Games sensitive to ram performance

[porina]

I've got an idea to do a test... but I'm out of the loop as to what games are considered to benefit from fast ram, and a search is not getting me very far... so, any pointers? Further problem might be I don't exactly have a massive list of modern heavy duty games...

[Oshino Shinobu]

Very few of them, basically. 

 

Battlefield games can benefit IIRC, but not many more.

[Enderman]

 

 

[ivan134]

Any CPU intensive game benefits from faster RAM. Witcher 3 and GTA 5 are examples.

[elago]

I think Digital Foundry found that when CPU becomes a bottleneck, RAM speed can help aliviate the problem. GTA V, Cities Skylines and AotS are CPU sensitive, Witcher 3 is another good game to test this in towns.

[Djole123]

The only think I know benefits from fast memory are AMD APUs, and even then to a certain degree.

[i_build_nanosuits]

16 minutes ago, porina said:

I've got an idea to do a test... but I'm out of the loop as to what games are considered to benefit from fast ram, and a search is not getting me very far... so, any pointers? Further problem might be I don't exactly have a massive list of modern heavy duty games...

 

13 minutes ago, ivan134 said:

Any CPU intensive game benefits from faster RAM. Witcher 3 and GTA 5 are examples.

this

watchdogs (watchdogs 2), crysis 3, fallout 4...witcher 3, cities skyline...these games...but it's only a couple FPS we are talking about...but if you can find faster RAM that are about the same price and you can deal with the aestetics of those sticks...get them regardless of the brand...RAM is RAM it wont do a night and day difference but faster sticks is good to have for those days where you need to push the CPU to the edge to get smoother gameplay.

[porina]

@elago @ivan134

 

I have Witcher 3 "free" with some nvidia card purchase in past, so will try with that. Never played it beyond the tutorial though... the controls confused me!

 

In short, elsewhere I found that bandwidth was important, latency was relatively unimportant, and module rank made a significant difference also. I'm wondering if that affects games too... having said all that, it will be a pain to setup to test so don't expect results quickly...

 

@Djole123 that's not the scenario I'm looking at, as I'm thinking of scenarios where you have a high-ish end GPU already. The same question could be asked for iGPUs though... I did testing previously with Broadwell's Iris Pro and found under demanding conditions there was a nice advantage to gain from going between 1600 and 2400 ram.

[0ld_Chicken]

I haven't tested extensively or anything but I ran some 1600 vs 2400 tests in gtav and witcher 3 when I got my 2400 kit, but it made a bigger difference with SLI than single card.  Also look at frame times too, I noticed a decent difference in max frame times for gta v which is a really noticeable measure

[porina]

6 hours ago, 0ld_Chicken said:

I haven't tested extensively or anything but I ran some 1600 vs 2400 tests in gtav and witcher 3 when I got my 2400 kit, but it made a bigger difference with SLI than single card.  Also look at frame times too, I noticed a decent difference in max frame times for gta v which is a really noticeable measure

Funny you should say that as I just picked up a 2nd 980Ti to test out SLI. Frame time isn't something I looked into before outside of VR so that'll be interesting.

 

I fired up my copy of Witcher 3 yesterday so that is about ready to go. I'm still debating getting GTA V as it is something I'd like to explore, although even with its age now its price is still high. Must be good!

[Imakuni]

2 minutes ago, porina said:

Funny you should say that as I just picked up a 2nd 980Ti to test out SLI. Frame time isn't something I looked into before outside of VR so that'll be interesting.

 

I fired up my copy of Witcher 3 yesterday so that is about ready to go. I'm still debating getting GTA V as it is something I'd like to explore, although even with its age now its price is still high. Must be good!

Give it some deep thinking as to what hardware it makes sense to test it with before you even begin.

 

Use reason for a second: if you are going for some 980ti SLI with a 6700k (which is easily 1k$) does it even make sense to buy cheap RAM to begin with? You are going with such a high end machine, might as well just drop a few miserable extra bucks and get good RAM, which kinda nulls the point of even asking if it's "worth" looking into performance gains.

 

Likewise, imagine your 6400 + RX 470 dude. Does it REALLY make sense to invest in a Z170 over the original H110, where you could have gotten a Gtx 1060 with that money instead? Not to say that it won't improve performance, but investing the cash elsewhere might make more sense.

 

Or perhaps a Titan XP SLI would be the best hardware to test? Avoid as much of a GPU bottleneck as possible.

 

Maybe just a single Titan XP, to avoid bottlenecks on the GPU while avoiding creating extra demands due to SLI.

 

 

 

Not to say that there's a right / wrong way of doing things. I just wanted to throw a few bones for your brain to munch at before you start dumping time into testing. It's also good to think about others' results and how they decided to tackle the problem.

[porina]

15 minutes ago, Imakuni said:

Give it some deep thinking as to what hardware it makes sense to test it with before you even begin.

Of anyone here I would have thought you might know be best...

 

I'm not suggesting any of the scenarios I'm testing should be used, but I just want to explore the impact of ram in gaming more deeply, with a slightly different angle than those previously. I need as good a CPU and GPU as I can get so they're not limiting, but at the same time I don't have an unlimited budget so no Titan XP for me. You've probably seen endless threads about what ram to put in a gaming rig, leading to the same arguments and links as evidence for or against a particular combination. I just want to try and put some more hard data into that myself.

 

On a similar note I was with a friend yesterday and I had a poke around her PC hardware. Her dad built it for her... and she knows nothing about what's in it. We kinda worked out it had a Titan X previously, but it made me wonder what CPU was in there. 5930k! I haven't figured out how to steal the system from her yet...  Fun part is, she says she never uses it, preferring her laptop... but her partner does some war game thing on it. I did get a quick Firestrike done, which was slightly but not significantly faster than my 6700k+980ti setup. I didn't see what ram was in there...

 

[porina]

Just started testing. I'm doing a mix of synthetic and actual games:

 

FireStrike default

TimeSpy default

Catzilla 720p

Cinebench R15

Doom (1080p, OpenGL + Vulkan, VSYNC off, need to double check other settings, built in frame rate display in one fixed spot from save file)

Witcher 3 (1080p Ultra, VSYNC off, framerate limit off, full screen instead of borderless, fraps 60s average in one fixed spot from save file)

 

I like the synthetics since they're easy to run and get a score. The two games fps will be taken at a single point, so it wont necessarily be representative of longer gameplay but I need to keep the variables controlled so that they aren't greater than what I'm trying to measure.

 

Test configuration:

Zotac GTX 1070 FE, 372.70 driver

i7-6700k fixed at 4.2 GHz

Asus Z170I

 

Ram is either Crucial Tactical 2666 2x4GB (single rank) tested at 2133 and 2666, and Kingston Hyper-X 2666 2x8GB (dual rank) tested at 2133 and 2666. Both dual channel mode.

 

This could take a while...

[porina]

Results are in... I should note, to equalise the ram as they have different timings, I slowed the Hyper-X to match the Tactical. At 2133 they were run at 15-15-15-35 and at 2666 16-17-17-36. I didn't check sub-timings. The Tactical served as single rank, and Hyper-X was dual rank.

 

Cinebench R15

	SR	DR
2666	924	925
2133	919	919

Ram speed seems to make a difference. Rank does not appear to be significant.

 

Firestrike physics

	SR	DR
2666	13403	13452
2133	13320	13347

There might be a small influence here, but we're talking in the ball park of 1%...

 

Catzilla physics

	SR	DR
2666	1931	1936
2133	1792	1873

Bit more of a difference here, showing some advantage to higher speed. Dual rank may have an advantage at lower speed. This could be consistent with past observations that ram performance only makes a difference when it is limiting. Possibly we're close to the threshold here with the 2666 in either case being fast enough, but 2133 slower enough that rank enters significance.

 

TimeSpy CPU

	SR	DR
2666	5270	5134
2133	5025	5171

Not such a clear picture here. I didn't re-run the benchmark several times to investigate, but possibly run to run variations are greater than any difference here, or is there a more subtle complicated scenario?

 

Doom OpenGL

	SR	DR
2666	179	179
2133	177	179

The single rank 2133 situation was slightly slower, but otherwise no difference between the lot.

 

Doom Vulkan

	SR	DR
2666	166	166
2133	165	166

Similar situation here as for OpenGL.

 

Witcher 3 average fps

	SR	DR
2666	81.3	81.1
2133	80.7	81.1

Again hard to interpret this, but I would say it seems to follow the same pattern as TimeSpy CPU results.

 

Witcher 3 max fps

	SR	DR
2666	84	87
2133	84	85

I'd be a bit cautious with this result, as the scene I used had clouds blowing through so that may cause variations even with a 60 second measurement time. This would suggest some advantage to dual rank but I personally I think there should have been some difference between the values for the single rank test.

 

 

Overall does this tell us much? That ram might or might not have a small difference in more CPU heavy tasks. At least for the scenarios I've looked at, I'd call it practically insignificant for regular gamers. Maybe it is something to dig deeper if you're a competitive benchmarker, who are probably already doing that anyway.