1600 mem vs higher for gaming

[rip]

I've searched already; so, I thought I'd ask the experts here:
current system = not enough for Witcher III
Witcher III requirements = http://www.pcgamer.com/the-witcher-3-system-requirements-announced/

My proposed system upgrade = i7 4790K, Z97 mb w/ 2 pci slots that I require for my sound editing. And, an AMD r9 290

I've been out of the loop for quite a while (i7 920) so my question is what difference does the RAM make....ie: 1600 vs whatever 1866, 3000, etc

 

[Ketchup]

The Haswell refresh likes faster RAM. I don't really care, but if you want every ounce of performance, get something fast:

http://www.corsair.com/en-us/blog/2014/march/haswellrealworld

 

[exar333]

1600mhz is plenty...

Generally you can get 1866/2133 for a small premium and can expect a few % more performance on games, but not huge. The only time it is really impactful is if you are using a APU and need the bandwidth for the iGPU.

I would suggest shooting for 16GB of 2000mhz or so and get the best deal you can, with good timings. Your build looks very solid and would be a great rig for TW3.

Only caveat is that there will be some nice new cards just around when TW3 is being released....the new 3xx AMD cards will be out around then (June-ish) and should provide a nice bump in performance. If you don't need the rig now, I would wait a couple months. Otherwise, if you have games to play now, go for it.

 

[bunnyfubbles]

for gaming on a single dGPU, no difference (i.e. almost always less than 2%): http://www.anandtech.com/show/7364/memory-scaling-on-haswell/7

although for other tasks and if you ever plan to go multi GPU, 1866 is probably a good sweet spot to shoot for as there should be some kits you could find near 1600 pricing, which you should definitely go no lower than (i.e. avoid 1333)

otherwise money spent on premium ram is money that could have gone towards more ram, better cooling, bigger/better/faster SSD, etc, any of which will have far greater impact on system performance/capability

The Haswell refresh likes faster RAM. I don't really care, but if you want every ounce of performance, get something fast:

http://www.corsair.com/en-us/blog/2014/march/haswellrealworld

can't really trust corsair to not have an obvious biased slant here...

 

[Cerb]

The returns from faster RAM tend to be greatly diminishing, unless you OC the CPU enough for RAM to be a real limiting factor. OTOH, if it's not much more expensive, as 2133MHz sometimes is, why not? Depending on program, 1600 or 1866 tend to be at the knee of the performance gain v. waste of electricity curve for Haswell, but the occasional program, and those same programs running on a faster CPU (overclocked by a fair margin) can make decent use of more. If you're not looking to overclock as far as possible, it should be the last thing you spend added money on.

 

[Dasa2]

if you are playing a cpu bottlnecked game like arma\dayz then faster ram like 2400c10 which costs ~$10-30 extra can add ~10% to the cpus performance
as Cerb said as you increase the cpu clock speed the gain from faster ram increases as it becomes more starved for bandwidth but with most games this will also move the limit to the gpu reducing the gains you see from faster ram

in gpu bottlnecked games (which is most games) like anandtech used in that ram review it doesnt mater how fast the cpu\ram is you will only see ~1-3% variance
once they moved to three 5870 dirt3 fps doubled and they were finally becoming cpu limited at which point they started to see 10-20% gains
dying light with two gtx980 at 1080p 100% view distance is a game i would like to see a cpu\ram review done on

the other thing to look out for is single sided dimms which is most high speed 4gb sticks
single sided 2400c10 is slower than double sided 2133c9
so that and lose timings is another reason 3000mhz kits wont show big gains

i would hold of on the upgrade
you may be surprised just how well your i7 920 handles the witcher3 if you can get a good oc out of it then it will be a lot faster than the phenom II X4 940
as for the gpu there is new cards on the way which may be here before the witcher

 

[Berryracer]

I upgraded my 32 GB 1600 MHz DDR3 RAM to 1866 MHz Kingston HyperX and the difference is mind blowing

 

[exar333]

I upgraded my 32 GB 1600 MHz DDR3 RAM to 1866 MHz Kingston HyperX and the difference is mind blowing

Not sure if serious. That would be a silly 'upgrade'.

 

[Berryracer]

Not sure if serious. That would be a silly 'upgrade'.

maybe I was experiencing the placebo effect????

 

[Dasa2]

maybe I was experiencing the placebo effect????

maybe not entirely if the 1600mhz kit was c11 and the 1866 is c9 but mind blowing still sounds like your stretching it
unless there was some strange bug affecting performance with the old kit

 

[DigDog]

no.

1866 costs the same as any lower speed, and offers the same results as the fastest ram on the market.

but its your money. feel free to throw it out.

 

[exar333]

maybe not entirely if the 1600mhz kit was c11 and the 1866 is c9 but mind blowing still sounds like your stretching it
unless there was some strange bug affecting performance with the old kit

True.

Only a few applications would see much of a jump;, usually 1-2% for most apps.

OP - 1600 is plenty, but usually the 1866 or even 2000mhz kits can be had for a ~$10 premium. I would say that is worth it, even just for a little more on re-sale down the road (if need be). A few percent for a few $ isn't bad, but I wouldn't drop much more on it unless you are strictly going for benchmark records. In that case, you would be looking for low-timing, 3000mhz+ kits. And those are $$$$.

 

[Valantar]

For discrete GPU users, recommending any kit over another is a tough call. In light of daily workloads, a good DDR3-1866 C9 MHz kit will hit the curve on the right spot to remain cost effective. Users with a few extra dollars in their back pocket might look towards 2133 C9/2400 C10, which moves a little up the curve and has the potential should a game come out that is heavily memory dependent. Ultimately the same advice also applies to multi-GPU users as well as IGP: avoid 1600 MHz and below.

AnandTech's article on Haswell memory scaling, gaming with a single GPU. Conclusions.

Don't waste your money on overpriced memory. But stay above 1600MHz, and avoid slow timings.

 

[Dasa2]

i have found 1600c11 to perform around 1333c9 level and 1866c9 to see ~13% increase in a cpu bottlnecked game like arma

http://forums.bistudio.com/showthre...rmance-comparison-1600-2133-up-to-15-FPS-gain
i have also seen people claim that moving from 1333c9 to high speeds of ram significantly improved load times but im not sure how much truth there is to this as i have yet to see it tested

in gpu bottlnecked games though like what anandtech tested it doesnt matter how much faster the cpu\ram is so its kinda useless linking it other than to say that a 4ghz 4770k is fast enough for old console ports with a 6950...
it doesnt show the affect of faster rma at all and the results would be much the same if they had a 4ghz 2600k in those tests for comparison against the 4770k there would be next to no performance difference between the two cpu due to the gpu bottlneck

 

[Valantar]

Why is everyone talking about such slow timings? I know they're more common, but please, try to find a set with the fastest available timings for its speed. It's worth the upgrade. 1600 is available at c7, so getting 1600c11 or even c9 is pretty dumb.

1866c8
2133c9
2400c9 (I can find good kits of both this and 2133c9 from g.skill very reasonably priced here in Norway at least)
2666c10 - which is where the price premium usually gets too high.

 

[UaVaj]

i have found 1600c11 to perform around 1333c9 level and 1866c9 to see ~13% increase in a cpu bottlnecked game like arma

http://forums.bistudio.com/showthre...rmance-comparison-1600-2133-up-to-15-FPS-gain
i have also seen people claim that moving from 1333c9 to high speeds of ram significantly improved load times but im not sure how much truth there is to this as i have yet to see it tested

in gpu bottlnecked games though like what anandtech tested it doesnt matter how much faster the cpu\ram is so its kinda useless linking it other than to say that a 4ghz 4770k is fast enough for old console ports with a 6950...
it doesnt show the affect of faster rma at all and the results would be much the same if they had a 4ghz 2600k in those tests for comparison against the 4770k there would be next to no performance difference between the two cpu due to the gpu bottlneck

anandtech results was a waste of time. testing memory with a gpu limited game.

anyway excellent results with a cpu limited game.

btw. can you share your testing method. thanks.

 

[Dasa2]

btw. can you share your testing method. thanks.

arma3 has two timedemo benchmarks you can download altis and stratis which i used with fraps to record fps
variance between runs was under 1fps
rest of the details should be in the first post i think but if there is anything specific you want to know just ask

did a similar test with thief which is partially cpu limited in the timedemo
http://forums.atomicmpc.com.au/index.php/topic/55771-cpuram-performance-in-thief/
in thief there was a big difference in fps between the first and second run but once files were preloaded repeat runs provided consistent results in line with the second run

 

[BFG10K]

You generally won't notice a difference outside of benchmarks and some extremely niche scenarios. If you have to ask, you probably don't need faster RAM.

 

[BarkingGhostar]

Is this for online gaming? I would imagine memory performance will have little benefit compared to the slowness of Internet connectivity. For standalone gaming, I would question the actual player's swiftnesss to benefit from better ram.

 

[rip]

If you have to ask, you probably don't need faster RAM.

Well, we all can't possibly be as sharp as you; can we?

 

[schmuckley]

Get better RAMs if you can find them
I'm running 2666.
Finding RAM like I have might prove troublesome these days, though.

 

[bunnyfubbles]

Well, we all can't possibly be as sharp as you; can we?

I don't think he was singling you out with that statement, its basically just a common trend in memory ever since memory controllers were integrated into the CPU and multipliers were unlocked, effectively removing memory as the bottleneck for performance or for achieving desired overclocks.

memory speed simply hasn't been a concern for the past several years, and things like capacity and voltage have been more substantial concerns as long as the kit was fast enough.

its basically just a common truth, i.e. "if anyone has to ask, they probably don't need faster ram"

Get better RAMs if you can find them
I'm running 2666.
Finding RAM like I have might prove troublesome these days, though.

for what purpose? the OP isn't chasing benchmarks

spending much more than what a kit 1866 commands is pretty much an inarguable waste of money, especially for someone whose concern is gaming on a single dGPU

 

[Dasa2]

for what purpose? the OP isn't chasing benchmarks

well it depends how you look at it
spending a bit extra on ram over 1866c9 may get you equivalent performance to ~250mhz higher cpu oc which is probably barely noticeable even in cpu limited games

but then you spend a bit extra on cooling to get another ~250mhz out of the cpu and when you put the two together you have a rather noticeable difference in the odd game that needs it
but still no difference in most games which are totally gpu bottlnecked

 

[PPB]

The only game I have seen a HUGE difference is BF4 MP 64Player full servers.

I was running 1333 CL11 (I know, stupidly slow) on Zavod 311 and the most CPU heavy portion of the map went easily sub 60FPS.

I changed the settings to 1600 CL9, and it was a night and day difference, that same portion went again to 75-80 fps.


So there might be a odd case where RAM timmings make a difference, but I cant imagine that benefit going a lot higher after you pass the 1600/1866 mark. And this comes from a TridentX 2400 10-12-12-31 user

 

[kasakka]

What about latency? Does 2133 CL11 trump 1600 CL9 or is the speed difference lost to higher latency?