The Skylake-X Mess Explored: Thermal Paste And Runaway Power

[MageTank]

Quote

Under normal operating conditions, the CPU, and not the motherboard, always throttles first.

Gee, it's like I have been saying that on this very forum, in this very section, ever since Der8auer launched that video. Was unaware we needed Tom's to make a review in order for that to be understood, lol.

 

The thermal paste used on these CPU's is not bad, and it's once again designed with longevity > performance. As you all can now see, their application method has improved drastically to the point that per-core variance is at an all time low, even on these CPU's with nearly a dozen cores. That should finally put the old argument to rest about the paste itself causing said variance, when the rest of us knew it was the gluing application.

 

As for whether or not these should have been soldered, that point has been beaten to death already. As much as I would have enjoyed seeing these chips soldered, so that your general consumer didn't have to delid these chips to get respectable thermal performance out of them, some enthusiasts actually prefer the ability to delid and use liquid metal (though those enthusiasts are few and far between). 

 

So, to recap: If you wish to use the enthusiast plaform, get a real motherboard and a real cooling solution. No more $200 boards and 212 evo's people. 

[done12many2]

6 minutes ago, PCGuy_5960 said:

Well, they completely lost the silicon lottery

 

People are going to believe what they want to believe.  Most people talking about Skylake-X will never own one or a Threadripper rig for that matter.

 

Most reviewers have a very basic understanding of overclocking, but since they have a channel, we believe everything they tell us.  

 

 

[MageTank]

1 minute ago, done12many2 said:

 

People are going to believe what they want to believe.  Most people talking about Skylake-X will never own one or a Threadripper rig for that matter.

 

 

 

Hey, I imagine I'd be able to afford one in a decade or two. I'll just be a little tardy to the party, that's all. 

[Valentyn]

Just now, MageTank said:

Hey, I imagine I'd be able to afford one in a decade or two. I'll just be a little tardy to the party, that's all. 

At least by then DDR4 prices will be decent and you can try running some 4266Mhz CL16 modules in Skylake X or Threadripper to see how it goes.

[MageTank]

Just now, Valentyn said:

At least by then DDR4 prices will be decent and you can try running some 4266Mhz CL16 modules in Skylake X or Threadripper to see how it goes.

I prefer my 3200 kit. For some reason, both @done12many2 and I both have better luck with these original B-Die kits than we do with higher factory binned kits. Mine can boot at 4000 on an ITX board. With the Z270 Apex, I won't settle for less than 4000 C15-15-15-30-2. 60GB/s bandwidth at sub-35ns. Mark my words, that's going to be a thing that happens, lol. 

[Valentyn]

1 minute ago, MageTank said:

I prefer my 3200 kit. For some reason, both @done12many2 and I both have better luck with these original B-Die kits than we do with higher factory binned kits. Mine can boot at 4000 on an ITX board. With the Z270 Apex, I won't settle for less than 4000 C15-15-15-30-2. 60GB/s bandwidth at sub-35ns. Mark my words, that's going to be a thing that happens, lol. 

I ordered two 4266Mhz G.Skill 16GB kits today. Will see if they first work in quad channel in my x99 deluxe.

Then likely use them in a Threadripper or Skylake X system to see what they can do. Saw some folks at overclock.net getting them at 3466Mhz on Ryzen.
Although considering I'm using launch week 2400Mhz DDR4 it'll all be an upgrade if it works well.

 

4000 C15-15-15-30-2 though....mental stuff!

[done12many2]

13 minutes ago, MageTank said:

I prefer my 3200 kit. For some reason, both @done12many2 and I both have better luck with these original B-Die kits than we do with higher factory binned kits. Mine can boot at 4000 on an ITX board. With the Z270 Apex, I won't settle for less than 4000 C15-15-15-30-2. 60GB/s bandwidth at sub-35ns. Mark my words, that's going to be a thing that happens, lol. 

 

Absolutely agreed.  There's a great deal of marketing involved in any G.Skill kit about the very well-know 3200 c14 and 3600 c15 (hard to find).  

 

3200 MHz c14 kit @ 3600 c15 w/ tweaks  LATENCY BABY!!!

 

 

 

3600 MHz c16 kit @ 3600 c16 w/ tweaks AND faster cache.  Will retest 3200 c14 kit with the faster cache as well. 

 

 

9 minutes ago, Valentyn said:

I ordered two 4266Mhz G.Skill 16GB kits today. Will see if they first work in quad channel in my x99 deluxe.

Then likely use them in a Threadripper or Skylake X system to see what they can do. Saw some folks at overclock.net getting them at 3466Mhz on Ryzen.
Although considering I'm using launch week 2400Mhz DDR4 it'll all be an upgrade if it works well.

 

4000 C15-15-15-30-2 though....mental stuff!

 

Unfortunately, the IMC on Haswell-E is what's going to be your limitation.  Most are stuck at 3200 with an exceptional few pulling off just over 3300 MHz with BLCK tweaking.  Broadwell-E is a bit better going upwards or 3400 MHz with stock BLCK and upwards of 3500 to 3600 MHz with some BLCK tweaking.

 

I'm sure that you are aware that 2 separate kits is usually a disaster for higher clock x99 memory overclocking.

 

16 minutes ago, MageTank said:

Hey, I imagine I'd be able to afford one in a decade or two. I'll just be a little tardy to the party, that's all. 

 

The difference between you and most of the people doing the talking is that you will actually make whatever you have run to it's fullest potential.  I find that most people in this forum know very little about pushing the envelope, but consider their knowledge level just enough to hype or discredit platforms at will.  We're smart enough to be dangerous to others.  

 

 

[Valentyn]

11 minutes ago, done12many2 said:

 

Absolutely agreed.  There's a great deal of marketing involved in any G.Skill kit about the very well-know 3200 c14 and 3600 c15 (hard to find).  

 

3200 MHz c14 kit @ 3600 c15 w/ tweaks

 

3600 MHz c16 kit @ 3600 c16 w/ tweaks

 

Unfortunately, the IMC on Haswell-E is what's going to be your limitation.  Most are stuck at 3200 with an exceptional few pulling off just over 3300 MHz with BLCK tweaking.  Broadwell-E is a bit better going upwards or 3400 MHz with stock BLCK and upwards of 3500 to 3600 MHz with some BLCK tweaking.

 

 

Oh I know that. Just want to get it to 3200 for now hopefully, and these were the same price as 3200-3600mhz kits so figured why not get them now, and then use them on Skylake X or Threadripper if they do work in quadchannel for me.

 

I'm hoping they'll play nice in quad considering it's technically two dual channel kits, but if not. All the way back to Germany they go.

 

[done12many2]

Just now, Valentyn said:

 

Oh I know tha. Just want to get it to 3200 for now hopefully, and these were the same price as 3200-3600mhz kits so figured why not get them now, and then use them on Skylake X or Threadripper if they do work in quadchannel for me.

 

I'm hoping they'll play nice in quad considering it's technically two dual channel kits, but if not. All the way back to Germany they go.

 

I like your can do attitude bud!  Let me know if I can be of any help as I've done quite a bit of tweaking with x99 quad channel.  Good luck man. 

[Valentyn]

2 minutes ago, done12many2 said:

 

I like your can do attitude bud!  Let me know if I can be of any help as I've done quite a bit of tweaking with x99 quad channel.  Good luck man. 

Yeah thanks mate!

Mindfactory.de are a sound bunch in Germany. 14 Days no hassle returns if you don't like an item, or have issues. Hopefully no issues with them though.

 

Also seen you over at overlock.net in the Skylake X thread. Been pursuing it all see what people are managing with the systems.

[TeoIzAwezome]

3 hours ago, Hunter259 said:

A few minutes aren't insignificant if you are doing rendering on a daily basis. 2 minutes per render, once a day every day for a year is 12.17 HOURS. That's a lot of time.

Then you get a Xeon/Opteron/Epyc/some ARM system.

[trag1c]

3 hours ago, Princess Cadence said:

You need something that does the workload as reliable as possible, finish a render a few minutes faster for instance hardly is enough reason to pay almost the double on a platform.

You would be surprised. That extra hardware cost can save you a ton of money by making your employee's much more productive. Especially in a field like game development where at a well respected studio your employee's are typically payed well above $100k USD to help retain talent.

 

For example say every business day (also assuming no holidays) you spend 1 hour performing a computational task. That works out to 260 lost hours. Assuming that person is making a $125k/yr (5 days a week at 8 hours a day) that works out to $60.09 an hour. Now you just wasted 260 * $60.09 = $15623.40 over the course of a year on one employee. Now say you spend $5000 on each employee's workstation but you cut that render time in half. That made the company more productive by 130 hours and also saved (15623.4 / 2 = 7811.70) $7811.70 - $5000 = $2811.70 per employee. Now if you have 100 employee's with similar tasks and pay that works out to $281k in savings for the company and 13000 more man hours available.

 

There is more than enough reason to pay for the much faster equipment for employee's.

[knightslugger]

3 minutes ago, trag1c said:

You would be surprised. That extra hardware cost can save you a ton of money by making your employee's much more productive. Especially in a field like game development where at a well respected studio your employee's are typically payed well above $100k USD to help retain talent.

 

For example say every business day (also assuming no holidays) you spend 1 hour performing a computational task. That works out to 260 lost hours. Assuming that person is making a $125k/yr (5 days a week at 8 hours a day) that works out to $60.09 an hour. Now you just wasted 260 * $60.09 = $15623.40 over the course of a year on one employee. Now say you spend $5000 on each employee's workstation but you cut that render time in half. That made the company more productive by 130 hours and also saved (15623.4 / 2 = 7811.70) $7811.70 - $5000 = $2811.70 per employee. Now if you have 100 employee's with similar tasks and pay that works out to $281k in savings for the company and 13000 more man hours available.

 

There is more than enough reason to pay for the much faster equipment for employee's.

on the other hand, paying additional employees at maybe $70,000 average to keep those machines performing at those levels with the know how might actually be more expensive, along with replacing insanely expensive burned out hardware.

 

I'm quite certain Xeon workstations are the better choice for company overhead.

[cj09beira]

those people multitask a lot, so while the pc is rendering they are probably doing something else that needs to be done, so the real benefit might be much much smaller, which leads to its more complicated than that and very dependent on the type of work being done.

in the case that the pc gets pawned a render server might be the best option. 

 

[leadeater]

9 hours ago, It's me! said:

As it stands, even a custom water-cooling loop has to throw in the towel at 250W, long before most motherboard voltage converters hit their limits. Under normal operating conditions, the CPU, and not the motherboard, always throttles first.

Not my loop 

 

Only limit I'd hit is how much heat dissipation an EKWB R4BE monoblock can handle.

[done12many2]

5 minutes ago, leadeater said:

Not my loop 

 

Only limit I'd hit is how much heat dissipation an EKWB R4BE monoblock can handle.

 

Not my loop either.  I was pulling 600w to 625w from the wall (~500 CPU + pumps/fans/system) running some of the wreckless shit that @MageTank had me running to test the "x299 VRM disaster".  I was definitely on the warm side (80c +) during some of those tests, but far from dangerous temperature wise.  Current wise is a whole different discussion.

[Drak3]

6 minutes ago, done12many2 said:

 

Not my loop either.  I was pulling 600w to 625w from the wall (~500 CPU + pumps/fans/system) running some of the wreckless shit that @MageTank had me running to test the "x299 VRM disaster".  I was definitely on the warm side (80c +) during some of those tests, but far from dangerous temperature wise.  Current wise is a whole different discussion.

Inb4 someone throws enough rad space to run 'passively' with only a few pumps. I'd say I'd do it, but funds and space to put that monstrosity are working against me.

[leadeater]

23 minutes ago, Drak3 said:

Inb4 someone throws enough rad space to run 'passively' with only a few pumps. I'd say I'd do it, but funds and space to put that monstrosity are working against me.

I do run my system passive 

 

Edit:

While under load with dual 290X's

[Drak3]

Just now, leadeater said:

I do run my system passive 

I mean the 500W power draw on CPU alone behemoth that @done12many2 has, and manage both passive and a metaphorically frosty at everything except AVX load. And still not scorching under AVX load.

[NumLock21]

In other news Intel Skylake-SP NDA lifts tomorrow

[Drak3]

Just now, NumLock21 said:

In other news Intel Skylake-SP NDA lifts tomorrow

Who cares when we can shit on the platform that brought us the best value CPU yet? The 7980XE? Huh?

Huh?!

 

HUH?!?!

[NumLock21]

1 minute ago, Drak3 said:

Who cares when we can shit on the platform that brought us the best value CPU yet? The 7980XE? Huh?

Huh?!

 

HUH?!?!

7980XE is for boys!

Skylake-SP is where the mens are at.

[Drak3]

Just now, NumLock21 said:

7980XE is for peasants!

Skylake-SP is where the mens are at.

I can't destroy mainboards with SLSP though.

[NumLock21]

Just now, Drak3 said:

I can't destroy mainboards with SLSP though.

What is this foreign language "SLSP"?

 

[Drak3]

Just now, NumLock21 said:

What is this foreign language "SLSP"?

 

SkyLake-SP