So, I set up my 3200mhz CL16 RAM to run at higher speeds.
So, I set up my 3200mhz CL16 RAM to run at higher speeds.
Hi, I own 16 GB of RAM that runs at 3200 mhz with the HyperX predator model. I successfully overclocked it to 3800 mhz CL17 with a CommandRate of 1 and voltages set to 1.37 V for vccsa and 1.15 V for vccio. Is this a good result? Also, is there any benchmark where I can see the improvement? I now get +2 FPS on average in some games. Thanks for your answers. My PC specs are: an Intel i7-8086K processor running at 5 GHz OC with 1.28 V voltage, an MSI Z370 A-PRO motherboard, HyperX RAM, a GTX 1080 Ti Gaming X card from MSI, and noctua NH-U14S cooling.
Klesa, If it's actually stable, that's awesome! Yes, you can test it out with a benchmark, but you can also figure out the real delay. Here is how to find the formula: 1 divided by 1 divided by 2 DDR in GHz multiplied by Column Latency equals True Latency in nanoseconds. So... Half Double Data Rate (DDR) becomes Single Data Rate (SDR). If DDR 3200MHz means 3.2GHz, then SDR 1600MHz is 1.6GHz. Column Latency (CL) is 16. That gives you 1 divided by 1.6 times 16 equals 10.00ns for True Latency And... Half Double Data Rate (DDR) becomes Single Data Rate (SDR). If DDR 3800MHz means 3.8GHz, then SDR 1900MHz is 1.9GHz. Column Latency (CL) is 17. That gives you 1 divided by 1.9 times 17 equals 8.95ns for True Latency Although you've increased your memory frequency by...
Klesa, If it's actually working smoothly, then that is great! Yes, you can test how fast things run, but you can also measure True Latency here. Here is the formula: one divided by one over half double DDR in gigahertz multiplied by column latency equals true latency in nanoseconds. So... Half Double Data Rate (DDR) becomes single data rate (SDR). If DDR 3200MHz equals 3.2GHz, then SDR 1600MHz is 1.6GHz Column Latency (CL) is 16 That means one divided by 1.6 times 16 is 10.00ns True Latency And... Half Double Data Rate (DDR) becomes single data rate (SDR). If DDR 3800MHz equals 3.8GHz, then SDR 1900MHz is 1.9GHz Column Latency (CL) is 17 That means one divided by 1.9 times 17 is 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%. Remember that only the best high-end memory modules can get close to the old latency wall at 8.0 nanoseconds. To put it in perspective, here are some common frequency and timing settings for comparison in order of True Latency: 3200 @ 16 is 10.0ns 3200 @ 15 is 9.38ns 4000 @ 18 is 9.00ns 3800 @ 17 is 8.95ns 3600 @ 16 is 8.88ns 3200 @ 14 is 8.75ns 3733 @ 16 is 8.57 ns 4000 @ 17 is 8.50ns With respect to your original 3200 @ 16: 3733 @ 16 gives a 533 MHz frequency increase (16.65 %) and 1.43 ns True Latency decrease (16.68 %). 3800 @ 17 gives a 600 MHz frequency increase (18.75 %) and 1.05 ns True Latency decrease (11.73 %). You might want to check if you can get it stable at 3733 @ 16, which is faster than 3800 @ 17, but you might need more voltage. 4000MHz @ 17 will probably be too slow for any lower voltage. Intel's rule for the IMC (Integrated Memory Controller) inside the CPU chip is 1.35 volts +/- 5%, which is slightly less than 1.42, so you have plenty of space to work with. A small tip: always make a copy of your entire system before trying any overclocking. Because BSODs are part of testing overclock stability, software crashes, especially those caused by unstable memory, are the fastest way to corrupt Windows files like photos and documents or save game data.
Hey, thanks a lot for your help. I'm going to try hitting 3733mhz CL16 today. Just one question: is the command rate important? Maybe lowering it to one could give me another tiny bit of extra voltage boost (+0.01)?
Klesa, if you can get it working smoothly on CR1 that would be great extra points. If you really need to push harder, just raise the voltage carefully and don't go above Intel's limit at 1.35 plus a little bit more, which comes out to 1.42. Here is a super helpful guide from Gamers Nexus, and even it has a video about Memory Timings so you can learn what CT means.
After testing today, I couldn't get a stable 3733mhz CL16 even not up to 3700mhz. But 3600mhz CL16 feels okay. Is the 0.07 ns latency drop worth losing bandwidth? Or should I stick with 3800 mhz CL17 (CR1: 17,20,20,39,595)?
So far as I can tell, not much has changed: 3800 clocks at 17 gives me 8.95 nanoseconds, while 3600 clocks at 16 comes to about 8.88 nanoseconds.