Threadripper OWNERS, stable OC, and vcore to do it.

[Markfw]

I have people in another thread doubting the average OC of threadripper, all core OC. I say its about 4.0 ghz at about 1.2-1.3 vcore.

 

[Markfw]

Mine is at 1.2 vcore 4 ghz

 

[Markfw]

I have reset the poll as anonymous users are not allowed to vote here.

 

[IEC]

I am running 24/7 at 4.0GHz 1.26V. It passed extended AIDA64 stress testing at 1.25V but I wanted a little extra just in case.

I can do 4.1GHz but it requires ~1.4V and I'm not willing to pump more than 1.3V into my chip for daily use.

 

[ddogg]

Just for the sake of this thread, I decided to see how low I could drop my voltage at 4Ghz. So far, I've been running ~1hr Aida64 at 1.22V so I'm fairly certain it should be stable at that. I tried 1.2V and it crashed within 5 minutes.
4.1 requires 1.381V.

 

[Markfw]

BTW, the main point of this thread is to prove some of the reviewers wrong, that it requires 1.4, 1.42 ot even 1.44 vcore to get a 1950X stable at 4.0 ghz. I say its about 1.25

 

[The Stilt]

The criteria to determine the "max OC" should be standardized. Otherwise this poll is utterly useless, when some users are posting their maximum CB15 frequencies (or worse), while the others something entirely different.
IMO >= 30min of Prime95 v28.10 with following settings should be set as the criteria:

Torture test > Custom > Min FFT size = 128, Max FFT size = 128, Run FFTs in-place ticked.

www.mersenne.org/ftp_root/gimps/p95v2810.win64.zip

The delta between the maximum benchable and the actual maximum stable frequency is rather large on Ryzen.

There are newer versions of Prime95 available, but based on my experience they all fail to fully stress Ryzen for some reason.
Linpack is totally useless as well unless you use >= 65k problem sizes (32GB of DRAM), because it essentially idles around 20% of the total time required to solve the problem.

 

[tamz_msc]

Member @DrMrLordX seems to be of the opinion that ycruncher also does a good job. So either that or P95 V28.10.

 

[DrMrLordX]

Yeah, try y-cruncher! Just make sure its one of the newer builds that supports "ADX".

 

[mixmasstrw]

@Markfw, @IEC , @ddogg Hi, what LLc Did you use to get stable 4G (1.20 - 1.25 V). What was your Vdroop. Could you share settings?

 

[mixmasstrw]

Done few attempts. For now I am running 4Ghz / 1.275V (Vcore) / LLC 3 (ASROCK Taichi) - voltage : max 1.275 min 1.262. Seams rock solid (prime95). Did not hit the wall because got no time to try go lower with the voltage. Might be some room for lowering the voltage.

 

[ddogg]

@Markfw, @IEC , @ddogg Hi, what LLc Did you use to get stable 4G (1.20 - 1.25 V). What was your Vdroop. Could you share settings?

I will check when I get home, it's been awhile since I tweaked with my settings. I know my core voltage is at 1.375 for 4.1ghz

 

[eek2121]

@Markfw, @IEC , @ddogg Hi, what LLc Did you use to get stable 4G (1.20 - 1.25 V). What was your Vdroop. Could you share settings?

Mode 2 on my MSI board works for me. Still messing with voltages, but 1.25 is stable.

 

[IEC]

Mode 2 on my MSI board works for me. Still messing with voltages, but 1.25 is stable.

I think mine was set the same way, just slightly higher voltage.

I'm currently running it stock on air cooling though.

 

[Markfw]

@Markfw, @IEC , @ddogg Hi, what LLc Did you use to get stable 4G (1.20 - 1.25 V). What was your Vdroop. Could you share settings?

I settled on 3.8 ghz@1.25 vcore due to power usage. I run 24/7/365@100% load.

 

[Bubben3k]

Stable 4,2 on all Cores. 1,5v
My AX 860i psu couldn't do it somehow and I ended up doing with my AX1200i, should be said I use 2 quad rads and have monoblock ex.(zenith) so I assume it would be unreasonable hot without proper cooling, but I wouldn't recommend 1,5v unless you have actual cooling on power phases and such, have a feeling you would drastically reduce the lifespan of the board if you didn't

 

[Kilobit81]

Mine is at 1.2 vcore 4 ghz

I've been trying to get 4.0 on my 1950X with a lower voltage than 1.40 VCore but I can't seem to do it. What is your S.O.C. Vcore voltage at? What do you have your LLCs set to?

 

[Markfw]

I've been trying to get 4.0 on my 1950X with a lower voltage than 1.40 VCore but I can't seem to do it. What is your S.O.C. Vcore voltage at? What do you have your LLCs set to?

See my latest. I settled on 3.8 ghz@1.25 vcore

 

[ub4ty]

1950x
1.2V @3.7ghz all core (max Wattage when all cores locked at 3.7) = 250 Watts
Cooled by Noctua. No aqua for my builds. Max temps on CPU around 65-68C will fully loaded along with GPUs.

I like to only put a mild power sane OC on my builds.
I have two other 1700s running at a similar config. I got golden samples after RMA'ing my prior 1700s due to the infamous segfault bug @ load on Linux.

3200 @CL 14 or 3466@CL 16 for 32GB of 4x8 G.skill 3600@CL16 sammy dies.
Build is locked in super tight. I have every pcie slot occupied + pcie 2.0 + all nvmes loaded down with gum sticks + sata + spinning rust. Huge air flow case and also a fan on the backside of the mobo cutout to keep the bum nice and cool.

She's a screamer lads.
Might be consolidating my two 1700 builds into another Ripper come Black Friday if there are any 1950x left. Need a deep discount $500 for 1950x and $200 for a mobo to make me budge. What is she tasked on? Something unworldly.

Kept my word on 'breaking the chains' on memecoin.
Hope everyone enjoyed the 'fall'.
A new fall has arrived and then winter.

Momma Su has ushered in a grand age.

Rumors are X499 in 2019.. Another inflection drop

 

[lightmanek]

After few years of average to bad silicon lottery I finally won!

My new TR1920X is a great chip!

So far I've managed to extensivly test it at:
- 3.9GHz and 1.16 Vcore (could try lower still)
- 3.6GHz and 1.1 Vcore (not extensively tested, but ran good amount of stress testing benchmarks)
- 4.0GHz and 1.25 Vcore (crashed at 1.2V during CB15, was benchmark stable at 1.225V)
- 4.1GHz and 1.3 Vcore (stress tested for the last 3 hours)

I also quickly tried 4.2GHz with 1.35 Vcore but that was fine in single threaded stuff, but crashed in heavy MT. Probably needs 1.39 to 1.42 Vcore to make it stable. I will try it more next time I have time, but I'm already super happy with anything 4GHz+ on 24 threads

My quad-channel memory is set to 3200Mhz TheStilt's fast timings 14-13-13-28 1T with optimized subtimings. This memory optimization from Auto 3200 setting alone gained me around 10%-15% in gaming type of workloads (example Cinebench OpenGL went from 105FPS to 119FPS).

According to HWMonitor my CPU SOC power draw during Frybench or Cinebench MT at 4.1GHz is around 200W and CPU temp stays under 60C (no offset) thanks to Enermax TR360 AiO.

What a monster of processor for £350.00

EDIT:

Picture from quick Linpack Stress Test. Setting 1.16V BIOS with drop to 1.13 under load. LLC Level 5 (lowest) on Taichi X399.

 

[Kilobit81]

Mine is at 1.2 vcore 4 ghz

S.O.C. Vcore or CPU Vcore?
I have been able to get 4.0 GHz with 1.32V CPU voltage and SOC Vcore set at default to 1.10V. I had posted earlier that I couldn't get to 4.0 GHz with less than 1.4 but I meant less than 1.3 on the CPU voltage. I keep my 1950X at 3.8 GHz with 1.28 CPU voltage because I feel 1.3V or higher is too much.

 

[Brek01]

Interesting, I Tried OC my threadripper when I first got it and couldn’t really get over 3.8 ghz to be stable, I was being conservative with voltages. However I revisited the Topic this week did some more research and have set core voltage to 1.3625 and ran some sims in Houdini which took around 7 hours, and no crashes running at 4.1, I am on air with noctua 140 blower, I added an additional 140 pull fan which I did Have previously. Peaking at 70 deg cel for a bout a second every now and again. So now backed off to 4ghz and very happy with results. Not used a synthetic bench mark as prefer to test with what I use system for as if it don’t crash it’s good enough for me.