Ryzen Owners Zen Garden

[tabascosauz]

On Ryzen 3000 both 3600 and 3700X ran at 88W max, I thought 65W on 5000 always was 76W. It's still the sweetspot on 5600X atleast. 84W gave me a measly 50MHz allcore at 5C higher temp and more noise.

5600X is really efficient and has less cores so maybe 76W is perfect. idk if there's new rules for 5000 though. 5600G and 5700G follow the old rules. There's also a 5900 OEM that's "65W", so 76W would be pretty much impossible for a 12 core to even keep up ST performance (more uncore losses for 2CCD, games will hit 70-100W).

 

[phill]

I hope I'm in the right section

Last night I finally got my 5950X setup up and running after having it sit idle on the side for too long for one reason and another... Can anyone suggest a few tweaks I could do, to help me get the best from the CPU/setup? I've some 3600MHz RAM in there rather than the normal 3200MHz stuff as it on a better deal at the time of me looking to buy it.. I believe it's Samsung as well, which I guess might help matters?

Primarily it'll be for crunching, gaming, folding and whatever else you guys could think of (I guess within reason ) so I'm looking for all core performance rather than single thread performance as most of the time it'll be pegged at 100% under load. It will be under a quad 560mm rad but at the moment, I've got it setup under a triple 420mm rad as it's not in the case and is just in my testing loop I put together last night.

For my 3900X/3950X machines I would normally set the vcore as low as I could go, then just let it boost and do it's thing on it's own but I wondered if the 5950X might be different at all? doing this for my 3900X manages 50C load temps on all 24 threads with about 1.10vcore (the other one I had isn't so good so needs a little more) they are both under 280mm AIO's on open benches...

Any suggestions anyone please?? Many thanks

 

[TheoneandonlyMrK]

I hope I'm in the right section

Last night I finally got my 5950X setup up and running after having it sit idle on the side for too long for one reason and another... Can anyone suggest a few tweaks I could do, to help me get the best from the CPU/setup? I've some 3600MHz RAM in there rather than the normal 3200MHz stuff as it on a better deal at the time of me looking to buy it.. I believe it's Samsung as well, which I guess might help matters?

Primarily it'll be for crunching, gaming, folding and whatever else you guys could think of (I guess within reason ) so I'm looking for all core performance rather than single thread performance as most of the time it'll be pegged at 100% under load. It will be under a quad 560mm rad but at the moment, I've got it setup under a triple 420mm rad as it's not in the case and is just in my testing loop I put together last night.

For my 3900X/3950X machines I would normally set the vcore as low as I could go, then just let it boost and do it's thing on it's own but I wondered if the 5950X might be different at all? doing this for my 3900X manages 50C load temps on all 24 threads with about 1.10vcore (the other one I had isn't so good so needs a little more) they are both under 280mm AIO's on open benches...

Any suggestions anyone please?? Many thanks

I tend to just use default settings , actually default not auto.

Then set a 96% max CPU usage in the balanced power settings of the OS , this gives about 3.8Ghz at 1.05 volts.
And a easy switch to max , which it handles well but heats the room a bit more.

 

[Mussels]

IMO, the 5800x was a panicked option at the last minute to beat intel in benchmarks, which is why its PBO values boost so high and it runs so hot vs everything else

There was room between the 5800 and the 5900x values, and they went balls to the wall instead

 

[phill]

I tend to just use default settings , actually default not auto.

Then set a 96% max CPU usage in the balanced power settings of the OS , this gives about 3.8Ghz at 1.05 volts.
And a easy switch to max , which it handles well but heats the room a bit more.

So default vcore you mean for the CPU? That's quite high isn't it to leave alone? I just think of the folding and try to make it run as cool as possible whenever it's under full load..

I'll give it a go Its not crunching at the moment, but I'll run some more tests and see what temps and such come back as

 

[eidairaman1]

I will not be doing any OC as the rig isnt setting benchmarks or gaming, it is a major upgrade from a 478 p4.

 

[TheoneandonlyMrK]

So default vcore you mean for the CPU? That's quite high isn't it to leave alone? I just think of the folding and try to make it run as cool as possible whenever it's under full load..

I'll give it a go Its not crunching at the moment, but I'll run some more tests and see what temps and such come back as

Yes but I mean the general oc profile in bios.

The 96% takes the core and volts down.

I'll be getting a rig or two running for the upcoming compo.

 

[phill]

Yes but I mean the general oc profile in bios.

The 96% takes the core and volts down.

I'll be getting a rig or two running for the upcoming compo.

I'll take a look Thanks @TheoneandonlyMrK

Well @TheoneandonlyMrK as promised....



Just doing a few tests with slightly faster memory at the moment.. The RAM in the system is 3600 but was only set to 3200. With default settings (as they are above) for some reason, running the GPU test in Cinebench R15, it crashed itself.. The other CPU tests seem to have been fine so far.... and as soon as I type that, weirdness happens

Still, what sort of voltages do you set for your hardware/s?

 

[eidairaman1]

Got the 5800 (OEM) for $262 instead of $362

 

[freeagent]

Still, what sort of voltages do you set for your hardware/s?

For ram I use anywhere from 1.35v to 1.6v occasionally.. where you want to be is between 1.35 and 1.45v, since you are at 1800 maybe bump SOC to 1.1v. You wont hurt anything.

 

[TheoneandonlyMrK]

I'll take a look Thanks @TheoneandonlyMrK

Well @TheoneandonlyMrK as promised....

View attachment 224431View attachment 224432

Just doing a few tests with slightly faster memory at the moment.. The RAM in the system is 3600 but was only set to 3200. With default settings (as they are above) for some reason, running the GPU test in Cinebench R15, it crashed itself.. The other CPU tests seem to have been fine so far.... and as soon as I type that, weirdness happens

Still, what sort of voltages do you set for your hardware/s?

I leave it on auto it pulls 1.35 loaded and in balanced profile so max CPU, but I do up the max allowed current in the motherboard vrm settings.
I found using docp flakey, at the moment I find setting Tcr to auto( the last higher timing) stablised my system it auto sets high though like 84.
With two sticks, I should spend time on it but recent GPU crashes distracted me.

Like I said I use Windows power save profile to cut usage for crunching, just set max CPU usage to about 96% (in advanced tab)and it's a nice chilled max ish.

As I said this holds mine at 3.8(150watts at the wall)

With a level easy max at hand.

 

[phill]

For ram I use anywhere from 1.35v to 1.6v occasionally.. where you want to be is between 1.35 and 1.45v, since you are at 1800 maybe bump SOC to 1.1v. You wont hurt anything.

I wasn't able to get 3400 stable quickly and that I did bump up the SOC to about 1.125v (which I think was read about 1.1v...) I need some more time with the RAM but there's all 4 sticks installed, 64GB.. Guessing it might need a fair bit of tweaks for that...

I leave it on auto it pulls 1.35 loaded and in balanced profile so max CPU, but I do up the max allowed current in the motherboard vrm settings.
I found using docp flakey, at the moment I find setting Tcr to auto( the last higher timing) stablised my system it auto sets high though like 84.
With two sticks, I should spend time on it but recent GPU crashes distracted me.

Like I said I use Windows power save profile to cut usage for crunching, just set max CPU usage to about 96% (in advanced tab)and it's a nice chilled max ish.

As I said this holds mine at 3.8(150watts at the wall)

With a level easy max at hand.

I've just tried testing it at 1.10vcore which crashed on boot up, then 1.15vcore seemed to do ok but was a little flaky. Eventually bumped to 1.20vcore (tested a few in between) and found the following...

So 100% CPU usage in the settings, gave me about 215w full system load, running CB R15 (I'll try others tomorrow, need some sleep but this was important to try and test! ) Max temps running the bench, was about 50C as well.
So when I set 96% CPU as you mentioned, it dropped the watts by about 5 to 8w total, so dropped it down to about 208w... Temps dropped about 2C at best (it seems CoreTemp doesn't work when I run CB and only updates afterwards!! Overall test score results, dropped by nearly 10% as well, fair big drop there....

Any limits with volts etc. that I need to be careful with at all? Just trying to make it as cool and quick as possible

Thanks again for the help so far!!

 

[freeagent]

215w *whistles*

She’s rippin

Voltage wise I think you are in the right neighborhood

I saw 240w once from my 5900X.. shake and bake.. I couldn’t control it

 

[tabascosauz]

215w *whistles*

She’s rippin

Voltage wise I think you are in the right neighborhood

I saw 240w once from my 5900X.. shake and bake.. I couldn’t control it

Full system load, not PPT

Can you imagine 210W @ 1.2V??? Hello alder lake lol

 

[mstenholm]

215w *whistles*

She’s rippin

Voltage wise I think you are in the right neighborhood

I saw 240w once from my 5900X.. shake and bake.. I couldn’t control it

Remember that @phill has a pump in his system.

 

[Mussels]

Right: i did dumb things

Upside: 11C drop

Downside: even after flushing the loop like 3 times, i'm still getting shit appearing in my CPU block. I must have a gunky radiator

It's extra effort to rip the socket off the mobo to help with pin safety, but worth it.

 

[phill]

215w *whistles*

She’s rippin

Voltage wise I think you are in the right neighborhood

I saw 240w once from my 5900X.. shake and bake.. I couldn’t control it

Full system load, not PPT

Can you imagine 210W @ 1.2V??? Hello alder lake lol

These are figures quoted from the wall, not the CPU It's AMD not Intel for the moment

I think CoreTemp reports about 120w or so from the CPU...?

I have a EK D5 pump, 3 140mm Noctua PPC fans in there as well plus my 1080 TI, but that's idle... Overall, very impressed with the power usage, even more so with 32 threads I will try and have another go tonight and see if we can get it boosting a bit higher. Temps are right about the 50C mark which is where I prefer them to be simply because they then have some space to move for the summer temps and such Plus less stress on the CPU for it being loaded 100% all the time

Right: i did dumb things

Upside: 11C drop

Downside: even after flushing the loop like 3 times, i'm still getting shit appearing in my CPU block. I must have a gunky radiator

It's extra effort to rip the socket off the mobo to help with pin safety, but worth it.

View attachment 224491
View attachment 224489View attachment 224490

There's no such things as dumb if there's improvements in my eyes

 

[Mussels]

Between the lapping, the better LM application (used a flat piece of card this time instead of generic cotton bud) and fixing the offset, the 75C spikes dropped to this:

(This is farting about in windows, browser, VLC etc... nothing heavy, but single thread boost is where the heat was coming in)

Oh and proof she's hitting 5050MHz at those temps:

 

[Nordic]

Forgive me father for I am being tempted into sin. I might have to find my 1500 / 2000/ 3000 grit sand paper.

 

[freeagent]

You are forgiven my son.

 

[Ibizadr]

That lapping was amazing and the am4 support to dont brake pins its a most do it. I think this is a good mod to help in temps

 

[tabascosauz]

Between the lapping, the better LM application (used a flat piece of card this time instead of generic cotton bud) and fixing the offset, the 75C spikes dropped to this:

View attachment 224493

(This is farting about in windows, browser, VLC etc... nothing heavy, but single thread boost is where the heat was coming in)

Oh and proof she's hitting 5050MHz at those temps:
View attachment 224494

Looking good. I would dare sacrifice my $800cad CPU to the sandpaper gods

Not much point now if you didn't also test beforehand, but for posterity if you want a much more consistent single thread temp test, use corecycler function in OCCT on one specific core, or the Corecycler script and edit the config file to only run on the core of your choice. At HWInfo's default polling interval 2000ms the Ryzen "spikes" are too fleeting to consistently capture in logging data, which is why sometimes you might see say 58C after booting your PC and other times 72C despite nothing appearing to have changed, or one day a game will suddenly decide to spike 10C hotter than it normally does.

On the other hand the default Core Clock metric is a little too optimistic about boosting, it will pick up every split second the CPU spent at 5050MHz even if it was way too short to ever have an effect on temps. If you use Snapshot Polling you can get a much more accurate idea of actual max boosting behaviour, without looking at Effective Clock.

Here's mucking about in Windows:

 

[dont whant to set it"']

I wanted to run this stress test only for about 15 minutes while I took a shower before having to leave for work, for witch I ended up leaving in a rush, so 11hours later, the meanwhile the attached screenshot.

 

[tabascosauz]

I wanted to run this stress test only for about 15 minutes while I took a shower before having to leave for work, for witch I ended up leaving in a rush, so 11hours later, the meanwhile the attached screenshot.

CPU-Z is just bugged like that if you hit Stress. The scores often don't make sense - it'll dip to like 0 then for the next 5 minutes proceed to slowly climb back up to normal score, then do it all again. Happens like every other run on my 5900X when I'm testing for thermals.

Stress mode should still be hitting the CPU the same while the score is bugging out, so as a temperature test that actually reaches thermal equilibrium quickly (without constantly dipping out like Cinebench), it still works well.

In your case though, it looks like something else is bugged as it's just not loading some of the cores, doesn't happen on mine. Seems like you did something to it for only to run on 4C/8T - safe mode?

 

[Mussels]

I was gaming while mining today, and with the extra heat from the GPU one core still hit 75C

The thing is... i was a good 40FPS up from where i was yesterday. No changes to power settings, simply fixing the cooling got me more perf at the same settings and temps. I need to cinebench...

15,130
PPT 108
EDC 99
TDC 68