Is this a notable overclock for ram?

[sandals100]

So I just finished overclocking my ram (properly this time) and I was able to get it to 2000mhz @ 1.6volts with the timings at 10,10,10,30.

 

The ram was originally at 1600mhz @ 1.5 volts with the timings at 9,9,9,24. 

 

Would you guys consider this overclock to be an improvement? If not please tell me why and how I can improve it.

[Oshino Shinobu]

I would say no for most things. I would always go for better timings over a little speed, it has a bigger effect in more applications. 

[sandals100]

I would say no for most things. I would always go for better timings over a little speed, it has a bigger effect in more applications. 

So would dropping it to lets say 1866 and the try and lower the timings?

[BiG StroOnZ]

I would say no for most things. I would always go for better timings over a little speed, it has a bigger effect in more applications. 

Very wrong.

 

Let me do the latency numbers for you:

 

DDR3-1600 CAS9 11.25ns

DDR3-2000 CAS10 10.0ns

 

He definitely improved his latency along with the increased bandwidth from the higher frequency. 

 

 

So would dropping it to lets say 1866 and the try and lower the timings?

 

No you're fine dont listen to him. That's a very good overclock. 

[Oshino Shinobu]

So would dropping it to lets say 1866 and the try and lower the timings?

If you could keep the original timings and get 1866MHz, maybe by raising the voltage from stock a little, then that would be a good balance I think. 

[Sauron]

I would keep 1600mhz, it makes no difference whatsoever from 1333mhz to 2133, only a few insignificant digits in synthetic benchmarks. It's not worth it.

[patrickjp93]

So would dropping it to lets say 1866 and the try and lower the timings?

http://www.newegg.com/global/au/Product/Product.aspx?Item=N82E16820231624&nm_mc=AFC-C8Junction&cm_mmc=AFC-C8Junction-_-na-_-na-_-na&cm_sp=&AID=10446076&PID=3938566&SID=

http://www.newegg.com/global/au/Product/Product.aspx?Item=N82E16820226508&cm_re=Mushkin_997169-_-20-226-508-_-Product

This is the timing to aim for in 1866.

Some people starting with those more premium sticks get down to CL 7.

On premium G.Skill Trident X 1600MHz I've gotten down to CL 6.

Very wrong.

Let me do the latency numbers for you:

DDR3-1600 CAS9 11.25ns

DDR3-2000 CAS10 10.0ns

He definitely improved his latency along with the increased bandwidth from the higher frequency.

No you're fine dont listen to him. That's a very good overclock.

FACEDESK*

The bandwidth of the RAM is not determined by speed alone, hence the PC 12800 vs 14900 between the 1600MHz sticks and the 1866 sticks. There's an onboard scheduler on the RAM which can only handle so much pre-determined, regardless of what clock you set it to.

This speedup can only help saturate it, but cannot push beyond that.

[sandals100]

 

Very wrong.

 

Let me do the latency numbers for you:

 

DDR3-1600 CAS9 11.25ns

DDR3-2000 CAS10 10.0ns

 

He definitely improved his latency along with the increased bandwidth from the higher frequency. 

 

 

 

No you're fine dont listen to him. That's a very good overclock. 

 

Haha this is hard because I have absolutely no knowledge about RAM.

 

If you could keep the original timings and get 1866MHz, maybe by raising the voltage from stock a little, then that would be a good balance I think. 

 

I raised the voltage already and I don't really want to go any higher than 1.6 volts...

I would keep 1600mhz, it makes no difference whatsoever from 1333mhz to 2133, only a few insignificant digits in synthetic benchmarks. It's not worth it.

[BiG StroOnZ]

I would keep 1600mhz, it makes no difference whatsoever from 1333mhz to 2133, only a few insignificant digits in synthetic benchmarks. It's not worth it.

 

Yeah ok, spread more misinformation:

 

Corsair BF4 Memory Benchmark:

 

 

 

BF4 1333MHz vs 2133MHz

 

Testscenario:

Domination Siege of Shanghai, on the roof (see video)

Private Server 1/32 Players, "Scooby Snacks BF4 PCW/CW"http://battlelog.battlefield.com/bf4/en/servers/show/pc/d9788992-49d3-403a-9c9d-b3f1dbce62e1/Scooby-Snacks-BF4-PCW-CW/

Video of the test scenario (short cut of the benchmark, benchmark is 180s long, video is lagging because my hard drive is not fast enough for fraps):

http://youtu.be/GGUlkTA-JII

For every RAM 2 runs.

 

 

 

Conclusion

• Less framedrops or less incursions
• FPS increase 58%, RAM MHz increase 60%
• -> in CPU-limited situation the fps increase is nearly the same as the ram MHz increase

Here are a few of my results...

 

System:

- 3930K @ 4.5Ghz (usually 4.8)

- Rampage 4 Extreme

- 16GB G.Skill Ripjaws Z DDR3 2133 9-11-10-27

- EVGA GTX 670 FTW 2GB @ 1326/7460

- Samsung 830 256GB SSD (OS/Sys/Apps)

- WD Blue 320GB (Music/Video)

- HGST 7K500 320GB (Apps)

- WD RE3 1TB (Games)

- Creative Titanium HD

- LG/Hitachi Slim DVD-R/W

- NZXT HALE90 850W PSU

- TRIPP-LITE ISO-BAR-4 ULTRA Line Conditioner/Surge Suppressor/Voltage Regulator

 

I used the above system for the below tests, with all clocks identical with just the RAM changing. The system is watercooled so no drops in Kepler Boost or anything like that.

 

*TESTS*

 

(RAM SPEED + TIMINGS - AVERAGE FPS - MINIMUM FPS - MAXIMUM FPS)

All tests done in 1080p using a Dell P2212Hb connected via DL-DVI-D, all game settings set at maximum unless otherwise noted.

 

HALF-LIFE 2 EPISODE 2 (CPU @ 3.4Ghz, GPU @ Stock FTW Speeds)

1600 6-7-7-19 - 198.5 - 131.8 - 233

1600 9-9-9-24 - 191.3 - 124.9 - 219

1866 8-8-8-24 - 204.1 - 137.7 - 242

2133 9-11-10-27 - 217 - 147.1 - 259

2133 9-10-9-26 - 226 - 154.3 - 266

2360 9-12-10-29 - 231 - 159.7 - 283

 

Half-Life 2 and it's countless derivatives (mods, etc) all seem to be CPU Bound at this point, as I see a perfectly linear relationship between a CPU's speed and FPS.

 

 

FAR CRY 3 (CPU @ 4.5Ghz, GPU @ 1326/7460)

1600 6-7-7-19 - 52.2 - 26.1 - 93

1600 9-9-9-24 - 50.8 - 24 - 86

1866 8-8-8-24 - 54.1 - 27.9 - 96

2133 9-11-10-27 - 57.4 - 31.1 - 105

2133 9-10-9-26 - 58.7 - 33 - 107

2360 9-12-10-29 - 60.2 - 35.5 - 108

 

The biggest thing with FC3 is the increase in smoothness. The slower memory feels choppy at times, such as when you get into a firefight. The faster memory never has this problem. Also, with 2133 and above, I get zero texture "pop in", yet it's present with lower memory speeds.

 

 

I did recordings of 9 games, 11 benchmarks, and timed start up/shutdown/opening (Firefox with 25tabs/Photoshop/Paint.Net/Chrome with 25tabs,and a half dozen other things), and a few other things.

 

There is not a single instance in which the 1600 9-9-9-24 didn't come in dead last. The differences ranged from "benchmark-noticeable" to "Wow that's a huge improvement".

 

 

I simply don't recommend getting 1600 when the option for faster memory is there, especially if you have an IVB platform. The tests above are X79, and I have done the same tests on a 3770K + GA-Z77X-UP7, 3770K + Maximus 5 Extreme, 3570K + Extreme6 (for Ivy), a 2700K + Maximus 4 Extreme-Z, 2600K + G3.Sniper3, 2500K + Extreme9 (for Sandy), a Phenom II X4 980BE + ASRock 990FX Fatal1ty, 1100T + Crosshair V Formula-Z, 960T + 990FX Sabertooth R2.0, 1090 + M5A99X, 965BE + Gigabyte 990FX UD5(UD7, can't remember), and 8350 + Crosshair V Formula, 8130 + Sabertooth, 6100 + Extreme6, 4100 + Extreme3.

 

These have been over the course of 16mo, and variables change, not all systems had all tests run (most only had 2-3 games and a few benchmarks), and it's not a controlled experiment. Still, the results are only compared against the results from the same system, so they are perfectly valid.

 

Every single system wanted the fastest memory possible, although the Phenom II systems had to be controlled for timings by ensuring that the actual latency in ns was better than the prior test (which means most of the Phenom II tests are more about timings for a given speed than speed itself, although 1800 7-8-7-26 was always the fastest, beating 1600 6-7-6-19 by 9.3% on average).

 

 

 

Heaven Benchmark

 

DDR3 1600

Min 11 FPS

Avg 54.7 FPS

Max 117.5 FPS

 

 

DDR3 2133

Min 28.6

Avg 55 FPS

Max 117.7 FPS

 

 

[Oshino Shinobu]

 

 

 

I raised the voltage already and I don't really want to go any higher than 1.6 volts...

I meant raising it from 1.5v if it wouldn't do it normally. Not from 1.6v (: 

[sandals100]

http://www.newegg.com/global/au/Product/Product.aspx?Item=N82E16820231624&nm_mc=AFC-C8Junction&cm_mmc=AFC-C8Junction-_-na-_-na-_-na&cm_sp=&AID=10446076&PID=3938566&SID=

 

http://www.newegg.com/global/au/Product/Product.aspx?Item=N82E16820226508&cm_re=Mushkin_997169-_-20-226-508-_-Product

 

This is the timing to aim for in 1866.

Woah thanks for those references. I will try and get my ram to those values.

[BiG StroOnZ]

FACEDESK*

 

The bandwidth of the RAM is not determined by speed alone, hence the PC 12800 vs 14900 between the 1600MHz sticks and the 1866 sticks. There's an onboard scheduler on the RAM which can only handle so much pre-determined, regardless of what clock you set it to.

 

So you are going to flat out lie and say frequency doesn't effect bandwidth at all. 

[Oshino Shinobu]

So you are going to flat out lie and say frequency doesn't effect bandwidth at all. 

You seem like a very angry and rude person. If you're going to correct someone, at least be polite about it. 

[sandals100]

I meant raising it from 1.5v if it wouldn't do it normally. Not from 1.6v (: 

 

Oh ok

 

Yeah ok, spread more misinformation:

 

Corsair BF4 Memory Benchmark:

 

 

 

BF4 1333MHz vs 2133MHz

 

Testscenario:

Domination Siege of Shanghai, on the roof (see video)

Private Server 1/32 Players, "Scooby Snacks BF4 PCW/CW"http://battlelog.battlefield.com/bf4/en/servers/show/pc/d9788992-49d3-403a-9c9d-b3f1dbce62e1/Scooby-Snacks-BF4-PCW-CW/

Video of the test scenario (short cut of the benchmark, benchmark is 180s long, video is lagging because my hard drive is not fast enough for fraps):

http://youtu.be/GGUlkTA-JII

For every RAM 2 runs.

 

 

 

Conclusion

• Less framedrops or less incursions
• FPS increase 58%, RAM MHz increase 60%
• -> in CPU-limited situation the fps increase is nearly the same as the ram MHz increase

 

 

 

That post took so long to load ahaha. 58% increase is huge from such a 'simple' adjustment!

[BiG StroOnZ]

You seem like a very angry and rude person. If you're going to correct someone, at least be polite about it. 

 

And you seem like the soft sensitive type because there was nothing angry or rude about what I said. Did you see any CAPS or exclamation points? 

[sandals100]

What have I started!?!

[patrickjp93]

I would keep 1600mhz, it makes no difference whatsoever from 1333mhz to 2133, only a few insignificant digits in synthetic benchmarks. It's not worth it.

This is sorely untrue, but only if you track CAS Latency correctly.

 

For Instance:

Assume the 4770 chip with clock rate of 3.7GHz  1/(3.7*10^9) = 0.2703 ns/cycle

1600 MHz CL 9 absolute latency = 9/800 = 11.25 ns

1866 MHz CL 9 absolute latency = 9/933 = 9.646 ns

 

Difference = 1.604 ns 

That's an extra 8 clock cycle down time just on a single core (you can't have fractional down cycles. That is huge over the course of a running program once memory gets involved. 

[patrickjp93]

So you are going to flat out lie and say frequency doesn't effect bandwidth at all. 

It does if and only if you weren't saturating the max possible of the onboard scheduler in the first place. A PC 12800 scheduler can easily be saturated at 1866 and 1600 if the CAS Latency is low enough.

 

When the controllers get better (the upper speed sticks with PC 14900 and above) then clock rates matter more.

[sandals100]

This is sorely untrue, but only if you track CAS Latency correctly.

 

For Instance:

Assume the 4770 chip with clock rate of 3.7GHz  1/(3.7*10^9) = 0.2703 ns/cycle

1600 MHz CL 9 absolute latency = 9/800 = 11.25 ns

1866 MHz CL 9 absolute latency = 9/933 = 9.646 ns

 

Difference = 1.604 ns 

That's an extra 8 clock cycle down time just on a single core. That is huge over the course of a running program once memory gets involved. 

 

 

I don't know how to manipulate the numbers like you just did but:

 

My 4670k is running at 4.1 ghz so I don't know how much of a difference I'd be seeing.

[patrickjp93]

I don't know how to manipulate the numbers like you just did but:

 

My 4670k is running at 4.1 ghz so I don't know how much of a difference I'd be seeing.

1/frequency = seconds per clock cycle. nanoseconds = 10^-9.

Difference in latency/clock cycle time rounded to ceiling (no fractional clock cycles).

 

It's easy math, just a bit to keep track of.

 

1/4.1*10^9 = 0.2439 ns/cycle

1600 MHz CL 9 absolute latency = 9/800 = 11.25 ns

1866 MHz CL 9 absolute latency = 9/933 = 9.646 ns

Difference is 1.604 ns / 0.2439nspc = 9 cycles of extra downtime because of fractions in each iteration rounding to whole clock cycles.

[BiG StroOnZ]

It does if and only if you weren't saturating the max possible of the onboard scheduler in the first place. A PC 12800 scheduler can easily be saturated at 1866 and 1600 if the CAS Latency is low enough.

 

When the controllers get better (the upper speed sticks with PC 14900 and above) then clock rates matter more.

 

Then explain why DDR3-1600 has a peak transfer rate of 12,800 and DDR-2133 has a peak transfer rate of 17,066 ⅔. 

[patrickjp93]

Then explain why DDR3-1600 has a peak transfer rate of 12,800, and DDR-2133 has a peak transfer rate of 17,066 ⅔. 

If and only if the onboard controllers on the sticks can handle that much traffic. You do know they have tiny CPUs on the RAM sticks which only do a tiny set of tasks, right? They put lower quality chips on the lower speed RAM. This is why some of G.Skill's sticks have gotten past 4GHz speeds and some haven't.

[sandals100]

So I dropped her down to 1866mhz with CL9 and I did notice an improvement in the benchmark. Gonna try CL8 once I get some sort of verification from you guys.

 

1866:

 

2000:

[Priller]

So I dropped her down to 1866mhz with CL9 and I did notice an improvement in the benchmark. Gonna try CL8 once I get some sort of verification from you guys.

 

1866:

 

2000:

 

Run the same benchmarks a few times and record the highest number seen and the average across 3 runs. That will give you a better idea of what is better or worse.

[Sauron]

-

 

Ok then, I was wrong. I thought I remembered a video from Linus where he said it was hardly worth it to get faster memory. Maybe the timings he achieved weren't all that great.