• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Ryzen 7 9800X3D

pBay said:
Currently have my 9800x3d with 2x scalar, +200mhz, -35 CO. Hitting 24300 or so in Cinebench R23. All core maxing out under full load around 72C and a vcore of 1.2v.

Is that a decent chip or pretty much what most people are seeing?
Click to expand...

That’s in-line with expectations, especially with 2x Scaler.

-35CO is also good.
 
My chip will do -40 for 90% of tasks, but absolutely s*it the bed with YCruncher. Probably would be stable at -35, but I backed off to -30.

Judging by all the forums I’ve read, surprisingly, -15 is about the average.
 
Last edited:
pBay said:
Currently have my 9800x3d with 2x scalar, +200mhz, -35 CO. Hitting 24300 or so in Cinebench R23. All core maxing out under full load around 72C and a vcore of 1.2v.

Is that a decent chip or pretty much what most people are seeing?
Click to expand...
Seems ok, I think many people can dial in -40 curve, +200 boost, boot and play a few hours of games.

What is not apparent though is when clock stretching is happening on this stepping we have, or where this is actually stable considering almost everyone is using a RAM OC (remember, every expo profile is an overclock)

Read lots of immediate success going straight to undervolts with max boost override, read almost as much stress test failures doing so.

And as a couple of people have rightly pointed out, we're not getting the best bins. AMD have always binned their best CCD complexes for the preferential CCD on the 16 core SKU.
 
I got this chip on Wednesday, and I literally haven’t been able to do a single solitary moment of gaming yet. Work, coupled with my brain not allowing me to ‘Expo and go’ can be annoying at times lol.

I can never get a new component without obsessing over squeezing out every drop possible.

As much as I love overclocking (especially RAM) I can’t wait to be done with it so I can have my system back
 
Cenuij said:
Seems ok, I think many people can dial in -40 curve, +200 boost, boot and play a few hours of games.

What is not apparent though is when clock stretching is happening on this stepping we have, or where this is actually stable considering almost everyone is using a RAM OC (remember, every expo profile is an overclock)

Read lots of immediate success going straight to undervolts with max boost override, read almost as much stress test failures doing so.

And as a couple of people have rightly pointed out, we're not getting the best bins. AMD have always binned their best CCD complexes for the preferential CCD on the 16 core SKU.
Click to expand...

If it passes about 10,000% coverage in Karhu (Karhu claim 6400% shows up over 99% of errors), it’s almost certainly ready to go.

CPU is the very easy part in comparison. Set an aggressive undervolt/curve and just test in OCCT for one hour. If it passes that, it’s probably most of the way there. Small Linpack/YCruncher and play some games to confirm. If it’s not stable? Very easy to rectify. Just back off and repeat. CPU overclocking/undervolting is easy peasy. It’s RAM overclocking that’s the hard part.

RAM overclock should always been done and nailed in first.
 
Last edited:
matty007 said:
I got this chip on Wednesday, and I literally haven’t been able to do a single solitary moment of gaming yet. Work, coupled with my brain not allowing me to ‘Expo and go’ can be annoying at times lol.

I can never get a new component without obsessing over squeezing out every drop possible.

As much as I love overclocking (especially RAM) I can’t wait to be done with it so I can have my system back
Click to expand...
I hear that, spent most of this week tuning and doing stress runs. And that's after picking kit for ezmode, not performance.

Here's how it goes ;)
---

.oO

Ryzen IMC is still a lottery

But boards and BIOS have matured

Still, life's too short, 6000/30 is almost a gimme

May as well buy 64G @ these timings

Oh, 6200 was ezmode as well, what's in the tank?

Well, we've booted CL28, it's only a few tweaks from stable right?

Glad I only spent 6 hours getting CL28 "stable", tweaking 128G DDR4 was much worse (it wasn't)

CL28 cleared X test after X hours, time to pwn the noobs!

Was that packet burst or IMC error correction that killed me?

Oh, this higher intensity shorter test throws errors.

Let's relax the main timing and focus on the IMC

Well, that was only 2 days of testing, not bad for this ezmode kit right? right...?

Let's bench the pwn the noobs game again, yep still zero CPU bottleneck but at least my ego knows it _must_ be packet burst that gets me killed ;)

Now I've marginally OC'd my ram OC, I can tune my CPU, nothing better than knowing my zero CPU bottleneck will be even more zero, winning!
---

Sad or best part of this is I've enjoyed pointlessly tweaking ram more than playing the current crop of games

¯\_(ツ)_/¯
 
Cenuij said:
I hear that, spent most of this week tuning and doing stress runs. And that's after picking kit for ezmode, not performance.

Here's how it goes ;)
---

.oO

Ryzen IMC is still a lottery

But boards and BIOS have matured

Still, life's too short, 6000/30 is almost a gimme

May as well buy 64G @ these timings

Oh, 6200 was ezmode as well, what's in the tank?

Well, we've booted CL28, it's only a few tweaks from stable right?

Glad I only spent 6 hours getting CL28 "stable", tweaking 128G DDR4 was much worse (it wasn't)

CL28 cleared X test after X hours, time to pwn the noobs!

Was that packet burst or IMC error correction that killed me?

Oh, this higher intensity shorter test throws errors.

Let's relax the main timing and focus on the IMC

Well, that was only 2 days of testing, not bad for this ezmode kit right? right...?

Let's bench the pwn the noobs game again, yep still zero CPU bottleneck but at least my ego knows it _must_ be packet burst that gets me killed ;)

Now I've marginally OC'd my ram OC, I can tune my CPU, nothing better than knowing my zero CPU bottleneck will be even more zero, winning!
---

Sad or best part of this is I've enjoyed pointlessly tweaking ram more than playing the current crop of games

¯\_(ツ)_/¯
Click to expand...

Brilliant lol
 
This will do for me....

Managed to reduce the primaries to 37. I could possibly tighten a couple of timings further, however, I'm riding close to the edge with this. Reducing SCL's by even 1 caused serious instability, and I also think the TRFC is close to limit. In regards to FCLK, I do believe 99% that it's stable at 2200, however, I am currently running 2133 to bench the diffrence between 2133 and 2200. My very early impressions however are that 2133 is very slightly faster in games. I will test further to confirm.

In regards to CPU, I tried a manual all core overclock at 5.6 and 1.35v, but was very unstable. Even dropping to 5.5Ghz didn't help. As such, I have gone with -

-30CO
+200 offset
10xScaler

Cinebench score is 24,000 on the dot.

I may in the future compare this profile with 6000CL26 GDM disabled. I do believe that this will be slightly faster though due to bandwith, and still running at a very respectable CL28.

As much as I love overclocking and constantly pushing, I am very happy with this, and am content.

FY1DamO.png
 
Last edited:
matty007 said:
If it passes about 10,000% coverage in Karhu (Karhu claim 6400% shows up over 99% of errors), it’s almost certainly ready to go.

CPU is the very easy part in comparison. Set an aggressive undervolt/curve and just test in OCCT for one hour. If it passes that, it’s probably most of the way there. Small Linpack/YCruncher and play some games to confirm. If it’s not stable? Very easy to rectify. Just back off and repeat. CPU overclocking/undervolting is easy peasy. It’s RAM overclocking that’s the hard part.

RAM overclock should always been done and nailed in first.
Click to expand...
So this is where things seem to be given on faith. And I'm starting to suspect that some of this faith is given on nothing more than word of mouth.

For example Karhu suggests that my CL28 is stable after an 8 hour run.

But then TM5 fails the same timings within something like 15-20 mins

I think it's not fair to say y-cruncher is just a short cursory test. The all test run defaults will use all instruction sets available and with as much ram as it can use.

So what's going on here? I don't know that the source is available for these two utils (Karhu/TM5) so programmer error could be a factor in one or both of them.

I should also point out that one of the popular usages of TM5 is with a custom profile that's supposed to show errors "early". I find that somewhat unconvincing.

Though I've not yet examined the profile and the methodology it configures, I'm certainly curious why a mem testing tool would not be at its most effective without a custom profile.
 
Cenuij said:
So this is where things seem to be given on faith. And I'm starting to suspect that some of this faith is given on nothing more than word of mouth.

For example Karhu suggests that my CL28 is stable after an 8 hour run.

But then TM5 fails the same timings within something like 15-20 mins

I think it's not fair to say y-cruncher is just a short cursory test. The all test run defaults will use all instruction sets available and with as much ram as it can use.

So what's going on here? I don't know that the source is available for these two utils (Karhu/TM5) so programmer error could be a factor in one or both of them.

I should also point out that one of the popular usages of TM5 is with a custom profile that's supposed to show errors "early". I find that somewhat unconvincing.

Though I've not yet examined the profile and the methodology it configures, I'm certainly curious why a mem testing tool would not be at its most effective without a custom profile.
Click to expand...

Stable should be stable, but here's the thing...if you are on the limit with your overclocks, it is very possible that multiple programs will have diffrent results. Different programs stress the ram with varying profiles and methods. All I can speak from is personal experience. Testmem5 with extreme config is very very good, however, I have always persoanlly found Karhu to find the same amount of errors, and most importantly, to find them faster. Also, I have never had Testmem5 find errors that Karhu couldn't, but, I have had Karhu find errors that Testmem did not. A lot of it is personal experience and preference.

Personally, if my RAM overclock is stable in Karhu for over 10 hours, I am VERY confident it is actually stable. If it actually isn't? Meh. I have nothing critical on my OS drive so it's not the absolute end of the world. I will also say though that I test my RAM quite stringently, and it's very rare for me to obtain 100% trust in my RAM OC, and it turning out to be unstable.

In regards to YCruncher, I did not mean it was designed as a short test, I am merely saying that if 1 hour of OCCT passes for me, then usually a short run of YCruncher is sufficent to confirm stability. I don't use YCruncher on it's own.
 
Ehh, personal experience of negative Curve Offsets on the 5800X, 5900X, 7700X and 7800X3D have taught me that I can be stable in benchmarks, stable in stress tests overnight (OCCT, Prime 95) and still get random hangs or crashes every week or so playing games. It stopped being worth it to me losing progress in a single player game or being disconnected from a Counter Strike match or a hard Destiny 2 strike just for a couple more FPS or slightly lower temperatures. I usually end up settling on -10 offset after starting at -30 but 2 of my CPUs were only completely bug free at -8.

I likewise tried the manual memory tuning from Buildzoid that is "guaranteed" to be stable and the automated memory tuning using Ryzen DRAM Calculator from 1usmus. Both were faster, could pass benchmarks and got me some decent ePeen scores but neither were stable (the DRAM calculator especially so).

I had a lot of fun tuning my chips over the last few years but ultimately I am looking for stability and therefore I think with the 9800X3D I am probably going to leave it at stock with EXPO since it has so much headroom above my monitor's max refresh rate. Sorry to be a party pooper
 
Got mine built last night. Coming from a 5700x.

EXPO enabled so 2x16gb running at 6000m/ts CL30. No PBO or anything yet. I'm seeing 35-40C at idle (NHD15), 55Cish in games, 78C in Cinebench.

Time Spy I got another 2k+ in total score (my CPU score went up 50%, obviously).

Cyberpunk saw massive gains at 3440x1440 with DLSS set to performance and RT on (3080 12gb). The benchmark was 95fps up from 80fps. I was struggling to clear 60fps in busy areas with lots of NPCs, now I'm comfortably at 80-85fps. Buzzing!
 
InVitriol said:
Would a Noctua NH-U12S with dual fans be enough to cool the 9800x3d? Without overclocking. It works fine for my 5800x3d. Want to stay air and perhaps safe myself some money on the upgrade lol
Click to expand...
yes i use a u12a and it cools it very well

edit: sorry forgot to mention that i use the offset bracket and with a minus 40 curve see temps at 62 after 30 mins in cinebench
 
Last edited:
InVitriol said:
Would a Noctua NH-U12S with dual fans be enough to cool the 9800x3d? Without overclocking. It works fine for my 5800x3d. Want to stay air and perhaps safe myself some money on the upgrade lol
Click to expand...

My take on it, at the moment, is that the chips do run hot, but this is not because of the cooler, rather it is because of their construction. Being layered, reduces the chips ability to conduct heat away. What that means is, there is probably less advantage to using a super-huge cooler. The 12S will probably work as well as a D15. I have no evidence to confirm that, it's just based off my understanding of the 9800X3D.
Certainly, what I would do is try it.
 
Got mine installed, filled and primed loop. I’ve not tweaked anything yet other than ‘EXPO 1’, but is 75 deg. C max ok after 10 mins of CPU-Z stress test? Yes I know this is a very basic stresser but as a quick check?

Loop = 3 x 360 thick rads, water temp got to 27 deg. C
 
You must log in or register to reply here.
Back
Top Bottom