ES or QS chip?Does anyone know that Tyan S8030 supports engineering samples? Can it be overclocked?
Yes, but the oldest BIOS version Agesa I can find on Tyan official website is 1.0.0.6, so I'm not sure whether this motherboard can be overclockedES or QS chip?
- If ES (ZS/2S), it all depends on the AGESA version more-or-less, see 1st post (hint: Agesa < 1.0.0.3)
- If QS (retail SKU ends with -04) should work with any board that supports 7002 series Epycs.
mainboards that came with rome support generaly don't have old AGESA, if there is no naples revision the old BIOS can get from you are out of luck.Yes, but the oldest BIOS version Agesa I can find on Tyan official website is 1.0.0.6, so I'm not sure whether this motherboard can be overclocked
Thank you for your replymainboards that came with rome support generaly don't have old AGESA, if there is no naples revision the old BIOS can get from you are out of luck.
only QS then, depends on the VRM power stages.
It's not only an overclocking issue, but if you try an ES chip in a board with BIOS with AGESA > 1.0.0.3 the CPU will not even boot/post.Yes, but the oldest BIOS version Agesa I can find on Tyan official website is 1.0.0.6, so I'm not sure whether this motherboard can be overclocked
Smallest QS is the 32-core which is around $800, but proceed with caution: we're pretty data deficient for running the 32-core QS at high clocks and voltages for extended periods of time and its unclear if they'll degrade.Hi all,
I started using EPYC for my workstation. I saw this thread and was wondering what some of my options may be. I have an H11SSL-I and a 7302p. Are there any 16 core QS part numbers I can purchase? Basically looking for an alternate (cheaper) upgrade path vs buying a 7F52. My 7302p stays pretty much locked at 3.3GHz but I'd like to get my single core speed as fast as I can.
Just simply couldn't afford Threadripper Pro and I got into this build sub $500 + RAM (which would have cost the same anyway)
How to set core voltage?Linux tools can be modified, however I'm working on a better detection, since Epyc and Desktop Threadripper 3000 series (Castle Peak) share the same basic cpuid, so another more complicated detection mechanism should be implemented - at least socket type detection is needed.
If you want, you can try the modified code, however for now it works on Epyc or on TR - not on both at the same time, since command IDs are different.
Attached is the modified version of my Zenstates-Linux fork: irusanov/ZenStates-Linux
It had been tested on a ES Epyc and it worked, but this version is unofficial until I sort out the cpu detection. Ofcourse, as always, everyone is free to contribute.
You can first try to set a different frequency and if it works, proceed following the guide and mess with rest of the things. For the test, please use the CLI, then you can use the GUI as well.
Frequency works in 25MHz steps.Code:sudo ./zenstates.py --no-gui --oc-frequency=2000
Be extremely careful with voltage. Make sure you have something to monitor it with - preferably a multimeter to check the actual voltage. Zenmonitor seems to be the best currently available on Linux, but I'm not sure if it displays correct info on Epyc: ocerman/zenmonitor
MD5: 0c2fe08b1dd06bfb29f834206d95ba18
They are 1.6 stock low to 2.6 high I think. I would already be happy to get to the retail levels of performance. Maybe just a matter of the right settings?you scores seems low, especially low unmodded, might need some bios tweaks. what are the clocks stock during all core load, i assume the ES is running lower than Retail, since they easily hit 110k+ on the retail launch.
i cant comment on exact settings for milan, didnt know it even worked for them, but i can say i hit 100k with 2x7742s, and the whole system drew close to 1000w if i remember correctly.
the biggest bottleneck on the server boards are usually the VRM which are tiny on all the boards, they are not made to go over specs, and overclocking does exactly that, i had to put a waterblock on mine, but also fans directly at the vrm could help.
the max i could get stable without sudden low scorres was 1.18v 32.5 multi for an r23 of 100153.
stock with some biostweaks i got 85k.
XGI to 18x nps at one or 4. and possibly some other tweaks i dont remember. there are some decent amd documentation on bios optimizations floating around
Try to hit stock all core Freq first, think it's around 2.8 2.9 during cinebench.They are 1.6 stock low to 2.6 high I think.
The tool I am using does not have multipliers listed but target all-core freq.
It offers
All-Core Freq - with "Lock Freq"
CPU Volt
Dual Socket
Limits for PPT (W), TDC (A), EDC (A).
Maybe not the right tool? I tried other tools, either they seem identical in settings, some say it's an unsupported processor.
And with some it's no clear to me what to do as it has too many options (eg SMU Debug Tool)
Have to check my BIOS settings, have not seen this yet.Try to hit stock all core Freq first, think it's around 2.8 2.9 during cinebench.
Also set the bios to power mode, not performance ( yes power is faster) and power target to max or put in 300w.
do you have all channels populated?
It's also possible the Milan es are just weaker than retail, while the 7742 -03 where as fast and faster overclocked.
That's very interesting I'll have to try it out, thanks!Also set the bios to power mode, not performance ( yes power is faster)