Overclock.net banner

2901 - 2920 of 3006 Posts

·
Registered
Joined
·
356 Posts
Brilliant stuff, just the kind of examples I have been looking for, great comparisons.

I've booted "X299-B" up, this has not had my 30 minutes of google / realbench / cinebench yet. (*Edit, it is NOT stable, google stress test proved it!)

7800X @ CPU x45 Cache x32 DDR-3600


Thaiphoon data on the RAM kit

ThaiphoonBurner166_CMK16GX4M4B3600C18.png


AIDA: 7800X

7800X_cachemem_CPU_x45_CACHE_x32_3600_17-18-18-36-CR1-UNTESTED.png

RAM KIT: Corsair Vengeance 3600 (CMK16GX4M4B3600C18 v4.24) (Link)

Read: 77.863
Write: 94.648
Copy: 72.587
Latency: 64.8


In comparison, the 10900X (running at 3733) is doing

10900X_cachemem_CPU_x48_CACHE_x31_3600_18-22-22-42-CR1-UNTESTED.png

RAM Kit: CMW32GX4M4D3600C18 4.32 / C-Die (link)

Read: 106.05
Write: 82.892
Copy: 87.027
Latency: 61.3


The difference in write performance is frustrating!
 

·
Old to Overclock.net
Joined
·
2,170 Posts
Brilliant stuff, just the kind of examples I have been looking for, great comparisons.

I've booted "X299-B" up, this has not had my 30 minutes of google / realbench / cinebench yet. (*Edit, it is NOT stable, google stress test proved it!)

7800X @ CPU x45 Cache x32 DDR-3600


Thaiphoon data on the RAM kit

View attachment 2488851


AIDA: 7800X

View attachment 2488853

RAM KIT: Corsair Vengeance 3600 (CMK16GX4M4B3600C18 v4.24) (Link)

Read: 77.863
Write: 94.648
Copy: 72.587
Latency: 64.8


In comparison, the 10900X (running at 3733) is doing

View attachment 2488855

RAM Kit: CMW32GX4M4D3600C18 4.32 / C-Die (link)

Read: 106.05
Write: 82.892
Copy: 87.027
Latency: 61.3


The difference in write performance is frustrating!
The 7000 series need more uncore voltage to maintain higher ram clock speeds. What are you running for uncore on your 7800X?

My old 7940X was a great overclocker, could do 4.8 with only 1.22 vcore. But had a horrible memory controller.

Good news is you have samsung e die. A good bin will do 16-17-17-36 320 TRFC at 1.35-1.4volts

Good thing is e die scales extremely well with voltage and is even more tolerant than b die. So to make 16-17-17 work just add more voltage. But for now I would start with 3600 (or 3733 or 3800, just add more voltage) 16-18-18-38 TRFC 360 at 1.45v. if that works start going down bins. Once you feel comfortable start working on TREFI (should be able to max out). Once tweaked properly you have high bandwidth and latency around 52ns depending on where you land.

And yes it is weird you have such high write speeds and the rest are lower. Could be a RAM timing that is left on auto in the BIOS.
 

·
Registered
Joined
·
356 Posts
I've just started tuning the 7800 based on what i learnt from @Jpmboy with the 10900.

My earlier voltages are a mile out. Currently running the following and it has passed 30m realbench.

Something like 1.115 CPU, 1.856 VCCIN, 0.88 SA. Uncore offset Auto.

Cache is on auto ratio, i'll tune that next.

2488897
 

·
Registered
Joined
·
787 Posts
Then the only option left is by checking the version number mentioned on the sticks which will indicate the die used. See below for nomenclature :

View attachment 2488730
I have some Corsair crap that is ver 4.31 and is claimed to be B-Die but is 3600C18 and let me tell you,It is some of the worst B-Die I have ever used.
 

·
Registered
Joined
·
77 Posts
I have some Corsair crap that is ver 4.31 and is claimed to be B-Die but is 3600C18 and let me tell you,It is some of the worst B-Die I have ever used.
Yeah they usually reserve the best bins for Dominator series so the Vengeance LPX and RGB series get average bins.

Also best bin B-Die do high frequency and low timings at rated speed/timings. For example a stick rated 3600C18 is bad bin , 3600C16 is a average bin and 3600C14 is the best bin. So need to pick sticks accordingly while buying.
 
  • Rep+
Reactions: sultanofswing

·
Registered
Joined
·
134 Posts
Yeah they usually reserve the best bins for Dominator series so the Vengeance LPX and RGB series get average bins.

Also best bin B-Die do high frequency and low timings at rated speed/timings. For example a stick rated 3600C18 is bad bin , 3600C16 is a average bin and 3600C14 is the best bin. So need to pick sticks accordingly while buying.
Hi there.
I might have something for you to try running 4000c15 T1.

tRCD is not scaling with voltage so have you tried to increase it while keeping CAS down?

Mine stable for 4000c15 T1 only if tRCD set to 19 while I can also be stable with tRCD set 18 albeit only with CAS 16. So loosening tRCD will allow you to bring down CAS is my experience.

Note that total of CAS + RCD will be the same so it may not increase bandwidth but I think will reduce latency.
Last night same setting pass 20000% karhu. Now tightening other timings.

The kit I’m using is 19-19-19-39 xmp so it make sense that it needs 19 tRCD. I’m not sure with other kit with better bin but try not to touch tRCD or even increase it first.

+0.175 uncore (may need more on yours depends on the IMC)
1.53 vDRAM
Temp doesn’t matter I had it up to 62 degC on purpose and it won’t budge.
I will double the tRAS when I’m done tuning so it should be 36.

2489129


By the way stay safe there.
 

·
Registered
Joined
·
4,319 Posts
Yeah my daughter's 7800X is delidded, purchased it delidded and lapped from another member. Running it at 4.6 on a 4 heatpipe air cooler, don't think that would be possible if it weren't delidded.

Your Corsair 4x4GB kit might be 4 gigabit samsung e die, 18-19-19 seems like it would. Maybe with 1.4-1.45 volts it could do 16-18-18 with a sub 360 TRFC. If it can maintain a frequency and low TRFC it will definitely be e-die.

Below are the screenshots I was talking about in the earlier thread.

This is my main system, 9980XE @ 4.8 / 3.2 cache / DDR4 4000 on a Asrock Taichi CLX.



Daughter's rig, 7800X @ 4.6 / 3.2 Cache / DDR4 3200 on a Asus Deluxe II



Here is the other member's ram settings. Running 3733 @ 16-18-18 and a TRFC of 300. From what I remember only samsung e-die can achieve that low of TRFC with that high a frequency.
about the trfc.. problem is heat and voltage... The lower it is?? the hotter it gets!!!! and the lower it is!!!! the more voltage it needs!!!! and the hotter it gets..!!!!
here are my results 7980xe @ 33 mesh

This is patriot viper 4400 b dies
2489178

This are Team Xtreme 3733 64GB kit b dies too "yeah i went way overkill on that run" or didnt i?
2489179

GB5

just bcuz some ppl around this forums :rollseyes:
certain old geezer probably would say is fake too hahahaha

@Jpmboy still havent try that 7980xe on the dark with the xoc bios to see how it reacts on the membw and latency and without specter and such mitigations enable out of the bios..
 

·
Registered
Joined
·
134 Posts
about the trfc.. problem is heat and voltage... The lower it is?? the hotter it gets!!!! and the lower it is!!!! the more voltage it needs!!!! and the hotter it gets..!!!!
here are my results 7980xe @ 33 mesh

This is patriot viper 4400 b dies
This are Team Xtreme 3733 64GB kit b dies too "yeah i went way overkill on that run" or didnt i?

GB5

just bcuz some ppl around this forums :rollseyes:
certain old geezer probably would say is fake too hahahaha

@Jpmboy still havent try that 7980xe on the dark with the xoc bios to see how it reacts on the membw and latency and without specter and such mitigations enable out of the bios..
Wow, I’ve yet to see 10980XE which can run above 4000 on memory at least on 3 samples I personally know even with kit above 4000. Seems 7980XE has much better IMC in that regard.

Could be because your board has 1 dpc?

Same for the cache, all 3 maxed out at 32. It can run 33 if I leave the core at stock but with core OC, even using Intel VID to make sure it has enough vcore to sustain stability, 33 was never stable. One sample can be stable however needs too high vcache and vccio to be used for daily.
 

·
Registered
Joined
·
4,319 Posts
Wow, I’ve yet to see 10980XE which can run above 4000 on memory at least on 3 samples I personally know even with kit above 4000. Seems 7980XE has much better IMC in that regard.

Could be because your board has 1 dpc?

Same for the cache, all 3 maxed out at 32. It can run 33 if I leave the core at stock but with core OC, even using Intel VID to make sure it has enough vcore to sustain stability, 33 was never stable. One sample can be stable however needs too high vcache and vccio to be used for daily.
Maybe, my first Micro2 behave the same as my second one they are exactly the same board in overclocking capabilities wasnt a dud but i spilled coffee by accident on it lol lucky i managed to get a second board and for what i paid for it ?? this board is awesome.. they were both brand new...Then i bought a dark from a user from here.. i have it in the back with a 7940x for now bcuz i dont want to dismantle my main yet its to complicate to dismantle lol and also im mining on it lol..

So im keeping myself busy with the 5900x & 6800xt in the open bench in the back. Having all this pcs around is like you get to a point of "wathever"..
that 7980xe is using adaptive on a negative offset.. 33x cache only needs 1.125mV but i have it at 1.135mV lil bump over.. now i want to try the 6800xt on the 7980xe to see the difference in fps specially in the dark where it have a xoc bios with no mitigations on it... to see cpu bottlenecks scenarios bcuz of 4x16GB DR @ 3800 low latency vs my 5900x @ 2x16GB DR 3800 low latency but just the thought of dismantling that pc is a pita..
 

·
OC...D
Joined
·
28,206 Posts
@Jpmboy still havent try that 7980xe on the dark with the xoc bios to see how it reacts on the membw and latency and without specter and such mitigations enable out of the bios..
been too busy with AMD eh? 😉
 
  • Rep+
Reactions: zGunBLADEz

·
Registered
Joined
·
3,101 Posts
been too busy with AMD eh? 😉
Busy is the right word with Amd 🤭

New agesa and bios "every" week so no time for actual using it. There is never a final overclock 😅. That's why we still use x299 ;)
 

·
Old and Decrepit
Joined
·
2,150 Posts
Yeah they usually reserve the best bins for Dominator series so the Vengeance LPX and RGB series get average bins.

Also best bin B-Die do high frequency and low timings at rated speed/timings. For example a stick rated 3600C18 is bad bin , 3600C16 is a average bin and 3600C14 is the best bin. So need to pick sticks accordingly while buying.
The kit I have is Corsair Dominator platinum 3600 C18 8x8GB RGB kit. It’s not the kit holding it back as I was running it with ease at 3800 CL16 on a 9940X and another Dominator platinum 3800 CL19 at 3800 CL16 on the same 9940X. Granted I’ve not dived into serious tweaking of tertiary timings but I can’t even run the 3800 kit at XMP on my sample of the 10980XE once it’s overclocked even with mesh at auto. I get the same out of either one of them, 3600 CL16 with VRam at 1.5V. 2 sticks closer to the DIMM.2 run 3-4C hotter than the rest. Sometimes it will lock up after it heats up and letting the latency out a point to CL17 amd dropping the VRam back to 1.4 works fine. The 3600 kit will run XMP at standard 1.35Vram. I’ve been on the fence about a 3600 CL14 or the tried and true 3200 CL14. Problem is 8 stick kits of 8x8GB are hard to come by. Of course I could buy 4X16 GB but my OCD of empty slots would drive me nuts worse than a minimal sacrifice in ram performance. I use the 64GB (never maxed out but up to 75%) but have no need for the 128 much less 256GB kits.

My take is the IMC on the 109XX is not as good as the 99XX or 79XX. The core and mesh clocks better for me though even though it has to be a static Vcore as the VID table is horrendous. I almost crapped myself the first time it lit it up. Everything in auto and jump bumped multi to 46 and the Vcore hit 1.414. I was waiting for the smoke to roll out.
 

·
OC...D
Joined
·
28,206 Posts
Busy is the right word with Amd 🤭

New agesa and bios "every" week so no time for actual using it. There is never a final overclock 😅. That's why we still use x299 ;)
I know, right? It's a never-ending quest for "ah, this is the sweetspot". (y)
 

·
Old and Decrepit
Joined
·
2,150 Posts
I know, right? It's a never-ending quest for "ah, this is the sweetspot". (y)
Biggest collection of tech geeks on the planet. 99.9% of the population are happy if it boots when they turn the on/off switch to the on position and don’t have a clue what a BIOS is. My mother is getting up in the years now but when she was about as old if a goat that I am now I tried to teach her the on/off principle and she got that far and now what? She falls behind keeping up with the family so I sent her an ASUS tablet with one Icon on it for social media. I’m like push the button then poke the only thing on the screen. I already set you up a user account, auto login and linked you to everyone in the family and your old friends that have been asking about you so you can see what’s going on. When it dies plug it in, give it a few hours then turn it on when you want to look again. That’s about as far as I got with her, 90 in August.
 

·
Registered
Joined
·
4,319 Posts
been too busy with AMD eh? 😉
not really is was quite quick to get a working "pbo/auto lol" setting on it. ram i just used the 32gb kit i tweaked using the 8700k and did an initial hci run and compare with other users which mind you i was getting better bw vs users pushing tighter timings and just call it a day after that that was it no more ram tweaks its not even worth it if i cant go above 1900fclk 1:1 theres no point or lower any timings system its literally capped @ 3800 16s/1900 fclk lowering the 16s will net you nothing.

But i have kernel panics with no wheas on static overclocks bcuz of the l3 cache bw speed. she will pass rog bench/ blender/x264 v2 stress test. but crash on idle like aint no joke, it will crash reboot as soon windows starts crash again it can happen 3-10-20 times in a row...it doesnt happen on pbo as the l3 speed its somehow capped it doesnt happen either on static overclocks on old bios where l3 cache is kind of capped around the 300s... it looks like amd is aware of this so they capping the bw speed on the l3 reads.. I find out the error bombarding the l3 cache reads on aida64 and crash it at will sometimes i do a few clicks and trigger it but when i disable the 2 cores i run an autoclick on aida mem bench on l3 reads to test stability.. I digged around a lil bit more and i find out a bunch of users with the same problem specially 5900x and 5950x where you have higher l3 cache numbers but they dont know is bcuz of the l3 cache..

funny tho it happen in reads around 1100-1200gb but if i disable 2 cores and use static of course this will lower the read bw on l3 cache and it will be stable no problems using static.. pbo is 15-20c hotter than static thats why i want static over pbo.. Even so im using curve -30 plus adaptive negative offset in top of the -30curve and still heats more than static... and its not like its alot of wattage im talking about 170-200watts but the density is real it heats more than my 7980xe at 350-400watts i can cool my7980xe @ 400w better than this 5900x at 200w lol.. So i paid amd to self "boost/overclock" automatic system if you can call it like that :/.

Shes fast but the tweaking factor of it?? Thing overvolts alot to benefit one core bro ONE CORE so you suffering bcuz of that. 20c of heat over for one core aint worth it in my book.. i can do 48x/47x right??? 20c cooler than using just pbo which maxes around 46-47x depending of ambients mostly 46.50x.. 47x is best case scenario where ambients are very cold lol... also 46x pbo is less performance than static 46x if you catch my drift.. this amd hardware love cold so the cold it is the less tdp they use.. they perf better the colder they are and they will adjust their tdp lower just bcuz of it.. System is already optimized for pbo/stable thats it no more tweaks need it but it runs hotter. My fight is why is crashing at random on static bcuz of that l3 cache thing..

pbo 49-51.50x one core 20c hotter 46x all core load in tests like blender for example.. cb15 single score is around 271 right? but thats just one core tho lol.. 46x would be the real single core perf instead.. anyway..

versus

48x ccx1 / 47x ccx2 better performance than all core pbo @ 46x, -20c less heat mind you lol.. vs a pbo using negative curve -30 & adaptive negative offset in top of it lol.. which it should be colder than automatic pbo lol.. so 48x 6 cores cb15 single score @ 48x is 264 which setup you prefer??
i take the static and colder and not so random performance myself.... im only loosing 3% and thats only in one core scenario benchmark tests like in real usage thats not even perceivable. you will only notice it in tests like cb or geekbench where they use one core tests even aida latency checks only use one core and thats core0 so if you core 0 is not the fastest you will get hit a penalty using pbo.. Now cb and gb is more aware of it but still.. Thing is on pbo you cant control how it decides to boost other than that 1 core...





Now the amd gpu she undervolts like a boss im in love with that one... The old geezer "block" me bcuz of it he dont believe it lol he said i was photoshopping the screenshots lol had to put a darn video thats when he realized he mess up and supposedly block me lol.. funny dude :p


anyway, saw this video the other day and this guy is quite fun to watch lol
my system is way optimized than his thats for sure ends of why i want to test the 6800xt on my x299..
he dont like evga boards either lol
 

·
Registered
Joined
·
356 Posts
Quick questions for the gang?

Background: I have moved my system from the workbench to the big box and my x48 overclock from earlier in the thread has stopped working. The system powers off under CB load and I am only stable on AUTO.

I have been trying to figure out what's going on. In theory, the main box has improved cooling and a more powerful PSU. These are the "main" components that changed between Workbench vs Main PC

EK 3x120 with push pull vs EK 3x140 push pull.
EK D5 Revo vs EK FLT 250
Corsair HX1000i PSU vs Corsair AX1200i PSU

The questions:

1) Would you expect to re-do your entire overclock if you used a different PSU?

2) Do these temperatures appear "normal" for a 10900K running on AUTO*, Watercooled on a 3x140 loop with push / pull?

30 minute Cinebench run, room temp ~17 max.

Cinebench_30.png

*Notes,
It looks to me like I am hitting the same termps on AUTO that I was hitting at X48 - indicating to me that the loop is underperforming?
x30 cache because this is the only overclock that seems to not cause a crash!
Pipework and inlet / outlet have been checked and re-checked
CPU block and paste cleaned and remounted a few times
 

·
Registered
Joined
·
356 Posts
When you say power off are you getting a BSOD or just BAM it’s like you pulled the plug?
Just like I pulled the plug, full power off and reboot.

Do those CPU temps seem high to you for Auto, like +10 too much?

I've just run a Google memory test on Auto, core temp ranges are 55-60. Comparing that with my post here, it's running +10.


I just don't get it. I'm starting to think that the CPU pump is acting strange. It is reporting full speed, but, movement in the res seems significantly different to the movement in the GPU res running at the same speed.
 

·
Old and Decrepit
Joined
·
2,150 Posts
That sounds like one of two things. PSU or thermal shut down on the CPU.
Pump could definitely cause it. Times like this is when a Flowmeter comes in handy. If it boots and is fairly stable until you load it down it could quite possibly be lack of flow. The block will actually dissipate enough if the heat to idle and run low load things. Is the block getting warm? How about the tubing coming in and out?
 

·
Registered
Joined
·
356 Posts
Indeed, it's one of the things that I have never considered using ... kinda wish i had one now!

It's really difficult to tell because i have a lot of anti kink springs, see pics for context.

CPU loop is top FLT240
GPU loop is bottom FLT240

I've just asked my better half to look at both pump / reservoir. She's non IT, but said it is obvious that the top pump is not moving as much as the bottom one, you can tell because the bubble in the top one is not being agitated.

IMG_20210526_202725_3.jpg

Both pumps are running 100% / 4800 RPM

The top bubble is static, the bottom bubble is agitated. I think this indicates the pump is under performing rather than a blockage.

IMG_20210529_153921_3.jpg

I have a spare pump. Not PWM but it will work for testing.
 
2901 - 2920 of 3006 Posts
Top