your thread was moved to the CPUs, Motherboards, and Memory section, please always post in the appropriate section *

[Guest]

12 minutes ago, wkdpaul said:

* your thread was moved to the CPUs, Motherboards, and Memory section, please always post in the appropriate section *

My bad! Will keep that in mind.

12 minutes ago, straight_stewie said:

Can you share your specific testing methodology? There are many factors that could cause significant differences in results, especially if you only ran the test once each.

I've tested this kind of thing numerous times before in multiple different games, TBH, but just on a personal level. The story is generally very similar. As far as the CS:GO test though I did play several matches with the exact configuration described. No variance, to speak of, apart from the huge framerate gap depending on the active memory speed.

[Middcore]

I fail to see the relevance of this. We're re-litigating what RAM people should have bought ~8 years ago, because (if I'm reading your argument right), graphics cards weren't fast enough for it to make a difference then but they are now? So you would have told a person in 2013 "Spend more money on faster RAM now because on the slim chance you're still using this system circa 2020 it will make a difference?" (And 2400mhz DDR3 would have cost a pretty penny back then.) How many people are still using such DDR3 systems with modern graphics cards in the current year? Not many, I'd reckon. This is the folly of the whole "future-proofing" mentality. 

[Chris Pratt]

There's two ways of seeing this. If you're just trying to fight some persistent notion that RAM speed doesn't matter even today, that's one thing. However, I don't think anyone today is saying that, and if they are, they should be smacked and ridiculed mercilessly.

 

Otherwise, you're relitigating old ideas around an old technology as @Middcorepointed out. Back then, it didn't really matter, and nothing is ever said with an eye to keeping a PC for the better part of a decade.

[Mister Woof]

Memory speed has typically been an area of highly diminishing returns relative to cost and comparatively to other components.

[Orian Pax]

2 hours ago, Akira13645 said:

tech news outlets (including LTT, IIRC) between roughly 2012 - 2014 or so that "RAM doesn't matter for gaming" or "you should buy the cheapest 16GB kit you can find"

I guess they had provided context in their videos.

And like @Mister Woof said I couldn't agree more.

[Guest]

9 hours ago, Chris Pratt said:

There's two ways of seeing this. If you're just trying to fight some persistent notion that RAM speed doesn't matter even today, that's one thing. However, I don't think anyone today is saying that, and if they are, they should be smacked and ridiculed mercilessly.

 

Otherwise, you're relitigating old ideas around an old technology as @Middcorepointed out. Back then, it didn't really matter, and nothing is ever said with an eye to keeping a PC for the better part of a decade.

It's not quite either I was trying to address TBH. Rather, I was sort of trying to disprove the fairly popular notion that RAM speed matters more at a baseline technical level (like as far as it actually impacts a given CPU, directly) than it did at any point in the recent past, in any situation where the graphics card being used is fast enough in the first place to "hit a wall" with slower memory.

 

As far as I can tell that is not the case. If DDR3-3200 was a thing that existed in more than extremely limited experimental forms, for example, I imagine I'd see even further framerate scaling with that versus DDR3-2400.

Edited June 14, 2021 by Guest
Fix grammar