Sign in with
Sign up | Sign in
Your question

Gaming performance, to overclock or to REALLY overclock - discussion

Last response: in Overclocking
Share
a b } Memory
June 8, 2010 5:35:04 PM

I thought some people might find these results as interesting as I did:

Currently and previously, I am running

BCLK 160
i7 920 - 3.35ghz
Ram - Dominator 1604mhz 8-8-8-24-1T 1.66v

Using heaven benchmark, this scores 1675. Not mentioning any other specs because they were unchanged in the test. I like to run Unigine Heaven a few times and take the last result and go with that, so I do that anytime I want accurate readings. (I use 2.0, not 2.1, because the score is different between the Unigine versions, and I have old scores that were taken before 2.1 release that I like to compare by.)

When I changed to overclocking a little more:

BCLK 170
i7 920 - 3.5ghz
Ram - Dominator 170Xmhz 9-9-9-25-2T 1.68v

Heaven scored it 1669. Other scores before that were 1665 and 1664. I did run memtest and prime95 on the stability for 30 minutes before I did the Unigine test and it was fine.
So it actually went down in performance on the heaven benchmark. I didn't test any other benchmarking programs, because heaven benchmark is the only one that applies to my interests atm. The process between the Unigine tests is the same everytime. I reboot the PC, run memtest and prime95 each time for 30 minutes before the test so that everytime I run it, its the same scenerio. Been doing it like this for months.

This confirms with some peoples discussions about whats more important... latency or mhz of ram. In this conclusion, latency is more important.

Or am I missing something? Thanks in advance

Note: I did try BCLK 180, 1804mhz, 3.65ghz, but I didn't like the 78-80c temps under prime95. Suprisingly, the 10 BCLK difference between 170 and 180 is 68c-70c vs 78c-80c.
a b } Memory
a b K Overclocking
June 8, 2010 5:53:33 PM

j
m
0
l
a b } Memory
June 8, 2010 5:58:00 PM

I hear ya.. The original benchmark of 2.8ghz on the i7 was 1637, then at 3.35 went up to 1675 so that overclock of just the processer while leaving the ram timings alone was successful. But when changing the ram timings and a slight more overclock, it had negative effects.

So I'm trying to say that while a CPU overclock makes things faster, changing the ram timings to make it stable to match the CPU might in some cases not be as great. I was reading how some guy made his ram into 18xx mhz by increasing the latency, but latency is more important when doing this paticular overclock.

Unfortionantly since the 920 is locked, it makes it difficult to overclock as you have to stabilize everything else.
m
0
l
Related resources
June 9, 2010 5:23:34 AM

I don't think oc-ing the cpu would give you much higher scores in Unigine because that was designed mostly for gpu testing. You would get higher scores out of oc-ing the cpu only if your gpu was very powerfull and the cpu was the bottleneck, but since you did not post what gpu you are using I don't know. (If you're gpu isn't very powerfull, that you can oc it to 5 ghz, it won't do much)
m
0
l
a b } Memory
June 9, 2010 6:27:38 AM

its SLI 250 geforces

note: the increase from cpu 2.8ghz to 3.35ghz increased from 1637 unigine score to 1675. (well tested, not a fluke)

I didn't include that part in the initial post, but its in my reply from earlyer
m
0
l
a b } Memory
a b K Overclocking
June 9, 2010 5:57:43 PM

From my knowledge of many tests involving DDR3 memory, both online and printed publications show consistently that latency is an important factor when running speeds from 1066 to 1600Mhz.
The other tendency I see is when ram is increased beyond these speeds, there is a point where memory speed shows higher benchmark scores over 1800Mhz. So, if you had 1333Mhz ram at CL7 and 1600Mhz at CL9 it is likely the lower latency ram would win.
However if you had ram at CL9 clocked at 2000Mhz, it would probably be faster than the 1333Mhz memory- but it would also be dependent on the brand of ram, cooling included, and other variables. Buying this much higher priced ram, one should be particularly choosy.

Parts that get too hot don't run efficiently. That's why sometimes the max overclock may give you less performance in benchmarks. There are other reasons as well, like the particular hardware you are using.

I have an inexpensive motherboard that happens to yield quite good overclocks. I have found however that overclocking the reference clock over 210Mhz does not translate into higher performance. It actually drops. This is not thermal dependent, as my system temps are chilly; much lower than my CPU temps or any other temps for that matter.

I think i5 and i7 CPU's are more sensitive to memory latency than core2, much like the AMD phenoms. This is an untested theory of sorts, mostly just an educated opinion.
By the way, this is a great topic for discussion. I encourage others to add their opinions and experiences here.
m
0
l
a b } Memory
June 9, 2010 9:07:14 PM

Thanks! That's really good to know. I'd love to see some other results to really fine tune where the exact line is of where mhz beats latency or vice versa on one brand of ram vs another, or motherboard vs motherboard, SLI vs quad vs single GPU.

m
0
l
a b } Memory
a c 100 K Overclocking
a b 4 Gaming
June 9, 2010 10:43:12 PM

You had a big difference between the RAM speeds... 1T vs 2T. 1T is significantly faster. I'd like to see these results redone at the same timing. Otherwise you can probably hit 1800mhz at CL9 (mine could and it's also rated 1600mhz CL8) so that means at 1700 CL9 there's a lot of room for improvement.

I don't have a proper memory benchmark but I do have Everest that can figure out the Read, Write, Copy, and overall Latency of the RAM. I found that R,W,C barely changed regardless of speed or latency. However, the total Latency did move a bit but still not much. In fact I found it was mostly dependant on CPU speed.

You should consider trying different base clocks but set the multiplier so that it's roughly the same CPU speed, and then test different RAM speed/timings. So say 160 (3.2ghz) 1600mhz CL8 at 175 (18x 3.2ghz) 1400mhz CL7, 180 (18x 3.24ghz) 1800mhz CL9 then you could really see how RAM on it's own affects things.
m
0
l
!