Ryzen Owners Zen Garden

[phill]

I think this might be the thread I'm needing!!

So I've been messing about with the Ryzen's I have here.. I've yet to try my two 1700X's and a 2700 but I've been trying my 2700X and my 3900X with some what I'd class as some interesting results, mainly on my 3900X....

So short story version...
Kept hitting 75 to 80C or sometimes more when using without any all core boost and just leaving on auto.. Managed to do a few simple CPU tests with running WCG as my testing bench as it warms up the CPU nicely and contributes a little something so... win win

So I've had the CPU for a while, left most at stock as I've just enjoyed it as it just boots, works and I can shut it down and it will boot up from cold, outstanding kit

So I've gone from auto volts to 1.0vcore set in the bios. I've been able to get a 4.20GHz boost, but when I've tried all cores with 4.20GHz, I've had to push 1.15vcore through the CPU, whilst the temps haven't been bad either I've till opted for the 1.0vcore option instead... Has anyone had this issue before at all?? (I mean all core boost working on auto but not when it's set at the same volts??)

Screen shot as I think this helps better

I should have grabbed it whilst it had done over 12 hours (damn OCD of mine...) but never mind... I hope you can see that ok..

So another question is, what vcore reading would you trust?? What I've set in the bios? What CPU-Z reads? What HWInfo reads? Or what CoreTemp reads?? Answers on a postcard please too....

The other twist was that the 3900X at auto settings was pulling about 280w from the wall full systems (no GPU load tho...) When it's set to 1.00vcore, it's around the 150w marker.... I've tested it with CB R20, there's a bit of a points difference from 1.00vcore/4.20GHz reading to the 1.10vcore/4.00GHz+ reading... Around 6000 points for the 1.00v setting and 7050 points for the 1.10vcore/4.00GHz setting.... I've recently updated the bios on my Asus Crosshair 8 Hero to 1201 (it was on 0605 I believe??) so a little out of date... Originally did the testing on the original bios the board shipped with. Was able to get the 3900X posting and in to Windows at 0.90vcore... Have I hit the jackpot or something??

Would love some others opinions on this one


On the 2700X it seems I can't set it lower than about 1.275vcore else it won't post... Are those CPUs normally about this sort of voltage or do they/can they go down lower? I see the motherboards LCD read out, gives it an error code of 8 I believe?? (Asus Crosshair 7 Hero board...) Just trying to do the same thing as my 3900X.. they are both on air cooling which is not going to allow them to reach their max but I'm looking for a little more efficiency as well as they are crunching machines (the 3900X is my gaming rig as well...)

Any sort of advice I'm all ears

 

[Toothless]

Okay, had a little fun today.

So it's liking 4.5ghz. I dropped the core a bit and during a CPU-z run, it went down to 1.225v (set in BIOS) but that was all it could do. It also doesn't like 1.25 for normal usage.

(^During the test^)


The lowest it'll go is 1.275v under full load, and seems to give a big middle finger to any LLC changes and doesn't like going that low for all applications like P95. So I'm sticking with a set 1.35v with it going to 1.3v under full load. Was worth a shot.

Also it tells me to fuck off with ANY ram changes. I attempted going to 3400mhz with 1200mhz fabric clock and it wouldn't post at all. 3200/C16 and 1100 is the best there.

Side note: the naming for this one was based off the fact it was on an outdated BIOS. Veraltet seemed to hit the spot. (Volt for my Z97 since the core voltage was 1.45v on my 4790k w/ stock cooler, and Dualie is for the dual cpus.)

 

[phill]

Are you on air @Toothless ? I thought I might have seen that in a post I went past this morning (it was late and I was tired, my brain might not have been engaged lol )

I've got to try that Fabric clock, I think it's all stuck at Auto at the moment.

I also had not much help with the Ryzen Master software, turns out, it wasn't doing much at all... I'd stick to the bios for the overclocking, think it's the best way

 

[tabascosauz]

@Toothless Half of 3200 is 1600, and half of 3400 is 1700...at 3400-1200 and 3200-1100 you're not even synced up 1:1. Inability to sync 1200 FCLK/UCLK is barely above JEDEC speeds and would be an instant QC fail for a Matisse CPU.

At those speeds, it's most definitely the RAM. It's always necessary to find out what sort of ICs you have under the heatspreaders, by opening up Thaiphoon. Without that info, you're getting nowhere. For everything except maybe B-die, you can't just take the freq up from 3200 without relaxing certain timings accordingly.

I'm guessing you have a run of the mill G.skill kit rated for 3200 16-18-18-38 @ 1.35V. ICs in those could be a lot of things.

@phill no, CPU-Z may tell you the manufacturer and whether it's single or dual rank, but not the actual IC. HWInfo will only tell you the current primary timings (eg 16-16-16-36-48-480-1T) and FCLK/UCLK/MCLK speed.

 

[phill]

@Toothless Half of 3200 is 1600, and half of 3400 is 1700...at 3400-1200 and 3200-1100 you're not even synced up 1:1. Synced 1200 FCLK/UCLK is barely above JEDEC speeds and would be an instant QC fail for a Matisse CPU.

At those speeds, it's most definitely the RAM. It's always necessary to find out what sort of ICs you have under the heatspreaders, by opening up Thaiphoon. Without that info, you're getting nowhere. For everything except maybe B-die, you can't just take the freq up from 3200 without relaxing certain timings accordingly.

I'm guessing you have a run of the mill G.skill kit rated for 3200 16-18-18-38 @ 1.35V. ICs in those could be a lot of things.

Could you get the info from CPU-Z or HWInfo @tabascosauz ??

 

[TheLostSwede]

Could you get the info from CPU-Z or HWInfo @tabascosauz ??

No, you need this http://www.softnology.biz/

 

[phill]

No, you need this http://www.softnology.biz/

Thanks @TheLostSwede

 

[Lorec]

Okay, had a little fun today.

So it's liking 4.5ghz. I dropped the core a bit and during a CPU-z run, it went down to 1.225v (set in BIOS) but that was all it could do. It also doesn't like 1.25 for normal usage.

(^During the test^)


The lowest it'll go is 1.275v under full load, and seems to give a big middle finger to any LLC changes and doesn't like going that low for all applications like P95. So I'm sticking with a set 1.35v with it going to 1.3v under full load. Was worth a shot.

Also it tells me to fuck off with ANY ram changes. I attempted going to 3400mhz with 1200mhz fabric clock and it wouldn't post at all. 3200/C16 and 1100 is the best there.

Side note: the naming for this one was based off the fact it was on an outdated BIOS. Veraltet seemed to hit the spot. (Volt for my Z97 since the core voltage was 1.45v on my 4790k w/ stock cooler, and Dualie is for the dual cpus.)

Whats up with those huge PPT, TDC and EDC values? Is that on PBO enabled?
Did You change those values in bios?

 

[Nordic]

Have you tried running a per CCD overclock? I run CCD0 @ 4.4 and CCD1 @ 4.2 with peak voltage set to 1.325v in Ryzen Master. Am running it like that for about a month now, really like the performance and the temps are lower also.

I have not tried that. I was thinking about trying to overclock each CCD individually to see what they can do individually.

 

[Toothless]

@phill NH-D15

@tabascosauz 3400 and 1200 was set in the bios. Instant nope.

@Lorec PBO off, those all were set to auto

 

[tabascosauz]

Whats up with those huge PPT, TDC and EDC values? Is that on PBO enabled?
Did You change those values in bios?

Those are the Ryzen Master limits. When you manual OC in RM, the gloves come off and those are basically the limits. Basically, much less of a safety net than PBO would provide.

Does work with IF properly set to 1600? I find it hard to believe that the Auto setting on an X570 board for 3200 RAM is not IF 1600. Where is your fabric clock setting located, under memory section?

 

[Nordic]

You can set PPT, TDC, and EDC much higher via the bios also.

 

[tabascosauz]

You can set PPT, TDC, and EDC much higher via the bios also.

The difference there being that PPT, TDC and EDC are usually found under the PBO menu. PBO is still utilizing the stock boost mechanisms, just with higher ceilings, so setting insanely high limits for those three often makes zero difference. Manual OC, however, has no safety limits with regards to voltage and temperature (apart from the usual Tjunction).

 

[heky]

The difference there being that PPT, TDC and EDC are usually found under the PBO menu. PBO is still utilizing the stock boost mechanisms, just with higher ceilings, so setting insanely high limits for those three often makes zero difference. Manual OC, however, has no safety limits with regards to voltage and temperature (apart from the usual Tjunction).

Might be, but i just dont see the problem of me setting a max voltage of 1.325v (which is only being used under really low loads), since the voltage goes down as it should under load! The manual oc also lowered the temps for me and upped the performance. So what safety limit is not there anymore that can hurt my chip the way i run it now?

 

[tabascosauz]

Might be, but i just dont see the problem of me setting a max voltage of 1.325v (which is only being used under really low loads), since the voltage goes down as it should under load! The manual oc also lowered the temps for me and upped the performance. So what safety limit is not there anymore that can hurt my chip the way i run it now?

I didn't say anything about Vcore. I was talking about PPT/EDC/TDC. There's plenty of discussion on FIT. Stilt's initial estimate of a 1.325V max safe static voltage turned out not to be the case when some owners took his advice and ended up with degraded silicon a few months later. Others have allegedly been running 1.4V all core and suffering no ill effects, though usually under water. Anyhow, there's no point getting into all of this again.

How much is Vcore "dropping"? If it's only dropping by 0.025-0.05V, that's just stock LLC vdroop behaviour.

 

[heky]

How much the voltage drops depends on the load. For example under CB R20 its 1.244v, AIDA64 FPU its 1.232v and under Prime95 smallFFTs its 1.213v.

Edit: Comparing the stock settings, without PBO, just plain stock settings, the voltage pushed for a CB 20R allcore load is 1.35v!!!!! It creates much more heat and lower scores! So i just dont get the reasoning behind why a stock chip, that gets 1.35v under CB R20 is safer, than my chip with a maximum voltage of 1.325v set in Ryzen Master that gets 1.244v under CB R20, runs cooler and gets better scores!

 

[tabascosauz]

How much the voltage drops depends on the load. For example under CB R20 its 1.244v, AIDA64 FPU its 1.232v and under Prime95 smallFFTs its 1.213v.

Okay, I see what you mean. AMD's replaced the set voltage box in RM with "peak voltage" instead. Doing so still appears to set exactly that voltage, however. Ryzen Master reports a varying idle voltage, but everything else just reports that manual voltage. Fire up P95 and it's just exactly that many volts at load. I'm not seeing any boosting going on with the manual OC setting.

 

[heky]

Yeah, but the thing is the voltage under Prime95 load can not be the "peak voltage" reported by Ryzen Master. The power draw (A) and the reported Wattage dont add up. Also the temps are lower, so that also doesnt add up. I read the voltage that HWInfo64 reports under SVI2 TFN and the IR35201 WRM controller shows the same values under different loads i wrote in the post above (1.213v Prime95). The temps also confirm that the voltage under load gets droped or else the temps would be much much higher.

 

[tabascosauz]

Yeah, but the thing is the voltage under Prime95 load can not be the "peak voltage" reported by Ryzen Master. The power draw (A) and the reported Wattage dont add up. Also the temps are lower, so that also doesnt add up. I read the voltage that HWInfo64 reports under SVI2 TFN and the IR35201 WRM controller shows the same values under different loads i wrote in the post above (1.213v Prime95). The temps also confirm that the voltage under load gets droped or else the temps would be much much higher.

Are you on RM 2.2.0.1543? On the main screen there's also an Average Core Voltage field, which is a bit like HWInfo's "effective" voltage. At idle, the Average is fluctuating well below 1.0V like it is normally, but as soon as I run P95 it comes up and matches whatever I've set as Peak Voltage. I haven't seen any kind of voltage dropping at load.

Also, I'm not 100% sure, but I can recall reading that EDC is just a limit, not a real-time current reading. I'll have to double check that. Using either RM's Average reading or HWInfo's SVI2 value, multiplying that by the Output from the CPU side of the IR35201 is roughly what CPU Power is reporting in HWInfo.

However, when I do test fixed clocks, including my stable 4.0/1.2V setting, I am seeing what you describe. SVI2, RM and IR35201 all report the same Vcore, but temps are lower, and IR35201 output amps x Vcore =! RM's stated wattage. On the other hand, they do add up to the power draw in HWInfo. So it may just be an RM thing.

 

[Nordic]

What is a good ram set to get for ryzen? Given that I have such a crappy ram set, it is something I keep thinking about upgrading even though what I have gives okay performance.

This is what I have now. 3600mhz Cas 18 XMP

 

[TheLostSwede]

What is a good ram set to get for ryzen? Given that I have such a crappy ram set, it is something I keep thinking about upgrading even though what I have gives okay performance.

This is what I have now. 3600mhz Cas 18 XMP

I can only recommend what I have, although a few others here have gotten them as well.
Hard to beat for the price. Not going to say it's the best memory out there, as it's far from it, but Ryzen 3000 CPUs seems to be happy with it.

Viper Steel DDR4 Performance Memory DRAM | Viper Gaming

Patriot’s Viper Steel memory modules are designed with true performance in mind. Built for the latest Intel® and AMD™ platforms, our Viper Steel DDR4 memory provides the best performance and stability for the most demanding computer environments.The Viper Steel memory utilizes a custom designed...

viper.patriotmemory.com

This is with 4x 8GB.

 

[tabascosauz]

What is a good ram set to get for ryzen? Given that I have such a crappy ram set, it is something I keep thinking about upgrading even though what I have gives okay performance.

This is what I have now. 3600mhz Cas 18 XMP

It's already Rev.E, one of the top ICs for Ryzen, easily on par with CJR. Only concern is quality and consistency as with every other IC put in Vengeance, because it's cheap Corsair. How about plugging in some DRAM calc profiles to see what it can do? I don't know if your sticks are single- or dual-rank.



Aside from Team, BZ has a new video up with a large selection of 16/32/64GB Crucial kits. Higher end Crucial kits with good freq/timings XMP profiles will probably be Rev.E, because Crucial is Micron, and Rev.E is Micron's best IC. But that's only if you've actually tested your current kit and found that it actually sucks ass, because if not, you're just paying more for what you already have. Albeit with probably better binning. You never know until you try.

Then there's G.Skill Trident Z / RGB / Neo kits, whatever they might be called, as long as they're rated for 16-19-19-39 @ 3600. CJR and DJR. All of these kits usually have good Ryzen compatibility.

 

[Nordic]

@tabascosauz I apologize if you remember this conversation, but I am going to refer you to my posts in this thread from last month. We determined that my memory does seem to be the issue. https://www.techpowerup.com/forums/threads/ryzen-owners-zen-garden.231658/page-85#post-4247124

 

[TheLostSwede]

@tabascosauz I apologize if you remember this conversation, but I am going to refer you to my posts in this thread from last month. We determined that my memory does seem to be the issue. https://www.techpowerup.com/forums/threads/ryzen-owners-zen-garden.231658/page-85#post-4247124

Oh wow, the latencies you're getting are indeed poop. No idea what's going on there. Any chance your IF is locked to some super slow speed or 1:2 mode by mistake? It's the only thing that makes sense for latencies that bad.

 

[Nordic]

Oh wow, the latencies you're getting are indeed poop. No idea what's going on there. Any chance your IF is locked to some super slow speed or 1:2 mode by mistake? It's the only thing that makes sense for latencies that bad.

It appears not. I am able to boot with 1833mhz FCLK. My memory won't do that though.

Edit: Since those posts last month, I discovered that I can run my memory at XMP but only with PBO off. At XMP settings I get 3600mhz Cas 18, Best time: 108.5, Random Latency: 78.14.