Well for I went over to the dark/red side yesterday and got an AMD Ryzen 5950X for AUD $800, then constructed this bad boy ( I had been an Intel bloke ).
Everyone knows "red" heads have more fun!!! ;)
I hope it works well for you. The auto for CPU boost and probably leaving the PBO disabled if you are doing any CPU crunching are probably good starting points.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Hi Mike, congrats on coming over to the "dark side". I see that you have a later kernel which has picked up a lot of new Zen features, like builtin sensors monitoring and improvements to the scheduler and I/O.
What brand and model of motherboard did you get?
Should be able to really blast through work.
I had been an AMD bloke for twenty years until somebody dropped a free Intel X99 workstation motherboard on me and I had to get up the Intel learning curve I had missed out on for twenty years.
Too bad it wasn't an ASUS board. Those are well supported in the 5.17 kernel.
Though if I remember correctly, you should have a WMI kernel driver for your Gigabyte board also.
You should investigate that and see if I remember correctly.
Or if you don't have sensor support, there is still the zenpower3 out of band kernel driver. That will give you voltages, currents, amperages and clocks for your 5950X.
zenpower-pci-00c3
Adapter: PCI adapter
SVI2_Core: 1.23 V
SVI2_SoC: 969.00 mV
Tdie: +74.4°C (high = +95.0°C)
Tctl: +74.4°C
Tccd1: +64.0°C
Tccd2: +76.0°C
SVI2_P_Core: 72.64 W
SVI2_P_SoC: 6.27 W
SVI2_C_Core: 59.29 A
SVI2_C_SoC: 6.18 A
Here zenmonitor3 reports about 145 W all up for the package to disperse/cool, with the cores turning over at around 4.3 GHz. A heater that calculates too ...
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
asusec-isa-0000
Adapter: ISA adapter
CPU Core: 1.09 V
CPU_Opt: 1732 RPM
Chipset: 2234 RPM
Water_Flow: 0 RPM
Chipset: +61.0°C
CPU: +49.0°C
Motherboard: +33.0°C
T_Sensor: -40.0°C
VRM: +44.0°C
Water_In: -40.0°C
Water_Out: +34.0°C
CPU: 101.00 A
george@GWG-PC3:~$
FYI, it is ~27॰C / ~80॰F ambient with the A/C on (@ 2:00 am it is 84॰F outside) and my 5950X is running at 4.425 GHz all core with a package power of ~142 to ~144. Keith thinks I won the silicone lottery on my CPU. :*) It is being cooled with a 360mm rad in push/pull configuration.
Mike, you said that was at full load so both dies should have been fully loaded.
You might investigate your TIM spread on the IHS. There shouldn't be that much spread between the two die temps.
With the normal "one good die - one mediocre die" that is typical with the two die cpus, normally the differential is only around 1-2 degrees at most.
Take a look at the TIM coverage especially on the bottom of the IHS where the chiplets are located. You should use the complete paste coverage method with the Ryzen cpus because the heat source is not centrally located anymore. The pea method of application is inadequate which was always used with Intel processors.
Here are some diagrams for location of the dies for reference.
The second image shows how my block input/output ports and water jet location is oriented along with the orientation of the cpu in the socket with motherboard component callouts. Makes for better visualization of where the dies are located in reference to where the TIM spread should absolutely be fully applied.
Because there is so much heat output from such a concentrated area, you get the best temps by using every square millimeter of the IHS for heat transfer.
Thank you Keith for your ideas and the diagrams. I thought something along those lines also.
Currently the TIM comes from the layer as provided by the manufacturer of the H150i CPU header ( in the shape of a square with rounded corners ). I'm rather reluctant to break the seal on the paste, clean and re-apply etc. So I've gently adjusted the tension on the four header mounting screws - with my fingers, so not too tight - and managed with a bit of experimentation to get Tccd1 and Tccd2 to within about four degrees C most of the time. Thus now sometimes Tccd1 > Tccd2 and sometimes vice versa, whereas beforehand Tccd2 was consistently well above Tccd1. I'm hoping therefore that it's mainly a matter of the alignment/contact of the face of the IHS with respect to the face of the underside of the header. But I think I may have to bite the bullet and re-dress the IHS face with new paste.
Interestingly there must be much noise in these measurements. The quiet ( E@H tasks paused ) vs busy ( 32 E@H tasks running ) difference in temperatures ( eg. Tdie ) is about 40 degrees C. So each E@H task has it's own thermal load to a given core, plus of course a necessarily different computational pattern per E@H task type. And the monitoring itself is at least one thread, there are other non-E@H threads too, scheduling algorithms come into play etc so this is quite fascinating ! ;-)
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Mike Hewson wrote: Well for
)
Everyone knows "red" heads have more fun!!! ;)
I hope it works well for you. The auto for CPU boost and probably leaving the PBO disabled if you are doing any CPU crunching are probably good starting points.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Well, I'll leave it for some
)
Well, I'll leave it for some time at stock standard to see how it performs.
To be a proper hoon : the cooler CPU header, the radiator fans and memory sticks all have an RGB glow. ;-)
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Hi Mike, congrats on coming
)
Hi Mike, congrats on coming over to the "dark side". I see that you have a later kernel which has picked up a lot of new Zen features, like builtin sensors monitoring and improvements to the scheduler and I/O.
What brand and model of motherboard did you get?
Should be able to really blast through work.
I had been an AMD bloke for twenty years until somebody dropped a free Intel X99 workstation motherboard on me and I had to get up the Intel learning curve I had missed out on for twenty years.
The motherboard is the
)
The motherboard is the Gigabyte B550 AORUS PRO AX. For AUD $250 a steal really.
A shout out for Scorptec : they insisted that it have the latest BIOS update before it left the shop. That way it works straight out of the box.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Too bad it wasn't an ASUS
)
Too bad it wasn't an ASUS board. Those are well supported in the 5.17 kernel.
Though if I remember correctly, you should have a WMI kernel driver for your Gigabyte board also.
You should investigate that and see if I remember correctly.
Or if you don't have sensor support, there is still the zenpower3 out of band kernel driver. That will give you voltages, currents, amperages and clocks for your 5950X.
Zenpower3 kernel module
Zenmonitor3 monitoring application
Thanks for those
)
Thanks for those links.
After installation, GTK3 et al, 'sensors' gives ( under full load, with ambient ~ 150 C ) :
iwlwifi_1-virtual-0
Adapter: Virtual device
temp1: +46.0°C
nvme-pci-0300
Adapter: PCI adapter
Composite: +37.9°C (low = -273.1°C, high = +81.8°C)
(crit = +84.8°C)
Sensor 1: +37.9°C (low = -273.1°C, high = +65261.8°C)
Sensor 2: +41.9°C (low = -273.1°C, high = +65261.8°C)
zenpower-pci-00c3
Adapter: PCI adapter
SVI2_Core: 1.23 V
SVI2_SoC: 969.00 mV
Tdie: +74.4°C (high = +95.0°C)
Tctl: +74.4°C
Tccd1: +64.0°C
Tccd2: +76.0°C
SVI2_P_Core: 72.64 W
SVI2_P_SoC: 6.27 W
SVI2_C_Core: 59.29 A
SVI2_C_SoC: 6.18 A
Here zenmonitor3 reports about 145 W all up for the package to disperse/cool, with the cores turning over at around 4.3 GHz. A heater that calculates too ...
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Cheers to you, too, Mike!
)
Cheers to you, too, Mike! I'm also glad that you have turned over a new leaf... so to speak.
Just for grins, here are my stats for my 5950X 'sensors':
george@GWG-PC3:~$ sensors
iwlwifi_1-virtual-0
Adapter: Virtual device
temp1: +51.0°C
zenpower-pci-00c3
Adapter: PCI adapter
SVI2_Core: 1.09 V
SVI2_SoC: 1.06 V
Tdie: +59.9°C (high = +95.0°C)
Tctl: +59.9°C
Tccd1: +59.2°C
Tccd2: +59.5°C
SVI2_P_Core: 74.96 W
SVI2_P_SoC: 14.93 W
SVI2_C_Core: 68.52 A
SVI2_C_SoC: 14.13 A
nvme-pci-0300
Adapter: PCI adapter
Composite: +45.9°C (low = -273.1°C, high = +84.8°C)
(crit = +84.8°C)
Sensor 1: +45.9°C (low = -273.1°C, high = +65261.8°C)
Sensor 2: +44.9°C (low = -273.1°C, high = +65261.8°C)
asusec-isa-0000
Adapter: ISA adapter
CPU Core: 1.09 V
CPU_Opt: 1732 RPM
Chipset: 2234 RPM
Water_Flow: 0 RPM
Chipset: +61.0°C
CPU: +49.0°C
Motherboard: +33.0°C
T_Sensor: -40.0°C
VRM: +44.0°C
Water_In: -40.0°C
Water_Out: +34.0°C
CPU: 101.00 A
george@GWG-PC3:~$
FYI, it is ~27॰C / ~80॰F ambient with the A/C on (@ 2:00 am it is 84॰F outside) and my 5950X is running at 4.425 GHz all core with a package power of ~142 to ~144. Keith thinks I won the silicone lottery on my CPU. :*) It is being cooled with a 360mm rad in push/pull configuration.
Just something to compare yours to.
Proud member of the Old Farts Association
Some nice numbers there,
)
Some nice numbers there, especially for the die temps. I think you have got a good slab of silicon there. ;-)
I think my core temps are up probably because of I/O : the PCIe bus is so loaded up. Of course it could be all the pretty blinking lights too :
.... I'll clean up all the cabling soon. That's a Corsair AIRFLOW 4000D mid tower box with a modular Corsair RM850X ( gold ) supply.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Mike Hewson
)
Mike, you said that was at full load so both dies should have been fully loaded.
You might investigate your TIM spread on the IHS. There shouldn't be that much spread between the two die temps.
With the normal "one good die - one mediocre die" that is typical with the two die cpus, normally the differential is only around 1-2 degrees at most.
Take a look at the TIM coverage especially on the bottom of the IHS where the chiplets are located. You should use the complete paste coverage method with the Ryzen cpus because the heat source is not centrally located anymore. The pea method of application is inadequate which was always used with Intel processors.
Here are some diagrams for location of the dies for reference.
The second image shows how my block input/output ports and water jet location is oriented along with the orientation of the cpu in the socket with motherboard component callouts. Makes for better visualization of where the dies are located in reference to where the TIM spread should absolutely be fully applied.
Because there is so much heat output from such a concentrated area, you get the best temps by using every square millimeter of the IHS for heat transfer.
Thank you Keith for your
)
Thank you Keith for your ideas and the diagrams. I thought something along those lines also.
Currently the TIM comes from the layer as provided by the manufacturer of the H150i CPU header ( in the shape of a square with rounded corners ). I'm rather reluctant to break the seal on the paste, clean and re-apply etc. So I've gently adjusted the tension on the four header mounting screws - with my fingers, so not too tight - and managed with a bit of experimentation to get Tccd1 and Tccd2 to within about four degrees C most of the time. Thus now sometimes Tccd1 > Tccd2 and sometimes vice versa, whereas beforehand Tccd2 was consistently well above Tccd1. I'm hoping therefore that it's mainly a matter of the alignment/contact of the face of the IHS with respect to the face of the underside of the header. But I think I may have to bite the bullet and re-dress the IHS face with new paste.
Interestingly there must be much noise in these measurements. The quiet ( E@H tasks paused ) vs busy ( 32 E@H tasks running ) difference in temperatures ( eg. Tdie ) is about 40 degrees C. So each E@H task has it's own thermal load to a given core, plus of course a necessarily different computational pattern per E@H task type. And the monitoring itself is at least one thread, there are other non-E@H threads too, scheduling algorithms come into play etc so this is quite fascinating ! ;-)
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal