[UPDATED] Der8auer States Some X299 Low End boards Have Bad VRM Coolers

[uplink777]

done12many2 & PCGuy_5960 - thank You guys, You've been of a great help. I started to lose faith in my choice. Thank You once more, well, mostly and mainly You done12many2.

[done12many2]

Just now, uplink777 said:

done12many2 & PCGuy_5960 - thank You guys, You've been of a great help. I started to lose faith in my choice. Thank You once more, well, mostly and mainly You done12many2.

 

You're welcome bud.  

[uplink777]

I'll try to put it together tomorrow/after tomorrow, and will update the thread My aim is stable 4.5 GHz on all cores for 365/7 load, though, would love to see 4.7 GHz w/o delid, and 5 GHz with delid. Wish me luck, I'm gonna need it, as always, when I change generation. Btw. @done12many2, how far did You manage to push Yours?
 

[PCGuy_5960]

4 minutes ago, uplink777 said:

I'll try to put it together tomorrow/after tomorrow, and will update the thread My aim is stable 4.5 GHz on all cores for 365/7 load, though, would love to see 4.7 GHz w/o delid, and 5 GHz with delid. Wish me luck, I'm gonna need it, as always, when I change generation. Btw. done12many2, how far did You manage to push Yours?

Nice, have fun building and overclocking bud!

 

BTW, if you want to tag someone make sure to add an "@" before their name, so that they get a notification

@done12many2

[uplink777]

Thank You very much @PCGuy_5960, the @ shortcut will come in handy :). Hehe, well, fun...most of the times it's week or two lasting agony spent in debugging and figuring out some early hardware issues, but I'll try to have some fun

 

Hmm, my tagging doesn't work that well  @PCGuy_5960 ah, now it does

[done12many2]

3 hours ago, uplink777 said:

I'll try to put it together tomorrow/after tomorrow, and will update the thread My aim is stable 4.5 GHz on all cores for 365/7 load, though, would love to see 4.7 GHz w/o delid, and 5 GHz with delid. Wish me luck, I'm gonna need it, as always, when I change generation. Btw. @done12many2, how far did You manage to push Yours?
 

 

I'm able to run 4.8 GHz stable (no delid), but I can't figure out how to completely remove the AVX offset on this damn motherboard.  The best I can do is reduce it to 100 MHz, which means that under any AVX load it's bouncing between 4.7 and 4.8 GHz.  Things can get a bit hot (~80c) during some Handbrake runs using H.265 so I'm definitely wanting to get my hands on the DDM-X to chop the top off.

 

Honestly, I'm having a lot more fun with memory right now.  The IMC on these Skylake-X chips are amazing compared to previous Intel HEDT chips.  I've already passed a 400% memory test on a memory configuration that brought my latency down to 52ns using stock RAM, VCCIO and VCCSA voltages.  Now I'm testing one at 51ns.  It's looking like this IMC is going to let me take this 3600 MHz memory kit into the 40ns range soon.  That is nuts for a quad channel setup.

 

 

RAM stability test was done at stock CPU clocks and voltages so disregard the temps, which weren't horrible anyways.  

 

 

[uplink777]

@done12many2 holy moses, got lost at memory overclocking :D. Never did that myself, not really. I always go just use UEFi default XMP profile and go right for the CPU. Hit the vcore with an adaptive offset and that's about that, set some basic increase in vcore, link all cores, increase or decrease the MP, turn off all the vendor OC enhancements and increment Turbo Boost vcore by 0.015 with each try and that's about my "expertise" in OC :).

Anyway, does the memory OC brings any real life performance enhancements in games/productivity [Adobe Creative Cloud, etc.], or just at server tasks, latency dependant specific programs. Or You do it just for fun?

 

And another one. The latency You mentioned. Is it the one in between the time when iMC tells memory module which column to access and the arrival of signal on RAM pin [CAS]. How exactly do You manage to run it down? BiOS/UEFi settings? You mentioned some repeated runs of some test, which brings it down, or? Or it's not the CAS? [CL]. Sorry, I'm kind of new to memory OC, and CPU OC as well, since I only use very basic methods.

[Drak3]

@done12many2 Talk more about your experience with X299 so far. The Corsair Lighting Node Pro I just got to replace my HUE+ is also a piece of shit and is REALLY pissing me off on top of what my week's been so far, and I need some cheering up.

[done12many2]

22 minutes ago, uplink777 said:

@done12many2 holy moses, got lost at memory overclocking :D. Never did that myself, not really. I always go just use UEFi default XMP profile and go right for the CPU. Hit the vcore with an adaptive offset and that's about that, set some basic increase in vcore, link all cores, increase or decrease the MP, turn off all the vendor OC enhancements and increment Turbo Boost vcore by 0.015 with each try and that's about my "expertise" in OC :).

Anyway, does the memory OC brings any real life performance enhancements in games/productivity [Adobe Creative Cloud, etc.], or just at server tasks, latency dependant specific programs. Or You do it just for fun?

 

And another one. The latency You mentioned. Is it the one in between the time when iMC tells memory module which column to access and the arrival of signal on RAM pin [CAS]. How exactly do You manage to run it down? BiOS/UEFi settings? You mentioned some repeated runs of some test, which brings it down, or? Or it's not the CAS? [CL]. Sorry, I'm kind of new to memory OC, and CPU OC as well, since I only use very basic methods.

 

There's nothing wrong with setting XMP and going with it.  I would encourage you to overclock in this order.  CPU core, memory and lastly CPU cache.  CPU core trumps all with regards to overall performance so you shouldn't allow an overclock in the latter two areas to interfere with you hitting the highest clock speed possible.  

 

Once you've determined the maximum stable CPU clock, then start messing with memory and CPU cache.  Since you already know that your CPU core is stable, if you start having issues, you know where to focus your attention.  Keep in mind that as you increase memory and cache performance you may need to add more Vcore to what was a stable CPU overclock.  This is due to the fact that you are now pushing more data over the cores and increasing load.  More load usually equals more Vcore.  

 

As far as AIDA64, about the only thing I like about it is the Cache & Memory Benchmark and the cache stress testing.  Other than that, I don't really use it for stability testing other than cache.  

 

The latency that I was talking about is within the Cache & Memory Benchmark.  It measure the read delay from the time that a read command is issue to it hitting the CPU ready for work.  The faster (lower latency) this happens, the less cycles on the CPU that  are wasted.  

 

Yes, memory tweaking helps a great deal.  Some applications will favor a lot of bandwidth while others favor tighter timings with lower latency.  Why not have both if you can? 

 

12 minutes ago, Drak3 said:

@done12many2 Talk more about your experience with X299 so far. The Corsair Lighting Node Pro I just got to replace my HUE+ is also a piece of shit and is REALLY pissing me off on top of what my week's been so far, and I need some cheering up.

 

How about something as simple as POST?  Talk about starting right from the beginning.  POST is as fast as it is with z270.  X99 is notorious for very long POST times and that is now a thing of the past.  If you set up a fast boot with x299, it will boot up in seconds like a z270 platform.  POST / booting is the only thing that I don't love about my 5960x/x99 setup.  Switching gear a bit.  Anyone that says that x99 is not longer relevant is a fucking idiot.  My x99 rig is still a beast!

 

There's more, but I need to keep you coming back for the good stuff.  

[Drak3]

1 minute ago, done12many2 said:

X99 is notorious for very long POST times and that is now a thing of the past.

You don't need to tell me. Between the 120GB 850 EVO and 256GB 950 Pro, my X99 takes approx. 20 seconds from power on to login screen, Windows 7 or 10. In contrast, the old 6800K with a WD Green 5200RPM spinner takes 18-22 seconds from power on to login screen, Windows 8, 8.1, and 10. XP was a different story.

 

4 minutes ago, done12many2 said:

As far as AIDA64, about the only thing I like about it is the Cache & Memory Benchmark and the cache stress testing.  Other than that, I don't really use it for stability testing other than cache.

I used AIDA to validate my lazy and conservative Game Boost (tm) OC when I first got the system. Passed everything Aida could throw at it, crashed installing Firefox. I don't trust it much anymore.

[VagabondWraith]

3 hours ago, done12many2 said:

 

I'm able to run 4.8 GHz stable (no delid), but I can't figure out how to completely remove the AVX offset on this damn motherboard.  The best I can do is reduce it to 100 MHz, which means that under any AVX load it's bouncing between 4.7 and 4.8 GHz.  Things can get a bit hot (~80c) during some Handbrake runs using H.265 so I'm definitely wanting to get my hands on the DDM-X to chop the top off.

 

Honestly, I'm having a lot more fun with memory right now.  The IMC on these Skylake-X chips are amazing compared to previous Intel HEDT chips.  I've already passed a 400% memory test on a memory configuration that brought my latency down to 52ns using stock RAM, VCCIO and VCCSA voltages.  Now I'm testing one at 51ns.  It's looking like this IMC is going to let me take this 3600 MHz memory kit into the 40ns range soon.  That is nuts for a quad channel setup.

 

 

RAM stability test was done at stock CPU clocks and voltages so disregard the temps, which weren't horrible anyways.  

 

 

Holy sh*t.  @MageTank will love those numbers. I've gotta pick up some nice quad channel b-die ram soon. ?

[done12many2]

27 minutes ago, VagabondWraith said:

Holy sh*t.  @MageTank will love those numbers. I've gotta pick up some nice quad channel b-die ram soon. ?

 

Do you want to hear the funny part? 

 

It's not even a kit certified for quad channel use and this beast of an IMC is running it like a boss. It's a 4 DIMM dual channel kit for z270.  Yet another Skylake-X fail.

[VagabondWraith]

1 minute ago, done12many2 said:

 

Do you want to hear the funny part? 

 

It's not even a kit certified for quad channel use this beast of an IMC is running it like a boss. Yet another Skylake-X fail.

Wonder if I can get mine OC'd to ~ 3600 with good timings. Never OC'd RAM before, but also never OC'd a CPU til I got my new sexy beast. From reading online, my kit has Samsung E-die RAM and some people have had luck getting more out of it.

[done12many2]

Just now, VagabondWraith said:

Wonder if I can get mine OC'd to ~ 3600 with good timings. Never OC'd RAM before, but also never OC'd a CPU til I got my new sexy beast. From reading online, my kit has Samsung E-die RAM and some people have had luck getting more out of it.

 

3600 should be a walk in the park.  There's 4300+ RAM already certified by Gigabyte for my board and they know more about this IMC than I do.  There's going to be a lot of sweet memory overclocking with x299.

 

I haven't had a chance to play with E-die yet.  All my kits are B-die. Keep me posted as I'm very interested in it. 

[MageTank]

24 minutes ago, done12many2 said:

 

Do you want to hear the funny part? 

 

It's not even a kit certified for quad channel use and this beast of an IMC is running it like a boss. It's a 4 DIMM dual channel kit for z270.  Yet another Skylake-X fail.

Technically speaking, "quad channel certification" just means the kits are cut from the same cloth, and offer the highest chance of compatibility. The dies are the same, they were from the same fab/stepping, and are likely going to hit similar limitations. I've actually overclocked both Samsung and Hynix kits on the exact same board, together, and still achieved some insane results when compared to XMP. What honestly impresses me the most, is the results you are getting on a board that isn't exactly designed with memory overclocking at the forefront. 

 

If this is what you are capable of on a "lackluster" memory overclocking board, imagine what an Apex will look like in your hands. Remember that goal I set for you? I know for certain the Apex will achieve it, lol. 

[Taf the Ghost]

6 hours ago, done12many2 said:

 

3600 should be a walk in the park.  There's 4300+ RAM already certified by Gigabyte for my board and they know more about this IMC than I do.  There's going to be a lot of sweet memory overclocking with x299.

 

I haven't had a chance to play with E-die yet.  All my kits are B-die. Keep me posted as I'm very interested in it. 

Do we know where the scaling falls off on DDR4 memory? Eventually the speed gets high enough that other bottlenecks get in the way.  It's a new platform with new CPUs, so I imagine it'll take some time to find the limits.  I'm just curious.

[uplink777]

14 hours ago, done12many2 said:

 

There's nothing wrong with setting XMP and going with it.  I would encourage you to overclock in this order.  CPU core, memory and lastly CPU cache.  CPU core trumps all with regards to overall performance so you shouldn't allow an overclock in the latter two areas to interfere with you hitting the highest clock speed possible.  

 

Once you've determined the maximum stable CPU clock, then start messing with memory and CPU cache.  Since you already know that your CPU core is stable, if you start having issues, you know where to focus your attention.  Keep in mind that as you increase memory and cache performance you may need to add more Vcore to what was a stable CPU overclock.  This is due to the fact that you are now pushing more data over the cores and increasing load.  More load usually equals more Vcore.  

 

As far as AIDA64, about the only thing I like about it is the Cache & Memory Benchmark and the cache stress testing.  Other than that, I don't really use it for stability testing other than cache.  

 

The latency that I was talking about is within the Cache & Memory Benchmark.  It measure the read delay from the time that a read command is issue to it hitting the CPU ready for work.  The faster (lower latency) this happens, the less cycles on the CPU that  are wasted.  

 

Yes, memory tweaking helps a great deal.  Some applications will favor a lot of bandwidth while others favor tighter timings with lower latency.  Why not have both if you can? 

Ah, I see. Well, I never got beyond CPU core, it's good to know I can pump up even CPU cache and mem as well. Well, I'll take some time to digest these information, and I'll start as You suggested, with CPU cache, right after I determine the CPU stable clock + it's befitting vCore.

Thank You @done12many2 so much for these informations and explanation. And yes, when You can have both, why not? That's right. I'll finally put those thermal sensors attached to Dominators to at least some use.

And one more question, if I may. What's the safe temp for Corsair Dominator DDR4 2400 MHz? Where can I push them? Running them at XMP profile [the less extreme one - called just XMP 1 on my X99 Deluxe II. TBH I have yet no idea how far it pushes them] is pushing them to about 45-60°C.

You seem kind of knowledgeable, which is rare nowadays. Well, at least from what I read so far in some of Your submissions here at ltt forums. It seems like You have the experience . I'll be off now for 1-2 days, still waiting for case [am moving from Fractal Define R5 Blackout to Be Quiet! Base 900 Pro with glass, due to better clearance for EK-XLC Predator 360 on top, due to incredible modularity of this case. I saw 3 vids doing review yesterday, You can actually move the mobo tray up and down up to 6 cm, which is an effing dealbreaker for me :). I just have to look away from the incredible annoying PSU installation regarding this case :D]

[Notional]

 

Der8auer posted an update. The point of blaming his PSU is moot as he tested with 3 different PSU. He also managed to recreate the issue with TTL.

[zMeul]

13 minutes ago, Notional said:

 

so his reporting sounds like  utter bullcrap because no one in the right mind will run very expensive CPUs over TjMax for long periods of time

do some boards have issues, sure, but putting extreme OCing in the same basket as typical use scenarios is utter crap

[Notional]

Just now, zMeul said:

so his reporting sounds like  utter bullcrap because no one in the right mind will run very expensive CPUs over TjMax for long periods of time

It's not the CPU's overheating, but the VRM. Mikrothrottling on the CPU is primarily due to the mud Intel uses as thermal grease inside the heat shield, as well as a poor cooler on the CPU. If people can afford such a CPU, they can afford a proper cooling solution.

[done12many2]

32 minutes ago, zMeul said:

so his reporting sounds like  utter bullcrap because no one in the right mind will run very expensive CPUs over TjMax for long periods of time

do some boards have issues, sure, but putting extreme OCing in the same basket as typical use scenarios is utter crap

 

This is exactly it.  What he's doing is he's trying to force the maximum power draw through heat.  As temps increase, so do VCore requirements.  It just becomes a run away effect at some point well before what he's actually doing.

 

If everyone tests just like they have in the past this wouldn't even be a talking point.  Der8bauer just found a way to exaggerate testing.  

 

I have never increased my CPU's thermal limits in BIOS.  Don't plan on starting now just to see if I can get a VRM to throttle.  Another no shit moment.  

 

It's nothing, but an extreme overclocker exposing weaknesses on cheap x299 boards.

 

Specifically at this time stamped portion of the video, he goes into further detail about the VRMs being good, but the cooling for those VRMs being poor.  He compares the better VRM cooling of the Asus Maximus z170 board to the VRM cooling of the cheap x299 boards that he chose for the video "disaster".  

 

 

I have a z170 board with a cheap ass VRM cooler on it and it will throttle with high clocks on 6700K and 7700K CPUs.  Does that mean that the VRMs on the entire platform are a disaster?  Fuck no.  It means I bought a cheap ass board and I got what I paid for.

 

The Gigabyte x299 board that he chose to use happens to be the one with the weakest VRM cooler in the Gaming line up.  Why?  I suspect the same reason he used the cheapest boards from the other manufacturers lines.  It allows him to create the "disaster" he needed for his title.  

 

Go one step higher on Gigabyte's Gaming line and you get a much better cooler.  

 

This is nothing new.  We're just taking notice of this shit now because someone created a "disaster" for the sake of clickbait.  

 

Gaming 3:

 

\

 

Gaming 7:

 

 

 

[zMeul]

20 minutes ago, done12many2 said:

This is exactly it.  What he's doing is he's trying to force the maximum power draw through heat.  As temps increase, so do VCore requirements.  It just becomes a run away effect at some point well before what he's actually doing.

 

If everyone tests just like they have in the past this wouldn't even be a talking point.  Der8bauer just found a way to exaggerate testing.  

 

I have never increased my CPU's thermal limits in BIOS.  Don't plan on starting now just to see if I can get a VRM to throttle.  Another no shit moment.  

 

Nothing, but an extreme overclocker exposing weaknesses on cheap x299 boards.  No shit.

what he's doing is utter pointless for the consumer, he only does this for extreme OCing for what 1H?

it's like buying a 10000HP car and driving it in a 30Km/H zone

[done12many2]

1 minute ago, zMeul said:

what he's doing is utter pointless for the consumer, he only does this for extreme OCing for what 1H?

it's like buying a 10000HP car and driving it in a 30Km/H zone

 

What he's doing is creating "the sky's falling" titles and topics for the sake of views.  Really sad.

 

Like I said, you can buy the cheapest of boards from any platform and they will have VRMs that you can get to throttle due to poor cooling.  This is nothing new.  

[Drak3]

24 minutes ago, done12many2 said:

What he's doing is creating "the sky's falling"

So he's Chicken Little?

[MageTank]

44 minutes ago, done12many2 said:

 

This is exactly it.  What he's doing is he's trying to force the maximum power draw through heat.  As temps increase, so do VCore requirements.  It just becomes a run away effect at some point well before what he's actually doing.

 

If everyone tests just like they have in the past this wouldn't even be a talking point.  Der8bauer just found a way to exaggerate testing.  

 

I have never increased my CPU's thermal limits in BIOS.  Don't plan on starting now just to see if I can get a VRM to throttle.  Another no shit moment.  

 

It's nothing, but an extreme overclocker exposing weaknesses on cheap x299 boards.

 

Specifically at this time stamped portion of the video, he goes into further detail about the VRMs being good, but the cooling for those VRMs being poor.  He compares the better VRM cooling of the Asus Maximus z170 board to the VRM cooling of the cheap x299 boards that he chose for the video "disaster".  

 

 

I have a z170 board with a cheap ass VRM cooler on it and it will throttle with high clocks on 6700K and 7700K CPUs.  Does that mean that the VRMs on the entire platform are a disaster?  Fuck no.  It means I bought a cheap ass board and I got what I paid for.

 

The Gigabyte x299 board that he chose to use happens to be the one with the weakest VRM cooler in the Gaming line up.  Why?  I suspect the same reason he used the cheapest boards from the other manufacturers lines.  It allows him to create the "disaster" he needed for his title.  

 

Go one step higher on Gigabyte's Gaming line and you get a much better cooler.  

 

This is nothing new.  We're just taking notice of this shit now because someone created a "disaster" for the sake of clickbait.  

 

Gaming 3:

\

 

Gaming 7:

 

 

 

 

I disagree with this entirely. While it is true, he is blowing a lot of it out of proportion, his intent still seems noble. His comparison of the Z170 Maximus heatsink to the X299 Strix is a valid one to make, as both are part of the ASUS ROG premium lineup and both were (hopefully) designed around the product stack intended to be used in them. The problem he sees, is that the smaller platform got a more robust cooling solution than the larger platform. He isn't trying to compare strong against cheap, but rather point out that what is on the bigger boards, isn't exactly enough for his standards.

 

What I found the most interesting about the video, isn't the whole "these boards will blow up!" rhetoric that people thinks he is spewing, but rather the fact that the throttling has gone unnoticed by most. Specifically the VCCIN throttle, which nobody mentioned before, which could have a pretty substantial impact on actual power draw metrics. When you are drawing far less power than you think you are, under what appears to be significant stress, it potentially misleads people into buying inferior power supplies and cooling solutions. Pointing out that "hey, you can potentially draw even more power if you remove this limitation" is fair warning in my eyes, as someone could stumble upon that accidentally when it's far too late. 

 

I understand you like that board of yours, but just because the higher end product stack is good, doesn't mean the rest is above criticism. It would be one thing, if he said "all gigabyte boards from this lineup are bad", but I didn't see him say that. Pointing out a potential flaw in people's testing methodology isn't him trying to profit off disaster. To people like myself, that actually explore the many various ways to stress ones system, it seems that he is trying to make sure people are doing so correctly. 5ghz Prime95 with a -500 AVX offset, isn't stressing 5ghz on Prime95. Just like 5ghz Prime95, rapidly throttling down to 1.2ghz and back to 5ghz, isn't stressing Prime95 at 5ghz. 

 

I said it before, and I'll say it again, the thermals he saw wasn't dangerous (aside from the warm 8 pin), and it really is being blown out of proportion, but it does not make what he says any less true, or insightful for those that need to know.