Overclock.net banner
2461 - 2480 of 3100 Posts

·
Registered
Joined
·
807 Posts
Thanks for taking a look JustinThyme! I'll walk through the steps I have been taking up until now and then include results of your suggested starting point:

My per-core overclocking methods
1) Set each individual core to 1.19
2) Start at 43x per core
3) Complete 20 minutes P95 26.6 non-avx Blend test + Aida64 Cache Stress 20 minutes
4) Bump up another multiplier for each core which stays below 70c (my target core temp when water reaches it's fully heat soaked temp of 32-33c) Stupid Southern California winters :)
5) Rinse repeat until a core reaches 70c OR there is BSOD/thread failure
6) Upon failure, add more voltage until stable or temps are reached

Now for the results of the 10 minute R23 run @ x47, 1.225v core, AVX Offset -0, LLC Level 4, CPU Current 200%, Mesh x28-31, and everything else on auto:
CPU Package max: 96c
VCCIN: 1.76v

VCCSA: 0.928v-0.944v
VCCIO: 1.04v
PCH Core: 1.0v
VRM Temp max: 89c
Cache Voltage: 1.209v-1.216v

All in all, it looks like the AUTO options gave the system a lot of juice. I have never seen my PC draw 900w from the socket before from just a CPU stress test. The AVX offset of 0 was fun to test but i'm not sure how important it is if I don't use a ton of AVX instructions in my day to day. Can anyone recommend another stress test for non-avx workloads? Or is it such that the system is deemed stable for all instructions if AVX passes? The issue here is that I think P95 Blend is not the best stress test since it's limited to 16 cores / 32 threads. I would like to get above 47x all-core WITH an AVX offset.

Putting a fixed cache voltage of 1.1 instead of Auto should shave some degrees off the package too, but it'll jump right back up once I start tuning DRAM.
I would check your CPU block mount. 96C is pretty toasty if you are using 2 XE 480s. What kind of spread are you getting between cores?

For VCCIN, I'd go for 1.82-1.85.If you are using HWInfo, check to see what your VCCIN drops to under load. My chip seems happy when the VCCIN voltage doesn't go below 1.75.

That is HOT for the VRMs. Maybe rig a fan over them or add a VRM water block to cool them down a bit.

The only way to "save power" on idle is to use adaptive voltage, which does not work too well for most of us due to the pre-programmed VID tables being so high. As an example, say you set your target adaptive voltage to be 1.225V with a 47 multiplier. If Intel's pre-prgrammed voltage for your chip at that multiplier is 1.32, that is what you will see. Your setting will be over ridden. I have tried this, as others have, and resorted back to a static voltage.

I tried the "By specific core" method and it took me ages and netted me no temperature improvements. I wouldn't waste too much time there. I would also not use the latest bios. The newer bios versions have security patches which really kill performance and may require more voltage to achieve a certain OC. I looked at the RVIE bios versions and I would try version 2002 which contains MC 29. You will have to use the bios flashback option to flash the bios. If the RVIE has 2 bios chips, you can always load 2002 onto the second chip and compare..
 

·
Registered
Joined
·
77 Posts
RTLs look good. You got a good silicon sample there. (y)
So there are several tweaks I see right off the bat. FAW can be set as low as 4x tRDD, and tRTP should be able to go as low as 8 (maybe 6). (which is really key as RAS >= CAS+RCD+RTP plus/minus 2 or the IMC will set an arbitrary RAS since the ras window must be open for all three operations to complete). Both changes together may require a slight bump in Vdimm. 4x8GB should also be able to run 1T, but that will not buy much without lower RFC further.
Try out the FAW and RTP suggestions. (Re)test stability. +25mV is usually all that might be needed.
here's a few examples (thumbnails) at 4000c16, 3600c15 and if the IMC is really good, 4200 with 64GB (8x8GB). I'd have to dig back to find snips when I had 32GB on this R6EO. What's really interesting is the GSAT M/s at 3600c15 (tightened) vs 4000c16 (also pretty tight). AID64 has them both at the same 50-52ns latency, but slightly higher read band for 4000
View attachment 2472718 View attachment 2472719 View attachment 2472720
This was very helpful. So I made the changes to tFAW and tRTP as well as tightened the tRFC and tREFI under a new timing 4000Mhz @ 16-16-16-36-2T , AIDA64 Reads jumped to 121GB/s and copy also bumped up to 104GB/s, i'm pretty impressed with the chip so far

2472833
 

·
Registered
Joined
·
3,353 Posts
This was very helpful. So I made the changes to tFAW and tRTP as well as tightened the tRFC and tREFI under a new timing 4000Mhz @ 16-16-16-36-2T , AIDA64 Reads jumped to 121GB/s and copy also bumped up to 104GB/s, i'm pretty impressed with the chip so far

View attachment 2472833
Nice results :)
Bump the Cache to 30 or even 32, and you are good to do whatever :D
 

·
Old and Decrepit
Joined
·
2,515 Posts
I would check your CPU block mount. 96C is pretty toasty if you are using 2 XE 480s. What kind of spread are you getting between cores?

For VCCIN, I'd go for 1.82-1.85.If you are using HWInfo, check to see what your VCCIN drops to under load. My chip seems happy when the VCCIN voltage doesn't go below 1.75.

That is HOT for the VRMs. Maybe rig a fan over them or add a VRM water block to cool them down a bit.

The only way to "save power" on idle is to use adaptive voltage, which does not work too well for most of us due to the pre-programmed VID tables being so high. As an example, say you set your target adaptive voltage to be 1.225V with a 47 multiplier. If Intel's pre-prgrammed voltage for your chip at that multiplier is 1.32, that is what you will see. Your setting will be over ridden. I have tried this, as others have, and resorted back to a static voltage.

I tried the "By specific core" method and it took me ages and netted me no temperature improvements. I wouldn't waste too much time there. I would also not use the latest bios. The newer bios versions have security patches which really kill performance and may require more voltage to achieve a certain OC. I looked at the RVIE bios versions and I would try version 2002 which contains MC 29. You will have to use the bios flashback option to flash the bios. If the RVIE has 2 bios chips, you can always load 2002 onto the second chip and compare..
DITTO on that. If you have two 480s you should NEVER see 96C running CB23 at those settings. I didn’t even see that when I was testing out of the case on the MOBO box top with a single 480 and an Xres140 for a pump/res combo on either the Optimus Sig V2 or the Magnitude. Even adding GPUs to the loop won’t change anything for CB23 but flow rates which I have no problems at all with 3 D5s in serial pumping 10L/min. The GPUs don’t even load up with that bench. You either have a flow problem or contact problem. There are only two blocks on the market ATM that IMO are capable and that’s Optimus and EK magnitude, HKIV pro lags behind a little. If you don’t already have one HK also sells a VRM block for that board that makes a massive difference. If you already have one then we have identified the problem to be flow as I ram the crap out of my R6E with the HK block on the VRMs and they never passed 45C. If you have a mono block yank that POS off and hang it on your wall of shame. This is something no one has seemed to master. I tried the new one EK came out with for the R6EE and found out why they didn’t respond to my request for numbers and no reviewer put up anything but a regurge Of EKs marketing announcement. It cools the VRMs just fine but sucks balls at cooling the CPU. Both are persnickety with the mount. Once you get it right the first time it’s like riding a bike. Too much TIM is detrimental to both. Pea blob right in the middle, ease the block down squarely and wiggle a little within the limitations of the mount. Keep pressure on it while getting two opposing mount screws snugged up but not torqued just yet. Then you can turn loose and put the other two on and tighten no more than half a turn at a time like you are putting on a car tire working your way diagonally until you bottom that sucker out. These chips like more pressure. Don’t be afraid to torque down the Optimus if that what you are using to as tight as you can without using a tool and on the Magnitude bottom it out, have to use a tool on those but don’t break out an impact wrench just tighten with a moderate amount of force until it stops turning. If it’s not the mount than it can be flow. Do you have a flow meter? It doesn’t take much of anything in the loop to clog the micro channels. For this reason I run an inline filter. Also loop prep is extremely important. I don’t care how much any company says they ship their rads flushed. Running water doesn’t count as it doesn’t get the flux and gunk out from the manufacturing process. Before I put any rad into service I use a weird but effective process. Some think I’m a nut job but the best rad cleaner by far is Coca-Cola. I sit the rad with ports facing up and a short extension or tube on both and slowly pour the Coca-Cola in and leave it over night. Wake up the next morning and before I even start flushing it out you can see green goop that bubbled out overnight. Then I connect a flex tube via a faucet thread barb tube that you can buy at any Home Depot and let the hot water rip for a good 15-20 minutes. When you are done the inside of the rad is as clean as it ever will be. I can look into the ports and see the end cap is squeaking clean with shiny copper. Of course after that a couple of DW fill, slosh around and dump cycles. I don’t use the mayhems that requires goggles for eye protection as its corrosive and all my blocks are nickel plated because copper is going to oxidize, not a matter of if, it’s when. Something milder like primochill loop prep is my preference and run that DW for a day then drain and dump it. Then 2-3 fill, run and hour and drain with DW. I go one step further than most and pull a plug in the loop and leave the res cap on and drain valve open and connect my 3 gal portable 120PSI compressor to it and blow out what’s left before I fill it with coolant. While you are pulling the block to check the mount it’s a good a time as any to crack open the block and check the micro channels are full of goop. Those thin micro channels are great for heat dissipation but come with the caveat that they are very easy to clog.....again in line filter. I can’t recommend a good flow meter enough. I’m not talking about the paddle wheel indicators, a meter that measures the volume of flow. Mine it connected to an aquaero controller with alarm set points so I know when the flow is boogered up. Last time it was the nickel plating from the Optimus cold plate guaranteed not to flake off all up I’m my filter clogging it up. In their defense I sent them photos of the cold plate fins and they sent me a new cold plate and said batch made it though unsecured by QC with bad plating. Regardless it was boogered and I replaced it.

There is definitely something not right!!
 

·
Registered
Joined
·
248 Posts
I would check your CPU block mount. 96C is pretty toasty if you are using 2 XE 480s. What kind of spread are you getting between cores?

For VCCIN, I'd go for 1.82-1.85.If you are using HWInfo, check to see what your VCCIN drops to under load. My chip seems happy when the VCCIN voltage doesn't go below 1.75.

That is HOT for the VRMs. Maybe rig a fan over them or add a VRM water block to cool them down a bit.

The only way to "save power" on idle is to use adaptive voltage, which does not work too well for most of us due to the pre-programmed VID tables being so high. As an example, say you set your target adaptive voltage to be 1.225V with a 47 multiplier. If Intel's pre-prgrammed voltage for your chip at that multiplier is 1.32, that is what you will see. Your setting will be over ridden. I have tried this, as others have, and resorted back to a static voltage.

I tried the "By specific core" method and it took me ages and netted me no temperature improvements. I wouldn't waste too much time there. I would also not use the latest bios. The newer bios versions have security patches which really kill performance and may require more voltage to achieve a certain OC. I looked at the RVIE bios versions and I would try version 2002 which contains MC 29. You will have to use the bios flashback option to flash the bios. If the RVIE has 2 bios chips, you can always load 2002 onto the second chip and compare..
Hey Rich,

Thanks for taking a look and providing your expertise around adaptive and per-core OC'ing. It seems that likely isn't the way to go. Mixed back of reports across the interwebs with people's per-core OC'ing results. You know how it goes, you have to find bits and pieces of people's stories to paint a semi-accurate picture before trying it for yourself. My 5 days of per-core overclocking attempts didn't get far with thermal optimizations...it might be nice to hit 4.9 or 5.0 on a few cores but there's zero chance of shaving off 10-15c by reducing the multiplier on a fraction of the cores.

I was able to drop the CPU Package temp down to 90c by removing some intake fans and letting the negative air pressure of the case (Inwin 909EK) do its thing. I previously had 3 intake fans supplying 8 exhaust fans and figured the radiator fans weren't getting enough air.

VCCIN under load right now is 1.745v. The good news? R23 crashes at x48 multiplier with 1.215vcore but runs for a full hour with 1.220v with flying colors. For now I'm happy with x48 and will target reducing temps before any other hopes of 4.9ghz.

After 10 minutes of R23, the delta between the coldest and hottest cores was 14c, which seems a bit high for a soldered chip? I checked the waterblock to make sure it was tightened all the way and it's on there SNUG as a bug. The cold plate also goes right down the middle of the die so that's fine too. The VRM's on the other hand may be seated poorly as I recall having some issues tightening them down during the build process but I have also read that the Asus R6E VRM's were suited for 8-14cores and struggled a bit more with the 18-core cpu's. Not sure if that's what's causing the super high temperatures or just poor seating/use of thermal pads. Maybe going to a thinner pad would make sense but that would need a full drain and partial disassemble which is not ideal after spending a week building over the holidays.

@JustinThyme - my room ambient temp is about 24c with PC idle temps sitting at 28c and load temps never exceeding 33c. I believe you have more radiators than me with 11 fans. Flow shouldn't be an issue for my PC as I'm running 2x D5's in serial... pressure for days.

Next steps for me are to flash back to 2002. When I bumped this motherboard up to 3201, it broke my 7900x OC with a vengeance and I couldn't get back to it due to thermals. My hunch is moving back to 2002 may reduce temps and provide some more headroom before I look to re-seat the CPU and VRM blocks.
 

·
Registered
Joined
·
77 Posts
Nice results :)
Bump the Cache to 30 or even 32, and you are good to do whatever :D
Cache is already running at 30 @ 1.01V Adaptive , AIDA64 never manages to read the full cache clock in the info section of benchmark . This will be my daily driver then :)
 
  • Rep+
Reactions: Nizzen

·
Old and Decrepit
Joined
·
2,515 Posts
Hey Rich,

Thanks for taking a look and providing your expertise around adaptive and per-core OC'ing. It seems that likely isn't the way to go. Mixed back of reports across the interwebs with people's per-core OC'ing results. You know how it goes, you have to find bits and pieces of people's stories to paint a semi-accurate picture before trying it for yourself. My 5 days of per-core overclocking attempts didn't get far with thermal optimizations...it might be nice to hit 4.9 or 5.0 on a few cores but there's zero chance of shaving off 10-15c by reducing the multiplier on a fraction of the cores.

I was able to drop the CPU Package temp down to 90c by removing some intake fans and letting the negative air pressure of the case (Inwin 909EK) do its thing. I previously had 3 intake fans supplying 8 exhaust fans and figured the radiator fans weren't getting enough air.

VCCIN under load right now is 1.745v. The good news? R23 crashes at x48 multiplier with 1.215vcore but runs for a full hour with 1.220v with flying colors. For now I'm happy with x48 and will target reducing temps before any other hopes of 4.9ghz.

After 10 minutes of R23, the delta between the coldest and hottest cores was 14c, which seems a bit high for a soldered chip? I checked the waterblock to make sure it was tightened all the way and it's on there SNUG as a bug. The cold plate also goes right down the middle of the die so that's fine too. The VRM's on the other hand may be seated poorly as I recall having some issues tightening them down during the build process but I have also read that the Asus R6E VRM's were suited for 8-14cores and struggled a bit more with the 18-core cpu's. Not sure if that's what's causing the super high temperatures or just poor seating/use of thermal pads. Maybe going to a thinner pad would make sense but that would need a full drain and partial disassemble which is not ideal after spending a week building over the holidays.

@JustinThyme - my room ambient temp is about 24c with PC idle temps sitting at 28c and load temps never exceeding 33c. I believe you have more radiators than me with 11 fans. Flow shouldn't be an issue for my PC as I'm running 2x D5's in serial... pressure for days.

Next steps for me are to flash back to 2002. When I bumped this motherboard up to 3201, it broke my 7900x OC with a vengeance and I couldn't get back to it due to thermals. My hunch is moving back to 2002 may reduce temps and provide some more headroom before I look to re-seat the CPU and VRM blocks.
regardless of what BIOs you run it’s not going to cause that high of a core temp. Ive run just about every variant for the R6E and the R6EE and the best of the best changed temps by 3C. It has more of an impact on performance at any given clock speed. 24 C ambient is precisely what my controlled environment sits at and I posted up a 10 min run with a single rad where the hottest core is 81C.

tight doesn’t necessarily mean a good contact and TIM spread. Only way to find out is pull the block. How many pumps doesn’t mean much if you have a restriction. I run 3 and my flow was cut in half, only way to know is adding a flow meter.
As for the VRMs a little advice that works well for me. I’m addition to the pads I put a thin layer on both sides of kryonaut. Not a bunch, just a tiny dab. You may be correct in that the 8 stage just isn’t enough. I can’t confirm as the biggest I ran was a 9940X but that is only 4 less cores. I’d expect some increase but running the HK VRM block I didn’t see more than 40C. There is not much of a chance if there being a flow restriction on that. It’s just an open channel for the most part with 3 passes across. I’m still putting my bet on a flow restriction, most likely in the CPU block. If one or the other was acting up it would be different but for both to go that high at the settings you posted screams flow restriction.

Jpm recently posted up where he yanked his Optimus block apart and found it pretty clogged up. A flow meter will go a long way. With 2 D5s you should be getting a minimum of 6L/min no matter what variant you are running, maybe a little more. I’ve been there with 3 pumps that were initially at 10L/min drop just under 5 running the same speed. Clear the restriction and I’m right back to 10L/min. I set mine to 4.7 sync all at 1.2Vcore manual override and I barely break 70C.
Changing BIOS may gain some performance and drop 3C but it doesn’t explain 35C.

All part of the challenge. But once you get it nailed you are good. Gotta get that cooling temp problem fixed before bothering with anything else. A spread of 15C isn’t unheard of. Some chips are better than others. But a lot of it is a good seat which may take extreme measures to flatten out like lapping to get a better surface to surface seat. I miss the old days where everything was flat and your biggest challenge was polishing the cold plate to a mirror finish.

if I’m not mistaken Rich yanked the Oring on this Optimus which will flatten the bow and got better results. I’m about to tear into mine and while I’m at it I’m going to try the magnitude flat cold plate that didn’t work very well on the 9940X and also yanking the oring on my Optimus block just to see which get me the best performance on this chip. What I’m running now is what was extensively tested on a 9940X and just put on the10980XE. Maybe need to fine tune it without taking it to anything that will void my warranty.
 

·
OC...D
Joined
·
28,265 Posts
This was very helpful. So I made the changes to tFAW and tRTP as well as tightened the tRFC and tREFI under a new timing 4000Mhz @ 16-16-16-36-2T , AIDA64 Reads jumped to 121GB/s and copy also bumped up to 104GB/s, i'm pretty impressed with the chip so far

View attachment 2472833
Yeah, a good start... the Latency is high(er) than it should be and that's likely because the RRD_L (long) is shorter that the RRD (short). Reverse these numbers: just set 4 for RRD and leave RRD_L on auto. Then FAW should be good at 16. If it posts, I'm betting things improve (and get more stable). Micron wrote a paper on the effect of FAW for cloud server bandwidth ) internode coms), which is very much in play for cpu-gpu coms (like in gaming). Also, with 4x8GB, the tWR and tCWL (write latencies) can usually be set at or below the CAS value. tWR can go as low as 12 in this configuration (but stability would need to be verified, if it can post with a tWR 4 below CAS.)
(y)
 

·
OC...D
Joined
·
28,265 Posts
Thanks for taking a look JustinThyme! I'll walk through the steps I have been taking up until now and then include results of your suggested starting point:
snip

Can anyone recommend another stress test for non-avx workloads? Or is it such that the system is deemed stable for all instructions if AVX passes? The issue here is that I think P95
You can disable AVX/FMA in the new version(s) p95 very easily. then it just runs SSEs
{paste}
Just use the latest Prime95 and use the commands below to disable FMA3 and AVX if you so wish. Explained in the undoco.txt file. Put the command into the local.txt file. This way you get rid of all bugs that have been fixed while giving you an option to test without FMA3 or AVX (FMA obviously won't work if you disable AVX).

CpuSupportsAVX=0 or 1
CpuSupportsFMA3=0 or 1
 

·
Old and Decrepit
Joined
·
2,515 Posts
Yeah, a good start... the Latency is high(er) than it should be and that's likely because the RRD_L (long) is shorter that the RRD (short). Reverse these numbers: just set 4 for RRD and leave RRD_L on auto. Then FAW should be good at 16. If it posts, I'm betting things improve (and get more stable). Micron wrote a paper on the effect of FAW for cloud server bandwidth ) internode coms), which is very much in play for cpu-gpu coms (like in gaming). Also, with 4x8GB, the tWR and tCWL (write latencies) can usually be set at or below the CAS value. tWR can go as low as 12 in this configuration (but stability would need to be verified, if it can post with a tWR 4 below CAS.)
(y)
Im gonna have to pick your brain on memory timings when I get to that. Right now still working on CPU clocks and temps. I’m doing OK but think I can do better. MO-RA3 420 shipped and arriving Friday. If that’s doesn’t get me where I want to be it’s off to the new Koolance 900W chiller. I’m hoping the MO-RA will do it because I really don’t feel like running another circuit up from the basement to the second floor.
 

·
OC...D
Joined
·
28,265 Posts
Im gonna have to pick your brain on memory timings when I get to that. Right now still working on CPU clocks and temps. I’m doing OK but think I can do better. MO-RA3 420 shipped and arriving Friday. If that’s doesn’t get me where I want to be it’s off to the new Koolance 900W chiller. I’m hoping the MO-RA will do it because I really don’t feel like running another circuit up from the basement to the second floor.
lol - the sum total of my brain-pan may be in the posts above! :oops:
Warning... the rabbit hole shows no mercy.
 

·
Old and Decrepit
Joined
·
2,515 Posts
lol - the sum total of my brain-pan may be in the posts above! :oops:
Warning... the rabbit hole shows no mercy.
Yeah Alice certainly proved the bottomless pit of the rabbit hole. I’ve been in the electronics industry for 35+ years and that’s certainly one thing I’ve learned. As soon as you think you have it all figured out you find out you know nothing. I teach classes on a pretty regular basis for the largest UPS power density and most complex machines on the planet. One of my biggest hurdles is getting those who think they know it all to check that at the door when entering my class. I leave the best of them with smoke pouring out of their ears with their biggest hurdle still being forgetting what you think you know and pay attention to reality. I get some that are so convinced they know the material better than me that I hand them my laser pointer amd control of the electronic files that have the system schematics and go sit down while they fumble trying to cover what’s going on. None of them have made it more than about 15 seconds. I am the authority on these machines and yet I still don’t know everything. Once in a while I’ll get a question that stumps me (always from one or the people that spent more time on trying to trip me up than paying attention) and I have a park it on the board for those questions and keep it rolling. I’ll have the answer for them before the next day and funny part is those are usually the people that don’t do very well on test day. No one fails my classes. They screw up and all they get is staying behind after everyone else has gone for the day. Don’t care if we are still there when everyone shows back up the next morning. The alternative is to refuse at which point I send them home, fail them for the class and call their managers who prepare their exit interviews before they get home, no highway option other than to the unemployment line. Even after all that I still learn more with every class I teach. Some want to dig too deep into a topic that there just isn’t time for in the curriculum and I tell them to take my word for it and consider it PFM. PFM? What’s that? Pure Fuggjng Magic! If you really need to get that deep I’ll arrange a few days in my schedule and a Skype invite. Be forewarned though that if you take it there you are not going to bail!
 

·
Registered
Joined
·
807 Posts
Hey Rich,

Thanks for taking a look and providing your expertise around adaptive and per-core OC'ing. It seems that likely isn't the way to go. Mixed back of reports across the interwebs with people's per-core OC'ing results. You know how it goes, you have to find bits and pieces of people's stories to paint a semi-accurate picture before trying it for yourself. My 5 days of per-core overclocking attempts didn't get far with thermal optimizations...it might be nice to hit 4.9 or 5.0 on a few cores but there's zero chance of shaving off 10-15c by reducing the multiplier on a fraction of the cores.

I was able to drop the CPU Package temp down to 90c by removing some intake fans and letting the negative air pressure of the case (Inwin 909EK) do its thing. I previously had 3 intake fans supplying 8 exhaust fans and figured the radiator fans weren't getting enough air.

VCCIN under load right now is 1.745v. The good news? R23 crashes at x48 multiplier with 1.215vcore but runs for a full hour with 1.220v with flying colors. For now I'm happy with x48 and will target reducing temps before any other hopes of 4.9ghz.

After 10 minutes of R23, the delta between the coldest and hottest cores was 14c, which seems a bit high for a soldered chip? I checked the waterblock to make sure it was tightened all the way and it's on there SNUG as a bug. The cold plate also goes right down the middle of the die so that's fine too. The VRM's on the other hand may be seated poorly as I recall having some issues tightening them down during the build process but I have also read that the Asus R6E VRM's were suited for 8-14cores and struggled a bit more with the 18-core cpu's. Not sure if that's what's causing the super high temperatures or just poor seating/use of thermal pads. Maybe going to a thinner pad would make sense but that would need a full drain and partial disassemble which is not ideal after spending a week building over the holidays.

@JustinThyme - my room ambient temp is about 24c with PC idle temps sitting at 28c and load temps never exceeding 33c. I believe you have more radiators than me with 11 fans. Flow shouldn't be an issue for my PC as I'm running 2x D5's in serial... pressure for days.

Next steps for me are to flash back to 2002. When I bumped this motherboard up to 3201, it broke my 7900x OC with a vengeance and I couldn't get back to it due to thermals. My hunch is moving back to 2002 may reduce temps and provide some more headroom before I look to re-seat the CPU and VRM blocks.
Figuring out case air flow can be a PITA. I know I changed my fans around a few times and finally settled on all 9 fans as intakes with no exhaust fans. This resulted in lower load temps due to cooler water temps. My case, a PC 011XL, has plenty of open space in the back so air exits pretty easily. For example, my case temps are 1C higher with the side and front panels on compared to all of the panels off.

48 is fine. You may get marginally better performance at 4.9 (if you can get it stable) but you will be fighting heat once again. I'm running all cores at 4.8 @ 1.235V and my highest temp is @81C. That's with 2 EK PE 360s and a XE360 all running Noctua IPPC 3000 fans. Other than full load, my PC is pretty silent.

I have a feeling 2002 will perform much better and with slightly cooler temps. The newer microcodes really put a hurting on performance and jack up the heat. I saw a notable increase in performance when I rolled my Omega bios back a few versions. I have since modded my favorite bios version, 3006, with an older microcode and it's been running great. I still want to lower temps even more, but I'm waiting to see how the MO RA work out for Justin.
 

·
Registered
Joined
·
807 Posts
regardless of what BIOs you run it’s not going to cause that high of a core temp. Ive run just about every variant for the R6E and the R6EE and the best of the best changed temps by 3C. It has more of an impact on performance at any given clock speed. 24 C ambient is precisely what my controlled environment sits at and I posted up a 10 min run with a single rad where the hottest core is 81C.

tight doesn’t necessarily mean a good contact and TIM spread. Only way to find out is pull the block. How many pumps doesn’t mean much if you have a restriction. I run 3 and my flow was cut in half, only way to know is adding a flow meter.
As for the VRMs a little advice that works well for me. I’m addition to the pads I put a thin layer on both sides of kryonaut. Not a bunch, just a tiny dab. You may be correct in that the 8 stage just isn’t enough. I can’t confirm as the biggest I ran was a 9940X but that is only 4 less cores. I’d expect some increase but running the HK VRM block I didn’t see more than 40C. There is not much of a chance if there being a flow restriction on that. It’s just an open channel for the most part with 3 passes across. I’m still putting my bet on a flow restriction, most likely in the CPU block. If one or the other was acting up it would be different but for both to go that high at the settings you posted screams flow restriction.

Jpm recently posted up where he yanked his Optimus block apart and found it pretty clogged up. A flow meter will go a long way. With 2 D5s you should be getting a minimum of 6L/min no matter what variant you are running, maybe a little more. I’ve been there with 3 pumps that were initially at 10L/min drop just under 5 running the same speed. Clear the restriction and I’m right back to 10L/min. I set mine to 4.7 sync all at 1.2Vcore manual override and I barely break 70C.
Changing BIOS may gain some performance and drop 3C but it doesn’t explain 35C.

All part of the challenge. But once you get it nailed you are good. Gotta get that cooling temp problem fixed before bothering with anything else. A spread of 15C isn’t unheard of. Some chips are better than others. But a lot of it is a good seat which may take extreme measures to flatten out like lapping to get a better surface to surface seat. I miss the old days where everything was flat and your biggest challenge was polishing the cold plate to a mirror finish.

if I’m not mistaken Rich yanked the Oring on this Optimus which will flatten the bow and got better results. I’m about to tear into mine and while I’m at it I’m going to try the magnitude flat cold plate that didn’t work very well on the 9940X and also yanking the oring on my Optimus block just to see which get me the best performance on this chip. What I’m running now is what was extensively tested on a 9940X and just put on the10980XE. Maybe need to fine tune it without taking it to anything that will void my warranty.
Yeah, I'd pull the blocks and check contact. Something is just not right.

An older bios version MAY help temps, but not in a jaw dropping way. Check those mounts!

Yes, I removed the center o-ring from my Signature V2 and it tightened up the core spread to ~7-9C between all the cores depending on what I'm running.
 

·
Old and Decrepit
Joined
·
2,515 Posts
Figuring out case air flow can be a PITA. I know I changed my fans around a few times and finally settled on all 9 fans as intakes with no exhaust fans. This resulted in lower load temps due to cooler water temps. My case, a PC 011XL, has plenty of open space in the back so air exits pretty easily. For example, my case temps are 1C higher with the side and front panels on compared to all of the panels off.

48 is fine. You may get marginally better performance at 4.9 (if you can get it stable) but you will be fighting heat once again. I'm running all cores at 4.8 @ 1.235V and my highest temp is @81C. That's with 2 EK PE 360s and a XE360 all running Noctua IPPC 3000 fans. Other than full load, my PC is pretty silent.

I have a feeling 2002 will perform much better and with slightly cooler temps. The newer microcodes really put a hurting on performance and jack up the heat. I saw a notable increase in performance when I rolled my Omega bios back a few versions. I have since modded my favorite bios version, 3006, with an older microcode and it's been running great. I still want to lower temps even more, but I'm waiting to see how the MO RA work out for Justin.
The MO-RA will be a project this weekend. Hopefully that will give me improvement. About has to as it will be with no case airflow restrictions. Ive not seen anyone running run nor a review that said otherwise. Again in the wintertime put that puppy in the window and I can almost pull off what Mr Too Short does but not quite seeing how he is in Calvary BC I believe. Way the hell up there where winter temps are well below zero for a good part of the winter. I may still end up getting a chiller for giggles. Im putting quick connects though a PCIE slot so I can just unplug one and plug in the other. Only caveat is Koolance insists on their coolant which is a watered down blue color and Im a bit stuck on my Mayhems pastel red. They just came out with a 900W version. Im not looking to go sub ambient, just keep the coolant temps at ambient which the MO-RA may be able to pull off.

In the meantime here is a list of modded BIOS files for all the X299 ASUS boards. The one Im running has MC 29. Whats on the website is whatever they came out with at the time with all the patches installed. Ive run pretty much all of them including stock on the R6E and R6EE. I was torn between the Omega and Encore after I borked the pins on my R6E. On one had the Omega still has a U2 port and for OS an intel 905P U2 is the shiznit for low que depth randowm read on small files....in other words all the puny little DLL files that are scattered everywhere. That gave me the best boot performance also. Im liking the M.2 Intel drives in VROC raid zero but the latency of the raid slows down the low que depth speeds but the sequential reads are 5GB/S Cant do that with the Samsung drives and boot from them. You can do a software raid and get good scaling but their low que depth numbers arent nearly as good and really go in the crapper once raided up. Good for large scratch files or media storage but no an OS. The reason I chose the Encore is the infineon VRMs, supposed to be the best there is.

ROG modded BIOS Tistou hasn't disappointed I did well with the MC 49 on the 9940X for the R6E. There are some not up there anymore that I have stored somewhere in my mess but ended on 2002 MC49, theres another link later in the thread for the Encore 702 with MC29 which is what Im running. There is also an app "Inspectre" where you can disable the Meltdown and Inspectre patches, if you cant find that then I can upload it.


But back to the basics, regardless of the BIOS your temps suggest something other than the BIOS, something not well in the loop. Id not only pull the blocks Id pop the cold plate off and check for shyte in there that doenst belong. Some coolants breakdown and collect, EK cryofuel has a rep for this. Dont know what you are running but as long as you dont have your Vcore in auto letting it go to mega volts you should nt get near those numbers on CB23 unless you have your PC in the oven with your turkey.
 

·
Registered
Joined
·
253 Posts
Im gonna have to pick your brain on memory timings when I get to that. Right now still working on CPU clocks and temps. I’m doing OK but think I can do better. MO-RA3 420 shipped and arriving Friday. If that’s doesn’t get me where I want to be it’s off to the new Koolance 900W chiller. I’m hoping the MO-RA will do it because I really don’t feel like running another circuit up from the basement to the second floor.
Already tried MORA 420 can't tame above 1.290v all cores with my warm ambient temperature, SO I plan to order a chiller from alphacool hope it can tame the heat ... (it should isn't? )
 

·
Registered
Joined
·
248 Posts
Yeah, I'd pull the blocks and check contact. Something is just not right.

An older bios version MAY help temps, but not in a jaw dropping way. Check those mounts!

Yes, I removed the center o-ring from my Signature V2 and it tightened up the core spread to ~7-9C between all the cores depending on what I'm running.
You and @JustinThyme were right. I spent the evening after work today draining and re-doing my CPU/VRM heatsinks. Here are the results:

3201 to 2002 BIOS:
CPU dropped from 90c to 87c after 10 minutes of Cinebench R23. I do these runs after my system water is fully heat soaked.
VRM max temp dropped from 89c to 82c.

Removed Optimus Foundation O-ring, Switched from Kryonaut to Conductonaut, Re-did the VRM thermal pads + some Kryonaut for good measure:
CPU dropped from 87c to 77c after 10 minutes of Cinebench R23. I wasn't able to heatsoak as high this time around, even after 60 minutes. The system is now more effective.
VRM max temp dropped from 82c to 65c.

Fan and water pump curves now need to be adjusted, a few more degrees can probably get squeezed out but we're well within the ranges you guys mentioned! May even go after 4.9ghz all core tomorrow :)
 

·
Registered
Joined
·
210 Posts
Already tried MORA 420 can't tame above 1.290v all cores with my warm ambient temperature, SO I plan to order a chiller from alphacool hope it can tame the heat ... (it should isn't? )
I have an Airplex Gigant 3360 with 26 1800rpm fans and also 4x360 rads with 12 noctua 3000rpm fans. Currently running 2xserial d5 and have 165 l/h flow max with qfittings, filter, 2xflow meters, pressure meter, gpu block in it.

This gives me in 26-27 deg ambient the ability to cool my settings of 2x5.1 4x5.0 12x4.8 and never go above 85 degrees on 5.1 or 5.0 cores in bios 0702 and a 3090 with 520w bios in loop too. I have a crappy SP 71 chip that requires 1.36 on 5.1 adaptive, around 1.32 on 5.0 adaptive and 1.25 on 4.8 fixed.

Fans on 60% speed only. Delta T ambient to water is 1 degree max 2 if standing for a long time since it is in a corner without any window closeby.

I have ordered 2 more D5 to try to improve the flow to 1 gpm 225ish l/h) at least. Will try 4x serial (will the pressure be too high for blocks / rads? 4x0.37psi) and 2xparallel + 2 serial. Also I will remove the quick fittings and filter I have to see what that improves.

How you guys get 600l/h is beyond me to understand unless you have 4 pumps a cpu block and rads only, nothing else in the loop.
My cpu block is techn and ekwb for strix (will be bykski and mp5 shortly).

In essense, maybe a MoRA 420 might not be enough for you to cool it properly ?
 

·
Registered
Joined
·
77 Posts
Yeah, a good start... the Latency is high(er) than it should be and that's likely because the RRD_L (long) is shorter that the RRD (short). Reverse these numbers: just set 4 for RRD and leave RRD_L on auto. Then FAW should be good at 16. If it posts, I'm betting things improve (and get more stable). Micron wrote a paper on the effect of FAW for cloud server bandwidth ) internode coms), which is very much in play for cpu-gpu coms (like in gaming). Also, with 4x8GB, the tWR and tCWL (write latencies) can usually be set at or below the CAS value. tWR can go as low as 12 in this configuration (but stability would need to be verified, if it can post with a tWR 4 below CAS.)
(y)
So just setting 4 for tRRD and leaving tRRD_L to Auto didn't work as IMC decided it has to set it as 3 to keep it below tRRD lol not sure why. So i had to actually reverse it manually i.e set tRRD to 4 and tRRD_L to 7. I could bring down both tWR and tCWL to 14. With these changes , in AIDA64 write and copy improved to 103.71GB/s and 106.95GB/s respectively.

Who would have thought 2 pairs of dual channel kits would scale this well in quad channel 4000 CL16 :). It's just pure luck that I have pair of Samsung B-Die's and a 10980xe IMC that just refuses to give up. Surprising thing is Latency remained the same even though write latencies were tightened.

2472930
 

·
Old and Decrepit
Joined
·
2,515 Posts
You and @JustinThyme were right. I spent the evening after work today draining and re-doing my CPU/VRM heatsinks. Here are the results:

3201 to 2002 BIOS:
CPU dropped from 90c to 87c after 10 minutes of Cinebench R23. I do these runs after my system water is fully heat soaked.
VRM max temp dropped from 89c to 82c.

Removed Optimus Foundation O-ring, Switched from Kryonaut to Conductonaut, Re-did the VRM thermal pads + some Kryonaut for good measure:
CPU dropped from 87c to 77c after 10 minutes of Cinebench R23. I wasn't able to heatsoak as high this time around, even after 60 minutes. The system is now more effective.
VRM max temp dropped from 82c to 65c.

Fan and water pump curves now need to be adjusted, a few more degrees can probably get squeezed out but we're well within the ranges you guys mentioned! May even go after 4.9ghz all core tomorrow :)
Awesomel!!
 
2461 - 2480 of 3100 Posts
Top