***The Official 5900X \5950X owners thread***

[Finners]

Also just recently got a 5900x, I ran Ryzen master and curve optimizer and it set every core.to.-29. Do I now have to either set that In the bios or run RM every boot

It will need setting in BIOS or run ryzen master on every boot.

How are you testing that? -29 on all cores would be rare

 

[Audioboxer]

Also just recently got a 5900x, I ran Ryzen master and curve optimizer and it set every core.to.-29. Do I now have to either set that In the bios or run RM every boot

Ryzen Master CO is near useless, there is very little chance -29 on every core is stable.

Everyone should use this instead https://github.com/sp00n/corecycler/releases

If you want to try and speed things up a little initially, open the config.ini, change runtimePerCore = 3m, and then scroll down to FTTSize and make it 720-720. chronic, to test your -29 hypothesis, restart and set it in the bios to -29 and then run Corecycler under my instructions. Let me know how many cores it reports fail



There is my curve on my 5950x. Yes, one of my best cores is at -1.

The other test to run through corecycler, is open the config.ini again, change stressTestProgram = YCRUNCHER, scroll all the way down to under [yCruncher] and change mode to 19-ZN2 ~ Kagari. This will use YCRUNCHER instead of PRIME95. Be prepared for reboots on this test when it hits an unstable core, unlike the PRIME95 output which often just fails in testing and moves on to next core.

To find out which core caused a reboot with YCRUNCHER, either go into the logs folder and find the latest log and see what core was in testing during the reboot, or check event manager for your WHEA error.

 

[uscool]

Also just recently got a 5900x, I ran Ryzen master and curve optimizer and it set every core.to.-29. Do I now have to either set that In the bios or run RM every boot

I prefer to do it the bios and not bother installing extra stuff in windows , prefer to be lean as possible also i'd be surprised if its stable at -29, my 5900x is at -24

 

[chroniclard]

Heh, yeah it confused me, it's just what it came up with. Cheers for info, will have another play when get time. Though I'm more than happy with it at stock. It's made coding ide's noticeably faster over 3900x.

 

[Audioboxer]

Heh, yeah it confused me, it's just what it came up with. Cheers for info, will have another play when get time. Though I'm more than happy with it at stock. It's made coding ide's noticeably faster over 3900x.

Curve can take a while to fully tune/get stable, but it is worth it.

To try and cut corners a little, I'd put your best 2/4 cores on -10~-15 (check these in AMD Ryzen Master ~ gold stars/silver circles), put the rest on -20 and even initially change corecycler core run time to 1~2 mins. See what cores fail over 1~2 hours, reduce them by 2, those that don't fail, increase them by 2. Rinse/repeat till you get something that seems pretty solid.

Then you move into longer testing, where it's worth putting the runtime back to 3~6 minutes, and allowing the PC to run overnight.

Worth doing a curve properly, because unstable cores can be fickle things. At some point randomly one day your PC might reboot, or even worse, reboot during a game or doing work.

If you ever OC your memory, then an unstable curve will wreak havoc on that. In other words, totally confuse people if it's their memory that is unstable or the CPU. Which is why it's basically mandatory to know your CPU is fully stable before you tackle memory OCing.

edit - I should remind everyone, that AMD Ryzen Master names cores from 1, the BIOS usually names cores from 0. So be mindful that if Ryzen Master has a gold star next to say, core 4, that is actually core 3 in the BIOS. Corecycler follows BIOS naming conventions, so if you're ever adding cores to "coresToIgnore" in the config.ini (if you know they're stable and want to cut down on runtime), it's from 0

 

[uscool]

ideally I should fine tune 5900x, instead of an dirty one lol its been stable for past year and in games it pretty much stays at 4900-5000mhz

 

[Audioboxer]

Games that are lighter on the CPU I tend to see boost spikes to 5050~5100. Those that use 15~25% of the CPU, it's more likely to be around 4900.

 

[233]

just picked up a bily bargain 5900x (upgrading from a 3600
any suggestions for a mild overclock? sitting in a Asus X570-I

 

[heatonpkmassive]

just picked up a bily bargain 5900x (upgrading from a 3600
any suggestions for a mild overclock? sitting in a Asus X570-I

Overclocking these chips doesn't gain as much in terms of both clocks and temps as using PBO, Curve Optimiser and tweaking voltage values does.

 

[DeliciousStorage]

Ryzen Master CO is near useless, there is very little chance -29 on every core is stable.

Everyone should use this instead https://github.com/sp00n/corecycler/releases

If you want to try and speed things up a little initially, open the config.ini, change runtimePerCore = 3m, and then scroll down to FTTSize and make it 720-720. chronic, to test your -29 hypothesis, restart and set it in the bios to -29 and then run Corecycler under my instructions. Let me know how many cores it reports fail



There is my curve on my 5950x. Yes, one of my best cores is at -1.

The other test to run through corecycler, is open the config.ini again, change stressTestProgram = YCRUNCHER, scroll all the way down to under [yCruncher] and change mode to 19-ZN2 ~ Kagari. This will use YCRUNCHER instead of PRIME95. Be prepared for reboots on this test when it hits an unstable core, unlike the PRIME95 output which often just fails in testing and moves on to next core.

To find out which core caused a reboot with YCRUNCHER, either go into the logs folder and find the latest log and see what core was in testing during the reboot, or check event manager for your WHEA error.

I was thinking about trying this but it seems like an awful lot of effort for 5% extra performance. I'm also worried that even if it passes in core cycler it might not be stable in other workloads and it could corrupt my files.

 

[Audioboxer]

I was thinking about trying this but it seems like an awful lot of effort for 5% extra performance. I'm also worried that even if it passes in core cycler it might not be stable in other workloads and it could corrupt my files.

Performance uplift is probably nearer 10~15% on a properly tuned curve, maybe even a bit more depending on your cooling system under some loads.

Corecycler pushes each core, independently, to its limits, so if it's stable you are fine. The kinds of loads it carries out are also incredibly stressful and for the most part well beyond ordinary workloads. That's the main point, make sure the CPU comes up stable under the heaviest workload possible.

But yeah, if you're happy as is and don't want to dedicate some time to the Curve Optimiser, don't worry about it!

 

[Tuvoc]

They are certainly a stonkingly good performer at stock. Unless you really need the extra speed, I'd say enjoy it as it is. OK I've run mine at stock for 18 mths now and I'm getting the tweaking bug, not because I need to but because I'm simply curious.

 

[DeliciousStorage]

Performance uplift is probably nearer 10~15% on a properly tuned curve, maybe even a bit more depending on your cooling system under some loads.

Corecycler pushes each core, independently, to its limits, so if it's stable you are fine. The kinds of loads it carries out are also incredibly stressful and for the most part well beyond ordinary workloads. That's the main point, make sure the CPU comes up stable under the heaviest workload possible.

But yeah, if you're happy as is and don't want to dedicate some time to the Curve Optimiser, don't worry about it!

What's your cinebench score score with your properly tuned curve, and what is it with curve optimizer disabled?

 

[chroniclard]

Ryzen Master CO is near useless, there is very little chance -29 on every core is stable.

Everyone should use this instead https://github.com/sp00n/corecycler/releases

If you want to try and speed things up a little initially, open the config.ini, change runtimePerCore = 3m, and then scroll down to FTTSize and make it 720-720. chronic, to test your -29 hypothesis, restart and set it in the bios to -29 and then run Corecycler under my instructions. Let me know how many cores it reports fail



There is my curve on my 5950x. Yes, one of my best cores is at -1.

The other test to run through corecycler, is open the config.ini again, change stressTestProgram = YCRUNCHER, scroll all the way down to under [yCruncher] and change mode to 19-ZN2 ~ Kagari. This will use YCRUNCHER instead of PRIME95. Be prepared for reboots on this test when it hits an unstable core, unlike the PRIME95 output which often just fails in testing and moves on to next core.

To find out which core caused a reboot with YCRUNCHER, either go into the logs folder and find the latest log and see what core was in testing during the reboot, or check event manager for your WHEA error.

Question then, can you just tune 1 core at a time, do you have to corecyler them all.

 

[muon]

How do you guys have such high offsets. No Zen 3 chip i've had, owned a few now, can go below -15 on all cores.

The 5900X I have now, -20 results in random reboots every few days, even though it is stable under load. It's off load the problems occur, i'm guessing when it trues to boost one or two cores.

 

[VersionMonkey]

How do you guys have such high offsets. No Zen 3 chip i've had, owned a few now, can go below -15 on all cores.

The 5900X I have now, -20 results in random reboots every few days, even though it is stable under load. It's off load the problems occur, i'm guessing when it trues to boost one or two cores.

If you set an all-core value it can only ever be as low as the "best" core.

But by using the core cycler script you can find how low you can go for each core individually.

For my 5950X, the best core will only go down to -8 - so that would be the limit for an all-core setting. However, some cores (and only some cores) will go as low as -27; which is why I've got a per-core curve set to get the best out of each core.

 

[muon]

How can that be the 'best' core if it requires a higher voltage.

You want your best cores to be to have the voltage frequency curve shifted up (frequency y-axis), so they can move further along it. Not shifted down where they are more likely to hit temp and tdp limits.

The negative offset as far as i'm aware reduces the voltage at each frequency level.

 

[opethian]

Tempted to get a 5900/5950 but hear that the new chips are several weeks away. I guess that means these will go down in price

 

[heatonpkmassive]

How do you guys have such high offsets. No Zen 3 chip i've had, owned a few now, can go below -15 on all cores.

The 5900X I have now, -20 results in random reboots every few days, even though it is stable under load. It's off load the problems occur, i'm guessing when it trues to boost one or two cores.

Has to be silicon lottery, mine does -30 on all 12 cores. I did start off at -10 on all and ran it on -30 on all but the best 2 cores, which I did -25 on. Then tried -30 on those two and it's ran fine and stable since.

 

[chroniclard]

Have been corecycling over the past few days, progressing........out of curiosity I set everything to -30 and it errored straight off.

What is the actual aim here, just to get the max undervolt for each core? Most of the cores are currently -26, so possibly not far off some overnight testing.