orl
Regular Member
Posts: 96
|
Post by orl on Mar 14, 2021 21:43:07 GMT -5
Good afternoon gents! Was on the prowl for a reference AMD 6800XT when I stumbled across a NIB 3900X for $299. Needless to say thats a little tough to pass up, after a long exchange at the checkout about incorrect pricing, I walked away with that price and they went and fixed the tags on the shelf... 
Anyhow, to get it running required a BIOS flash on my Asus Tuf x470 and the results were interesting to say the least, the whole bios changed substantially and a lot of things were moved around. In the mean time while I am tinkering with this new BIOS, I wanted to reach out to the clockers here and see if you guys have any experience with them and or any detailed guides around this particular CPU. As I understand with early reading, clocking the 3900 and 3950 CPUs is a very unique beast that doesnt quite play by all the same rules... IE: You can end up with a paper weight quickly.
What you guys got?
|
|
|
|
Post by Vinster on Mar 14, 2021 22:30:15 GMT -5
|
|
orl
Regular Member
Posts: 96
|
Post by orl on Mar 14, 2021 22:46:32 GMT -5
Ty Vin, I missed that thread. This CPU is an impressive beast and I am really excited to play with it more, it seems to have a much more solid IMC than my 2600 did as well, so really looking forward to pushing this memory higher which is what I started with this time to get a feel for, as it was my major limiter with the 2600 and 4 sticks.
|
|
|
|
Post by freeagent on Mar 14, 2021 22:57:38 GMT -5
Nice man! Enjoy!
|
|
|
|
Post by Vinster on Mar 14, 2021 23:20:15 GMT -5
happy I could help
Vin
|
|
orl
Regular Member
Posts: 96
|
Post by orl on Mar 14, 2021 23:36:51 GMT -5
Settled with 3600Mhz on mem right now as reading informed me it will change from 1:1:1 as soon as you go over it and that introduces a whole new set of problems for performance when ratios begin changing, the simple fact that my memory went to 3600 with ZERO fuss already has me super happy with this chip, as my 2060 did not like much over 3200 without SERIOUS tweaking having 4 DIMM populated.
Additionally, while my custom loop was carrying the load well enough, I seen the CPU is running via optimized defaults at 1.47v, seemed oddly high and sure enough as I found its not supposed to be anywhere near that, but many motherboard manufacturers have that as the default, which is a head scratching wonder right there. Offset it down to about 1.3v in the interim before clocking it up. Now the CPU is hanging out happily at lower load temps so that will be nice for the OC push.
Really need to find me some decent tools for hardware monitoring in linux, CLI is great and all but I do miss having things all in one easy GUI in this case. May temporarily just install windows again for this process to use the tools and reinstall linux once I am "content". If I can snag 4.3 on this monster and still have fair cooling, this will be one happy individual.
|
|
|
|
Post by ShrimpBrime on Mar 15, 2021 9:24:38 GMT -5
That kind of VID has been the norm since 1st release.
Still trying to figure out the concern for XFR voltage. Its a single core light load. No where near the EDC or PPT max current threshold.
It seems people are more worried about v-core rather than the amp draw for some reason.
In reality, 1.47v doesn't mean shit honestly. If thats what it is default, then there shouldn't be any issues.
Its seen in bios while using a single core, the chip is just boosting its XFR specs. Should be a non issue.
|
|
orl
Regular Member
Posts: 96
|
Post by orl on Mar 15, 2021 10:20:36 GMT -5
Shrimp, unfortunately AMD and Linux are currently in despair of each other and my core voltages are disabled from reading while the OS is live, the voltage threw up red flags for me when seeing it though in the BIOS. However adding the offset of -0.15 has resulted in a much lower overall operating temperature (which is expected). Are you saying these voltages are being collected outside of normal operation then via XFR and are not actual operational outside of the boost?
Im pretty new to these 3000 series chips and especially ones with this many cores. What I did in this case was - offset 0.15 and then added a regular LLC to it leaving PBO online for now.
|
|
orl
Regular Member
Posts: 96
|
Post by orl on Mar 15, 2021 11:48:40 GMT -5
Interestingly, using the Phoronix tools and running Blender | Barbershop | CPU-Only provided some insight into whats going on behind these voltages and XFR. The tests were ran in sequence with no break or cooling time between them. All tests performed at default speeds and 3200Mhz on memory.
With Offset Voltage of -0.15:
CPU Temp Max - 65C
3 Run Averaged Score - 476s
With Default Voltage
CPU Temp Max - 71C
3 Run Averaged Score - 421s
I find this quite interesting and it really points a nice light on what AMD has done with XFR actually. While temps are higher about 9%, performance gains were about 12% using all defaults and not manually adjusting the voltages with PBO enabled.
|
|
|
|
Post by ShrimpBrime on Mar 15, 2021 12:29:58 GMT -5
When running XMP DOCP enabled, it runs the cpu in a performance mode which also includes increased CPU SOC which will create additional heat.
So you would test all defaults.
Then with XMP enabled
PBO tested enabled with both XMP enabled and disabled.
Most restriction will be EDC at full load no matter what settings you change in regards to all core loads vs XFR single core loads.
|
|
