Jump to content

I've seen this topic go both ways in terms of gains and losses to performance and no clear guidelines for do's and don'ts if interested in trying, From what I heard originally was decoupling has some sort of penalty, atleast with past Ryzen I'm unsure about the current situation but I've heard the penalty can Be made up into gains, is this function worth experimenting with, what are the penalty's if any, what voltages for managing stability?

                          Ryzen 5800X3D(Because who doesn't like a phat stack of cache?) GPU - 7700Xt

                                                           X470 Strix f gaming, 32GB Corsair vengeance, WD Blue 500GB NVME-WD Blue2TB HDD, 700watts EVGA Br

 ~Extra L3 cache is exciting, every time you load up a new game or program you never know what your going to get, will it perform like a 5700x or are we beating the 14900k today? 😅~

Link to comment
https://linustechtips.com/topic/1479080-11-or-decoupled-fclkmemory/
Share on other sites

Link to post
Share on other sites

On Ryzen 3000/5000 there was a massive latency penalty when running decoupled, and unless you managed to clock the FCLK very high with memory speeds in excess of 4400MT/s+ you couldn't overcome the performance decrease in most situations. For daily operation in general you're better off sticking to ~3800MT/s so you can stay in 1:1 mode for the best latency. 

 

On Threadripper, there was already a massive latency penalty, so running memory decoupled doesn't really have much of a downside, though when you're running the amount of memory you're likely to be running with a Threadripper rig you'll likely be stuck running at speeds where 1:1 makes more sense. 

 

On Ryzen 7000 theres no such thing as 1:1 mode as the MCLK and FCLK have been fully decoupled from each other, so you want to clock both as high as you can. 

 

9 minutes ago, lotus10101 said:

what voltages for managing stability?

On Ryzen 3000/5000 (haven't personally messed around with Threadripper or Ryzen 7000 yet so can't give any commentary on how those work), the biggest one is the SOC voltage, this will tend to sweet spot between between 1.1 and 1.2V on most chips, with 1.15V generally being considered "good enough" for 95% of CPUs. The VDDG voltages also help with some memory/FCLK overclocking, though in my experience the motherboard sets them well enough that they're not worth touching. The 1.8V rail also helps with FCLK overclocking, but this is more for competitive overclocking than daily operation as I've never had it get settings that were unstable to fully stable by raising it, only to mitigate the performance penalties from running unstable settings in some workloads. 

Link to post
Share on other sites

There's no 1:1 on the 7000's ? 

 

Wouldn't that mean they have a penalty, or was that penalty built in to the previous gen?

                          Ryzen 5800X3D(Because who doesn't like a phat stack of cache?) GPU - 7700Xt

                                                           X470 Strix f gaming, 32GB Corsair vengeance, WD Blue 500GB NVME-WD Blue2TB HDD, 700watts EVGA Br

 ~Extra L3 cache is exciting, every time you load up a new game or program you never know what your going to get, will it perform like a 5700x or are we beating the 14900k today? 😅~

Link to post
Share on other sites

5 hours ago, RONOTHAN## said:

On Ryzen 3000/5000 there was a massive latency penalty when running decoupled, and unless you managed to clock the FCLK very high with memory speeds in excess of 4400MT/s+ you couldn't overcome the performance decrease in most situations. For daily operation in general you're better off sticking to ~3800MT/s so you can stay in 1:1 mode for the best latency. 

 

On Threadripper, there was already a massive latency penalty, so running memory decoupled doesn't really have much of a downside, though when you're running the amount of memory you're likely to be running with a Threadripper rig you'll likely be stuck running at speeds where 1:1 makes more sense. 

 

On Ryzen 7000 theres no such thing as 1:1 mode as the MCLK and FCLK have been fully decoupled from each other, so you want to clock both as high as you can. 

 

On Ryzen 3000/5000 (haven't personally messed around with Threadripper or Ryzen 7000 yet so can't give any commentary on how those work), the biggest one is the SOC voltage, this will tend to sweet spot between between 1.1 and 1.2V on most chips, with 1.15V generally being considered "good enough" for 95% of CPUs. The VDDG voltages also help with some memory/FCLK overclocking, though in my experience the motherboard sets them well enough that they're not worth touching. The 1.8V rail also helps with FCLK overclocking, but this is more for competitive overclocking than daily operation as I've never had it get settings that were unstable to fully stable by raising it, only to mitigate the performance penalties from running unstable settings in some workloads. 

 

I agree, I have a AMD /900X 12 core Zen4 and it truly seem like doing "synced FCLK" doesn't do anything, in games or in Cinebench, like has the same result, the only thing that changes is that the power consumption seems to be lower. I run my SOC at 1.1v 2x 16GB 5600Mhz CL36 DDR5.

Intel NUC 13 | i3 1315U2x 8GB 3600/CL16

 

 

AMD 7950x3d | Sapphire 7800XT Nitro | 2x 16GB Corsair Vengeance 5600Mhz CL36 1R | MSI B650-P Pro Wifi | Custom Loop


AUNE X8 Magic DAC + SS3602

AUNE X7s PRO Class-A | Sennheiser HD58X

Schiit Rekkr ELAC BS243.3

Link to post
Share on other sites

38 minutes ago, lotus10101 said:

There's no 1:1 on the 7000's ? 

 

Wouldn't that mean they have a penalty, or was that penalty built in to the previous gen?

Yeah back in August AMD touched on it. 

 

Quote

Now as per Robert, we now know that the default FCLK for AMD Ryzen 7000 CPUs is set to 1733 MHz. Robert also states that memory overclocking is a little bit different with Ryzen 7000 since 1:1:1 (FCLK:UCLK:MCLK) isn't important anymore. It is mentioned that to achieve the best results, you should leave the FCLK to auto and overclock the memory modules and memory controller in 1:1 mode. There will be some corner cases where users will be able to get better performance results by hitting over 2 GHz FCLK speeds but those aren't a big priority, as AMD mentions.

https://wccftech.com/amd-confirms-ryzen-7000-ddr5-6000-memory-sweet-spot-auto-fclk-recommended-for-overclocking/ 

Ryzen 7 7800x3D -  Asus RTX4090 TUF OC- Asrock X670E Taichi - 32GB DDR5-6000CL30 - SuperFlower 1000W - Fractal Torrent - Assassin IV - 42" LG C2 - Windows 11 Pro

Link to post
Share on other sites

hm, interesting

I'm just curious why previous gen would get penalized 

                          Ryzen 5800X3D(Because who doesn't like a phat stack of cache?) GPU - 7700Xt

                                                           X470 Strix f gaming, 32GB Corsair vengeance, WD Blue 500GB NVME-WD Blue2TB HDD, 700watts EVGA Br

 ~Extra L3 cache is exciting, every time you load up a new game or program you never know what your going to get, will it perform like a 5700x or are we beating the 14900k today? 😅~

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×