Info Dual EPYC 7601 build. Comments and questions ?

[Markfw]

So, here is is. Dual EPYC 7601 32 core CPU's and 256 gig ECC registered DDR4 2666 ram, with a 1 tb NVME and a 2 tb spinner drive. 128 threads of greatness ! About $3400 for 2 CPU's, memory and the motherboard. About the same prices (and possibly performance) of the new 3970x. But I do have 8 memory channels. Then again, it only runs at 2400 mhz. But it only runs 50c 100% loaded. I can see why they run so slow after seeing the manual and the 1u heatsinks they were intended to run using.

 

[Trotador22]

And how would I do that ????

Enter BIOS, go ADVANCED then NB Configuration, then cTDP Control and enter 200.

 

[Markfw]

Pretty incredible Mark. I did notice you still need to upgrade your sig. ;-)

Its in my sig. end of line 2

 

[Markfw]

Enter BIOS, go ADVANCED then NB Configuration, then cTDP Control and enter 200.

OK, with my cooling, I entered 220. I saw 3000 once, but most of the time its only holding 2600. Still way better than 2400.
Here is the current, 2700

 

[Trotador22]

Nice!, you can try with cTDP=225 but 2700Mhz is the best official all cores boost you will get.

If you want to overclock beyond specs you have to use ZenStates, go here https://forums.servethehome.com/index.php?threads/overclock-your-amd-epyc.23422/, you can go to post 115 to start seeing what you need for your setup.

My H11DSI dual EPYC 7551 is currently at 3 GHz all cores in Rosetta and also crunching Milkyway in a Radeon VII. At these clock/voltages the CPU vrm temperature becomes the critical/limiting factor. You'd better installing Supermicro Superdoctor utility to monitor all your board/cpu sensors. The problem of your setup (and mine) is that the coolers do not let room enough to put a fan powerful enough onto the Cpuvrms heatsink (between both cpu sockets). The thread linked has lots of information you will enjoy.

Your CPU utilization graph looks strange to me, in my systems it used to be a flat line at the 100% utilisation level. Rosetta is currently not stressing a lot CPUs, with WCG MCM I could not go much beyond 2.6 GHz before going above 90ºC in the CPUvrms. CPUs themselves are OK with fans at 1500rpm. So you have to check depending of the project.

 

[Hitman928]

OK, with my cooling, I entered 220. I saw 3000 once, but most of the time its only holding 2600. Still way better than 2400.
Here is the current, 2700

Can you do

cat /proc/cpuinfo | grep MHz

In the terminal and post a screenshot of the printout?

 

[Markfw]

Can you do

cat /proc/cpuinfo | grep MHz

In the terminal and post a screenshot of the printout?

Nothing. I also tried lower case mhz
But like I said lscpu say 2600 to 2700 consistently.

Edit: I removed the " | grep MHZ" and its just the output from lscpu. Its MHz.

 

[Hitman928]

Nothing. I also tried lower case mhz
But like I said lscpu say 2600 to 2700 consistently.

Edit: I removed the " | grep MHZ" and its just the output from lscpu. Its MHz.

Yeah, it's case sensitive. Doing it this way will just show you what each individual core is at, was just curious if every core was boosting the same amount.

 

[StefanR5R]

This shows the current clocks of all CPUs:
cat /sys/devices/system/cpu/cpufreq/policy*/scaling_cur_freq

This condenses above output to the fastest and slowest:
cat /sys/devices/system/cpu/cpufreq/policy*/scaling_cur_freq | sort | sed -e 1b -e '$!d'

(This works only if a cpufreq driver is loaded.)

Edit, using /proc instead:
grep MHz /proc/cpuinfo | cut -d: -f2 | sort | sed -e 1b -e '$!d'

 

[Markfw]

This shows the current clocks of all CPUs:
cat /sys/devices/system/cpu/cpufreq/policy*/scaling_cur_freq

This condenses above output to the fastest and slowest:
cat /sys/devices/system/cpu/cpufreq/policy*/scaling_cur_freq | sort | sed -e 1b -e '$!d'

(This works only if a cpufreq driver is loaded.)

Thanks ! The slowest was like 2583 and the fastest was like 2623, very close.