Overclock.net banner
Cancel
Create thread New Forums

AMD Ryzen Threadripper Owners Club - 1950X | 1920X | 1900X

18 reading
Jump to Latest
3,241 - 3,260 of 4,337 Posts
#3,241 ·
OrionBG said:
It is quite possible that you baked the CPU and the transistor connections inside the silicon have degraded slightly... 1.57v is way too much voltage vor that chip.
You can test with a little bump of the SOC voltage but if it doesn't help then you are most probably out of luck...
Click to expand...
He would have had to have had it at that voltage for quite a while. :) My chip degraded, but that was days of me tinkering just to see what Threadripper can do. It still does 4.0 @ 1.275 though, it used to be around 1.2V for 4.0. I've since stopped overclocking though. The machine is silent and only uses around 100 watts of power browsing the web, listening to music, etc. (since P-States are finally working on this board, it goes from around 1850MHz up to 4.2 GHz as needed). Still wish it could go a bit lower on the minimum speeds. You don't really need much speed if you are watching a hardware accelerated video, browsing the web, or idling at the desktop.
 
#3,242 ·
betam4x said:
OrionBG said:
It is quite possible that you baked the CPU and the transistor connections inside the silicon have degraded slightly... 1.57v is way too much voltage vor that chip.
You can test with a little bump of the SOC voltage but if it doesn't help then you are most probably out of luck...
Click to expand...
He would have had to have had it at that voltage for quite a while. /forum/images/smilies/smile.gif My chip degraded, but that was days of me tinkering just to see what Threadripper can do. It still does 4.0 @ 1.275 though, it used to be around 1.2V for 4.0. I've since stopped overclocking though. The machine is silent and only uses around 100 watts of power browsing the web, listening to music, etc. (since P-States are finally working on this board, it goes from around 1850MHz up to 4.2 GHz as needed). Still wish it could go a bit lower on the minimum speeds. You don't really need much speed if you are watching a hardware accelerated video, browsing the web, or idling at the desktop.
Click to expand...
Do you know why I can't use 1.290 anymore at 4.0 and need 1.380 ?
 
Save
Reply Quote Rep+ Rep+
#3,243 ·
x7007 said:
Do you know why I can't use 1.290 anymore at 4.0 and need 1.380 ?
Click to expand...
assuming your temps are under control, I'd say degradation is the culprit, but without more information it'd be hard to pin down... made any changes lately.. bios update and such?

on another note, I don't own threadripper, but I follow this thread and all this degradation talk is a bit scary... wondering if I need to back off my 1700 now...
 
Save
Reply Quote Rep+ Rep+
#3,244 ·
Minotaurtoo said:
assuming your temps are under control, I'd say degradation is the culprit, but without more information it'd be hard to pin down... made any changes lately.. bios update and such?

on another note, I don't own threadripper, but I follow this thread and all this degradation talk is a bit scary... wondering if I need to back off my 1700 now...
Click to expand...
More than likely, he just needs to tweak his LLC settings. Keeping LLC on auto isn't good for overclocking. It causes you to need to pump more voltage in to keep things stable. For example, if I left LLC on auto, my chip would need 1.375V or somewhere around there to do 4.0. However, by correctly setting the LLC, I get away with a much lower voltage.

Also, anything below 1.425V is definitely fine though I'd try to stay at 1.35 or below, within AMD's spec. 1.425-1.45 is pushing it, and I wouldn't run the chip beyond that. MSI boards can't even handle that much for the 1950X anyway. My chips throttle above 1.425 in prime95 unless I turn off PROCHOT.
 
#3,245 ·
betam4x said:
More than likely, he just needs to tweak his LLC settings. Keeping LLC on auto isn't good for overclocking. It causes you to need to pump more voltage in to keep things stable. For example, if I left LLC on auto, my chip would need 1.375V or somewhere around there to do 4.0. However, by correctly setting the LLC, I get away with a much lower voltage.

Also, anything below 1.425V is definitely fine though I'd try to stay at 1.35 or below, within AMD's spec. 1.425-1.45 is pushing it, and I wouldn't run the chip beyond that. MSI boards can't even handle that much for the 1950X anyway. My chips throttle above 1.425 in prime95 unless I turn off PROCHOT.
Click to expand...
^

All off this.

If you don't do LLC control, you're going to have to bump up voltages, and that's not a smart move. Auto is only useful if you are running stock settings and don't want to bother.
 
#3,246 ·
Soooo, after using TR for close to 9 months, what is the consensus on "max safe vcore"? I am sporting 1950x @ 3975mhz @ 1.37v and LLC on auto - goes up to 1.417 with offset under load. This seems to be both realbench stable and prime95 small FFT 8h test stable.

I want to get to 4.0ghz really bad, because 4.0 is the magic number and 25mhz will infinitely boost by productivity (I am sure of it).

Hence my question about max safe stable.

Thanks.
 
#3,247 ·
Kriant said:
Soooo, after using TR for close to 9 months, what is the consensus on "max safe vcore"? I am sporting 1950x @ 3975mhz @ 1.37v and LLC on auto - goes up to 1.417 with offset under load. This seems to be both realbench stable and prime95 small FFT 8h test stable.

I want to get to 4.0ghz really bad, because 4.0 is the magic number and 25mhz will infinitely boost by productivity (I am sure of it).

Hence my question about max safe stable.

Thanks.
Click to expand...
According to AMD it's like under 1.35v and under 68c.
 
#3,249 ·
Kriant said:
Well, I got the under 68c part even when under prime95 full load. I saw 1.35v before - is that "max recommended" or "max safe"? I know Intel has that duality going in their tech sheets for example. But I couldn't find a definitive answer with my google-fu skills.
Click to expand...
Every chip is different, there is no technically safe when overclocking. If your running a sane daily voltage, bad degradation won't come before you sell off the platform. You might need some more volt to hit the same clocks. I use to need 1.32 to hit 4ghz now I need 1.35. Just the way it is. I think anything under 1.4 is fine for daily.
 
#3,253 ·
#3,254 ·
Kriant said:
Soooo, after using TR for close to 9 months, what is the consensus on "max safe vcore"? I am sporting 1950x @ 3975mhz @ 1.37v and LLC on auto - goes up to 1.417 with offset under load. This seems to be both realbench stable and prime95 small FFT 8h test stable.

I want to get to 4.0ghz really bad, because 4.0 is the magic number and 25mhz will infinitely boost by productivity (I am sure of it).

Hence my question about max safe stable.

Thanks.
Click to expand...
You HAVE to tweak the LLC. Many chips can do 4.0 at 1.2-1.275V @ 4.0 with the correct LLC. My chip degraded quite a bit from pushing voltages and that's not something you want to have happen. I went from 1.2V to 1.275 for 4.0 just because I kept playing around in the 1.475-1.675 range and higher clocks.

On a side note:

I got my chip to 4.3-4.4 GHz and ran benches on it, even got it to 4.5 and ran CPU-Z, it beat the 7980xe in both single and multicore easily (516 single core or something like that, multicore I can't remember)...running anything else made it crash. I don't really care much because I'm ebaying this platform (along with the board and the RAM) the second the 2950x comes out and it still does 4.0 @ 1.275. Right now I'm running @ stock with AMD C&Q enabled and min processor state set to 0%. I get speeds anywhere from 1.95GHz to 4.2 GHz depending on what I'm doing. I only OC when gaming for a while. Luckily I can still do 4.1 @ 1.35.

Anyone else going to get the 2950x when it comes out? I am doing a completely new build. Not happy with the MSI motherboard and I want Samsung memory instead of Hynix, so I figure I'll do a full upgrade, even if clocks only receive a marginal boost. I hope they have VRMs with full on heatsinks like some of the x470 boards. I also hope the chips overclock better.
 
#3,255 ·
Kriant said:
Soooo, after using TR for close to 9 months, what is the consensus on "max safe vcore"? I am sporting 1950x @ 3975mhz @ 1.37v and LLC on auto - goes up to 1.417 with offset under load. This seems to be both realbench stable and prime95 small FFT 8h test stable.

I want to get to 4.0ghz really bad, because 4.0 is the magic number and 25mhz will infinitely boost by productivity (I am sure of it).

Hence my question about max safe stable.

Thanks.
Click to expand...
I'm not as radical as @betam4x though I do push a bit farther than most when overclocking. I also don't keep my systems for long -- the longest I've had a system is 26 months.

I agree with @nycgtr that 1.4v is a "safe" daily driver voltage for most Threadripper overclockers using ambient cooling (assuming you can control the CPU and VRM heat) though I would consider 1.425v as the maximum for my own system.

I test stability with three main programs:
RealBench v2.56 with the full amount of RAM selected for eight hours.
Prime95 v29.4b8 Small FFT preset for eight hours.
Y-Cruncher default stress test for eight hours.

If it doesn't pass all, it's not stable. I noticed with Heatripper Threadkiller that streaming TV shows while stress testing with Y-Cruncher will stutter/pause if the voltage isn't high enough, so I do that now as well when I change speed or voltage.

For your system, I would suggest trying for a lower voltage with CPU LLC at Level 3 or even Level 2. Are you using P-State overclocking or just doing it through the regular overclocking BIOS page?
 
Save
Reply Quote Rep+ Rep+
#3,256 ·
betam4x said:
Anyone else going to get the 2950x when it comes out? I am doing a completely new build. Not happy with the MSI motherboard and I want Samsung memory instead of Hynix, so I figure I'll do a full upgrade, even if clocks only receive a marginal boost. I hope they have VRMs with full on heatsinks like some of the x470 boards. I also hope the chips overclock better.
Click to expand...
I'm considering it. If I do, I will likely keep the rest of my system the same as I already have Sammy-B memory and a board I'm happy with. I did have two VRM MOSFETs with no cooling for the first week due to bad QA (the factory thermal pad was too short), so I may get a new board too just as a precaution.
 
Save
Reply Quote Rep+ Rep+
#3,257 ·
MNMadman said:
I'm not as radical as @betam4x though I do push a bit farther than most when overclocking. I also don't keep my systems for long -- the longest I've had a system is 26 months.

I agree with @nycgtr that 1.4v is a "safe" daily driver voltage for most Threadripper overclockers using ambient cooling (assuming you can control the CPU and VRM heat) though I would consider 1.425v as the maximum for my own system.

I test stability with three main programs:
RealBench v2.56 with the full amount of RAM selected for eight hours.
Prime95 v29.4b8 Small FFT preset for eight hours.
Y-Cruncher default stress test for eight hours.

If it doesn't pass all, it's not stable. I noticed with Heatripper Threadkiller that streaming TV shows while stress testing with Y-Cruncher will stutter/pause if the voltage isn't high enough, so I do that now as well when I change speed or voltage.

For your system, I would suggest trying for a lower voltage with CPU LLC at Level 3 or even Level 2. Are you using P-State overclocking or just doing it through the regular overclocking BIOS page?
Click to expand...
My chip would require 1.375-1.4 if I didn't change LLC to 3 (no vdroop). With LLC at 3 (1 is highest on my board, ugh, 3 keeps vcore the same no matter what), I can drop it down to 1.2xx. I would say that's a pretty drastic improvement.
 
#3,258 ·
betam4x said:
My chip would require 1.375-1.4 if I didn't change LLC to 3 (no vdroop). With LLC at 3 (1 is highest on my board, ugh, 3 keeps vcore the same no matter what), I can drop it down to 1.2xx. I would say that's a pretty drastic improvement.
Click to expand...
Mine still has vdroop at Level 3, just not as much as on Auto. It's nowhere near 0.1v of vdroop though, even on Auto. What board are you using?
 
Save
Reply Quote Rep+ Rep+
#3,259 ·
MNMadman said:
Mine still has vdroop at Level 3, just not as much as on Auto. It's nowhere near 0.1v of vdroop though, even on Auto. What board are you using?
Click to expand...
MSI, the LLC settings are inversed on mine, with 1 being the highest and 10 being the lowest. Level 3 results in no vdroop at all, which I don't see any drawbacks to. Level 1 actually adds voltage under load, which is rather odd. Let's fry some chips! I guess some overclockers may have a need for it. It could be useful for lower idle voltages as well. However, I find this board handles LLC very poorly on 'auto'. Can't wait for x499 to probably go for asrock or something else.
 
#3,260 ·
nycgtr said:
Every chip is different, there is no technically safe when overclocking. If your running a sane daily voltage, bad degradation won't come before you sell off the platform. You might need some more volt to hit the same clocks. I use to need 1.32 to hit 4ghz now I need 1.35. Just the way it is. I think anything under 1.4 is fine for daily.
Click to expand...

Then I got to wonder - did I epically lost in the silicon lottery; or 4.0 ghz and 4.1 ghz users with low vcore test their system only with cinebench for stability. P95 requires 1.439v under load on my chip to run stable (stable in my book = 3 random time tests for 4h each, followed by 8h of realbench). Cinebench, gaming, realbench - all require somewhat less, but I don't and can't consider that stable.
 
3,241 - 3,260 of 4,337 Posts
About this Discussion
4.3K Replies
286 Participants
Last post:
Offler
Overclock.net
posts
28M
members
562K
Since
2004
The premier forum for overclocking experts and enthusiasts. Discuss hardware optimization, custom builds, benchmarking, cooling solutions, and pushing the boundaries of computing performance. From beginner guides to extreme overclocking, join our technical community to master system tuning.
Full Forum Listing
Explore Our Forums
Intel CPUs Hardware News NVIDIA PC Gaming Water Cooling