9800x 3d vs 12900k - Battle of the Century

[mkppo]

The HDV can't go over 8200. Or maybe it's just my ram, dunno. Can't make it even boot. Still pretty nuts for the price.

The HDV from what i've seen in general can do 7800-8000 on dual CCD's and on single CCD's 8000-8200 if your IMC is great. But disabling GDM at 8000 is a no go and sometimes 7800 too but what more can you expect from a $100 board..

If yours does 8000, set UCLK:FCLK synced to 2000 and see if it's faster for gaming than whatever you get with 2:3 FCLK:UCLK which in the 9800X3D seems to be cap out at 2133:3200 on average but 2200:3300 seems to be getting more common now. VSOC requirements go down too with both FCLK:UCLK at 2000 too so once you dial that in the CPU power budget also increases slightly. Then there's the usual timings and subtimings and honestly given the scaling on the 9800X3D you don't even need to go ultra tight, like 98% there with absolute stability is fine I guess but what fun is it if you don't push it to 11..

ODT's seem to be set slightly weirdly too, ProcODT at 40 ohms and DRAM DQs at 48 improved stability for me compared to auto. Then there's the obscure Addrcmdval (or whatever setting its called) which helps match tphyrdl's or just make it train harder. Mine also needs three taps of Rx (receiver side) but two on Tx are fine for 8000. There were a few other minor adjustments other than the usual timings/subtimings which helped me stabilize at 8000 but it's absolutely on the edge with my 9950's pretty poor IMC and it being dual CCD which hits the IMC harder.

I guess I haven't really followed the whole thread but you seem to be having stuttering for which I can't really be of too much help and others surely addressed the issue. All i know is POE2 seems to be pretty bad with stutters in general and HogL too before patches. Judging by the length of the thread though, i suspect it's something else.

 

[RootinTootinPootin]

I've been waiting for the X870 Tachyon ever since I heard a whiff of it but honestly got tired of waiting for so damn long and got a B650-HDV to tide me over for now. Mobo manufacturers be doing jack shit for years and literally released TWO boards with good memory overclocking ever since X870's launch, both of which were super limited in stock. Hell, the Tachyon was borderline vaporware, I tried to buy one for ages but short of getting it from Japan it was damn near impossible. Gene was available for a while, but there were a few revisions with A04 being the best one of the lot and I missed out while trying to get that specific revision and soon enough they were gone and people were asking stupid prices for it.

I'm receiving an Aorus B650E Tachyon for testing purposes.

 

[phanbuey]

I'm receiving an Aorus B650E Tachyon for testing purposes.

I think gigabyte has some of the best ram training for AMD - their boards consistenlty spank everyone else on AM5 at dialing in subtimings.

 

[ShrimpBrime]

I think gigabyte has some of the best ram training for AMD - their boards consistenlty spank everyone else on AM5 at dialing in subtimings.

All the B series boards look really good in a lot of ways. Makes me wonder why they even bother with X series chipsets...

 

[RootinTootinPootin]

I think gigabyte has some of the best ram training for AMD - their boards consistenlty spank everyone else on AM5 at dialing in subtimings.

That's why it was sent to me for testing..hehehehe..

All the B series boards look really good in a lot of ways. Makes me wonder why they even bother with X series chipsets...

Haven't seen anybody need 20 USB high speed ports and 2 chipsets..I am kinda happy already with how modest boards makers do with their I/O plates.

 

[JustBenching]

Some whack results - I don't know what the heck is going on with this CPU, woke up today and run some tests, now temperatures seem to have dropped a LOT. I changed absolutely nothing in the bios (im running stock with -20 CO) - room temperature is always at 19-22C moderated, run a CPU Z and CPU sits at 66-71C. @ 120w. A week ago it was hitting 90-92C at the same 120w....

So, temps fixed themselves I guess?

 

[oxrufiioxo]

AMD dropped by while you were sleeping and swap one with a better bin.... You didn't see the note that said we know you want that sexy AF 9070XT bruh....

 

[JustBenching]

AMD dropped by while you were sleeping and swap one with a better bin.... You didn't see the note that said we know you want that sexy AF 9070XT bruh....

Man, that's a huge difference, makes the damn thing usable now. Good to know it wasn't the cooler, I guess something with early bioses / early agesas, changes settings on reboots or I don't even know what the hell is happening. Still hitting 84C in CBR23 but at least now it's pulling the full 150-160w, before it was throttling at 96C. It's even scoring 23400 at stock settings.

 

[oxrufiioxo]

Man, that's a huge difference, makes the damn thing usable now. Good to know it wasn't the cooler, I guess something with early bioses / early agesas, changes settings on reboots or I don't even know what the hell is happening.

I've seen most max out around 85C but in a 100% workload.... Down into the mid high 70s with a 360 aio so your 92 always seemed off to me unless it was OC'd

Although i will say anyone who has worked with a decent amount of Zen cpu knows they can vary quite a bit I ran into a few 5800X that were trash and some 2600X that were worse than 2600....

 

[JustBenching]

I've seen most max out around 85C but in a 100% workload.... Down into the mid high 70s with a 360 aio so your 92 always seemed off to me unless it was OC'd

Im still hitting 84C on CBR23, I edited the previous post.

But the best part is, it's doing all that with the fan at 38% (850RPM). Before it was maxed out at 100%. Im at a loss...

 

[oxrufiioxo]

Im still hitting 84C on CBR23, I edited the previous post.

But the best part is, it's doing all that with the fan at 38% (850RPM). Before it was maxed out at 100%. Im at a loss...

Super strange I would be baffled as well... Although my R5 7600 ran oddly warm so you just never know with Zen some chips behave oddly.

 

[JustBenching]

This is a video from today

Spoiler

And this is from a week ago

Spoiler

Looks like the decreased temps also reduced the voltage required by quite a bit. Both runs are with the fans maxed out just for iso comparisons.

 

[oxrufiioxo]

This is a video from today

Spoiler

And this is from a week ago

Spoiler

Looks like the decreased temps also reduced the voltage required by quite a bit. Both runs are with the fans maxed out just for iso comparisons.

Wattage is close enough but voltage is def lower.

 

[JustBenching]

Wattage is close enough but voltage is def lower.

Yeah it's boosting a bit higher.

Maybe the LM needed some time to settle

 

[oxrufiioxo]

Yeah it's boosting a bit higher.

Maybe the LM needed some time to settle

I have zero experience with LM but have heard it can be a pita.


Is that even necessary for ambient cooling lol. Doesn't that destroy the IHS or cause pitting etc.

 

[JustBenching]

I have zero experience with LM but have heard it can be a pita.


Is that even necessary for ambient cooler lol. Doesn't that destroy the IHS or cause pitting etc.

It destroys the looks on the IHS yes, but conductivity gets better. The LM itself gets "absorbed" by the IHS closing the gaps - basically makes it flatter. You can use LM for a month, then clean it and use normal paste, now your temps should be better than if you used normal paste from day one. I've been using LM only both for CPUs and GPUs since 2017.

For example, there is no way a u12a can do this without LM. This is 330 watts @ 95c. No deliding, stock ihs, no LGA bracket. This is custom loop territory.

Spoiler

The HDV from what i've seen in general can do 7800-8000 on dual CCD's and on single CCD's 8000-8200 if your IMC is great. But disabling GDM at 8000 is a no go and sometimes 7800 too but what more can you expect from a $100 board..

If yours does 8000, set UCLK:FCLK synced to 2000 and see if it's faster for gaming than whatever you get with 2:3 FCLK:UCLK which in the 9800X3D seems to be cap out at 2133:3200 on average but 2200:3300 seems to be getting more common now. VSOC requirements go down too with both FCLK:UCLK at 2000 too so once you dial that in the CPU power budget also increases slightly. Then there's the usual timings and subtimings and honestly given the scaling on the 9800X3D you don't even need to go ultra tight, like 98% there with absolute stability is fine I guess but what fun is it if you don't push it to 11..

ODT's seem to be set slightly weirdly too, ProcODT at 40 ohms and DRAM DQs at 48 improved stability for me compared to auto. Then there's the obscure Addrcmdval (or whatever setting its called) which helps match tphyrdl's or just make it train harder. Mine also needs three taps of Rx (receiver side) but two on Tx are fine for 8000. There were a few other minor adjustments other than the usual timings/subtimings which helped me stabilize at 8000 but it's absolutely on the edge with my 9950's pretty poor IMC and it being dual CCD which hits the IMC harder.

I guess I haven't really followed the whole thread but you seem to be having stuttering for which I can't really be of too much help and others surely addressed the issue. All i know is POE2 seems to be pretty bad with stutters in general and HogL too before patches. Judging by the length of the thread though, i suspect it's something else.

Can boot 7800 instantly with auto settings - which is impressive - better than what my 700$ unify X and 800$ apex could pull. For anything above that it needs some manual tuning but it can get to 8200 stable. I haven't really tried much for higher than that so not sure if it can pull it.

Haven't really tested performance though but between my current 6400c28 and whatever 8000 I end up running, i think it will be a wash.

 

[x4it3n]

Overclocking as in PBO or are you also touching ram? Have you tried a static OC?

My lows do struggle if I mess with the ram settings, and also if I enable "Low latency mode" it makes the sutters a bit more obvious - PBO doesn't have any negative impact, though.

Yeah OC with +200MHz PBO. My RAM is still the same I have 2x32GB 6400MHz 30-38-38-30 that I optimized and is working fine without the OC.

I'll try to repaste my CPU just in case because I noticed that I get high temperatures when there's an all-core load and sometimes it can go high during a very quick moment and then go back down.

 

[JustBenching]

Went back to the 12900k now, I noticed that one also stutters in veilguard. So not a 9800x 3d problem specifically. It's still very infrequent. Next one ill try POE.

Now going back, 3 things I noticed.

1) The 9800x 3d is in fact snappier in windows contrary to what I initially thought. Both using windows 10. Not anything to write home about really, but it feels a tiny bit worse.

2) 12900k is much much faster at shader compilation. Much faster than their MT performance difference would tell you. Since both games im playing now (stalker 2 / veilguard) actually compile shaders every single time you try them, the difference between the 2 cpus is nuts. (maybe down to memory bandwidth?)

3) I really missed the 12900k temperatures. Currently it's sitting at 29C - compared to 45-49C I had on my 9800x 3d for simple tasks. Holy cow

 

[Degreco]

The 9800x 3d is in fact snappier in windows contrary to what I initially thought. Both using windows 10. Not anything to write home about really, but it feels a tiny bit worse.

Interesting that this changed. What are your Latencymon scores for the 12900k vs 9800x3D? Is there a huge difference?

I found out that my 14600KF has better latency in Latencymon than my 7800x3D in W11 24H2. Before knowing that I felt no difference when doing work on both rigs. After knowing the results the 7800x3d for real felt subjectively less snappy (outside of games). Of course, knowing the results forms the perception

 

[JustBenching]

Interesting that this changed. What are your Latencymon scores for the 12900k vs 9800x3D? Is there a huge difference?

I found out that my 14600KF has better latency in Latencymon than my 7800x3D in W11 24H2. Before knowing that I felt no difference when doing work on both rigs. After knowing the results the 7800x3d for real felt subjectively less snappy (outside of games). Of course, knowing the results forms the perception

Well your 14600 should be snapper than my alderlake and your 7800x 3d should be less snappy than the 9800x 3d so it kinda makes sense. Overall they are both overkill for simple tasks so doesn't really matter that much at the end of the day.

 

[mkppo]

I couldn't tell much of a difference in snappiness when comparing stock vs tuned 9950x where I shaved a bunch of latency. I can, however, tell the difference in my two OS boot drives. The P5800X is noticeably snappier compared to the 980 Pro, but that's to be expected I suppose.

 

[Cowboystrekk]

I disabled hw prefetcher in bios, that improved snappyness

 

[Cowboystrekk]

Went back to the 12900k now, I noticed that one also stutters in veilguard. So not a 9800x 3d problem specifically. It's still very infrequent. Next one ill try POE.

Now going back, 3 things I noticed.

1) The 9800x 3d is in fact snappier in windows contrary to what I initially thought. Both using windows 10. Not anything to write home about really, but it feels a tiny bit worse.

2) 12900k is much much faster at shader compilation. Much faster than their MT performance difference would tell you. Since both games im playing now (stalker 2 / veilguard) actually compile shaders every single time you try them, the difference between the 2 cpus is nuts. (maybe down to memory bandwidth?)

3) I really missed the 12900k temperatures. Currently it's sitting at 29C - compared to 45-49C I had on my 9800x 3d for simple tasks. Holy cow

Did you try disabling hw prefetchers and all viritulization? This goes for 12900k aswell. I got rid of the infrequent rare stuttering on my 7800x3d doing so. As for shadercompiling I think that is case where 2200fclk (increase BW by over 10% vs 2000fclk) and going 2:1 at 7800+ would help as it increases BW at write by 10%+. A 2 ccd cpu that scales with read and copy aswell would be far better and be more comparable to 12900K. It would probably lower compiling by 15-20% on 9800X3D, but 12900K would still be far superior.

As for temps AM5 will always suck unless you go direct die due to thick heatsink.

You have some tempissues I dont understand. My 7800x3d which is harder to cool than 9800x3d idles at 37-38C with 20C ambient. I have quite good airflow and using peerless assassin and ptm7950.

 

[JustBenching]

Did you try disabling hw prefetchers and all viritulization? This goes for 12900k aswell. I got rid of the infrequent rare stuttering on my 7800x3d doing so. As for shadercompiling I think that is case where 2200fclk (increase BW by over 10% vs 2000fclk) and going 2:1 at 7800+ would help as it increases BW at write by 10%+. A 2 ccd cpu that scales with read and copy aswell would be far better and be more comparable to 12900K. It would probably lower compiling by 15-20% on 9800X3D, but 12900K would still be far superior.

As for temps AM5 will always suck unless you go direct die due to thick heatsink.

You have some tempissues I dont understand. My 7800x3d which is harder to cool than 9800x3d idles at 37-38C with 20C ambient. I have quite good airflow and using peerless assassin and ptm7950.

It's just my cooler that doesn't like am5, it's a u12a. Works better on intel apparently.

 

[Cowboystrekk]

It's just my cooler that doesn't like am5, it's a u12a. Works better on intel apparently.

I had D15 earlier and the peerless assassin works better on am4 and am5 in my experience.