Overclock.net banner
Cancel
Create thread New Forums

[Official] Zen 5 X3D Owner's Club (9800x3D / 9900x3D / 9950x3D)

Tags
3d vcache 9700x3d 9800x3d 9950x3d zen 5
22 reading
Jump to Latest
1,701 - 1,720 of 9,494 Posts
#1,701 ·
Pedros said:
although its good to have low timings, test these on your games... you may suffer stutters even if the mem is stable, the timings may be off ... from experience :)
Click to expand...
yeah it wasnt that good of a kit couldnt do CL26 even at 6400 , so i switched again! To another ANPEC pmic kit but now its finally able to run decent A-die timings!
also in YC so maybe worldrecord this week:

Font Rectangle Screenshot Parallel Technology
 
#1,703 ·
Pedros said:
what do you mean?
do you think there's no heat accumulation ? or where you expecting it should be at ambient temp? sorry can't understand that comment ...
Click to expand...
Nothing to understand. It was just a statement in a positive way that it's idling with core temps 6c above ambient. Just to show it's possible with mediocre AIO cooling. Someone asked about it so. I don't understand how people all idle in the 50's (unless they mean tdie and not cores).
 
  • Rep+
Reactions: fabianojosiribeiro
Save
Reply Quote Rep+ Rep+
#1,704 ·
Imprezzion said:
I don't understand how people all idle in the 50's (unless they mean tdie and not cores).
Click to expand...
Everyone looks at the CPU (Tctl/Tdie), you at the cores. :)
 
  • Rep+
Reactions: fabianojosiribeiro
Save
Reply Quote Rep+ Rep+
#1,706 ·
I changed the settings some and i got my cpu to run abit cooler and use less vcore.

PBO - MB limits
+200 mhz offset
Scalar +6
Curve optimizer -18
ECLK 101.5

Image


Image


This is after looping CB23 for like ~10mins.

Using an ASUS Prime LC 360 A-RGB aio, but i changed out the included fans for three Phanteks T30, mostly because the included Asus fans sounded bad, it was some weird scraping noise coming from them. As the Phanteks T30 have a pretty okay humming noice.
 
  • Rep+
Reactions: Arizor
Save
Reply Quote Rep+ Rep+
#1,709 ·
Hi lads. Is anyone kind enough to share their corecycler config using kagari to test cores separately? It seems I'm dumb and I can't make it work.
I'm trying to fine tune the CO a bit more instead of using just -30 in all 9800x3d cores.
 
#1,710 ·
Luggage said:
scalar x10 = you are ok with x10 shorter life span
Click to expand...
lol that's an exaggeration

whatever you do on your cpu the voltage will still be below 1,4v in any transition and that's the important thing

degradation on such a new product is known only to amd at this point but from 1,35v (o scalar) all core under load to 1,37v (10x scalar) is not that you are killing your cpu in a month

just keep attention to temperatures and the max voltage (and of course don't push too much amps through it, voltage is a factor but not the only one) and that's it in these scenarios
 
  • Rep+
Reactions: uddi86
Save
Reply Quote Rep+ Rep+
#1,711 · (Edited)
Taichi also have a new beta bios out now.

1. Support AMD Ryzen7 9800X3D CPU overclocking feature.
2. Add ‘Gaming Mode’ option.

www.asrock.com

ASRock X870E Taichi

Supports AMD Ryzen™ 9000, 8000 and 7000 Series Processors; 24+2+1 Phase Power Design, 110A SPS; 4 x DDR5 DIMMs, Supports Dual Channel, up to 8200+ (OC); 2 PCIe 5.0 x16; Graphics Output Options: 2 USB4 Type-C, 1 HDMI; Realtek ALC4082 5.1 CH HD Audio Codec,, ESS SABRE9219 DAC, WIMA Audio Caps; 1...
www.asrock.com www.asrock.com

www.asrock.com

ASRock X870E Taichi Lite

Supports AMD AM5 Socket Ryzen™ 9000, 8000 and 7000 Series Processors; 24+2+1 Phase Power Design, 110A SPS; 4 x DDR5 DIMMs, Supports Dual Channel, up to 8200+ (OC); 2 PCIe 5.0 x16; Graphics Output Options: 2 USB4 Type-C, 1 HDMI; Realtek ALC4082 5.1 CH HD Audio Codec, ESS SABRE9219 DAC, WIMA Audio...
www.asrock.com www.asrock.com

 
  • Helpful
Reactions: owikh84
Save
Reply Quote Rep+ Rep+
#1,713 ·
Taichi also have a new beta bios out now.

1. Support AMD Ryzen7 9800X3D CPU overclocking feature.
2. Add ‘Gaming Mode’ option.

www.asrock.com

ASRock X870E Taichi

Supports AMD Ryzen™ 9000, 8000 and 7000 Series Processors; 24+2+1 Phase Power Design, 110A SPS; 4 x DDR5 DIMMs, Supports Dual Channel, up to 8200+ (OC); 2 PCIe 5.0 x16; Graphics Output Options: 2 USB4 Type-C, 1 HDMI; Realtek ALC4082 5.1 CH HD Audio Codec,, ESS SABRE9219 DAC, WIMA Audio Caps; 1...
www.asrock.com www.asrock.com

www.asrock.com

ASRock X870E Taichi Lite

Supports AMD AM5 Socket Ryzen™ 9000, 8000 and 7000 Series Processors; 24+2+1 Phase Power Design, 110A SPS; 4 x DDR5 DIMMs, Supports Dual Channel, up to 8200+ (OC); 2 PCIe 5.0 x16; Graphics Output Options: 2 USB4 Type-C, 1 HDMI; Realtek ALC4082 5.1 CH HD Audio Codec, ESS SABRE9219 DAC, WIMA Audio...
www.asrock.com www.asrock.com

[/QUOTE]
Is there a powerlimit on this board?
 
Save
Reply Quote Rep+ Rep+
#1,714 ·
Very interesting and frustrating random instability whenever seemed so stable then was again on my 9800X3D.

RAM 2 X 32GB Corsair DDR5 6200 sticks with XMP enabled. Use some of Buildzoid custom timings though more conservative and do not go full blown with them and they worked great no issues on 7800X3D with 2 X 16GB RAM kit

Set SOC to 1.225V manually and VDDIo to 1.275V.

Set FCLK to 2067MHz to keep 1:3 ratio with RAM 3100 clock to 2067 FCLK which is 1:3

With no PCO nor curve optimizer enabled ran TestMem PCBDestroyer config and Extreme Anta and passed flying colors. Also ran Y Cruncher VT3 and VST and passed flying colors. Also ran Postmark MemTest and passed twice 4 runs and passed OCCT Large Dataset variable which stressed IMC and passed no issues a few times.

That was all with no PBO nor curve optimizer to ensure my RAM and FCLK were stable. FCLK ran performance scaling with AIDA64 compared to 2000 top ensure it was better and not error correcting and it was

Then it was time to dial in all core overclock

Have +200MHz clock override.

Have PBO Scaler X3

Curve optimizer all core -28



Get speeds of 5295 to 5370MHz running CInbebench R23 and stable and score of 23951 to 24000 fully stable no crashes nor WHEAs a few runs

Then I run a series of OCT Large Dataset variable, Prime95 Small and Smallest FFTs SSE and AVX2 for 1 hour and max temp 88C clocks 5GHz AVX2 and 5.25 to 5.3GHz SSE and no issues

Then I decide to run Large FFTs SSE and AVX2. On the AVX2 run within about 30 minutes the 2nd last core gets a rounding error and oh darn I was so close.

So ok I back CO to -25 on 2nd last core and weird it seemed to just stop test before it even started on that core indicating maybe still instability. Then ran it again and it went further than 30 minutes all cores no errors, but still not comfortable.

So I wondered did I need more VDDIO or is it CO. I figured to try putting VDDIo and DRAM voltage back on auto. DRAM voltage always at 1.4, but had to manually set it to play with VDDIO. If on auto, it sets VDDIo to 1.4 and cannot change it.

SO when I set VDDIO to 1.34, I start over CO -28 all cores then run OCCT Large Dataset variable CPU + RAM test and 15 minutes in a whole bunch of CPU core errors like over 1000 15 minutes in. So I put VDDIO back to 1.275 and now OCCT passed full hour run no errors. Then I wonder was 1.4V VDDIo causing instability or was it just a random fluke. So put VDDIO back to 1.4V and run OCCT Large Data set variable CPU+RAM again and OMG it passes 1 hour run no errors.

So random instability it appears probably related to curve optimizer settings and most tests do not show it unless I get unlucky/lucky referring to lucky in terms of the fact to catch them.

Anyone seen anything like this?? Was the VDDIO thing 1.4 vs 1.275 just a coincidence and had nothing to do with it since it passed next time at 1.,4 And can higher VDDIo make thing unstable or no. I know SOC at 1.4 is dangerous which is why I set it to 1.225 (ok up to 1.25 or even 1.3 but lower it can be the better I hear), but I think VDDIO is ok up to 1.5 so thought it was fine on 1.4 auto with XMP turned on.
 
#1,716 ·
thomasck said:
Thanks again @konawolv. I haven't touched any clock in all these gens apart from the 1800X so I am bit lost in what to do. Clock override would not work with the 7800x3d (at least not in my setup) so I did not bother trying. I will test out with the 9800x3d. Also, allow me to ask, shall I fiddle with curve shaper too?

I just installed the 9800x3d few hours ago, dialled previous timings (with a great addition to latency for some reason, it's "funny" going from 58ns with the 7800x3d to 66ns with the 9800x3d) and now I'm testing the stability of it. So far so good.

Timings for science. Cadbus block is untouched as there's some new parameters there. VDDG is auto.

View attachment 2683103
Click to expand...
Its my understanding that curve shaper enables fine tuning of voltages and boost based upon high, medium, and low workloads.

As an example of why this would be useful, ill go back to my time tuning my 5800x. When doing CO tuning on the 5800x, it would rarely have issues on highly taxing multi threaded workloads, but would instead crash when web browsing, or doing something light with the PC. This lead to some people gathering data about their boosting and voltage behavior based upon load. The result was that their ryzen 5000 sample would experience a larger voltage reduction on light workloads when applying negative CO to their CPU, and would see only a minor voltage reduction at medium and high workloads. This was the cause of instability of lightly threaded workloads when applying CO, and it was ultimately the problem which in turn became the genesis of the ryzen core cycler utility.

So, at the time, i took that knowledge and applied a +mv offset to my cpu to boost the voltage applied across the entire range in order to cover the lightly threaded workloads voltage drop. This stabilized my CPU and boosted my score.

Now, enter curve shaper, and this basically allows for that type of tuning across all 3 phases of different types of workloads, as well as per core.

with aaaallllll that said, i dont think its worthwhile on 9800x3d to attempt to tune this unless youre attempting eclk overclocking. Ive not messed with eclk overclocking, on my 7900x even a 101 was completely unstable.
 
  • Rep+
  • Love
Reactions: Luggage, ilmazzo and derlarbz
Save
Reply Quote Rep+ Rep+
#1,717 ·
konawolv said:
Its my understanding that curve shaper enables fine tuning of voltages and boost based upon high, medium, and low workloads.

As an example of why this would be useful, ill go back to my time tuning my 5800x. When doing CO tuning on the 5800x, it would rarely have issues on highly taxing multi threaded workloads, but would instead crash when web browsing, or doing something light with the PC. This lead to some people gathering data about their boosting and voltage behavior based upon load. The result was that their ryzen 5000 sample would experience a larger voltage reduction on light workloads when applying negative CO to their CPU, and would see only a minor voltage reduction at medium and high workloads. This was the cause of instability of lightly threaded workloads when applying CO, and it was ultimately the problem which in turn became the genesis of the ryzen core cycler utility.

So, at the time, i took that knowledge and applied a +mv offset to my cpu to boost the voltage applied across the entire range in order to cover the lightly threaded workloads voltage drop. This stabilized my CPU and boosted my score.

Now, enter curve shaper, and this basically allows for that type of tuning across all 3 phases of different types of workloads, as well as per core.

with aaaallllll that said, i dont think its worthwhile on 9800x3d to attempt to tune this unless youre attempting eclk overclocking. Ive not messed with eclk overclocking, on my 7900x even a 101 was completely unstable.
Click to expand...
Why is it not worth to tune the 9800X3D unless you have ecLK.

I mean I purchased the X670E Carbon (which does not have eCLK specifically cause 9800X3D was unlocked and thought eCLK would be useless. Apparently not the case. Should I return it (still in return window) and get a board with an eCLK as it is much easier to get better boost speeds.

Without CO nor PBO scaler best speeds I get running benchmarks are 5150MHz and often 5040 to 5090MHz, not even the all core 5.2GHz all core boost it advertises. And has a 5.25GHz single core boost at stock.

-28 CO and +200MHz made it good with all core boost in CInebench 5350MHz or around that, but not truly stable overall, and seemingly impossible to catch instability issues as they only occasionally show up as tests 80-90% time pass flying colors.

I suppose I could try manual static overclock, but man these new CPUs (not just this gen but last few gens from AMD and even last 2 gens form Intel) are not meant for that and someone posted that their CInebench results were so much worse even though it did not crash meaning something is going on.
 
#1,718 ·
konawolv said:
Its my understanding that curve shaper enables fine tuning of voltages and boost based upon high, medium, and low workloads.

As an example of why this would be useful, ill go back to my time tuning my 5800x. When doing CO tuning on the 5800x, it would rarely have issues on highly taxing multi threaded workloads, but would instead crash when web browsing, or doing something light with the PC. This lead to some people gathering data about their boosting and voltage behavior based upon load. The result was that their ryzen 5000 sample would experience a larger voltage reduction on light workloads when applying negative CO to their CPU, and would see only a minor voltage reduction at medium and high workloads. This was the cause of instability of lightly threaded workloads when applying CO, and it was ultimately the problem which in turn became the genesis of the ryzen core cycler utility.

So, at the time, i took that knowledge and applied a +mv offset to my cpu to boost the voltage applied across the entire range in order to cover the lightly threaded workloads voltage drop. This stabilized my CPU and boosted my score.

Now, enter curve shaper, and this basically allows for that type of tuning across all 3 phases of different types of workloads, as well as per core.

with aaaallllll that said, i dont think its worthwhile on 9800x3d to attempt to tune this unless youre attempting eclk overclocking. Ive not messed with eclk overclocking, on my 7900x even a 101 was completely unstable.
Click to expand...
Thanks for the write up! I won't mess with eCLK as my board (Carbon x670e) does not have this chip. I will have to read about curve shaper, as it is a new feature. I left Karhu running overnight and I'm golden with the RAM timings, 40K no errors, but I also don't know if that software stills relevant to 9000 series.
On the CPU side, well, give what you said, I think I will just let it rip. Perhaps just a small adjustment to the high frequencies using curve shaper, but before that I need to read more about it.
What sort of tools are you using to test stability of the cores? Thank you once again.

@Wolverine2349 what cinebench version are you using? I've got the same board and I can run some tests to compare results. eCLK or not, depends on the balance of cost vs performance that you are going to get by upgrading the mobo - I have no idea how much a mobo with eCLK would cost on top of the carbon.

With the carbon, all I done was CO -30 and +200 override and I consistently get 5415mhz while playing and some benchmarks. That's probably temperature related too. I've got a beefy CL so it might help sustaining higher clocks.
 
#1,719 ·
thomasck said:
Thanks for the write up! I won't mess with eCLK as my board (Carbon x670e) does not have this chip. I will have to read about curve shaper, as it is a new feature. I left Karhu running overnight and I'm golden with the RAM timings, 40K no errors, but I also don't know if that software stills relevant to 9000 series.
On the CPU side, well, give what you said, I think I will just let it rip. Perhaps just a small adjustment to the high frequencies using curve shaper, but before that I need to read more about it.
What sort of tools are you using to test stability of the cores? Thank you once again.

@Wolverine2349 what cinebench version are you using? I've got the same board and I can run some tests to compare results. eCLK or not, depends on the balance of cost vs performance that you are going to get by upgrading the mobo - I have no idea how much a mobo with eCLK would cost on top of the carbon.

With the carbon, all I done was CO -30 and +200 override and I consistently get 5415mhz while playing and some benchmarks. That's probably temperature related too. I've got a beefy CL so it might help sustaining higher clocks.
Click to expand...
CIneBecnh R23.2

I also get lower clocks OCCT Large Dataset variable on stock. Using Noctua NH-D15 G2

I am able to return the Carbon as within the return window.
 
#1,720 ·
thomasck said:
Thanks for the write up! I won't mess with eCLK as my board (Carbon x670e) does not have this chip. I will have to read about curve shaper, as it is a new feature. I left Karhu running overnight and I'm golden with the RAM timings, 40K no errors, but I also don't know if that software stills relevant to 9000 series.
On the CPU side, well, give what you said, I think I will just let it rip. Perhaps just a small adjustment to the high frequencies using curve shaper, but before that I need to read more about it.
What sort of tools are you using to test stability of the cores? Thank you once again.

@Wolverine2349 what cinebench version are you using? I've got the same board and I can run some tests to compare results. eCLK or not, depends on the balance of cost vs performance that you are going to get by upgrading the mobo - I have no idea how much a mobo with eCLK would cost on top of the carbon.

With the carbon, all I done was CO -30 and +200 override and I consistently get 5415mhz while playing and some benchmarks. That's probably temperature related too. I've got a beefy CL so it might help sustaining higher clocks.
Click to expand...
I stick with Aida64 (CPU, FPU, Cache for CPU, and RAM for memtest) and y-cruncher 2.5b. I used to use core cycler, but its evolved so far, that im not able to currently use it effectively.

Youre probably better than me with the ram tuning, i dont find much value in testing out impedances and most of my voltages are auto (outside of the obvious).

I will say though, that i recommend fclk 2200 even at 6400 mhz. This is because of the nature of having massive amounts of l3 cache. for me, 6400/2133 had lower latency, but 6400/2200 had higher throughput. When you have scenarios where ram is accessed less frequently, the throughput is more important. Latency is key when something is accessed repeatedly. With the extra l3 cache, the CPU hits the cache way more than it needs to pull from the ram.

FCLK also manages access to the PCIE lanes, which speeds up data transfer between your GPU and SSD's.
 
  • Rep+
Reactions: derlarbz
Save
Reply Quote Rep+ Rep+
1,701 - 1,720 of 9,494 Posts
About this Discussion
9.5K Replies
519 Participants
Last post:
Tatili
Overclock.net
posts
28M
members
562K
Since
2004
The premier forum for overclocking experts and enthusiasts. Discuss hardware optimization, custom builds, benchmarking, cooling solutions, and pushing the boundaries of computing performance. From beginner guides to extreme overclocking, join our technical community to master system tuning.
Full Forum Listing
Explore Our Forums
Intel CPUs Hardware News NVIDIA PC Gaming Water Cooling