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

[10tacle]

anyone else get the feeling that intel has been sitting on this architecture for a while trying to milk every last drop out of its last tick-tock-tock strategy? seems like intel was waiting on a time to actually need this current architecture to battle with amd.

Yes. Intel has been dragging their feet with every tick-tock since Sandy Bridge, which was without question a substantial leap over the first i-Core series (Bloomfield). For a memory refresher, here are a couple of productivity app benchmarks from Tom's from 2011. Note the large spread between the i7 950 and i7 2600K. As you can tell in these, even the first generation i-Core was a massive leap over the quad core Core 2 series.

https://img.purch.com/o/aHR0cDovL21lZGlhLmJlc3RvZm1pY3JvLmNvbS81L1UvMjc0OTYyL29yaWdpbmFsL0hhbmRCcmFrZS5wbmc=
https://img.purch.com/o/aHR0cDovL21lZGlhLmJlc3RvZm1pY3JvLmNvbS82L0ovMjc0OTg3L29yaWdpbmFsL1dpblppcC0xNC5wbmc=

 

[icycool]

There seems to be an awful lot of anger in this thread.

Is it because Intel aren't living up to their own hype, or because they might not be living up to AMD's Threadripper threat?
These are complex waters to navigate and everyone has valid points and questions according to how they use their own system.
From what I can glean from this thread is that.
i9-7900X needs water cooling as a bare minimum.
i9-7900X with stock clocks and water cooling will use considerably more than the TDP under a stress test (occt, prime95, burn-in X).
i7-7820X is 20W less than the 10 core, but who knows how that fares in actual testing.

I currently have an i7-4930K, one of the few newish models with solder, disabled HT for games.
Not owned an AMD since Athlon 800, the P4 3.2HT after that was so much better.

Personally I was interested in the i7-7820X, with a Gigabyte gaming 7 or 9 mobo (they don't seem to have any professional looking ones anymore .
i7-7700K wouldn't be such a bad thing. High clocks for games, but not much in the way of multitasking with HT disabled (game core affinity stuff) and lack of ram slots/enhanced options.

The Ryzen, 8 physical cores is all I should ever need, but sadly not so great in games.

Alternatively, I could save lots of money and just try to OC the 4930K + Corsair h110i, increase vCore I guess?

The X79-UD4 mobo is getting a little old, still works great, but I feel it's struggling to keep up 2 GFX cards and a 144Hz screen.

First world problems, I know, but every time I get a new comp every few years, it does seem a bit quicker, less latency like..

Conclusions & Predictions.

Intel were blindsided by AMD, so they chopped down a Xeon, went cheap on the cooling and don't know if Mesh is already on Xeon, but did that too.
This resulted in a product that goes well if you can keep its thermals under control and don't do any heavy benchmarking.

The use of thermal paste on top-end models, and lets call an 8core consumer model top end for Intel, is disturbing and depressing, especially when you look at the price tag $859 in OZ.
And don't forget the reduced 28 lane PCI-E bus.
i9-12core+ will likely have solder since the results show that even with 10 cores, cooling is iffy.
AMD's Threadripper should run cooler regardless given the bigger surface area, even if they go stingy and use TIM and be cheaper while doing it, at the cost of a small percentage differential.

It's a good day when you can say you have had enough of Intels price gouging and now sub-standard solution in favour AMD who have been trying for years and now actually look like the better product.

I might try overclocking in the meantime and wait for the chips to fall where they may from the blue and the red team, but to be honest, just like the last Intel generation (7700K aside), this one doesn't seem to offer much but pain for 2 extra cores (even the X299 chipset seems outdated).

 

[Hooride]

6Ghz 7900x
https://youtu.be/KdD_QJ95chY?t=1m59s

 

[JamesSneed]

Icycool, I just picked up a Ryzen 1800x after waiting on and reading the reviews of the 7820x. Unless you want very high FPS at 1080p the Ryzen R7's are pretty awesome. Something about having 8-cores, makes games feel smooth as silk. I think its even more than 1% lows etc but anyhow the experience on an 8-core machine with gaming is just lovely.

 

[Karadjgne]

Would have been nice if dude had actually explained a little more, that video basically slammed tom's as idiots who don't know what they are talking about. Vast majority of even 'enthusiasts' don't run LN2 systems, and thats what it takes to get anything close to 6.01GHz. On any cpu basically.

 

[the nerd 389]

Would have been nice if dude had actually explained a little more, that video basically slammed tom's as idiots who don't know what they are talking about. Vast majority of even 'enthusiasts' don't run LN2 systems, and thats what it takes to get anything close to 6.01GHz. On any cpu basically.

While everything you've said is true, it's also worth mentioning that the CPU probably lasted a few minutes at that speed before going belly-up. That's par for the course with LN2.

 

[icycool]

Icycool, I just picked up a Ryzen 1800x after waiting on and reading the reviews of the 7820x. Unless you want very high FPS at 1080p the Ryzen R7's are pretty awesome. Something about having 8-cores, makes games feel smooth as silk. I think its even more than 1% lows etc but anyhow the experience on an 8-core machine with gaming is just lovely.

James, I have a 144Hz screen, and before getting a second video card (primarily for mining), the screen was the limiting factor, but with the second 1080Ti, now the CPU appears to have an impact. SLI scaling/gsync/microstutter strangeness are all new to me, but nVidia inspector seems like it has potential - or my CPU really is causing dramas.

The red team is certainly looking good and it's interesting you say that 8 cores helps, I was still under the impression games mainly used four
I might look into IPC between my old 4930 and the newer chips to see if overclocking will at least match up the speed to what's now available. The downside to OC is stability and power draw, but with the new Intel bleh stuff, power doesn't seem to be a concern anymore

As a side note, ETH mining seems like a waste of time, and I managed to blow up a PSU by overclocking my GPU ram and underclocking the core.. no idea what happened. I was just in the middle of investigating how to unlock the core voltage control in MSI Afterburner and BANG!

Friend sent this to me.. https://i.redd.it/rkwou70jut8z.jpg I was not impressed but still laughed anyway

 

[Adroid]

Quite apparent this is a blatant failure of a chip. The "response" to their only competition in the world putting out something decent for the first time in years... Been producing chips that run hotter than they need to since ivy bridge.

Thermal paste not working very good??? Let's try more of it for a few more years on end... Please stop the if it's half broken lets not fix it mentality.

How about some innovation Intel? Make something that runs cooler. Or something that runs faster. Or something designed for a desktop instead of a laptop. Or something noticeable better than the last generation that doesn't just focus on reducing footprint. Most of us can afford another few pennies a day for a processor designed for a desktop. Just don't make it a space heater. Increase the footprint a little and produce something that runs fast, stable, and cooler. Wow us please? It's been too long.

 

[ElMojoMikeo]

Oh dear I see people are getting into right old state on this. I have posted somewhere above these issues are not new. The faster you clock the more energy you need to maintain the higher clock rates. The price is heat. Intel can quite rightly turn around and say that these are professional level processors that gamers love to overclock for fun. How you cool the raging beast is your problem. I think we should be looking to a future that will require a step up in cooling generally as the heat generated is now stretching our current levels of tech/cost. Buy a 1,500 processor and stick a 200 cooler on it. Really, I think those days have gone. Science has been looking at this heat thing for a long time and simply will not go away.
I have been in this situation for years with my old X79 based 3970X you have to balance out the energy going in, to the capability of the cooler in maintaining heat levels below throttling. That's it there is no magic bullet. More expensive cooling technology is now going to very high on our list of things for future builds.

 

[John_561]

Yeah, well, there's enthusiast workarounds here. First and foremost a fast and easy delid with a rockit, then some metal ultra. We'll be pushing those things further and faster up to the wall like we always do, finding their limits. They're monster chips. Treat em like monsters and they'll be fine. YOU DON'T LEAVE A MONSTER ON A THERMAL LEASH, DO YOU?!

 

[ElMojoMikeo]

That's the problem with how us overclockers sometimes think. Intel make these processors just for us to play with. Wrong, they are really powerful processors used in many applications. True sometimes depending on the weather Intel make a little nudge in our direction. But only we are mad enough to play for fun or from necessity. If overclocking is done based on good knowledge it is safe and rewarding. I would rather not have to overclock. So with the right cooling a fast processor without overclocking capability that knocks my current overclocked setup into the sticks is sound thinking from Intel. Until now I have had no reason to change.

 

[Karadjgne]

Delidding a $100 cpu is one thing, delidding a $1500 cpu is another matter entirely. If you have kind of money to potentially throw away, be my guest, but most of us are struggling just to justify the initial expenditures in the first place.

 

[dudmont]

You wouldn't see me de-lidding and do the LMU treatment(I tried it once with a cheapy haswell celeron, epic fail). As Dirty Harry says, "a man's got to know his limitations". I think though, there might be a market for someone to explore on selling de-lidded and LMU jobs on I9s, if you were good enough. Word would get around that xperson was the person to talk to about an I9 that was "fixed". What to charge would be the issue, how do you factor in the issue of failures?

 

[ElMojoMikeo]

Exactly my point. Because I don't see any really smart coolers out there that I would trust my current setup with I have had to overclock four 3970X bit by bit knowing that what I am doing is right before doing it. Lots of research and time. However I do enjoy it too. However, $1500 is a lot so I thought I should add my experiences on dealing with it. I should say that I am using H100i that I found best at the time. But not even they maxed out could keep the temps in range under full load using Prime95. Having said that we are into code breaking type code here so massively processor intensive. So if your doing this all day long then you can afford the cost of keeping it cool. But in reality many applications don't require that level of cooling so I balanced on the H100i only because there is no closed loop predictive system that is affordable.

 

[JamesSneed]

Icycool, I just picked up a Ryzen 1800x after waiting on and reading the reviews of the 7820x. Unless you want very high FPS at 1080p the Ryzen R7's are pretty awesome. Something about having 8-cores, makes games feel smooth as silk. I think its even more than 1% lows etc but anyhow the experience on an 8-core machine with gaming is just lovely.

James, I have a 144Hz screen, and before getting a second video card (primarily for mining), the screen was the limiting factor, but with the second 1080Ti, now the CPU appears to have an impact. SLI scaling/gsync/microstutter strangeness are all new to me, but nVidia inspector seems like it has potential - or my CPU really is causing dramas.

The red team is certainly looking good and it's interesting you say that 8 cores helps, I was still under the impression games mainly used four
I might look into IPC between my old 4930 and the newer chips to see if overclocking will at least match up the speed to what's now available. The downside to OC is stability and power draw, but with the new Intel bleh stuff, power doesn't seem to be a concern anymore

As a side note, ETH mining seems like a waste of time, and I managed to blow up a PSU by overclocking my GPU ram and underclocking the core.. no idea what happened. I was just in the middle of investigating how to unlock the core voltage control in MSI Afterburner and BANG!

Friend sent this to me.. https://i.redd.it/rkwou70jut8z.jpg I was not impressed but still laughed anyway

That image, funny, poor GPU.

Yeah if you google around you can find lots of people running 8-core systems AMD Ryzen or Intel and many do note on the smoothness(I started looking after I was noticing it on some games I have 300+ hours in). The FPS isn't much higher so I was wondering if others "felt" this way. I suspect having free CPU resources is attributing to more than just the 1% lows like many reviews are trying to point at. Since gaming involves mouse, keyboard, other OS things like interrupts etc, that stuff doesn't get drawn on screen per say. Someone pretty savvy at performance tuning on Windows could probably pinpoint where the waits come from, interrupts etc.

 

[JackNaylorPE]

Be nice if I could view the graphs and images but with each passing week, more and more images on THG are not viewable and despite numerous inquiries, no answer to date on why.

 

[JackNaylorPE]

What's wrong with using Prime? It does a good job of testing the thermal limits of a CPU. You wouldn't test the limits of a weight lifters strength with 5 pound dumb bells. You need to go all out.

Because it's like testing out the towing capacity of your light pickup truck to pull your 14' boat trailer by towing the Space Shuttle up the Rockies.

1. If you use the later versions you risk damage from the voltage boost that occurs when modern instruction sets are present.

2. if you use the older versions, you have proven that your OC is stable ... as long as modern instruction sets are not present.

3. Without any rhyme nor reason, you artificially limit the OC of the CPU. Yes, if you built your PC to get your name on overclocking site leader boards, then of course, P95 is an essential tool. But if you built your PC to "run programs" than the best tool to test overclock stability and your ability to get the maximum OC at the temperature you have decided is "safe", then testing with a suite of programs that re[resents what you will do what that PC is the "logical" test method.

If I can hit 4.7 Ghz at 75C with RoG Real Bench, and 4.6 Ghz at 79C, what is going to be my 24/7 OC ? It's going to be 4.7 GHs because RB is going to put a far greater load on the CPU than anything it will ever see again in its life time.

 

[ElMojoMikeo]

As JackNaylorPE has said and as I have said above Prime95 is a perfect test for loading all the cores. The reason that it does this is because there is little external I/O going on when these calculations are been made.

The cores don't get a break as you would in normal applications that read and write data to and from external storage devices. Even process and thread switching gives the cores a break.

So Prime95 is as bad as it will get for any processor. Even the most power hungry games will not come close to stressing the cores like Prime95. They might put more stress across the whole system but not no the processor.

The reason is that the cores will be mapped for graphics, sound, secondary storage and the main scheduler. So some cores will be more stressed than others, some processes on cores will handle the high latency user inputs again this all means you will never reach anywhere near the stress on the processor package. I have encountered people who think that the processor is only going at max if all the cores are at a 100%.

When in fact trying to get a core that has been mapped for high latency use any where near a 100% is obviously going to take a lot disk I/O and some very, very fast typing. To overclock properly you have to learn exactly what is happening only read the Intel, AMD and NVIDIA stuff . Don't believe everything on overclocking sites check it with Manufacture documentation. Better still, ask them for the references they used to do the overclock.

I say this because I have seen the most amazing amount of BS on them.
A good overclock is one that operates in the zone that Intel leave as safety headroom on all their processors. I have bought insurance from Intel for my four overclocked 3970X which is kind of saying that if you overclock sensible we will cover you.

So there is one in the eye for the corporate why would you want to overclock it. The size of this headroom is dependent on many factors. The first is obviously the processor itself. When the silicon has been grown and sliced. They are analysed for imperfection the process then plans how to economically cut it into as many as possible. The best of these will be used in the top of range processors as these will have the largest cache sizes.

Most of these will end up Xeon, i7 xxxxX and now i9 xxxxX processors. The rest are graded and go into other processors. This is why the price is high on these processors. This is further complicated by slight variations in yield as some lumps of silicon will make more high end processors than other lumps. This is a hugely simplified description and still long.

What does this mean for overclockers? Well the silicon in your processor is almost unique. It is a lottery as to what you actually get. To avoid any problems Intel set a large amount of headroom on clock frequency. You can take two identical processors and almost be sure that they will not overclock the same. The better the quality of the silicon the higher the clock frequency it can take. This is why you hear of some crazy clock speeds that one person can get and others can't get close. Of course cooling plays it's part too.

So the stock clocks are the only safe place to start from as you don't know how much headroom is actually available until you feel your way up the frequency range. I have managed just over 5GHz and stable but the temps started making me nervous. I had no intention of running prime at that speed. I have found a good overclock for day to day use for my processor is 4.5GHz. When I am doing anything processor intensive I knock the clocks down to 4.3GHz. So the cooler has a chance to stay on top of the temperatures.

Edit Tese: added some paragraphs to this wall of text.

 

[JQB45]

Intel did this before. At the time the AMD64 "hammer" CPU's had just came out and continued to role out. In the end Intel Persevered at the top of the performance heap with $1000 dollar chips that consumed oodles of power: Pentium 4 Extreme Edition.

Mean while AMD charged a decent price for a decent product. The first computer I built from scratch was a AMD Athlon64 ClawHammer 3000+ (2.0GHz, 754 Socket) with 1GB of RAM a 80GB HDD and a horrible NVidia 5200U. I so wish I had brought the ATI 9600 back then.

So Intel at the time retained the performance crown at the very top end, but at the cost of extreme amounts of energy use.

BTW - my AMD Athlon 3000+ still runs till this day. 14.5 years after I brought it.

 

[bit_user]

I suspect having free CPU resources is attributing to more than just the 1% lows like many reviews are trying to point at. Since gaming involves mouse, keyboard, other OS things like interrupts etc, that stuff doesn't get drawn on screen per say. Someone pretty savvy at performance tuning on Windows could probably pinpoint where the waits come from, interrupts etc.

Based on some benchmarks (in the past year) of certain games running on non-hyperthreaded dual-core systems, I think some games/engines have issues with lock contention. This is probably a performance problem in their code. It was greatly aided by going to 4 threads, even on dual-core.

So, I'd tend to agree that code which is not threaded very carefully could be aided in both throughput and latency by simply having more cores and hardware threads available.

 

[ElMojoMikeo]

Indeed Intel did this before big time. The king of all power consumers was my 3970X until the launch of this monster. The story of how Intel got here is interesting as It did happen in the 39xxX series processors. There was the 3930X and 3960X then without much warning the 3970X max temp 91C and a TDP of 150W at base clock. I have power limits set at around 240W. I had no idea the problems I would have on keeping it cool. As the 3970X was the last of the 32nm X processors. The smaller lithography at 22nm seemed to make the problem go away on the 49x0X. The 5xxxX and 6xxxX all seemed to be less power hungry and generate much less heat. Then came the The Skylake-X and here we go again. If the current affordable coolers on the market can't keep my six cores under 91C. It makes 105C hilarious. We need some new coolers for these bad boys. If they bring out a new cooler for this Skylake-X then with the right adaptor I could see how fast these really can go. Before I replace them with the Skylake-X unlocked version when Intel see we can control the temperature of this version. I am ahead of the game already?

 

[Wisecracker]

I don't have a dog in this hunt but I've always been a bit 'meh' on P95 as a stress tester. I prefer more of a 'multi-tasking' approach.

I'll capture/record a video stream, run the CB OpenGL test, play some music, surf the web, etc, with OCCT cranking in the background. If I can keep temps/volts in a happy range with a smooth experience at 100% utilization, especially overnight, I'm good.

I also believe it's more practical to expand more typical usage scenarios, i.e., crank, sleep/hibernate, wake, crank, sleep, crank ... etc.

I understand it's a little old school, but my experience has convinced me this is more likely to reveal corruption or config issues, blue screens, whatever ...

 

[bloodroses]

Intel did this before. At the time the AMD64 "hammer" CPU's had just came out and continued to role out. In the end Intel Persevered at the top of the performance heap with $1000 dollar chips that consumed oodles of power: Pentium 4 Extreme Edition.

Mean while AMD charged a decent price for a decent product. The first computer I built from scratch was a AMD Athlon64 ClawHammer 3000+ (2.0GHz, 754 Socket) with 1GB of RAM a 80GB HDD and a horrible NVidia 5200U. I so wish I had brought the ATI 9600 back then.

So Intel at the time retained the performance crown at the very top end, but at the cost of extreme amounts of energy use.

BTW - my AMD Athlon 3000+ still runs till this day. 14.5 years after I brought it.

At one point AMD had an FX Athlon that was quite a bit more expensive than any Pentium. When you have the fastest, you charge a premium; regardless of the company.

 

[JamesSneed]

Intel did this before. At the time the AMD64 "hammer" CPU's had just came out and continued to role out. In the end Intel Persevered at the top of the performance heap with $1000 dollar chips that consumed oodles of power: Pentium 4 Extreme Edition.

Mean while AMD charged a decent price for a decent product. The first computer I built from scratch was a AMD Athlon64 ClawHammer 3000+ (2.0GHz, 754 Socket) with 1GB of RAM a 80GB HDD and a horrible NVidia 5200U. I so wish I had brought the ATI 9600 back then.

So Intel at the time retained the performance crown at the very top end, but at the cost of extreme amounts of energy use.

BTW - my AMD Athlon 3000+ still runs till this day. 14.5 years after I brought it.

At one point AMD had an FX Athlon that was quite a bit more expensive than any Pentium. When you have the fastest, you charge a premium; regardless of the company.

That is true but right now even If AMD was faster I think they would undercut a bit since there market share is so low. Now when they have Zen+ if they are still faster I would expect the prices to go up.

 

[JQB45]

It would be interesting to repeat these test in a proper case with the cpu cooler and additional fans moving air in and out of the case to see how the VRM's react when they have proper airflow.