AMD Raven Ridge support?


Recommended Posts

unRAID runs on Raven Ridge as it does on Summit Ridge Ryzens. There are no built-in drivers for the Radeon Vega iGPU, but neither are there any built-in drivers for other AMD or nVidia GPUs. If, say, you have a working Ryzen-based unRAID server and you replace the processor with a Raven Ridge APU it will work the same way and you can connect your monitor to the motherboard connector and see either the text console or GUI. The motherboard BIOS will need to be a very recent version though.

Link to comment
1 hour ago, baconborn said:

Okay, so is what I was told about the Linux kernel version just misconstrued information then?

 

Also, since the new Ryzen APUs do work, is it possible to push plex transcoding onto the Vega iGPU?

 

The accelerated Vega iGPU drivers are in Linux 4.15 which would be needed for Plex transcoding presumably.  We only include the Intel iGPU drivers at this time but we're open to including Vega iGPU drivers when we upgrade kernels, Plex supports it and the Vega iGPU drivers don't conflict with normal dedicated AMD video cards that would normally be used for passthrough.

  • Like 1
Link to comment

By the way, I have solved the problem of the freezes with the C6 State.

Just undo all the changes you have made to disable C6 in unraid, BIOS, etc., go to the BIOS and increase the SOC and CPU voltage by 0,1V, if it doesn't work you can try increasing to 0,15 the SOC.

I have 3 days of uptime with this solution, before, without disabling C6 the it just lasted a few hours.

 

This Light overvolt will save energy and increasy the life of your CPU vs disabling C6 State. In addition you can try to increase the clocks a little bit to benefit from this overvolting.

 

http://www.silence.host/node/1

 

@Squid You could update your plugin with a link to this solution.

@limetech in case you are interested.

Edited by L0rdRaiden
Link to comment
4 hours ago, L0rdRaiden said:

By the way, I have solved the problem of the freezes with the C6 State.

Just undo all the changes you have made to disable C6 in unraid, BIOS, etc., go to the BIOS and increase the SOC and CPU voltage by 0,1V, if it doesn't work you can try increasing to 0,15 the SOC.

I have 3 days of uptime with this solution, before, without disabling C6 the it just lasted a few hours.

 

This Light overvolt will save energy and increasy the life of your CPU vs disabling C6 State. In addition you can try to increase the clocks a little bit to benefit from this overvolting.

 

http://www.silence.host/node/1

 

@Squid You could update your plugin with a link to this solution.

@limetech in case you are interested.

 

Are you the author of that silence.host page?  If so (or even if not) the people over at this thread would be really grateful if this is a true solution  (and they will pick apart if not):

https://bugzilla.kernel.org/show_bug.cgi?id=196683

 

Link to comment
On 17/03/2018 at 1:17 PM, L0rdRaiden said:

This Light overvolt will save energy and increasy the life of your CPU vs disabling C6 State.

 

Can you explain the reasoning behind this statement, please? Why does disabling the C6 state decrease the life of the CPU, and why does overvolting increase the life of the CPU and save energy? Doesn't overvolting cause more stress and heat?

Link to comment
5 hours ago, John_M said:

 

Can you explain the reasoning behind this statement, please? Why does disabling the C6 state decrease the life of the CPU, and why does overvolting increase the life of the CPU and save energy? Doesn't overvolting cause more stress and heat?

Maybe I went to far with that statement but if C6 is disable your cores will be always awake and with voltage, with C6 voltage can be 0 freq as well so you probably will have less heat an stress despite the small over voltage introduced

 

Take a look to the voltages in this comment and the one before

https://bugzilla.kernel.org/show_bug.cgi?id=196683#c255

Edited by L0rdRaiden
Link to comment
3 hours ago, L0rdRaiden said:

so you probably will have less heat an stress despite the small over voltage introduced

While turning off cores saves energy, it should be noted that large temperature variations gives more stress than constant temperature.

 

However, since the silicon is so small and there is a head spreader on top, I would assume that there will not be too large differences in temperature between different parts of the chip so the main temperature variations will be when the CPU is fully busy compared to when it's idle.

Link to comment
2 minutes ago, pwm said:

While turning off cores saves energy, it should be noted that large temperature variations gives more stress than constant temperature.

 

However, since the silicon is so small and there is a head spreader on top, I would assume that there will not be too large differences in temperature between different parts of the chip so the main temperature variations will be when the CPU is fully busy compared to when it's idle.

For me the interesting point is that most probably the power consumption will be lower, which in my case is a important thing.

Link to comment
6 minutes ago, John_M said:

 

I watch that thread and have been trying the "power supply idle control" option that was recently introduced with a BIOS update on a non-unRAID computer. But it doesn't explain why not allowing a processor to enter C6 state is bad for it.

Is not "bad" for it, it just it will run with more voltage and freq when the core is iddle because C6 is disable so the core can't go even lower in freq and voltage. Then you can take your own conclusions.

 

Disabling ONLY C6 is not a bad solution if your motherboard suports it, although I prefer a small overvolt and leave everything enable.

If you motherboard ONLY allows you to enable or disable ALL C states then I must recommend to overvolt a little de CPU to avoid freezes, and leave C states enabled

 

A little bit more info here http://www.silence.host/node/1

Edited by L0rdRaiden
Link to comment
4 minutes ago, John_M said:

But it doesn't explain why not allowing a processor to enter C6 state is bad for it.

It should not be bad for the processor not not be allowed to enter C6 - the only difference is if there is a little bit of power save to be gained.

Link to comment
3 minutes ago, pwm said:

It should not be bad for the processor not not be allowed to enter C6 - the only difference is if there is a little bit of power save to be gained.

 

I know. That's what I said. However,

 

On 17/03/2018 at 1:17 PM, L0rdRaiden said:

This Light overvolt will save energy and increasy the life of your CPU vs disabling C6 State.

 

Link to comment
2 minutes ago, John_M said:

 

I know. That's what I said. However,

 

 

I know - and at really low energy levels there shouldn't be any practical differences to expected lifetime. It takes high temperatures or large temperature swings to start to get any practical difference in expected lifetime. When the chip is operating at the lower power-savings levels, it has an expected lifetime that is orders of magnitude outside the economical lifetime.

 

But if I understand this thread correctly, the "overvolting" mentioned isn't actually overvolting to get the processor to run at a higher speed than normally supported but actually avoiding the processor to undervolt and fail to maintain internal state of processor and/or RAM at the lower power levels.

Link to comment
3 minutes ago, pwm said:

But if I understand this thread correctly, the "overvolting" mentioned isn't actually overvolting to get the processor to run at a higher speed than normally supported but actually avoiding the processor to undervolt and fail to maintain internal state of processor and/or RAM at the lower power levels.

 

I have a point about this, if you overvolt the CPU at max frequency it will probably drain the max voltage allowed, so is a waste of performance not to try to overclock it as well. It's just a though not a statment.

In my case I have increased the freq 200Mhz

Edited by L0rdRaiden
Link to comment
9 minutes ago, pwm said:

I know - and at really low energy levels there shouldn't be any practical differences to expected lifetime. It takes high temperatures or large temperature swings to start to get any practical difference in expected lifetime. When the chip is operating at the lower power-savings levels, it has an expected lifetime that is orders of magnitude outside the economical lifetime.

 

So the assertion that disabling C6 shortens the life of the processor in any meaningful way is false.

 

11 minutes ago, pwm said:

But if I understand this thread correctly, the "overvolting" mentioned isn't actually overvolting to get the processor to run at a higher speed than normally supported but actually avoiding the processor to undervolt and fail to maintain internal state of processor and/or RAM at the lower power levels.

 

Isn't that just an alternative way of preventing the processor entering the C6 state, since by the slight overvolt you are not allowing it to reach the low voltage that C6 requires? Or is it that case that C6 in itself is not the problem but once in that state the voltage can drop even lower - to a critically low value, in fact, that causes the freeze?

Link to comment
Just now, L0rdRaiden said:

I have my doubts about this if you overvolt the CPU at max frequency it will probably drain the max voltage allowed, so is a waste of performance not to try to overclock it as well. It's just a though not a statment.

In my case I have increased the freq 200Mhz

When you increase the supply voltage, the processor can feed a higher current through a fixed resistance. So it will go quicker to charge/discharge the internal capacitances making each transistor turn on/off faster. But with the exception of leak currents (which depending on manufacturing process etc really shouldn't be ignored), the processor doesn't really get hotter because each transistor draws a higher current but for a shorter time - it's a fixed capacitance to charge/discharge. In some situations, the higher current can result in issues - lots of issues with older aluminium-based processors where the electrons did now and then knock away an aluminum atom resulting in electromigration.


But when you then overclock, you will increase the number of transistors that changes state every second. So the overclocking will result in a higher power consumption. This both means more heat to drain away but also increases the problem with electromigration - i.e. the currents on the chip traces will result in a larger wear.

 

This is a reason why overclocked processors after a while can become more and more unstable.

 

Anyway - what seems to happen here is that in the lowest power states, AMD decides to drop one or more of the voltages too low. If it's the core voltage for the processor, or the bus voltage to the RAM or the RAM voltage itself. A tricky part is that all chips and RAM are individuals - datasheets normally shows quite large variations - so getting too close to the limit and some chips will continue to work while other chips will start to produce bit errors.

Link to comment
4 minutes ago, John_M said:

 

So the assertion that disabling C6 shortens the life of the processor in any meaningful way is false.

 

 

Isn't that just an alternative way of preventing the processor entering the C6 state, since by the slight overvolt you are not allowing it to reach the low voltage that C6 requires? Or is it that case that C6 in itself is not the problem but once in that state the voltage can drop even lower - to a critically low value, in fact, that causes the freeze?

As I told you alread read this, and try to understand what they say now.

https://bugzilla.kernel.org/show_bug.cgi?id=196683#c254

 

Quote

> One thing I noticed is that with C6 disabled through the script or using the
> "Typical Current" power supply option in BIOS in my case MB sensors never
> report Vcore less than 0.8V, which indicates that indeed it doesn't reach
> lower idle states even if only "package C6" is disabled. On default settings
> it will frequently report 0.4-0.5V range when idle.

I'm reaching exactly this 0.4-0.5V range with overclocking (+200 MHz) and C6 completely enabled - and I didn't see the problem any more since a lot of weeks now (before not even after an hour). My PSU  officially is definitely not C6/C7 capable because it's too old.

 

Edited by L0rdRaiden
Link to comment
1 minute ago, John_M said:

Isn't that just an alternative way of preventing the processor entering the C6 state, since by the slight overvolt you are not allowing it to reach the low voltage that C6 requires? Or is it that case that C6 in itself is not the problem but once in that state the voltage can drop even lower - to a critically low value, in fact, that causes the freeze?

There is a long list of rules for when the processor may use different power levels. In this case, I think the overvolting tricks the chip to reach the lowest levels even if the voltage is a bit too high - but it now reaches the lowest states with enough voltage to run stable. When allowed to enter the C6 state with AMD-approved voltages, it seems to get too little voltage for all installations to be stable.

 

All newer processors are making use of multi-voltage programmable voltage regulators, often called PMIC. And you reprogram them as the different parts of the system switches between different power levels. But you also needs to take into account both the component tolerances of the regulator and the component tolerances of the powered chips. If you get too close to the tolerances you can get the same chip instability as if you have an overclocked chip - but in this case caused by a chip being slightly undervolted.

  • Like 1
Link to comment
34 minutes ago, John_M said:

 

I have read it. I said so. But it doesn't address my question.

LOL it address your question, you didnt understand it, I will translate it for you.

C6 State is fine, AMD processors are fine, old PSU have problems providing low voltages, so by doing a small OV the processor can enter in C6 State (if enable), and consume a lower voltage than with C6 disable, but probably a sightly higher than with a compatible PSU, no OV and C6 enable.

 

So in my case my processor can enter in c6 and stay at 0,4 0,5 volts

in you case you  are stuck at 0,8v and can't go lower

and with c6 enable and a compatible  PSU you can get 0,4v or even less.

 

If overclock reduces the life of the processors due to the voltage, to run at a lower voltage the processor should life increase, so if your machine is usually iddle C6 can help. Although taking into account how much they last, is irrelevant, even an OC processor can last "forever" with a good disipator.

In the case where you have to disable all C states because your motherboard doesn't give you any other option, it will affect even more.

 

Edited by L0rdRaiden
Just to edit a probably false statement.
Link to comment
4 minutes ago, L0rdRaiden said:

old PSU have problems providing low voltages

 

Nobody had mentioned power supplies until now but since you have, the freezing issue also affects people with modern (so-called "Haswell-compatible") power supplies. But if the slight overvolt helps with that problem then it might be an even more useful way to combat the freezing.

Link to comment
2 minutes ago, L0rdRaiden said:

old PSU have problems providing low voltages

The PSU never supplies low voltages. The PSU always supplies the "established" voltages and the motherboard will mostly use the 12 V line to create the actual voltages used by the processor core, the memory bus, the RAM etc.

 

Early PSU had issues with really low currents, because a DC/DC needs a load to be able to regulate the voltage. A DC/DC is using a switching circuit that is like a bucket of charge that is "loaded" on the primary side and then "unloaded" on the secondary side. The DC/DC adjusts the output by either changing how many buckets/second it passes or the size of the buckets (changing frequency or pulse width). But there is a minimum load required on the secondary side because the DC/DC can't stop completely. And if each bucket transports more energy than what is consumed, then the output voltage will keep rising until the overvoltage protection steps in.

 

Newer PSU have built-in loads,  such as clamping diodes that will dynamically add load when the voltage goes up over the zener voltage, because that's a requirement for newer Intel chips. So this thread shouldn't really be about low loads on the PSU. But what is suggested from the different "overvolting" threads is that AMD have been too optimistic and programs the PMIC to output too low voltages in the lowest power states which in this case can give the same instability issues as if the processor was overclocked.

 

When blocking the lowest power states, the chip stays at 0.8 V which is enough to be stable. That has been the "traditional" workaround.

 

When allowing the lowest power states, the chip supply voltage drops to 0.4 V - but it will then be the number of decimals that decides if the system is stable or not. Note that a 5 % tolerance on 0.4 V is +/- 20 mV or 0.02 V - and the guys in the kernel thread haven't measured the supply voltage using a real multimeter and/or oscilloscope connected to the DC/DC. The supervision chip doesn't have the necessary precision or resolution to be able to display these small voltage differences at the lowest power levels.

Link to comment

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.