|
SC2 and RAM Speed, what you need to know.
We all know that RAM clock and timings are usually not super critical for gaming. [1][2](Exceptions mostly including IGP performance.) However, SC2 is much more CPU dependent than a lot of games, and that stuff has to go somewhere before your display. To satisfy my own curiosity, I've decided to bench some RAM clocks and timings, to see if it's worth paying a little more for low voltage memory to OC it and tighten timings.
Obviously, some of this will be CPU dependent due to recent processors having the memory controller on-die instead of using the motherboard, so don't take my results as gospel for your system.
Methodology: Replay of maxed supply vs maxed supply of Zerglings A-Moved towards each other on the Unit Tester custom map, using FRAPS to record framerates. All graphics
settings on low except CPU settings on Ultra and Models on high. 30 Second FRAPS bench run
with Start at the same time each test, following the replay camera.
Test Rig: i7 930 HT disabled, Turbo disabled. 6GB OCZ Reapers at listed clocks/timings
SLI GTX 460 @ 820Mhz
Memory timings to be tested: 9 9 9 30 and 6 9 6 30.
Not sure why OCZ suggests the 8 9 8 and 7 9 7, but they do, and I stuck with the
formula for CL6 and it still worked. Since 1333 CL9 is sort of "standard" for DDR3,
I'm going to call that framerate 100%, and performance for the others will also be
shown as percentage of that performance, for easier assessment.
FPS Shown in Min/Max/Average
2.8Ghz CPU Clock
1066 CL9 FPS: 28/133/61.6--95%
1066 CL6 FPS: 28/134/62.6--96.6%
1333 CL9 FPS: 30/139/64.8--100%
1333 CL6 FPS: 30/139/65.2--100.6%
1600 CL9 FPS: 31/142/67.2--103.7%
1600 CL6 FPS: 31/142/67.3--103.8%
For giggles, the same test with my normal OC @3.8Ghz w/ 3600 Uncore.
1600 CL6 FPS: 50/193/95.1--146.7%
If anyone wants to make a pretty graph of this after I finish getting results, I'll happily add it with credit to the artist. I'm terrible at that sort of stuff.
Conclusion: SC2 pretty much couldn't care less what your timings are. Clock speed, on the other hand, makes a few percent difference in the thick of the action, and those couple of frames minimum can be a big difference when they're fairly low like the high supply situations. Minimum framerate can be what really counts, since that's where you suffer the most, and in the thick of the fight, the average definitely seems to benefit from faster RAM.
Since DDR3 1333 CL9 tends to run ~$8 US/GB (for the rationally priced stuff), I'd say if you can keep it to ~$10 US/GB to get some low voltage RAM, or better binned memory tested at 1600Mhz, it might be worth the step up.
Aside from that, you're still looking better off with a healthy overclock on your CPU than you are with faster memory. If you're looking at any more than a few dollars premium, you'd do better to spend the money on a faster CPU, or a better mobo and cooler to OC with.
Obviously, this is for the people who care about price/performance in the slightest. If you happen to have a license for 3DMark, if your desktop has more benchmark utilities than games, obviously you don't care much anyways, because you'll get something, anything, out of it. So it's worth it. But if you can only raise the clock or tighten the timings, but not both... for SC2, at least, the answer is clear.
|
interesting proposition. I'm not a huge tech guy so I have no idea what implications ram timings and speed will have so I'm looking forward to seeing the results.
|
I'd be interested to see how hyper threading affects performance in SC2.
|
On July 19 2011 02:57 Sotamursu wrote:
I'd be interested to see how hyper threading affects performance in SC2.
Considering SC2 is only optimized for two cores, I'd expect little to no benefit.
http://www.techspot.com/review/305-starcraft2-performance/page13.html
If anything, I could see performance suffering if it tried to use HT. Although if you could stabilize HT at a high enough OC, you might be able to free up more CPU for streaming purposes and still have playable framerates if it made a difference.
That's pure theory, and not something I feel like doing to my CPU. Temps at my OC with HT don't do so well.
|
One thing that seems to affect SC2 in reference to RAM is alt+tabbing.
I noticed when playing on bigger screens 32"+ there is a big difference in timings between alt+tabbing between 4GB and 8GB.
There may be other factors involved, but all I know is with everything stock, there is a noticeable difference.
I've noticed this difference between 4GB vs 8GB 1600 or 1333, but not tested 1600 vs 1333. This was done across various CPUs: i7 930, i5 750, Athlon II X2 245, Phenom II X2 555 (and unlocked to B55 as well)
It's not a big deal if you don't alt+tab much but I figure I'd add that in there.
|
Eh, Alt-tabbing is definitely going to be affected a bit by RAM, yeah, but I was mostly concerned with in-game performance.
With that, though, I'd guess it's pure quantity more than anything, as well as Pagefile speed if applicable.
Especially since we're always suggesting builds to people, and making RAM suggestions, it seemed entirely reasonable to have a benchmark with direct reference. I was surprised a bit, really. I expected to see a bit more out of it with how CPU intense SC2 is.
|
Good work.
This lines up with the general advice around the Tech Support forum - just get the cheapest DDR3 you can find (of course, this was based on benchmarks from reputable sources).
To sum it up, I would say DDR3 1600 is worth it over 1333 if you are already getting the strongest SC2 CPU and OC'ing it. (I did this - got a 2500k and also found DDR3 1600 1.5V 8GB for $55.)
Higher frequency memory tends to grow in price in a more exponential rather than linear fashion.
As an interesting aside, the higher cost of high frequency memory is also why I think AMD's Llano is in a bad spot - the GPU performance increases drastically with higher speed memory, but considering the price (and the socket and other factors) you're usually better off with another CPU and discrete GPU.
|
It does line up with what we tend to say, but I was surprised by how much, since SC2 is so CPU intensive. Of particular interest is how little the timings matter. If nothing else, this gives us something better to link to than the xbitlabs and techreport benches that are good, but not SC2.
|
Hm, I'm running i7 2600K, 2GB Asus Radeon HD 6950 with 8gb g.skill @ 1066 I believe (1650x1050 WFS); unless I'm streaming, the lowest FPS I get (4v4) is around 140. SC2 doesn't give a flying fork about your RAM.  W7 hogs 2GB by default though, so I can see why there could be a noticeable difference between 4 and 8.
|
@Dagobert
Different architecture, different IMC, and I did point out that the difference from RAM was negligible compared to CPU gains.
Anyways, the difference is small enough you have to run a bench to find it, and the gains are going to be more important at the mid-range, rather than the upper.
|
oh cool, thats what i was looking for, thanks for the info
|
damn...looks like i'm reverting back to 1600 clock 
Thanks for the benches!
|
On July 25 2011 09:52 Bambipwnsu wrote:damn...looks like i'm reverting back to 1600 clock Thanks for the benches!
Were you running tighter timings at a lower clock? Frequently that will be roughly comparable to looser and higher clocks on newer architectures, specifically Nehalem and Sandy Bridge. So if SC2 performance is what you want, it's definitely better to take clock over timing, and not worth it to increase VDIMM above safe ranges to increase timings.
Bear in mind, there's some margin for error on those tests, since there was a human element involved, but I tried to minimize it as best I could, and at any rate the results were consistent enough that I think I did a reasonable job of it. I didn't get any blatantly anomalous results.
|
What I had tried was lowering to 1333 and tightening the timings to 7-6-7-21 if i recall but I wasn't too sure of the stability. It seemed to be stable on prime95 for an hour or so, but when I played BFBC2 to test the computer a bit, it just BSODed. The CPU at that point was already OCed and stable, and I had bumped up the voltage just incase.
I had dropped down to 1333 since I was BSODing as soon as i did anything to the timings @1600. Should I even consider bumping up the voltage or is it just not worth it?
|
On July 25 2011 10:40 Bambipwnsu wrote:
What I had tried was lowering to 1333 and tightening the timings to 7-6-7-21 if i recall but I wasn't too sure of the stability. It seemed to be stable on prime95 for an hour or so, but when I played BFBC2 to test the computer a bit, it just BSODed. The CPU at that point was already OCed and stable, and I had bumped up the voltage just incase.
I had dropped down to 1333 since I was BSODing as soon as i did anything to the timings @1600. Should I even consider bumping up the voltage or is it just not worth it?
What's your memory kit, CPU, and mobo? People here will tell you different things, and only you can decide whether it's worth it for you. Unless you're tinkering OCD like me, and obsessed with squeezing all possible performance, or you're close to not performing well enough, it probably isn't worth it. skyR would tell you a definite no, Myrm would probably give you a solid maybe, not really sure what Wom would say, jacosajh would probably say about the same thing I am.
|
ASUS M4A79XTD EVO 790X ATX
Corsair XMS3 CMX4GX3M2B1600C9 4GB 2X2GB DDR3-1600 CL9-9-9-24 1.5V
AMD Phenom II X4 955 BE C3
I got really lucky with my chip in that I was able to OC it to 4.0ghz with a 0.02V increase. As far as I know this is the best OC I saw at that voltage when i went and checked on the benchmarks of others on OC.net
The ram's still giving me trouble though. I wasn't able to tinker with it as much as i'd like.. (I built and tested the computer from 2am till about 6am when I had to pack my bags and catch a flight) There's definitely room for improvement in this regard but I'm really aiming for something 24/7...so if uping voltage on ram is harmful I'd most likely stay away from that.
|
I'm painfully unfamiliar with Ph2, since when I built the PC before this one, they had just come out and were being smashed by C2, and when I built this one, i7 was raping face.
Is the memory controller on the mainboard or the CPU? If it's on the motherboard, you may well be able to push more voltage through the RAM safely. Anyways, my benchmarks are irrelevant for you, you'd do better to use my methodology to do your own testing.
|
Ah so depending on the cpu/mobo/ram there could be completely different behavior/benches for the same game...as in not even following the same pattern?
Once I get home, ill definitely do some tests and post what I get...which unfortunately will be in a while =/
|
On July 25 2011 11:59 Bambipwnsu wrote:
Ah so depending on the cpu/mobo/ram there could be completely different behavior/benches for the same game...as in not even following the same pattern?
Once I get home, ill definitely do some tests and post what I get...which unfortunately will be in a while =/
Yeah, performance behavior is very much hardware dependent.
|
|
|
|
|
|
EPT
