Overclock.net banner

1 - 12 of 12 Posts

·
Registered
Joined
·
18 Posts
Discussion Starter #1
So I bought myself the following components for a custom build:

- Ryzen 7 1700
- Asrock Taichi X370 (BIOS v3.20)
- Noctua NH-D15 SE-AM4
- G.SKILL Flare X Series 16GB F4-3200C14D-16GFX (Samsung B-die)
- GTX 1050 Ti

I have never overclocked a system before getting this machine. I watched a lot of videos on overclocking ryzen and felt like I had a reasonable grasp of the concept.

I started out with loading the flareX XMP-3200 profile and ran memtest86 from a USB stick. I ran it continuously for 16hrs and it reported 0 errors. So, naturally, I thought "great, time to overclock the CPU".

I had seen several videos stating that 3.8Ghz on the 1700 is a very reasonably goal with the right components, so I figured I might as well just go for the gold (why not?) and immediately started out at 3.8GHz with a vcore of 1.3625V which is one configuration that I saw on one overclocking video. I also set LLC to level 2 for both Vcore and SoC. This worked fine to my surprise. My testing included cinebench, OCCT Small dataset, OCCT linpack, Aida64, and Prime95 blend. Aside from cinebench I ran all of the others for 8 hrs _each_ without issue. Suffice it to say by this point I was on cloud 9. I couldn't believe how easy my first overclocking was!

I was trying to find other tests to run (because I wanted to be 1000% certain the system was stable before moving all of my data from my old machine over to it) so I ran the other two Prime95 tests for about 2hrs each. Again, no problems. Then I noticed the medium/large dataset options in the OCCT tool. So I tried running the large dataset option and this is where everything fell apart!! The large dataset ran for 2m7s before crashing and the entire PC rebooted. So I tried running again, because I thought it might have been some kind of fluke, but sure enough it crashed again right around the 2m mark. Note that the crashing was unrelated to CPU temp - my CPU core temp has never broken 60C @ 1.3625V vcore which is far from the thermal junction limit of 95C according to AMD.

I also thought that OCCT crashing was incredibly odd considering I had just run Prime95 blend for 8hrs straight with 0 issues. From what I read prime95 blend is supposed to be THE most extreme crazy (and unrealistic workload) stability test that you can run! But it would seem that the OCCT large dataset test contradicts the rest of the overclocking community! XD

In any case, I tried fiddling with a bunch of stuff including increasing vcore, messing around with various LLC settings, increasing the SoC voltage, and disabling the DDR XMP profile. I spent about 3 days trying every possible combination I could come up with but nothing would let my system pass the OCCT large dataset at 3.8GHz. Needless to say I was dismayed!

So I finally relented and tried decreasing the clock rate to 3.7GHz. I thought for sure this would work. But nope! OCCT Large dataset crashed again! So I kept reducing the frequency all the way down to 3.1GHz and it still crashed!! 0.o I was seriously bummed out at this point. I thought I had a dud CPU/mobo/whatever in my hands.

So I then disabled ALL overclocking completely (i.e. pure stock BIOS configuration) and ran the OCCT large dataset. It ran continuously for 8hrs without issue. I then ran all of the other stress tests listed earlier for 8 hrs each. All of them passed. OK, so clearly my base non-overclocked system seems fine. So did I just end up with literally THE worst silicon lottery pick ever?? I refused to believe that! I mean there is bad luck and then there's just nonsense.

So I then tried a different approach to overclocking. I first loaded the DDR XMP-3200 profile, kept all CPU overclocking disabled, left everything else at auto including LLC, etc, and ran the OCCT large dataset test again. It ran continuously for 8hrs without error. I tried running the other tests listed earlier for 2hrs each and they all passed. Alright, so my DDR overclock seems to be very stable.

With that working I now tried increasing the CPU frequency to 3.1GHz. I did not increase vcore at all, didn't touch SoC voltage, left LLC on Auto. The default vcore voltage on my mobo is 1.1875V and SoC voltage defaults to 1.0V. I ran OCCT large dataset with this overclock configuration for 2 hrs. It worked!! I was stoked!

Now here are where things get SUPER bizarre. I subsequently kept slowly increasing the CPU overclock frequency in 100MHz steps - each time running OCCT large dataset for 2hrs. I thought for sure it would start crashing, but it NEVER did. I got all the way up to 3.8GHz without _ever_ increasing the CPU vcore, the only thing I did have to adjust when I hit 3.5GHz was increasing the SoC voltage to 1.15V (I just picked a number that I had seen on another youtube overclocking video).

Once I got to 3.8GHz I ran OCCT large dataset for 8hrs straight overnight. No errors! I then ran all of the tests listed earlier for 8hrs each (including OCCT medium dataset and memtest86 again). Still no errors! Not only that but the CPU temp sits happily around 48C under extreme load.

So my question is...........how in the heck is this even physically/electronically possible???? How can this CPU be running at 3.8GHz with stock vcore voltage??? I mean there is "winning the silicon lottery" and there is what's within the realm of physical reality and reason. This is so unrealistic and unbelievable to me that I keep thinking there must be something that I'm doing wrong or something that is failing somewhere that I just haven't found yet. At this point I've been _trying_ to break it (i.e. make it crash) and I can't. No matter what I run or for how long it just keeps on going like the energizer bunny. 0.o

There is some other weirdness as well. When running HWINFO64 the CPU vcore voltage sits around 1.16V ! Even lower than the actual vcore voltage shown in the BIOS. Additionally, even though I have specified my SoC voltage as 1.15V in the BIOS, HWINFO64 shows it sitting steadily at 1.18V. You can see this in the attached image.

P.S. Some notes:
- None of my overclocking messes with the base clock. It has always been at 100MHz.
- I also didn't mess around with anything to do with p-states because I don't understand them and didn't want to break something.
- I'm running windows 10
- I have the AMD Ryzen Balanced power plan enabled
- I have not used the AMD ryzen overclocking tool for windows at all (I don't even have it installed)

overclock_settings.png 286k .png file
 

Attachments

·
Administrator
Joined
·
64,361 Posts
Sorry bud, no magic here. You set your Vcore to Auto, this means it certainly will not be at stock settings. The motherboard will adjust the Vcore as necessary for the OC. This can mean the Vcore is higher than whats recommended. I would download HWINFO and see what your VCORE is under auto. Remember a cool CPU does not mean a happy one, its the first step but too much Vcore over extended periods will degrade CPU lifespan.

Don't frett though, its unlikely you have. Done any damage at this point.
 

·
Registered
Joined
·
433 Posts
I build my Ryzen system 2 days ago and it looks like my 1700X is a decent chip as well. I OCed it to 4Ghz and with RAM at 2133Mhz it ran Intel Burntest with only 1.260V (full load). Now I OCed RAM to 3066Mhz and I need 1.300V to get 4GHZ on CPU stable (IBT and PRIME95 1344K). Didnt get to play and work much with the yet so will see how it goes the upcoming days.

After 1h of IBT temps are max. 68°C with an 8 year old Noctua U12P which is pretty decent as well.
 

·
Registered
Joined
·
18 Posts
Discussion Starter #5
Quote:
Originally Posted by ENTERPRISE View Post

Sorry bud, no magic here. You set your Vcore to Auto, this means it certainly will not be at stock settings. The motherboard will adjust the Vcore as necessary for the OC. This can mean the Vcore is higher than whats recommended. I would download HWINFO and see what your VCORE is under auto. Remember a cool CPU does not mean a happy one, its the first step but too much Vcore over extended periods will degrade CPU lifespan.

Don't frett though, its unlikely you have. Done any damage at this point.
I think you must not have actually read my post. I specifically mention the values that HWINFO64 is showing me for vcore (which is very low) and I even provide a screenshot.
blinksmiley.gif
 

·
Registered
Joined
·
433 Posts
Quote:
Originally Posted by jwdonal View Post

I think you must not have actually read my post. I specifically mention the values that HWINFO64 is showing me for vcore (which is very low) and I even provide a screenshot.
blinksmiley.gif
In HWinfo in think CPU Core Voltage (SVI2 TFN) is the right sensor to show for vcore.
 

·
Registered
Joined
·
1,611 Posts
Quote:
Originally Posted by jwdonal View Post

I think you must not have actually read my post. I specifically mention the values that HWINFO64 is showing me for vcore (which is very low) and I even provide a screenshot.
blinksmiley.gif
Your screenshot is showing CPU VID, which is not vcore. They are different things.
 

·
Registered
Joined
·
18 Posts
Discussion Starter #8
Quote:
Originally Posted by abso View Post

In HWinfo in think CPU Core Voltage (SVI2 TFN) is the right sensor to show for vcore.
Yes, that is shown in the screenshot. I have annotated it and attached it. It is hovering stable around 1.16v as I stated in my OP.

Vcore is also shown as 1.16v in the CPU-Z app (also shown in the screenshot that I attached).

overclock_settings_vcore_annotated.png 353k .png file
 

Attachments

·
Registered
Joined
·
18 Posts
Discussion Starter #9
Quote:
Originally Posted by MooMoo View Post

Your screenshot is showing CPU VID, which is not vcore. They are different things.
Yes, the screenshot shows vcore in 2 different locations. See my previous post. Am I missing something?
 

·
Registered
Joined
·
433 Posts
Mb the new ones need a bit less voltage now compared to models from early 2017? Mine is from Week 42. Also CPU-Z shows me total nonsens Vcore. 0.3-0.6V at 4GHZ. Only HWinfo CPU Core Voltage (SVI2 TFN) shows me realistic Vcore readings.
 

·
Registered
Joined
·
338 Posts
CPU-z shows correct for Taichi, 1.168 is very good for 3800 especially for a 1700. The newer batches of Ryzen are indeed better than the old ones.
 

·
Administrator
Joined
·
64,361 Posts
Quote:
Originally Posted by jwdonal View Post

Quote:
Originally Posted by ENTERPRISE View Post

Sorry bud, no magic here. You set your Vcore to Auto, this means it certainly will not be at stock settings. The motherboard will adjust the Vcore as necessary for the OC. This can mean the Vcore is higher than whats recommended. I would download HWINFO and see what your VCORE is under auto. Remember a cool CPU does not mean a happy one, its the first step but too much Vcore over extended periods will degrade CPU lifespan.

Don't frett though, its unlikely you have. Done any damage at this point.
I think you must not have actually read my post. I specifically mention the values that HWINFO64 is showing me for vcore (which is very low) and I even provide a screenshot.
blinksmiley.gif
Fair enough, this is why using the phone to assist is not always great lol.
 
1 - 12 of 12 Posts
Top