Thread: The quest for more...

Another update:

Code:

G. Skill Ripjaw 1600mhz 9-9-9-24 ram. (F3-12800CL9D-4GBRL)

SPD Settings:


                Trp*  Trrd  Trcd*  Tras*  Twr   Twtr   Trtp  Trc   Tcl*  CRate*   Volts       *=main setting
STD @ 533   :   7     4     7      20     8     4      4     27    7     2T       1.5
STD @ 610   :   8     4     8      22     10    5      5     31    8     2T       1.5
STD @ 686   :   9     5     9      25     11    6      6     34    9     2T       1.5


Test 1 ~669 :   7     4     7      20     8     4      4     27    7     1T       1.5         PASSED
Test 2 ~669 :   6     4*    6      18     7     3      4*    21    6     1T       1.5         FAIL TO POST
Test 3 ~669 :   6     4*    7      18     8     Auto   4*    27    6     1T       1.6         PASSED
Test 4 ~669 :   5     4*    7      18     8     Auto   4*    27    5*    1T       1.6         PASSED
Test 5 ~669 :   5     4*    7      17     8     Auto   4*    22    5*    1T       1.6         PASSED


Test 6 ~669 :   4*    4*    7      17     8     Auto   4*    22    5*    1T       1.6         FAIL TO POST
Test 7 ~669 :   4*    4*    7      17     8     Auto   4*    22    5*    1T       1.65        FAIL TO POST


*=minimum value
PASSED = 1 hour min, prime95 blend




Generally:
tRAS = tCL + tRCD + tRP
tRC = tRAS + tCL

Thats right!!! Got it stable at 5-7-5-17-22-1T eat that expensive ram!!!!!

edit: had the wrong volts listed

I think standard is a relevant term. It really depends on how your memory is rated and what it is capable of handling.

Personally, I always go the other approach by loosening up the timings and overclock the memory to a higher frequency. I have run benchmarks doing both and honestly, the results are minimal either way. Most of the time I just leave my memory at stock frequency and push the CPU/GPU instead.

Originally Posted by rush2049

Another update:

Code:

G. Skill Ripjaw 1600mhz 9-9-9-24 ram. (F3-12800CL9D-4GBRL)

SPD Settings:


                Trp*  Trrd  Trcd*  Tras*  Twr   Twtr   Trtp  Trc   Tcl*  CRate*   Volts       *=main setting
STD @ 533   :   7     4     7      20     8     4      4     27    7     2T       1.5
STD @ 610   :   8     4     8      22     10    5      5     31    8     2T       1.5
STD @ 686   :   9     5     9      25     11    6      6     34    9     2T       1.5


Test 1 ~669 :   7     4     7      20     8     4      4     27    7     1T       1.5         PASSED
Test 2 ~669 :   6     4*    6      18     7     3      4*    21    6     1T       1.5         FAIL TO POST
Test 3 ~669 :   6     4*    7      18     8     Auto   4*    27    6     1T       1.6         PASSED
Test 4 ~669 :   5     4*    7      18     8     Auto   4*    27    5*    1T       1.6         PASSED
Test 5 ~669 :   5     4*    7      17     8     Auto   4*    22    5*    1T       1.6         PASSED


Test 6 ~669 :   4*    4*    7      17     8     Auto   4*    22    5*    1T       1.6         FAIL TO POST
Test 7 ~669 :   4*    4*    7      17     8     Auto   4*    22    5*    1T       1.65        FAIL TO POST


*=minimum value
PASSED = 1 hour min, prime95 blend




Generally:
tRAS = tCL + tRCD + tRP
tRC = tRAS + tCL

Thats right!!! Got it stable at 5-7-5-17-22-1T eat that expensive ram!!!!!

edit: had the wrong volts listed

What performance gains are you getting with these timings vs stock? Do you have benchmarks to show? Like I said, my experience with memory is that you don't see much improvement in performance.

Well from what I am seeing the biggest performance boost is when you are multitasking in intensive applications at the same time. For me that is running Netbeans while executing the latest compiled code on the same machine, with who knows what else running at the same time.

Quicker response times lets it context switch quicker..... as your ram access pattern will be most non-procedural during those times.

The only way I can find to illustrate this is by running two memory intensive benchmarks at once....

I will go grab two ram benchmarks and do some testing at
1333 5-7-5-17-22-1T
vs
1600 9-9-9-24-2T (what my ram was sold to me as)

I can say even before doing any testing, right upon boot I can open chrome quicker than before. and that 'snappy' feel is there.... you know what I am talking about? That almost preempting my clicks snappiness.... and I don't have an SSD...... but I have used one, its a similar feeling.....

I ran the benchmarks, and besides my laziness.... basically from what I can tell looking at all these graphs and numbers..... No improvement on any of these graphs, besides the fluctuations between runs..... if I really wanted to fake myself out I could look at the best run with tight timings and the worst test with loose timings, but that is just general variance that normally occurs.

But I can say everything feels snappier. Especially mouse movement... like I bought a new mouse or something.

So heavyG you suggest I go the opposite direction? Loosen my ram timings and go for maximum mhz? How loose should I go? Is there a point where its not worth my time? Like if I get 2000mhz with CAS 11 is that worse than slower and tighter? I don't know where the correlation is.

That is tough to answer. I typically try to keep the timings stock and push the frequency as high as I can. If it starts to become unstable, I push up the voltage a bit until it posts and can run stable Prime95. I used to run my 1600 MHz memory at 1920 MHz. Even then, I only saw small improvements in benchmarks and small gains in FPS for games, but when it comes to getting the most out of your system, every little bit counts.

Well, I quickly did some higher mhz tests for the ram.

Put my front side bus up to 300, then lowered the cpu, nb, and ht multipliers so that they were slightly overclocked. Then I tested various mhz ratings. I got 1600 mhz at 8-8-8-24-32-2T, and I got 1800 mhz at 10-10-10-30-41-2T (which is the loosest I can set for ram). Anything higher I couldn't achieve with 1.7v, which I am not comfortable going over for my ram.

Then I switched to CPU overclocking again. I tried getting 4.2ghz straight away, but that was a bit too ambitious.

I lowered the FSB to 287 and the multi at x14. So I got it up and running at 4019 Mhz, or there about. Since I got some dedicated mosfet cooling it has worked better. I am stress testing with prime95 as I type this. It is hitting 57-58C on the 'core' sensor *known to be off*. And is hitting 65-66C on the 'cpu' sensor *near the processor*. So I can say I am most likely under the thermal max of 62C.... just I don't know by how much. These temps are when running small FFT's in-place.

Under prime95 blend testing I never go over 50C on the core, and 60C on the cpu sensors.

So, I am going to let this run for a few hours, if all is good then I will see what mhz options I have on my ram, and time it quick as possible for that mhz.....




Oh this is exciting, I love that I broke the 4ghz barrier finally.

I ran a quick bench: P8768 3DMarks http://3dmark.com/3dm11/1430757

Though I was folding while eating supper and returned to a bluescreen.... so I upped the cpu voltage 3 more notches and am trying again.

Well I struggled with it for most of the morning, I gave up till I get better fans....

I could game a bit and get through a bench like 3dmark, but when taxed 100% by prime95 or intelburntest it would BSOD.... and I can't up the voltage without hitting thermal max.

back to 3.5 ghz.... sigh...