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

[nguyen]

Ordered a 9800X3D myself, few reasons why but that's beside the point, really looking forward to seeing the CPU bound uplift in certain games like Space Marine 2.

Does anyone have any specific RAM kits to recommend? Looks like I'll be pairing it with a Gigabyte B650I AX, and from what I'm gathering 6000-6400 CL28-32 is around where I should be aiming, I'm not really interested in RAM tweaking so a set and forget option suits me, the shorter the better and zero need for RGB.

Locally sourced options in Australia highly preferable, if that helps.

Been playing with my 9800X3D for a few days, in term of FPS it is kicking ass, particularly in PUBG @ 4K competitive settings the 9800X3D is around 30-40% faster than the 13700K (UE4 games just love that 3D-Vcache)

I don't think RAM tuning is that important with 9800X3D anyways, I have roughly the same FPS in PUBG with the RAM at Jedec specs (4800MT)

My new kit

 

[JustBenching]

Been playing with my 9800X3D for a few days, in term of FPS it is kicking ass, particularly in PUBG @ 4K competitive settings the 9800X3D is around 30-40% faster than the 13700K (UE4 games just love that 3D-Vcache)

I don't think RAM tuning is that important with 9800X3D anyways, I have roughly the same FPS in PUBG with the RAM at Jedec specs (4800MT)

My new kit
View attachment 371530

If we don't tune the ram we will lose that 1 fps, what are you talking about man?

/s

 

[Cowboystrekk]

Got some information from the horse's mouth itself (cough amd cough), the ideal target should be 6600c26 1:1 IMC with 2200 FCLK. Should be doable at 1.7 VDD and 1.6 VDDQ. Needs active cooling (obviously). Ill target that first and if it flops ill go for 8k.

Problem is, how do you stress test this? You start just with FCLK, check if that's stable, and then target the ram / IMC afterwards?

8400/2100 is msybe even better. Nice latencydrop! But few cpus can do 8400.

I would fine tune ram first, then try higher fclk. Often you must reduce fclk at 1.25v+ soc, try staying below that.

 

[JustBenching]

8400/2100 is msybe even better. Nice latencydrop! But few cpus can do 8400.

I would fine tune ram first, then try higher fclk. Often you must reduce fclk at 1.25v+ soc, try staying below that.

So FCLK is only linked to VSOC in terms of stability? Meaning, let's say I run everything stock, in order to increase FCLK the only voltage rail I need to play with is Vsoc?

 

[Cowboystrekk]

So FCLK is only linked to VSOC in terms of stability? Meaning, let's say I run everything stock, in order to increase FCLK the only voltage rail I need to play with is Vsoc?

In general yes. On my setup I must lower fclk to 2167 if soc is 1.25v or higher. Higher soc snd vddp will increase ram oc potential. If ypu run 8000:2000 and 2000fclk you get great latency, 8400:2100 and 2100fclk even better. 2200fclk will beat them both on BW, but lose on latency. Ironically higher fclk needs more soc than lower fclk.

On a 7800x3d I build with terrible imc 6000/2000 was stable at 1.26v soc. If I increased to 2133fclk it failed on TM5 due to soc requirements rising. Higher fclk stresses imc more.

 

[rv8000]

So FCLK is only linked to VSOC in terms of stability? Meaning, let's say I run everything stock, in order to increase FCLK the only voltage rail I need to play with is Vsoc?

VDDG voltages can help stabilize fclk.

 

[phanbuey]

Shadow of the benchmarks run with some slowish ram at 6000 30-38-38-96

I seem to be getting better scores at 6000 cl30 than 6400 at the same timings....

So 6000/2000/3000 > 6400/2133/3200 .... not sure why - both sets stable.

 

[rv8000]

View attachment 371555

Shadow of the benchmarks run with some slowish ram at 6000 30-38-38-96

I seem to be getting better scores at 6000 cl30 than 6400 at the same timings....

So 6000/2000/3000 > 6400/2133/3200 .... not sure why - both sets stable.

How did you test stability? Linpack Xtreme was previously a good indicator of fclk stability; as long as each loop was returning the same gflop results it was a good indicator fclk was actually stable.

 

[phanbuey]

How did you test stability? Linpack Xtreme was previously a good indicator of fclk stability; as long as each loop was returning the same gflop results it was a good indicator fclk was actually stable.

1600% on ramtest 1.0 and then i usually leave it on for a week working and playing games.

 

[JustBenching]

Im getting my stuff today, pray to Lisa I get a good bin.

 

[RootinTootinPootin]

Im getting my stuff today, pray to Lisa I get a good bin.

report your overall system latency on a full tweaked setup..

 

[JustBenching]

report your overall system latency on a full tweaked setup..

You mean the Aida memory test? It's 47.5ns on my 12900k.

Regarding the 9800x3d, I don't think im building today, probably tommorow afternoon.

 

[Cowboystrekk]

What MB did you order?

 

[JustBenching]

What MB did you order?

X670 aorus elite and the HDV/M2. I'll test em both and see what's what

The 2 combatants

Spoiler

The world reknown grifter called framechasers released his 9800x 3d review yesterday. I didn't pay much attention but something picked my interest. He only scored 62 ups in factorio on a "MAX OC 9800x 3d". That felt weird to me cause im getting 63 on the 12900k. So I posted this yesterday on twitter

Spoiler

Today another user that paid more attention and looked through the factoriobox database discovered that framechasers was running 4800 jedec ram on the 9800x 3d

 

[wolf]

6000 CL30 is the least amount of headaches/Bang for bucks champion imho.

I would just go with the top tier 6400MT/s kits, whatever you can get in Australia today.

I don't think RAM tuning is that important with 9800X3D anyways, I have roughly the same FPS in PUBG with the RAM at Jedec specs (4800MT)

Thanks for the advice all, I settled on Lexar ARES 32GB Kit (2x16GB) 6000MHz, CL30-36-36-68. Seems like a kit with quite tight subtimings between everything I was looking at, and I'll be quite content with a set and forget approach knowing I'm right around the sweet spot.

 

[Cowboystrekk]

X670 aorus elite and the HDV/M2. I'll test em both and see what's what

The 2 combatants

Spoiler

View attachment 371642

The world reknown grifter called framechasers released his 9800x 3d review yesterday. I didn't pay much attention but something picked my interest. He only scored 62 ups in factorio on a "MAX OC 9800x 3d". That felt weird to me cause im getting 63 on the 12900k. So I posted this yesterday on twitter

Spoiler

View attachment 371654

Today another user that paid more attention and looked through the factoriobox database discovered that framechasers was running 4800 jedec ram on the 9800x 3d

which factorioscript did he run? Link to the twitterconversation? He defends himself by saying that Factorio will misread desync eclk\bclk and set jedec to 4800. Still does not explain low score.

 

[JustBenching]

which factorioscript did he run?

Flame sia 50k.

Bud he is a grifter.

 

[Cowboystrekk]

Flame sia 50k.

Bud he is a grifter.

Yeah, that is for sure. I commented on his youtubevideo that other 9800X3Ds got 78. It seems he deleted or hid my comment. There were a 14900KS user getting 76 aswell.

 

[JustBenching]

Yeah, that is for sure. I commented on his youtubevideo that other 9800X3Ds got 78. It seems he deleted or hid my comment. There were a 14900KS user getting 76 aswell.

He just blocked me over this

Spoiler

Now with that said, the argument that you don't need ram for the x3d chips is kinda whack. You get a 25% performance going from jedec to tuned. It's not as huge as with other platform (you can see 50+% gains) but it's still a lot.

 

[Cowboystrekk]

He just blocked me over this

Spoiler

View attachment 371734

Shame. He can't be met with criticism. You either follow his cult or get blocked. What a grifter!

 

[amit_talkin]

Flame sia 50k.

Totally off-topic but I got 45 UPS on my 5700x3d with 3600 c16 RAM ( Total 8x2GB ). Just in case anyone wants to compare

 

[JustBenching]

I have to admit I had my doubts about the b650 HDV/M2, it looks like crap from the pictures but it's actually pretty decent up close. Doesn't feel like cheap crap.

On the other hand, the whole box contains a single screw, the manual and a sata cable. But at this price it's expected

 

[Akkedie]

I would love if you could test Troy Total War, it's so CPU-bound (but also scales incredibly well multi-core with grass extreme) it's a great benchmark. You can even do it at high res, it's almost impossible to get GPU-bound with any mid-range GPU (excluding abusing stuff like 8x MSAA @ 4K). Lemme know, I'll prepare you a replay which you can run.

 

[Cowboystrekk]

I have to admit I had my doubts about the b650 HDV/M2, it looks like crap from the pictures but it's actually pretty decent up close. Doesn't feel like cheap crap.

On the other hand, the whole box contains a single screw, the manual and a sata cable. But at this price it's expected

I have the hdv and love it. Quite easy to overclock ram on due to 2dimm only. 8k ram is easier on this than even high end 4dimm-boards. It lacks some features, but has descent vrm, postcodeleds and looks quite nice. My definite recommendation for budget B650s.

 

[kDude]

I don't know if it's a big ask but do you perhaps have kingdom come installed?
I'm doing research on my next cpu upgrade and a thread came up about something called "amdip" which seems to be a bit of a controversial topic.

He runs trough a specific gate that's causing him a dip that supposedly intel doesn't suffer from.
If you have time could you perhaps go in the same spot in-game with your 9800x3d and see if you get the same dip?I was told the man in the video doesn't know how to tune amd cpu's and that its's probably flck related.
Thanks in advance.