Ryzen Owners Zen Garden

[mechtech]

Hello, it's been awhile since I have done any tweaking (1700) days. I've been pretty happy with the 5700X, especially considering what I paid for it.

Just wondering how to save more power.........specifically idle and light loads..........if it's even possible? UPS tells me about 88-90W 2d desktop nothing open and about 66W monitor off. I tried enabling eco mode in bios and that did nothing (or appears to anyway).

 

[tabascosauz]

It's a desktop, man..........lol. 90W for idle desktop is good for a 1CCD part with a dGPU that is also idling normally. I'm at 120W usually due to my 165Hz monitors' draw.

If you're idling at about 15-20W on Package Power in HWInfo there's not a whole lot you can do, that's about normal already. You can try cutting down on extra VSOC to try and lower your idle SOC Power draw a bit, but unless we're talking major drops (ie. 1.1V down to 0.9V) you won't save much on power. You can try some other stuff like lowering your PLL (1.8V default but called different things depending on vendor), but it doesn't do a whole lot for power or anything else for that matter.

Play with SOC/Uncore OC mode (either CBS or AMD OC menu can't remember) to see if it makes a difference, usually not. Make sure C-states are enabled. etc

Eco Mode only adjusts limits. It's the same as punching in the values yourself under PPT/TDC/EDC. It makes zero difference for idle state.

There are a couple things that will for sure lower your idle power:

• Get a slower 5700G, because the APUs idle at 5-10W. Or Intel.
• Get rid of your dGPU lol
• Start getting rid of fans and SATA drives
• If your dGPU idles at above 20W, consider getting a GPU that is more efficient at idle.
• Switch to a Rembrandt or Phoenix laptop and enjoy sub-10W idle.

I don't think those are good solutions for this problem that isn't really a problem.

 

[mechtech]

It's a desktop, man..........lol. 90W for idle desktop is good for a 1CCD part with a dGPU that is also idling normally. I'm at 120W usually due to my 165Hz monitors' draw.

If you're idling at about 15-20W on Package Power in HWInfo there's not a whole lot you can do, that's about normal already. You can try cutting down on extra VSOC to try and lower your idle SOC Power draw a bit, but unless we're talking major drops (ie. 1.1V down to 0.9V) you won't save much on power. You can try some other stuff like lowering your PLL (1.8V default but called different things depending on vendor), but it doesn't do a whole lot for power or anything else for that matter.

Play with SOC/Uncore OC mode (either CBS or AMD OC menu can't remember) to see if it makes a difference, usually not. Make sure C-states are enabled. etc

Eco Mode only adjusts limits. It's the same as punching in the values yourself under PPT/TDC/EDC. It makes zero difference for idle state.

There are a couple things that will for sure lower your idle power:

• Get a slower 5700G, because the APUs idle at 5-10W. Or Intel.
• Get rid of your dGPU lol
• Start getting rid of fans and SATA drives
• If your dGPU idles at above 20W, consider getting a GPU that is more efficient at idle.
• Switch to a Rembrandt or Phoenix laptop and enjoy sub-10W idle.

I don't think those are good solutions for this problem that isn't really a problem.

Thanks, Understood. Just wasn't sure as my 1700 system was less. The 6800 at idle in the catalyst or whatever says 5-7W. I know the 5k series is a different beast, low fan speeds are a nice, even more so than the watts..............but with my low flow heat box they go hand in hand.

I guess reading about how gigbayte boards tend to push the voltage to the cpu got me interested, I've done a bit of reading about -20 offset, about so much other stuff.

Is there a cpu voltage which a stock 5700x shouldn't need to really go over without being unstable? And I guess with idle out of the way, what about pinned out?? I remember I used to be able to reduce voltage on my olf PII 955 BE from 1.35-1.40V to 1.20V

 

[tabascosauz]

Thanks, Understood. Just wasn't sure as my 1700 system was less. The 6800 at idle in the catalyst or whatever says 5-7W. I know the 5k series is a different beast, low fan speeds are a nice, even more so than the watts..............but with my low flow heat box they go hand in hand.

I guess reading about how gigbayte boards tend to push the voltage to the cpu got me interested, I've done a bit of reading about -20 offset, about so much other stuff.

Is there a cpu voltage which a stock 5700x shouldn't need to really go over without being unstable? And I guess with idle out of the way, what about pinned out?? I remember I used to be able to reduce voltage on my olf PII 955 BE from 1.35-1.40V to 1.20V

1700 was still monolithic, as are all the APUs. Thus power-wise they behave exactly the same as Intel at idle. But if you want chiplets, this is what you have to deal with - a significant wattage that is always lost to an unknown source (not reported as cores or SOC power), presumably the interconnect over substrate. Those 6800 numbers look good.

The Gigabyte boards should be fine these days with regards to Vcore, it was more prevalent back in Z390/X470 days. The GB overvolting can still apply to VSOC, and always applies to VDIMM, but those are pretty easy to manually compensate for on AM4 if present, and are not dynamic in nature like Vcore.

Thus there's not really any reason to be paying attention to what is "safe" Vcore and what is not. If your scores are fine, temps are reasonable in the circumstances, and Power Reporting Deviation in HWInfo is around 95-105% at full load (any all core load), then things are fine. Vcore varies wildly by design with nature of load, type of instructions, current draw, # of cores/threads online. Let the boost algorithm do its job, unless something is clearly out of wack (ie. 1.6V logged peaks for Vcore).

Curve Optimizer (e.g. -20) is not a fixed value in millivolts. Neither does it usually directly interact with Vcore. It is a lot like dragging down points on the V-F curve for newer Intel CPUs in XTU - "undervolting" will allow you to achieve lower Vcore/temps/power for a given clock, or more clock for a given Vcore (usually the latter if you don't touch power limits).

 

[mechtech]

1700 was still monolithic, as are all the APUs. Thus power-wise they behave exactly the same as Intel at idle. But if you want chiplets, this is what you have to deal with - a significant wattage that is always lost to an unknown source (not reported as cores or SOC power), presumably the interconnect over substrate. Those 6800 numbers look good.

The Gigabyte boards should be fine these days with regards to Vcore, it was more prevalent back in Z390/X470 days. The GB overvolting can still apply to VSOC, and always applies to VDIMM, but those are pretty easy to manually compensate for on AM4 if present, and are not dynamic in nature like Vcore.

Thus there's not really any reason to be paying attention to what is "safe" Vcore and what is not. If your scores are fine, temps are reasonable in the circumstances, and Power Reporting Deviation in HWInfo is around 95-105% at full load (any all core load), then things are fine. Vcore varies wildly by design with nature of load, type of instructions, # of cores/threads online. Let the boost algorithm do its job.

Curve Optimizer (e.g. -20) is not a fixed value in millivolts. Neither does it usually directly interact with Vcore. It is a lot like dragging down points on the V-F curve for newer Intel CPUs in XTU - "undervolting" will allow you to achieve lower Vcore/temps/power for a given clock, or more clock for a given Vcore (usually the latter if you don't touch power limits).

Right on, so stock default bios settings as per day 1 and carry on.

 

[freeagent]

But you must touch power limits, moar

You must exploit that CPU for every MHz it has to offer

 

[tabascosauz]

them: you can't hear pictures, silly
me:

View attachment 306135

If you really wanted to (and your board decides to cooperate instead of going haywire), you probably could probably try punching in a manual Vcore. It might keep normal Precision Boost behaviour at your specified Vcore. It just doesn't make a lot of sense to do it - you probably strictly need somewhere between 1.3-1.5V to sustain the 5700X's stock single core 4650MHz Fmax, but a stock 5700X runs all-core at like 1.1V or even lower. You can see why fixed Vcore doesn't work.

Dynamic Vcore doesn't really work either (e.g. -0.050V offset), it's still one value at all times. If you also dial in your limits and other PBO parameters, negative offset can help reduce temps and improve all-core performance but single core is likely to suffer because that's what leverages the top end of the Vcore envelope by design.

 

[freeagent]

Yup..

I gave up being gentle. Mine don't like that. Have to be forceful

They like it!

 

[tabascosauz]

I got curious the other day because Dragon Range brings 2CCD to laptops. Lo and behold it idles much better than Raphael and is certainly workable for the target segment (thick and chunky laptops), but obviously worse than Phoenix and not appropriate for thinner laptops.

But maybe only because mobile lets both UCLK and FCLK fluctuate low at all times, and also only 1:2 UCLK for DDR5. So probably no real change in the near future as to chiplet idle power on desktop.

Still, maybe changes are coming soon to the Fabric design, seeing as Raphael runs out of bandwidth in memory-heavy situations at around ~10 cores loaded.

 

[kashika]

What are your thoughts on what I should do if I am still having USB issues with my 5800x? I tried updating to many of the new bios versions and even tweaking my VDDG voltages. I think the issue is just a bad I/O chipset on the motherboard itself because a Ryzen 3100 has the same issues on the same board. My board is out of warranty (MSI B550 gaming plus). I don't know if I should switch to AM5 or Intel or roll the dice with another AM4 board? It sucks the 5800x has been great besides the USB issues I just don't know what the best move is here.

 

[Klemc]

1634 BETA...
... is Agesa 1.0.0.7b new BIOS.

I'll wait stable.

ROG STRIX B650E-E GAMING WIFI | Gaming motherboards｜ROG - Republic of Gamers｜ROG France

The ROG Strix B650E-E Gaming WiFi is a future-ready B-series beast, featuring 16 + 2 power stages, hyperspeed DDR5 memory, and next-gen PCIe® 5.0 connectivity. It features four M.2 slots with heatsinks, USB 3.2 Gen 2x2 support, WiFi 6E, AI Cooling II, AI Networking, Two-Way AI Noise Cancelation...

rog.asus.com

 

[Nordic]

But you must touch power limits, moar

You must exploit that CPU for every MHz it has to offer

Hell yeah brother! I am finally beginning to find the 7800x3d's weak points. I am testing the following now:

• Core 0 = -50
• Core 1 = -60
• Core 2 = -60
• Core 3 = -40
• Core 4 = -45
• Core 5 = -60
• Core 6 = -60
• Core 7 = -60

I can't find many user examples of people getting down to -45 but it is hard to tell what is normal from random searches. I may have won the silicon lottery.

 

[TheLostSwede]

Not sure how much it matters any more, but I don't seem to have much luck when it comes to getting CPUs with an early core as the top performing core...

 

[tabascosauz]

Not sure how much it matters any more, but I don't seem to have much luck when it comes to getting CPUs with an early core as the top performing core...

View attachment 306208

It hasn't mattered on 1CCD CPUs ever since the CCX became 8 cores. CPU-Z bench might be one of the only holdouts that only knows how to use Core 0. 99% of the time CPPC and the scheduler know what to do.

Besides, the "rankings" don't mean much. It's not hard to get a CPU on which a lower rank core is able to run higher CO offset to functionally surpass the starred cores.

 

[Nordic]

Not sure how much it matters any more, but I don't seem to have much luck when it comes to getting CPUs with an early core as the top performing core...

View attachment 306208

Interesting. My Ryzen master doesn't even show the star or circle. I thought they just got rid of the preferred cores. Do you have the option to do the auto curve optimizer? I wanted to use that as a baseline before manual tuning but the option doesn't even appear. I reinstalled Ryzen master three times too.

 

[3x0]

Interesting. My Ryzen master doesn't even show the star or circle. I thought they just got rid of the preferred cores. Do you have the option to do the auto curve optimizer? I wanted to use that as a baseline before manual tuning but the option doesn't even appear. I reinstalled Ryzen master three times too.

Have you changed CPPC options in BIOS?

 

[TheLostSwede]

Interesting. My Ryzen master doesn't even show the star or circle. I thought they just got rid of the preferred cores. Do you have the option to do the auto curve optimizer? I wanted to use that as a baseline before manual tuning but the option doesn't even appear. I reinstalled Ryzen master three times too.

Haven't started playing around with things properly, as I'm only running this on top of a desk at the moment, will be at least another month or so until I'll have a complete system.

 

[Count von Schwalbe]

What are your thoughts on what I should do if I am still having USB issues with my 5800x? I tried updating to many of the new bios versions and even tweaking my VDDG voltages. I think the issue is just a bad I/O chipset on the motherboard itself because a Ryzen 3100 has the same issues on the same board. My board is out of warranty (MSI B550 gaming plus). I don't know if I should switch to AM5 or Intel or roll the dice with another AM4 board? It sucks the 5800x has been great besides the USB issues I just don't know what the best move is here.

@Mussels has done some testing, he could probably help you out.

What is your VSOC?

 

[Nordic]

Have you changed CPPC options in BIOS?

I have not knowingly changed this setting in the BIOS. I cannot even find this setting in the BIOS. I have a Gigabyte b650 motherboard. Any idea's where they might have placed it?

 

[3x0]

I have not knowingly changed this setting in the BIOS. I cannot even find this setting in the BIOS. I have a Gigabyte b650 motherboard. Any idea's where they might have placed it?

Does your BIOS have a search function? It's called Collaborative Processor Performance Control

 

[Nordic]

Although my performance is demonstrably quite good, something isn't right. I was looking at Ryzen Master and it is not displaying my CO settings input into the BIOS correctly. It should read as:

• Core 0 = -47
• Core 1 = -60
• Core 2 = -60
• Core 3 = -42
• Core 4 = -47
• Core 5 = -60
• Core 6 = -60
• Core 7 = -60

Not only does it not show my cores set to -60, Core 3 should show -42 not Core 4.

I have had the best luck finding errors/instability using Y-Cruncher with the "22-ZN4 ~ Kizuna" test mode.

Does your BIOS have a search function? It's called Collaborative Processor Performance Control

My BIOS does not have a search function but I found it upon a third inspection under SMU options. It is set to Auto but could also be set to Cache, Frequency, or Driver. I tried setting it to each of these options but Ryzen Master does not show preferred cores.

 

[tabascosauz]

Although my performance is demonstrably quite good, something isn't right. I was looking at Ryzen Master and it is not displaying my CO settings input into the BIOS correctly. It should read as:

• Core 0 = -47
• Core 1 = -60
• Core 2 = -60
• Core 3 = -42
• Core 4 = -47
• Core 5 = -60
• Core 6 = -60
• Core 7 = -60

Not only does it not show my cores set to -60, Core 3 should show -42 not Core 4.

I have had the best luck finding errors/instability using Y-Cruncher with the "22-ZN4 ~ Kizuna" test mode.

My BIOS does not have a search function but I found it upon a third inspection under SMU options. It is set to Auto but could also be set to Cache, Frequency, or Driver. I tried setting it to each of these options but Ryzen Master does not show preferred cores.

Didn't the Ryzen Master notes already state that it caps out at -50?

If supported, you can double check with PBO2 Tuner to see what it says your settings are.

All you found in the BIOS were the settings intended for load shifting on 7950X3D and 7900X3D. If they work properly then they should not work at all for a 7800X3D. That's not the CPPC or CPPC Preferred Cores switch, which are just on/off.

I have not seen in any BIOS where CPPC and CPPC Preferred Cores was spelled out, only as an acronym. In the past on Gigabyte it should either be on Tweaker page under Advanced CPU settings or maybe somewhere under the advanced menus. If Preferred Cores is disabled, all 8 cores should all display as rank #1 in most software. However, maybe it's just a Ryzen Master thing on the new CPUs. I disable Preferred Cores and Ryzen Master still shows me my stars.

 

[Nordic]

Didn't the Ryzen Master notes already state that it caps out at -50?

If supported, you can double check with PBO2 Tuner to see what it says your settings are.

I didn't know that Ryzen master didn't show the full range for 7000 series CPU's. I assume that will be fixed eventually. It is also displaying the values for each core incorrectly. It does not show preferred cores for me but it does for TheLostSwede.

PBO2 Tuner does not appear to be supported.

HWinfo will show the preferred cores even if Ryzen Master does not.

 

[tabascosauz]

I didn't know that Ryzen master didn't show the full range for 7000 series CPU's. I assume that will be fixed eventually. It is also displaying the values for each core incorrectly. It does not show preferred cores for me but it does for TheLostSwede.

PBO2 Tuner does not appear to be supported.

HWinfo will show the preferred cores even if Ryzen Master does not.

But that's the thing, do you have confirmation that those BIOS values are really applying in firmware? e.g. in timings sometimes you can set x number, BIOS would ignore it and apply y number, but would still show x number in the box every time you visit BIOS. Hard to tell without another software to check. If you tested with something like corecycler then a non-negligible improvement in effective clock could prove that there is real range beyond -50, but that would be assuming:

• you aren't already hitting 5050 at -0
• you aren't already hitting 5050 at -50
• silicon has the ability to scale past -50

idk if 7800X3D has access to boost clock override, if 5050 or 5250 is the hard limit

Sounds like ryzen master might just be bugged still. HWInfo is showing that you are using Preferred Cores, otherwise they would look like:

 

[Nordic]

But that's the thing, do you have confirmation that those BIOS values are really applying in firmware? e.g. in timings sometimes you can set x number, BIOS would ignore it and apply y number, but would still show x number in the box every time you visit BIOS. Hard to tell without another software to check. If you tested with something like corecycler then a non-negligible improvement in effective clock could prove that there is real range beyond -50, but that would be assuming:

• you aren't already hitting 5050 at -0
• you aren't already hitting 5050 at -50
• silicon has the ability to scale past -50

im not sure if 7800X3D has access to boost clock override, if 5050 is the hard limit

Sounds like ryzen master might just be bugged still. HWInfo is showing that you are using Preferred Cores, otherwise they would look like:

View attachment 306286

Ryzen Master is having several issues. That's for sure.

I'm not sure what other software can do that. PBO Tuner's ReadMe suggests that HwInfo can be used to see CO values. I am on the latest version of HwInfo64 and don't see them shown.

5050 is the hard limit. I can't go higher without adjusting bclk. This motherboard doesn't have an eclk option.

Having tested with a few different programs and different instruction sets, there does not appear to be any significant difference in effective clock speed between -50 and -60 co on the cores that can achieve those settings.