[SOLVED] So I overclocked my 3200 mhz CL16 memory.

[Klesa]

Hi, I own 16 GB of 3200 mhz CL16 (HyperX predator) and I managed to overclock it to 3800 mhz CL17 CommandRate 1 at 1,37 Volts (vccsa 1,2V, vccio 1,15V). Is it a good result? Also is there any benchmark where I can see the improvment? I now have +2FPS on average in some games. Thanks for answers.

My PC specs:
I7 8086K (5GHZ OC 1,28V)
Msi z370 a-pro
HyperX 16GB KIT DDR4 3200MHz CL16 Predator Series
Msi GTX1080Ti Gaming X
NOCTUA NH-U14S

 

[CountMike]

Hi, I own 16 GB of 3200 mhz CL16 (HyperX predator) and I managed to overclock it to 3800 mhz CL17 CommandRate 1 at 1,37 Volts (vccsa 1,2V, vccio 1,15V). Is it a good result? Also is there any benchmark where I can see the improvment? I now have +2FPS on average in some games. Thanks for answers.

My PC specs:
I7 8086K (5GHZ OC 1,28V)
Msi z370 a-pro
HyperX 16GB KIT DDR4 3200MHz CL16 Predator Series
Msi GTX1080Ti Gaming X
NOCTUA NH-U14S

https://www.aida64.com/downloads

 

[CompuTronix]

... 3200 mhz CL16 ... overclock it to 3800 mhz CL17 ... 1.37 volts ...

Klesa,

If it's indeed stable, then nice job!

Yes, you can benchmark it, however, you can also calculate True Latency. Here's the formula:

1 divided by 1/2 DDR in GHz multiplied by Column Latency = True Latency in nanoseconds.

So ...

Half Double Data Rate (DDR ) = Single Data Rate (SDR)
If DDR 3200MHz = 3.2GHz, then SDR 1600MHz = 1.6GHz
Column Latency (CL) = 16

Therefore 1 / 1.6 x 16 = 10.00ns True Latency

And ...

Half Double Data Rate (DDR ) = Single Data Rate (SDR)
If DDR 3800MHz = 3.8GHz, then SDR 1900MHz = 1.9GHz
Column Latency (CL) = 17

Therefore 1 / 1.9 x 17 = 8.95ns True Latency

Although you've increased your memory frequency by 600MHz or 18.75%, you've also decreased your True Latency by 1.05ns or 11.73%.

Keep in mind that only the very best high-end memory modules can approach the long-standing latency wall at 8.0 nanoseconds. To put it into perspective, here's a few common frequency / timing combinations for comparison in order of True Latency:

3200 @ 16 = 10.0ns
3200 @ 15 = 9.38ns
4000 @ 18 = 9.00ns
3800 @ 17 = 8.95ns
3600 @ 16 = 8.88ns
3200 @ 14 = 8.75ns
3733 @ 16 = 8.57ns
4000 @ 17 = 8.50ns

With respect to your original 3200 @ 16:

3733 @ 16 = 533MHz frequency increase (16.65%), 1.43ns True Latency decrease (16.68%)
3800 @ 17 = 600MHz frequency increase (18.75%), 1.05ns True Latency decrease (11.73%)

You might want to see if you can get it stable at 3733 @ 16, which is faster than 3800 @ 17, however, you might need an increase to 1.38 volts. 4000MHz @ 17 will probably be out of reach at any voltage. Intel's specification for the IMC (Integrated Memory Controller) within the CPU silicon Die is 1.35 volts +/- 5%, which is slightly less than 1.42, so you have ample headroom.

A word of advice; always perform a total system backup prior to attempting any overclock. Since BSODs are inherent to testing overclock stability, software crashes, especially those caused by unstable memory, are the most expedient way to corrupt Windows, as well as files such as photos, documents or game saves.

CT

 

[Klesa]

Klesa,

If it's indeed stable, then nice job!

Yes, you can benchmark it, however, you can also calculate True Latency. Here's the formula:

1 divided by 1/2 DDR in GHz multiplied by Column Latency = True Latency in nanoseconds.

So ...

Half Double Data Rate (DDR ) = Single Data Rate (SDR)
If DDR 3200MHz = 3.2GHz, then SDR 1600MHz = 1.6GHz
Column Latency (CL) = 16

Therefore 1 / 1.6 x 16 = 10.00ns True Latency

And ...

Half Double Data Rate (DDR ) = Single Data Rate (SDR)
If DDR 3800MHz = 3.8GHz, then SDR 1900MHz = 1.9GHz
Column Latency (CL) = 17

Therefore 1 / 1.9 x 17 = 8.95ns True Latency

Although you've increased your memory frequency by 600MHz or 18.75%, you've also decreased your True Latency by 1.05ns or 11.73%.

Keep in mind that only the very best high-end memory modules can approach the long-standing latency wall at 8.0 nanoseconds. To put it into perspective, here's a few common frequency / timing combinations for comparison in order of True Latency:

3200 @ 16 = 10.0ns
3200 @ 15 = 9.38ns
4000 @ 18 = 9.00ns
3800 @ 17 = 8.95ns
3600 @ 16 = 8.88ns
3200 @ 14 = 8.75ns
3733 @ 16 = 8.57ns
4000 @ 17 = 8.50ns

With respect to your original 3200 @ 16:

3733 @ 16 = 533MHz frequency increase (16.65%), 1.43ns True Latency decrease (16.68%)
3800 @ 17 = 600MHz frequency increase (18.75%), 1.05ns True Latency decrease (11.73%)

You might want to see if you can get it stable at 3733 @ 16, which is faster than 3800 @ 17, however, you might need an increase to 1.38 volts. 4000MHz @ 17 will probably be out of reach at any voltage. Intel's specification for the IMC (Integrated Memory Controller) within the CPU silicon Die is 1.35 volts +/- 5%, which is slightly less than 1.42, so you have ample headroom.

A word of advice; always perform a total system backup prior to attempting any overclock. Since BSODs are inherent to testing overclock stability, software crashes, especially those caused by unstable memory, are the most expedient way to corrupt Windows, as well as files such as photos, documents or game saves.

CT

Hi, thanks for help. I will try today to get to 3733mhz CL16. Just have one question now. What about the command rate? Is it worth decreasing it to one for an additional +0.01 voltage bump?

 

[CompuTronix]

Klesa,

If you can get it stable at CR1, that would be a bonus. If need be, you can increase the voltage, but don't exceed Intel's spec at 1.35 + 5%, which is 1.42.

Here's an excellent Guide from Gamers Nexus, which also includes a video: What Are Memory Timings?

CT

 

[Klesa]

So after some testing today I wasn't able to get 3733mhz CL16 stable, not even 3700mhz. But 3600mhz CL16 seems stable. Is the 0.07 ns latency decrease worth the loss of bandwidth? Or should I stay at 3800 mhz CL17 (17,20,20,39,595) CR1?

 

[CompuTronix]

As shown in my 1st post, there's very little difference:

3800 @ 17 = 8.95ns
3600 @ 16 = 8.88ns