• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Resolved.

Status
Not open for further replies.
My temps are inline with a buddies temps who did use paste on his, also air-cooled.

So I am sticking with the graphene pad.

Once we matched the same - offset, running the Cinebench results in temps within 3C of each other. My temps were actually cooler than his, but he has an SFF with a smaller heatsink than mine.
I can accept a small difference of a couple C if it's the difference caused by using the Graphene pad vs Paste. Besides, at -30 my temps stay in the 50's, about 57C on all core load when running Cinebench. That's plenty cool. Idle's at 44C. That's at -30, Thermalright Graphene pad (no paste), Noctua D15S Chromax Black with a 120mm noctua chromax fan added so that it has 2 fans.

To get the performance numbers up I need to fiddle with memory timings or overclocking. Haven't tried to do any of that yet.
 
Last edited:
Was thinking of swapping out the Wifi card on the TUF 670E but seems ok with the latest drivers , the connection seems to be more stable than the AX200 , always locked iat 1201/1201 for RX/TX vs the AX200 which would do 2402/2402 best case but often fall back to less than 1000 mbps tx/rx (my router is not 6E, regular wifi 6 RT AX88U), and actual peak throughput is similar if a tad lower at around 700-830 mbps from my router (90-105 MB/sec for LAN transfers) which is good enough for me. It wasn't a PnP install like the Intel drivers though so make sure to do a bios flashback to support the 9800 and grab the latest drivers on a usb stick if someone is building with a similar board.
 
Last edited:
GoodBoy said:
patched to 24H2, changed it to -20 allcore offset. Temp is a little higher 77.5C, but score basically the same:. the GPU score is higher by less than 3%
View attachment 698780

I think it's my ram speed/timings I bought the 8000 set, and after I built it I read that the higher speed runs at half the multiplier or something to that effect. idc if it doesn't hurt gaming perf. but if lowering the ram speed improves gaming perf them I'd do it. Need to find that speed to performance info again...
Click to expand...
ddr5 6000 is the sweet spot. it's 1:1 and you can run pretty aggressive timings. above this is a waste unless you go really high to offset the non 1:1 performance penalty
 
GoodBoy said:
Anyone air-cooling a 9800x3d, what max temps are you seeing?

Mine hit 96C playing Marvel Rivals.
Click to expand...
Highest I've seen is 79°C (steady state during CPU stress test, all settings stock). I've got very good airflow though... MX4 thermal compound + Peerless Assassin 120SE. Also my room is fucking freezing.
 
Last edited:
Blackstone said:
So I’m at 1383 in Cinebench multicore, every core pegged at 5,418MHz (5.4 GHz), max temp 84, max voltage 1.256. DDR5 6000 CL28.

I’m using Scalar 10X and a custom optimizer curve. -20 optimizer but i think i add another 10 to only mid and higher frequencies.

Am I on the right track and what is the best way to stress test it?
Click to expand...
shouldnt multi be like 23k?
 
Blackstone said:
This is my first time using Cinebench, and I downloaded the 24 version which, based on my brief research just now, scores differently and lower.
Click to expand...
ah. idk that one, yet. whats r23 say?
google says 1300+ is normal. so youre probably gtg.
 
He's using Cinebench 2024. Totally different point scale.


Blackstone, I'm getting 1444 so pretty close to you. Try -30 curve optimizer, or even more, if you have not yet. Don't assume it will be unstable until you prove it is unstable. The closer to -40 you can get and still be 100% stable the better temps will be. My chip appears stable in everything I've thrown at it so far at -37 or -38 which may be unusually good but I've seen lots of posts of people in the -30 range.

Make sure your FCLK is 2000. Make sure you have memclock=uclock.

Using a Noctua D15S with 2 fans, My CPU temp stabilizes at about 71C in Cinebench 2024 multicore in a room that's 68F. So probably about 75C in a 72F room. You're a little hot, which could be either the cooler you are using, or the motherboard's default voltage settings.

Make sure you are on the latest BIOS.

Check the MB default VSOC voltage in HWINFO. When the Cinebench run is going VSOC is 1.212. If your board is pushing it to 1.25 or higher under load, try taking 50mv out. My board was overdoing it.

Also disable the iGPU in bios. No reason for it to be on.
 
Blackstone said:
If I drop my DDR5 6000 C30 down to C28, do I need to adjust other memory specific settings to stabilize it?
Click to expand...
if you drop cas by 2pts, drop the other 4 main settings by 2 and leave the sub timings on auto.
 
To answer your RAM questions, it depends. Best just to bench it both ways. Lower timings is sometimes slower because of errors.

Your performance numbers may be just about perfect. I think I'd just work on temps if I were you.

My CPU is stable at 2066 FCLK and I run my DDR5 6400 at 6200 to keep the multiple even and latency low.

Like I said, I think I'd focus on temps I think your CPU score is just about maximized short of doing advanced things like async base clock and going past 5.4Ghz. Not worth it IMHO.

Oh, and I'm now with everyone else in this thread: put that scalar back to 1x and stop worrying about it. it won't make a measurable difference to your OC on the X3D you don't want to risk shortening the life of the CPU. Mine is now back at 1x and I can't tell any difference since the CPU bounces off the 5.4Ghz limit anyway.
 
Yes scalar does improve stability. The scalar settiing is probably counteracting the curve optimizer in certain ways and masking the low frequency issue. (Please read back earlier in the thread for some deeper dive on scalar, others can do a better job than me of describing what it's doing under the hood, what FIT is, etc.).

What you describe is exactly what will happen on many/most CPUs with very large negative offsets in Curve Optimizer. The system can easily end up unstable at light load because voltage is too low at light load.

I've gotten lucky with this chip and have not had to use shaper, but you are also exactly correct: Shaper is the proper and best tool to fix that problem, NOT scalar.

Scalar essentially is to enhance the ability to overclock in certain situations but at the direct risk of the CPU not lasting as long.
 
LittleBuddy said:
DPC latency is higher on the 9800X3D than the 5800X3D and 7800X3D currently, this will likely be fixed with updates though.

5800X3D average DPC latency: 1.0µs
7800X3D average DPC latency: 0.6µs
9800X3D average DPC latency: 1.3µs
Click to expand...
FYI -- AGESA 1.2.0.2b improved the 9800X3D DPC latency. It is now in-line with the 7800X3D at 0.6µs.
 
Anyone tried flashing 3067 for TUF 670E? Keep getting an invalid bios file error despite downloading and extracting a few times. I've done this process a 100 times so I'm pretty sure the file asus has up here is messed up:

1734990909165.png
 
Nvm, user error, there is apparently a non Wifi version of the board that I inadvertently downloaded the bios for. Good thing it didn't take, seems like this new AGESA is more finicky about RAM based on reports all over. Even on my setup EXPO I fails to post, EXPO II seems ok so far...getting too old for this shit. Why hasn't any Mobo mfg. implemented a way to carry over bios settings from one version to the next, manually setting 15 parameters and wasting time to see what takes is no longer my cup of tea...
 
undertaker2k8 said:
Why hasn't any Mobo mfg. implemented a way to carry over bios settings from one version to the next, manually setting 15 parameters and wasting time to see what takes is no longer my cup of tea...
Click to expand...
I'm with you there. I like to keep up with my bios as well, but having to change the settings every time is something I dread.
 
Put the darn m2 single sided bumpers on my 4.0 SSDs (990 4 tb, 980 2 tb and 770 1 tb), even the slight bend was bugging me. The SN700 and Hynix 3.0 ones don't need heatsinks or these pads, since there is no HS to press down on them and the ASUS Q-latch is actually very, minimal force m2 retention mechanism.
 
  • Like
Reactions: Niner
like this
Last edited:
Blackstone said:
If I drop my DDR5 6000 C30 down to C28, do I need to adjust other memory specific settings to stabilize it?
Click to expand...
Depends on your ram kit. I would set Expo settings and then manually change your memory speed. This guide was super helpful. am5 and DDR5 have a few more quirks then AM4/DDR4


View: https://youtu.be/dlYxmRcdLVw?si=Wc17-8t9PUanMohW

Note its for Hynix ram but same applies minus timings meant for Hynix. Betting they have a guide for your ram. Pretty much all memory kits will have Samsung chips, Hynix, or Micron dies.

This will also helpful for PB02 tuning

View: https://youtu.be/FaOYYHNGlLs?si=x_GYZAhKdcAVgbot
 
Advil said:
... I'm getting 1444 .... Try -30 curve optimizer, or even more, if you have not yet. Don't assume it will be unstable until you prove it is unstable. The closer to -40 you can get and still be 100% stable the better temps will be. My chip appears stable in everything I've thrown at it so far at -37 or -38 which may be unusually good but I've seen lots of posts of people in the -30 range.

Make sure your FCLK is 2000. Make sure you have memclock=uclock.

Using a Noctua D15S with 2 fans, My CPU temp stabilizes at about 71C in Cinebench 2024 multicore in a room that's 68F. So probably about 75C in a 72F room. You're a little hot, which could be either the cooler you are using, or the motherboard's default voltage settings.

Make sure you are on the latest BIOS.

Check the MB default VSOC voltage in HWINFO. When the Cinebench run is going VSOC is 1.212. If your board is pushing it to 1.25 or higher under load, try taking 50mv out. My board was overdoing it.

Also disable the iGPU in bios. No reason for it to be on.
Click to expand...
I'm on the latest bios from Dec 4 for the Crosshair x870e Hero.

I found the Fabric clock and set it to 2000. I am unsure what you mean by memclock=uclock.

I have uclock ON. The memclock is 1800, fabric clock is 2000. Did you mean set memclock to equal fclock?

My cinebench score didn't budge moving fclock from 1800 (where it defaulted) to 2000.

I bought 8000 DDR:

G.SKILL Trident Z5 RGB Series (Intel XMP 3.0) DDR5 RAM 48GB (2x24GB) 8000MT/s CL40-48-48-128 1.35V Desktop Computer Memory UDIMM - Matte Black (F5-8000J4048F24GX2-TZ5RK)​

It was the only 8000Mhz ram on the mobo compatibility list. Didn't know that too fast hurts performance. But I am thinking I can slow it down and tighten timings. I want to set it 1:1 as a techtuber mentioned for best performance, but he didn't say how to do that. If it's just memclock=fclock let me know I can set that. If i change memclock, do I need to tweak any other memory timings? Right now all of that is on auto.

Trying this now, bumped memclock to 2000:
1735337761700.png

Advil said:
My CPU is stable at 2066 FCLK and I run my DDR5 6400 at 6200 to keep the multiple even and latency low.

...

Oh, and I'm now with everyone else in this thread: put that scalar back to 1x and stop worrying about it. it won't make a measurable difference to your OC on the X3D you don't want to risk shortening the life of the CPU.
Click to expand...
This is what I need to know how to set. Run the DDR down at the even multiplier.

Can all of that be controlled with Ryzen Master?

Thanks, your posts have been helpful.
 
Last edited:
Typically these are set in bios. I don't ever use Ryzen Master for this and don't know what some of those values mean shown like that. I don't know if Ryzen Master's RAM speed is shown as actual or double data rate. I don't know what on/off means in relation to uclock mode.

I'm also no expert on the best settings for DDR5 8000 you are going to have to do some benchmarking.

What I'm running in bios is DDR5 at 6200 and Fclock at 2066 which is an even multiple of 3 of the effective RAM speed and with the BIOS setting "memclock=uclock" which means the memory controller is running the same speed as the RAM (the actual ram speed of 3100 not the effective speed which is 3100*2). I get both decent performance and good latency like this.

You have DDR5 8000 so the RAM's actual clock speed is 4000. There's no way the memory controller is going to run at 4000Mhz so you'll have to set the other option which is memclock=uclock/2 which makes the memory controller run at half the RAM speed in your case 2000Mhz. You can't run the memory clock speed the same as the memory controller speed, but you are still running an even multiple which in theory sets you up for a good result. The idea is for the outrageously fast speed of the RAM to make up for the fact you can't run 1:1.

Benchmarks I've seen don't show any major differences but most of the ones I've seen give a tiny edge to the DDR5 8000.

A good starting point for you that MIGHT "just work" is DDR5 8000, Fclock 2000, Memlock=uclock/2.
 
Advil said:
Typically these are set in bios. I don't ever use Ryzen Master for this and don't know what some of those values mean shown like that. I don't know if Ryzen Master's RAM speed is shown as actual or double data rate. I don't know what on/off means in relation to uclock mode.

I'm also no expert on the best settings for DDR5 8000 you are going to have to do some benchmarking.

What I'm running in bios is DDR5 at 6200 and Fclock at 2066 which is an even multiple of 3 of the effective RAM speed and with the BIOS setting "memclock=uclock" which means the memory controller is running the same speed as the RAM (the actual ram speed of 3100 not the effective speed which is 3100*2). I get both decent performance and good latency like this.

You have DDR5 8000 so the RAM's actual clock speed is 4000. There's no way the memory controller is going to run at 4000Mhz so you'll have to set the other option which is memclock=uclock/2 which makes the memory controller run at half the RAM speed in your case 2000Mhz. You can't run the memory clock speed the same as the memory controller speed, but you are still running an even multiple which in theory sets you up for a good result. The idea is for the outrageously fast speed of the RAM to make up for the fact you can't run 1:1.

Benchmarks I've seen don't show any major differences but most of the ones I've seen give a tiny edge to the DDR5 8000.

A good starting point for you that MIGHT "just work" is DDR5 8000, Fclock 2000, Memlock=uclock/2.
Click to expand...
Ok thanks. Hovered over the uclock button and it gives a tooltip:

1735359499750.png


So I just need to turn this off to get uclock=memclock/2, then try the 2000 memclock again.

I tried memclock 2000 earlier, and system wouldn't boot, and had to clear cmos to get it booted.

Edit: well that still didn't work. If I move memclock to 2000 I get a 15 post code and it never boots. 15 is a memory problem code.

Would it still be worth trying the uclock=memclock/2 ?
 
Last edited:
Correct. Off, which is what it is now, is the setting you need. uclock=memclock/2. But this should also be being set in bios and not Ryzen master.

Something is very wrong with your RAM speed setting.

From the quick reading I did online Ryzen Master memory speed is actual clock. Which means 1800 is (3600MT/s). Your RAM's EXPO profile should be for 4000 (8000MT/s).

You have got to get into BIOS and see what your RAM is actually set to at post, and see if your EXPO profile is enabled for DDR5 8000. If it is, Ryzen Master should be showing your RAM speed as 4000.
 
Well Ryzen Master is a POS.

It's not reading what the actual speeds are, it just knows what settings you set in it and apply. I had it set to 2000 memclock 2000 fclock, wasn't stable. Went into Bios and set to auto. Boot up, Ryzen Master still shows 2000 2000.

So yeah, doing it all in the Bios now.

Tried the extreme tweaker drop down settings DOCP I, DOCP II, DOCP Tweaked. None of them will post. Had to put it back to auto, manually set PBO curve to - 30, to get my shit running again. I tried it with all PBO modifications off too, didn't help. Guessing the AGESA or this mobo still not going to play nice with DDR5-8000.

It's 2:1 mem timings means 1800 = 3600 then with DDR it's x2 again, so its running at 7200. Edit: Wait I think you might be right. But these speeds are what it's doing when I set it to Auto. Manually setting DDR5-8000 = nope. I even tried this while also setting the option below it to Asynchronous, thinking it's was possibly the uclock memclock ratio thing, didn't help. It's shouldn't Auto ddr5-8000 off of the QVL to 3600 speeds... so again, back to thinking that the platform/AGESA is still half baked... Probably going to need a bios update to get it working at the rated speed. In the meantime I will probably try some other options, maybe 6200. I have to say I dislike how the Memory settings work on this system as compared to XMP setting on my old Intel system. That just works. This just... doesn't.

I watched this whole thing, it's pretty informative even tho it takes him awhile to explain it all. He doesn't start talking about ddr5-8000 until about 39 minutes, but the previous information is relevant.


View: https://www.youtube.com/watch?v=Xcn_nvWGj7U
 
Last edited:
I am within the return window still on this ddr5-8000 Ram since it was purchased leading up to Christmas, so I just ordered a similar set specifically with EXPO and 6000 speeds off of the QVL, will see how it performs and likely return the ddr5-8000.
Ordered this G.Skill set: F5-6000J3238G32GX2-TZ5NR
 
GoodBoy said:
Well Ryzen Master is a POS.

It's not reading what the actual speeds are, it just knows what settings you set in it and apply. I had it set to 2000 memclock 2000 fclock, wasn't stable. Went into Bios and set to auto. Boot up, Ryzen Master still shows 2000 2000.

So yeah, doing it all in the Bios now.

Tried the extreme tweaker drop down settings DOCP I, DOCP II, DOCP Tweaked. None of them will post. Had to put it back to auto, manually set PBO curve to - 30, to get my shit running again. I tried it with all PBO modifications off too, didn't help. Guessing the AGESA or this mobo still not going to play nice with DDR5-8000.

It's 2:1 mem timings means 1800 = 3600 then with DDR it's x2 again, so its running at 7200. Edit: Wait I think you might be right. But these speeds are what it's doing when I set it to Auto. Manually setting DDR5-8000 = nope. I even tried this while also setting the option below it to Asynchronous, thinking it's was possibly the uclock memclock ratio thing, didn't help. It's shouldn't Auto ddr5-8000 off of the QVL to 3600 speeds... so again, back to thinking that the platform/AGESA is still half baked... Probably going to need a bios update to get it working at the rated speed. In the meantime I will probably try some other options, maybe 6200. I have to say I dislike how the Memory settings work on this system as compared to XMP setting on my old Intel system. That just works. This just... doesn't.

I watched this whole thing, it's pretty informative even tho it takes him awhile to explain it all. He doesn't start talking about ddr5-8000 until about 39 minutes, but the previous information is relevant.


View: https://www.youtube.com/watch?v=Xcn_nvWGj7U
Click to expand...

To be fair, you went with some of the highest clocked memory that AM5 (mostly?) works with right out of the gate. FWIW, the fact that memory training brought the setting back down to something stable when you set 8000 is actually pretty impressive. Silicon lottery is still very much a thing for memory overclocks.

6000 C30 (or C28) is 97 to 99% of the performance in games most of the time, had you gotten 8000 working.
 
You have made the right choice to send back the 8000. What no one mentions on the packaging is that ultra high speed RAM is almost never plug and play. You're going to have a way better time and a more enjoyable and usable machine with the DDR5 6000.
 
arestavo said:
To be fair, you went with some of the highest clocked memory that AM5 (mostly?) works with right out of the gate. FWIW, the fact that memory training brought the setting back down to something stable when you set 8000 is actually pretty impressive. Silicon lottery is still very much a thing for memory overclocks.

6000 C30 (or C28) is 97 to 99% of the performance in games most of the time, had you gotten 8000 working.
Click to expand...
Well, if I manually try setting the ram speed in bios to DDR5-8000, the system doesn't post. Post code 15.
I have to clear the cmos then go in and set it to auto, then it boots. I cleared cmos yesterday like 10 or more times... I'm no genius when it comes to the 50 some odd settings for the ram. I made a bad assumption that ram on the QVL would work at the rated speed. If it had worked, great. I'd be very happy with it. I'm used to Intel systems where the XMP setting just works. At least in multiple systems it has worked every time at the ram's rated speeds. It was running at DDR5-3600 speed... when I paid for 8000 or something close, I mean shit. Thirty fucking six hundred? The new ram will be nearly twice as fast.

I do think that possibly it could start working with an AGESA update. Or, it might have been on the QVL by mistake. The ddr5-8000 ram I ordered right off of the QVL is XMP, not EXPO. So there is a good chance that has something to do with it. And no EXPO 8000 ram on the QVL? Probably because the shit doesn't work.. lol.

It's $500 worth of Ram I will return, the new stuff is less than half of that and I bet it is faster.

I just want the ram to run at the rated speeds...
 
Last edited:
GoodBoy said:
Well, if I manually try setting the ram speed in bios to DDR5-8000, the system doesn't post. Post code 15.
I have to clear the cmos then go in and set it to auto, then it boots. I cleared cmos yesterday like 10 or more times... I'm no genius when it comes to the 50 some odd settings for the ram. I made a bad assumption that ram on the QVL would work at the rated speed. If it had worked, great. I'd be very happy with it. I'm used to Intel systems where the XMP setting just works. At least in multiple systems it has worked every time at the ram's rated speeds. It was running at DDR5-3600 speed... when I paid for 8000 or something close, I mean shit. Thirty fucking six hundred? The new ram will be nearly twice as fast.

I do think that possibly it could start working with an AGESA update. Or, it might have been on the QVL by mistake. The ddr5-8000 ram I ordered right off of the QVL is XMP, not EXPO. So there is a good chance that has something to do with it. And no EXPO 8000 ram on the QVL? Probably because the shit doesn't work.. lol.

It's $500 worth of Ram I will return, the new stuff is less than half of that and I bet it is faster.

I just want the ram to run at the rated speeds...
Click to expand...
Ahh, I misread then.

Yes, 6000 MT/s RAM is for you. And FWIW, my 6400 G.Skill set is XMP and works fine with the XMP setting (but I now run slightly tuned 6000 buildzoid timings) - yet it wasn't 8000 which is quite a lot more strain on the IMC.
 
Status
Not open for further replies.
Back
Top