Question about running Haswell-E at stock but with 2800Mhz Memory.

[Franzi]

I'm going to get a new X99 based build together the next days and since my daily work will depend on it I wont overclock this machine. The memory I ordered however is a Corsair DP 2800Mhz kit. I read that at speeds above 2400Mhz selecting the X.M.P. profile will automatically set a 125 strap. Is there anything I need to take care of or just leave all CPU settings on auto and select X.M.P. for the memory? The CPU in question will be a 5930k and the board an ASUS RVE. I know, i know, its a very OC-specific board and I don't really need these extra features, but its whats available to me at the moment for a reasonable price. Basically I just need to know how to proceed to use the memory at its rated 2800Mhz speed while leaving the CPU at Intel stock settings. Many thanks in advance!

 

[Flapdrol1337]

why not keep it at stock? with quadchannel and a non overclocked sixcore it won't do anything for performance.

 

[Lepton87]

You can totally select manual mode and keep the BCLK at 100MHz although I don't know why would you do that. I just left it at XMP and the strap is 125 with a 29 multiplier instead of 100 with 36 multi with MCE enabled, 25MHz higher is nothing.

 

[BonzaiDuck]

You can totally select manual mode and keep the BCLK at 100MHz although I don't know why would you do that. I just left it at XMP and the strap is 125 with a 29 multiplier instead of 100 with 36 multi with MCE enabled, 25MHz higher is nothing.

Well, dude, I've been watching your progress with the 5820K.

Have you run any stress-tests on the processor at what I assume is close-to-spec speed? It would be interesting to know the ambient, idle and load temperatures -- six-core average if you can provide it.

I say that . . . because you're cooling it with an NH-D14 heatpipe cooler . . . for me -- a known point of reference . . .

Just curious, if you're so inclined to provide the info . . .

 

[Lepton87]

Well, dude, I've been watching your progress with the 5820K.

Have you run any stress-tests on the processor at what I assume is close-to-spec speed? It would be interesting to know the ambient, idle and load temperatures -- six-core average if you can provide it.

I say that . . . because you're cooling it with an NH-D14 heatpipe cooler . . . for me -- a known point of reference . . .

Just curious, if you're so inclined to provide the info . . .

I was very low on TIM so I didn't clean the cooler and reapply the TIM, because I had only very little diamond paste left, I left the old imprint and on top of that applied some TIM on the IHS. I knew beforehand that it may well be very under-optimal and indeed the temp seems abnormally high, it reaches 90C under IBT with only 1 fan so I think the TIM is to blame. I have only some Noctua NT H1 TIM at hand which is slightly inferior to that IC diamond, I'm considering cleaning the IHS and the cooler and applying the NH1 or just adding some NH1 to that IC diamond, I don't know how well mixing TIMs works but I may try it until I buy some more of that IC diamond. Do note that the temp is much lower in every other load test and the difference is high, over 10C but even 80C seems too much for stock. It hovers around 40C at idle but the fan is running at only 50% which makes the system VERY quiet, the difference compared to my old system is very high and very noticeable, the silence really makes much more of a difference than I thought. The silence is way more noticeable them the performance. The control over the fans is very well thought of on this board and they even bundle an OC panel which seems awesome, I haven't tried how it works yet.

UPDATE:
Adding a second fan had a massive effect on the temperature, it lowered it to 81C from 90C so certainly one fan wasn't cutting it. Next I'll take off the radiator to see the imprint on the CPU and if it doesn't wholly cover its IHS I'll add some of that noctua TIM. I've never mixed TIMs before so I'm curious how that works out.

 

[BonzaiDuck]

I was very low on TIM so I didn't clean the cooler and reapply the TIM, because I had only very little diamond paste left, I left the old imprint and on top of that applied some TIM on the IHS. I knew beforehand that it may well be very under-optimal and indeed the temp seems abnormally high, it reaches 90C under IBT with only 1 fan so I think the TIM is to blame. I have only some Noctua NT H1 TIM at hand which is slightly inferior to that IC diamond, I'm considering cleaning the IHS and the cooler and applying the NH1 or just adding some NH1 to that IC diamond, I don't know how well mixing TIMs works but I may try it until I buy some more of that IC diamond. Do note that the temp is much lower in every other load test and the difference is high, over 10C but even 80C seems too much for stock. It hovers around 40C at idle but the fan is running at only 50% which makes the system VERY quiet, the difference compared to my old system is very high and very noticeable, the silence really makes much more of a difference than I thought. The silence is way more noticeable them the performance. The control over the fans is very well thought of on this board and they even bundle an OC panel which seems awesome, I haven't tried how it works yet.

UPDATE:
Adding a second fan had a massive effect on the temperature, it lowered it to 81C from 90C so certainly one fan wasn't cutting it. Next I'll take off the radiator to see the imprint on the CPU and if it doesn't wholly cover its IHS I'll add some of that noctua TIM. I've never mixed TIMs before so I'm curious how that works out.

For thermal performance, the CLU and Indigo Xtreme metal-pads have the lowest thermal resistance once they have set. the Indigo doesn't set until temperatures rise above 70 or 80C.

Following thereafter in performance is any highly-loaded nano-diamond formulation, with IC Diamond being an established choice. You can buy your own diamond powder, and mix it with any silicone or grease-based TIM. Someone even invented a little DIY crank-operated mixer for it -- a little round plastic case for contact lenses. A custom-shaped paper-clip operates as the crank and paddle assembly.

After that, you have all the branded or mainstream TIMs, some with aluminum oxide.

The problem with the liquid metal or metal-pad products: At least with the metal pads, the gallium-indium-whatever makes an amalgam with copper -- possibly even with nickel if you don't lap the nickel-plate off your processor. It requires extra care to extract your heatsink from the top of the processor when you need to remove it . . .

The diamond paste can be re-used, perhaps by adding a small drop of mainstream TIM and a little extra fresh diamond. For the rest, I'd go after the IHS and heatsink with isopropanol on a rag, and maybe even give the processor a fresh lapping.

For what you spend on these processors like the 5820K, one could hesitate lapping the IHS if you intend to resell it or you are inclined to finagle an exchange of some kind. But getting rid of the nickel is worth a C degree or two with some older processors I know of -- around the 110 to 120 TDP range.

 

[Lepton87]

For what you spend on these processors like the 5820K, one could hesitate lapping the IHS if you intend to resell it or you are inclined to finagle an exchange of some kind. But getting rid of the nickel is worth a C degree or two with some older processors I know of -- around the 110 to 120 TDP range.

I always thought lapping had more to do with how convex or concave your IHS is than getting rid of that thin layer of nickel. I will check how even my IHS is on a ruler and if it's visibly not flat then I might attempt lapping but if it's even I won't bother, I want to have the option to upgrade to Broadwell-E if this was the last CPU I would use on this mobo I would be more inclined to lap.

 

[BonzaiDuck]

Just another thought: If 81C is the temperature from the leading core sensor, the average could be around 76C. If it's the average, it's my opinion that we need more effective cooling on these beast-E's.

I'd just wonder how much better off I would be with two radiators in series with double-D5 pumps. 10C drop would be a start, but I really think 20C would be adequate for passing those tests and getting maybe 4.5Ghz or more . .

 

[Lepton87]

Just another thought: If 81C is the temperature from the leading core sensor, the average could be around 76C. If it's the average, it's my opinion that we need more effective cooling on these beast-E's.

I'd just wonder how much better off I would be with two radiators in series with double-D5 pumps. 10C drop would be a start, but I really think 20C would be adequate for passing those tests and getting maybe 4.5Ghz or more . .

It's the temperature from the hottest core, I used Real Temp TI to get the readings. Well, at least that's what I think is the hottest core, I now realized it only shows the temperature of 4 cores and package which are 76/71/81/72 and package 81C, why only 4 cores?

 

[BonzaiDuck]

It's the temperature from the hottest core, I used Real Temp TI to get the readings. Well, at least that's what I think is the hottest core, I now realized it only shows the temperature of 4 cores and package which are 76/71/81/72 and package 81C, why only 4 cores?

Outdated version of the program? Is there a menu where you can select or deselect cores for monitoring? That looks like an average of 75C for the available core data. But ya hafta wonder . . .

How about CPUID HWMonitor? If the latest version is worth its salt, it will show all six cores and show the temperatures correctly. Latest versions of HWMonitor had always shown the same readings as RealTemp or CoreTemp.

Still -- an indication of how far you could go with the NH_D14, even with the sum of minor "improvements." This should also be close to the temperatures you'll get with an H110i AiO cooler.

All that being said . . . . I thought I'd seen a lot of variation in temperatures among different chips in the IB and Hassy cores, but that would also derive from the use of TIM. These E chips are supposed to use the indium solder.

I guess the next step is to see if BIOS tweaks can lower the load voltage while remaining stable. That would give a better idea.

 

[Lepton87]

I guess the next step is to see if BIOS tweaks can lower the load voltage while remaining stable. That would give a better idea.

I'm sure it can but I would rather OC on the stock voltage than lower the voltage at stock clocks. 81C gives some headroom, it's under artificial load never encounter anywhere else, so anything under 95C works for me. BTW. 3625MHz isn't under Intel spec under all cores load anyway.

 

[BonzaiDuck]

I'm sure it can but I would rather OC on the stock voltage than lower the voltage at stock clocks. 81C gives some headroom, it's under artificial load never encounter anywhere else, so anything under 95C works for me. BTW. 3625MHz isn't under Intel spec under all cores load anyway.

It might have been for the Z97 chipset, or maybe the X99. There are some auto-volting features which volt the processor more than it needs at load speeds. ShintaiDK mentioned it -- here or in another thread somewhere recently.

You'll find out directly before I do, if some such aspect is the case. You definitely have a little room there if you want to push the temperatures into the 90's.

The HardOCP review posted the CoreTemp maximums for ~4.7Ghz under a Prime95 test with Koolance EXOS liquid cooling. The leading maximum temperature was 80C, with an average of about ~75C. I would suspect that the Koolance performs better than an H110i or the NH-D14 (or D15).

But even so. They OC'd to 4.7. Looking again . . . They report a load core voltage of 1.32 and a CPU input voltage of 1.9V. Umm . . that was for 4.6Ghz. For 4.7, they had to increase the input voltage to 1.94. They didn't offer a lot of tweaking details, but they were using DDR4-2666 RAM.

Also -- the HardOCP review -- covering all three E processors -- observed that the Koolance kit seemed saturated with the other two models. There was nothing more they could do to lower temperatures, or they could only get a stable clock speed of 4.5 for the X and 5930K processors.

 

[Lepton87]

33x125 4125MHz works fine but it it totally saturates the cooling system, max temp about 93-94C I didn't try to lower the voltage it's left at stock (1.2V) but CPU-ID shows 1.251V but it must be wrong cause it doesn't change at all which is impossible. It seems it really needs water to OC properly, I'm going to check how it does at 4.3GHz but I'll leave it at that with an air cooler.

 

[escrow4]

I'm buying a 5820K next week with a Noctua UH 14S that I already have. How far beyond stock voltage (if any) do you need for a solid 4.0GHz?

 

[EvilNodZ]

I'm buying a 5820K next week with a Noctua UH 14S that I already have. How far beyond stock voltage (if any) do you need for a solid 4.0GHz?

i Got 4.0GHz at 1.190V on my 5820K but mines aint a great overclocker. Others should require less Volts.

 

[BonzaiDuck]

i Got 4.0GHz at 1.190V on my 5820K but mines aint a great overclocker. Others should require less Volts.

Do you have any additional evidence of this, perhaps from forum posters at other OC web-sites? There is going to be a frequency distribution of volts and speeds for these chips -- provided that the overclocking crowd are all consistently following optimal practices and settings. Not only am I totally unfamiliar with the features of the new X99 boards, but many of you are still getting familiar with it. So "optimal" settings as a pretty sketchy concept at the moment.

Also, some mobo-maker or other entity must have tested enough of the E processors (particularly the 5820K) to provide a crude profile of expectations: "X% OC to 4.4; Y% OC to 4.6 . . . [etc. etc.]

 

[Lepton87]

4250MHz at 1.25V but it just set that voltage at AUTO setting, I'm checking if its stable with 1.2V.

UPDATE:

So far so good at it lowered the temperature by 9C at load.

 

[njdevilsfan87]

Edit : Wrong thread.

 

[Lepton87]

4250MHZ 1.2V stable in IBT after 10runs, I'm going to try 1.18V now. Temperature dropped from AUTO settings by 8C from 96C to 88C. I use IBT which uses LINX, I know that no other program will come close to that power draw but I couldn't have a chip that would crash under a load that wouldn't crash at stock, no rational explanation, just not something I would be comfortable with.

UPDATE:

1.18V is unstable at 4250MHz so 1.2V it is, something in between might be stable but I won't bother checking I don't want to operate at the absolute lowest stable voltage. I always do that in some increments in that case it was 0.2V.
1.2V stable 1.18V not stable. So you should do 4GHz with 1.18V probably just fine as mine is almost stable at 4.25GHz with 1.18V.

 

[BonzaiDuck]

4250MHZ 1.2V stable in IBT after 10runs, I'm going to try 1.18V now. Temperature dropped from AUTO settings by 8C from 96C to 88C. I use IBT which uses LINX, I know that no other program will come close to that power draw but I couldn't have a chip that would crash under a load that wouldn't crash at stock, no rational explanation, just not something I would be comfortable with.

"Maximum" or "High" setting in IBT? Even with "High," 10 runs suggests complete stability is either near or accomplished.

The temperature improvement is significant. I'm wondering if other, low-tech nit-picky improvements could net you at least another 5C with the D14. And I assume you're still referencing the lead thermal sensor, or at least the highest among the cores shown by your monitoring software?

On the "nit-picky" angle. Assume my 2600K was operating with ~144W at load -- compared to the TDP as a baseline. I was able to reduce the loaded and v-drooped temperature by 5C degrees at a controlled ambient by a ducting mod and a Gen-Typhoon AP-30 exhaust fan with an acoustic mod built into the ducting mod. I can only say that I took pains to watch the temperatures between the tests, and I am sure that this improvement is what I saw.

That being said, I still think water-cooling is the way to go on this. But your info as you report it is very encouraging.

 

[Lepton87]

"Maximum" or "High" setting in IBT? Even with "High," 10 runs suggests complete stability is either near or accomplished.

The temperature improvement is significant. I'm wondering if other, low-tech nit-picky improvements could net you at least another 5C with the D14. And I assume you're still referencing the lead thermal sensor, or at least the highest among the cores shown by your monitoring software?

On the "nit-picky" angle. Assume my 2600K was operating with ~144W at load -- compared to the TDP as a baseline. I was able to reduce the loaded and v-drooped temperature by 5C degrees at a controlled ambient by a ducting mod and a Gen-Typhoon AP-30 exhaust fan with an acoustic mod built into the ducting mod. I can only say that I took pains to watch the temperatures between the tests, and I am sure that this improvement is what I saw.

That being said, I still think water-cooling is the way to go on this. But your info as you report it is very encouraging.

Neither 10GB with 6 threads, it was a quick(not really, 1750sec) check but it seems to be stable, it didn't immediately crash with 1.18V, almost passed 1 run so that suggests that it should be fine, I'm going to leave it working overnight at max memory usage but I'd be surprised to see it crash. I always report the temperature of the hottest core. I can check how much the power usage drops from 1.25V to 1.2V, I have a meter. Earlier reported temps are from 4GB usage, 10GB usage is much more strenuous, 6C more and 20Gflops!!! I think AIR is enough if you aim for 4-4.4GHz, anything more needs water IMHO. LINX at max memory usage is a good 10-15C higher than any load a useful app could generate so even 95C in LINX are fine IMHO.

 

[e-ricesucks]

overclock ftl
19 more posts for pm

 

[BonzaiDuck]

Neither 10GB with 6 threads, it was a quick(not really, 1750sec) check but it seems to be stable, it didn't immediately crash with 1.18V, almost passed 1 run so that suggests that it should be fine, I'm going to leave it working overnight at max memory usage but I'd be surprised to see it crash. I always report the temperature of the hottest core. I can check how much the power usage drops from 1.25V to 1.2V, I have a meter. Earlier reported temps are from 4GB usage, 10GB usage is much more strenuous, 6C more and 20Gflops!!! I think AIR is enough if you aim for 4-4.4GHz, anything more needs water IMHO. LINX at max memory usage is a good 10-15C higher than any load a useful app could generate so even 95C in LINX are fine IMHO.

I say that's a plausible view of it. But this question has been recently visited in other threads about Haswell K processors in general. There is a complaint that the AVX2 aspect of the stress-test software is pushing these temperatures higher than what had been shown before an AVX2 instruction set. [I assume from how this feature is variously described that it is a new instruction-set component for the Haswell-gen processors.]

Understandably, the discussion moved toward the issue of "how much stress-testing was necessary," or whether certain tests were any less useful than more strenuous tests. The author of OCCT asserts that his own "OCCT CPU" test component runs cooler but will reveal more errors and sooner than the LinPack test included as an OCCT option.

IDontCare opined that there has to be some "standard." This implies that using limp stress-tests has limited value.

If one has a successful test run that pushes the temperatures to 95C and within the throttling limit for the processor, you could argue exactly the same thing: It passed the stress test but the processor will never achieve those temperatures under any normal usage.

I might be inclined to agree with that, but I speculate that frequent and extensive stress-testing under those conditions might actually shorten processor life-span.

That's why I agree with IDontCare on the value of robust testing that has relatively short duration. For instance, OCCT's author suggests 3.5 hours for his test; others advise 25 to 30 iterations of IBT or LinX. These test durations are stark contrast to established Prime95 usage for which a user may prefer a duration of 18 hours or more.

Anyway -- we agree about the cooling. You can actually "get somewhere" with overclocking under NH-D14/D15 or Corsair H110i cooling, but you might prefer some custom-water solution with more radiator and cooling capacity.

Lower voltage means less thermal wattage, but conversely -- lower temperatures are likely to trim something off a voltage requirement for an overclock target.

 

[Lepton87]

Lower voltage means less thermal wattage, but conversely -- lower temperatures are likely to trim something off a voltage requirement for an overclock target.

I don't see that a lot, unless we are talking about sub-zero temps. The temparature drop is very considerable but the increase in OC potential is just not there.
http://www.tomshardware.com/reviews/visiontek-cryovenom-r9-295x2,3951.html

It's a GPU but it happens to CPUs as well, I can make a winter-cooling to see if HW really needs less V for a given frequency. The biggest advantage is pushing lots of voltage and have it still not overheat.

 

[Deders]

Always clean the old tim off thoroughly before applying new.