x299 OCing

8 Pack · 8 Sep 2017 at 11:30

james_2k said:
ok, just reduced input voltage from the auto (1.95) to 1.8v and, wait for it, load temps dropped by 20c. vrm temps went from 77 to 67 also.. that's kind of mental. would be really nice if anyone else out there could verify this with their own setup??

ok, so after testing, its stable BUT, the performance seems to have halfed.. cinebench score at least. how bizzare ey?

No because FIVR is inefficient when using too little on the input so the cpu loses efficiency.

james_2k · 8 Sep 2017 at 12:31

makes sense, was weird though!

sill, love my 4.8/5.0 cpu now hehe

Silent_Scone · 8 Sep 2017 at 12:36

Don't lower it too much, the CPU will throttle. LLC has been changed on this platform to accommodate for Intel's throttling mechanism. At 1.8v the CPU will be throttling. That's why Auto rules seem higher than on previous gen.

Disco_P said:
Threadripper

The sales haven't been all that great, actually

8 Pack · 8 Sep 2017 at 15:44

Efficiency was off on all FIVR based CPU with VCCIN too low.

Silent_Scone · 8 Sep 2017 at 18:36

8 Pack said:
Efficiency was off on all FIVR based CPU with VCCIN too low.

Yeah, but now Intel throttle both frequency and vcore to keep TPD within spec, as people are starting to notice (or not notice as is the case)

james_2k · 8 Sep 2017 at 18:39

Silent_Scone said:
Yeah, but now Intel throttle both frequency and vcore to keep TPD within spec, as people are starting to notice (or not notice as is the case)

haven't seen any frequency nor voltage throttling? this is @ all cores at 4.8 bar the two at 5.0

is yours doing this?

don't get why you are replying to 8 pack who was talking about my experiment with reducing Vccin (input voltage) and the subsequent performance loss.

8 Pack · 8 Sep 2017 at 18:41

My primary concern is efficiency.

Silent_Scone · 8 Sep 2017 at 20:03

james_2k said:
haven't seen any frequency nor voltage throttling? this is @ all cores at 4.8 bar the two at 5.0

is yours doing this?

don't get why you are replying to 8 pack who was talking about my experiment with reducing Vccin (input voltage) and the subsequent performance loss.

Was responding to you, and it's difficult to detect. None of the monitoring tools poll the CPU in a way that will. If you measure the VCCIN voltage you'll notice it actually moves around or more importantly raises above the set value. This is the auto rules counteracting Intel's throttling mechanism if VCCIN drops below the trip point. That's why you were seeing such lower temps.

james_2k · 8 Sep 2017 at 20:13

oh, i see, so it forces the issue by lowering vccin? does that mean ultra high perf if we stick 5v into the cpu?

Silent_Scone · 8 Sep 2017 at 20:58

james_2k said:
oh, i see, so it forces the issue by lowering vccin?

In a way, yes. Hoping one of the monitoring apps are updated to poll the CPU in a way that will show you the reduction in frequency and voltage. Currently, they don't

james_2k · 8 Sep 2017 at 23:13

ive been mainly doing some back to back cinebench runs as that generates quite a lot of heat (and is a good stability test). no perf reductions in 20+ consecutive runs