Confused and potential problems with 3900x overclocking and behavior

SilverSliver said:
360 rad, gentle typhoons, 66c during cinebench r20. Idle state blips and gaming are 4.25ghz and all core full load is 4.075ghz
Click to expand...
Seems like a Bios kink, to me. Launch reviews reported different behavior, due to using different motherboards. Probably just have to wait it out, for an update.
 
On mine, using custom water I was getting 78c at 4.3GHz on all cores using a manual overclock. Using boost clocks it typically stayed around 68c full load.
 
Running Cinebench R20 all cores at 3.975ghz and 65c for a temp scoring 7022. R15 is 4ghz all cores and also 65c for a temp and scoring 3106.........AMD Wraith for a cooler and ambient temps about 68f
 
Last edited:
I went ahead and put two sticks of GSkill 3200 that I had from my old build in my new build, replacing the 3600. Set XMP and booted up. Boost clocks are 100 to 150mhz higher now, and scores in R15/R20 did not change. This is really weird.

3200 vs 3600 ram.jpg
 
Last edited:
SilverSliver said:
I went ahead and put two sticks of GSkill 3200 that I had from my old build in my new build, replacing the 3600. Set XMP and booted up. Boost clocks are 100 to 150mhz higher now, and scores in R15/R20 did not change. This is really weird.

View attachment 175633
Click to expand...
That is weird, are you seeing the higher boost clocks as well? Faster speed ram limits boost clocks?
 
noko said:
That is weird, are you seeing the higher boost clocks as well? Faster speed ram limits boost clocks?
Click to expand...
he said boost clocks are higher ;)

Most likely better compatibility with that RAM, for that mobo and bios version.
 
chameleoneel said:
he said boost clocks are higher ;)

Most likely better compatibility with that RAM, for that mobo and bios version.
Click to expand...
Yes but that is from 3dMark, he said he does not use Ryzen Master. Only two programs that are accurate for real time monitoring I know of, CPUz and Ryzen Master. Plus his scores were the same. ;)
 
noko said:
Yes but that is from 3dMark, he said he does not use Ryzen Master. Only two programs that are accurate for real time monitoring I know of, CPUz and Ryzen Master. Plus his scores were the same. ;)
Click to expand...
Physics scores were a little better.

100mhz extra on the CPU isn't going to affect most of your graphics scores in 3D.
 
noko said:
Yes but that is from 3dMark, he said he does not use Ryzen Master. Only two programs that are accurate for real time monitoring I know of, CPUz and Ryzen Master. Plus his scores were the same. ;)
Click to expand...

I'm running ryzen master now, it's the only way to get some semblance of PBO running. I've verified higher clocks in ryzen master and cpuz. Rivatuner osd is accurate as well.
 
  • Like
Reactions: noko
like this
SilverSliver said:
I went ahead and put two sticks of GSkill 3200 that I had from my old build in my new build, replacing the 3600. Set XMP and booted up. Boost clocks are 100 to 150mhz higher now, and scores in R15/R20 did not change. This is really weird.

View attachment 175633
Click to expand...

I ran this also a while back and noticed your physics score and test are lower than mine for the same cpu, not sure what that means but something is holding the cpu back.

Capture.JPG
 
noko said:
That is weird, are you seeing the higher boost clocks as well? Faster speed ram limits boost clocks?
Click to expand...

If there was a higer booster clock, I didn't notice it when I ran my ram at 3200 and 3600. Running ram at 3600 still gave me higher Geekbench scores.
 
  • Like
Reactions: noko
like this
SilverSliver said:
I went ahead and put two sticks of GSkill 3200 that I had from my old build in my new build, replacing the 3600. Set XMP and booted up. Boost clocks are 100 to 150mhz higher now, and scores in R15/R20 did not change. This is really weird.

View attachment 175633
Click to expand...

That is really odd. I wouldn't have even thought it would behave that way.
 
Dan_D said:
That is really odd. I wouldn't have even thought it would behave that way.
Click to expand...

I have a different clocking bug, but occasionally upon boot, my 3800x will get stuck at around 1ghz and be super slow. Putting the computer in/out of sleep or rebooting fixes it. Nothing in event logs, and everything in Ryzen Master looks fine when this happens. It's just stuck at a lower clock.

Needless to say, there are some issues AMD needs to get fixed.
 
Dan_D said:
That is really odd. I wouldn't have even thought it would behave that way.
Click to expand...

Here is a better comparison. I just swapped and did back to back runs. Scores are better with 3600 ram, but you can see the boost clocks are lower. PBO is off.

3200 vs 3600 ram.jpg
 
Does anyone out there notice they get much higher clocks and consistently when they enable single CCD mode on their 3900x?

For those that dont know CCD is AMD speak for Core Complex Die, or basically since 3900x has two CCDs you disable one rendering your CPU into a 6 core. A 6 core with a shit ton of Cache haha but I notice that the clocks are much higher and I score higher points on CB 15 ST.

I wonder how one can improve ST performance on a 3900x while using both CCDs?

Or is it just an inherent performance penalty of using the fabric?

All my testing now is done with broke ass PB2 turned off. Just using auto overclocking (natural boost algorithm in other words)
 
tangoseal said:
Does anyone out there notice they get much higher clocks and consistently when they enable single CCD mode on their 3900x?

For those that dont know CCD is AMD speak for Core Complex Die, or basically since 3900x has two CCDs you disable one rendering your CPU into a 6 core. A 6 core with a shit ton of Cache haha but I notice that the clocks are much higher and I score higher points on CB 15 ST.

I wonder how one can improve ST performance on a 3900x while using both CCDs?

Or is it just an inherent performance penalty of using the fabric?

All my testing now is done with broke ass PB2 turned off. Just using auto overclocking (natural boost algorithm in other words)
Click to expand...

I thought it was two CCD on 3900x, one CCD had the two full CCX's meaning 8 cores, and one only had one CCX activated which is where you get your extra 4 cores.

Additionally, the cache isn't shared outside of each CCX, the increased cache size of 3900x doesn't really stack anything. You still have the same amount of cache per CCX, which is 16MB.

As to why you are getting higher clocks, it's because less heat in the entire package.
 
Mchart said:
I thought it was two CCD on 3900x, one CCD had the two full CCX's meaning 8 cores, and one only had one CCX activated which is where you get your extra 4 cores.

Additionally, the cache isn't shared outside of each CCX, the increased cache size of 3900x doesn't really stack anything. You still have the same amount of cache per CCX, which is 16MB.

As to why you are getting higher clocks, it's because less heat in the entire package.
Click to expand...

No you misunderstand the whole thing I was trying to say but then again I sucked at describing it in the detail I should have.

1CCD is 6 cores on the 3900x. Otherwise it would give me 8 cores. I only get 6 thus the math supports the CCD being one single die out of the two having 6 cores each.

The cache is still significant at 32MB per die. The 3600 has 32MB of cache on its die and so does the 3900x (per die x2 total dies @ 64MB l3 Cache...oops game cache). This is far more than the 2700x which had only 16MB of l3 cache. Hence my comment about having boatloads of cache.

Even Tom's supports this and Tom's is NOT enthusiast grade :p

upload_2019-7-21_23-12-10.png
 
Last edited:
Yes, you are correct, I just looked. I thought they cut an entire CCX, instead they just cut out a bad core per CCX, it makes more sense.

That being said, that doesn't change what I said about the cache. The cache is per CCX, and if you disable a CCD you are correct that you still have 32MB (16MB per CCX). Although at that point you should have just bought a 3700x/3800x as you'll have the high clocks and two extra cores.
 
You must log in or register to reply here.
Back
Top