4790K, LinX question

[Tsavo]

New 4790K, tried LinX with latest Linpacks. Didn't futz around in the BIOS. Used Intel Xtreme Utility or whatever, set all cores to 4.4 GHz. Cooler is CM 212+.

Instant 100C and throttling. Volts put themselves up to almost 1.3

Set volts to 1.2, temp is still 93C.

Fan speed ramps up to full blast as it should. TJC is Noctua NT-H1.

Is this normal for LinX and a subpar cooler? Prime 95 maxes out at 68C.

 

[deanx0r]

LinX/OCCT are just bad stressing tools when it comes to Haswell in heat output. Use XTU stress or benchmark tool instead to give your a better idea how your CPU will perform under load.

 

[Tsavo]

LinX/OCCT are just bad stressing tools when it comes to Haswell in heat output. Use XTU stress or benchmark tool instead to give your a better idea how your CPU will perform under load.

Ah, ok. XTU gets it around 67C...which is acceptable to me.

 

[Nothingness]

Prime 95 maxes out at 68C.

IIRC a recent Prime95 (one that uses AVX2) with short FFT should get you as high as Linpack.

 

[Tsavo]

I meant to say CM 212 EVO...which seems to be a bit reedy for this application, perhaps. It's quite possible that the mounting bracket springs are knackered. They have been clamped down for 3 years prior. Easy fix to test.

70C after an hour of gaming.

 

[crashtech]

4.4GHz isn't really much of an overclock that it should even need manual voltage control, imo, it's just the max single core turbo. So you might turn on multicore enhancement and leave all the voltage settings on auto and see if it runs a bit cooler yet.

 

[BonzaiDuck]

4.4GHz isn't really much of an overclock that it should even need manual voltage control, imo, it's just the max single core turbo. So you might turn on multicore enhancement and leave all the voltage settings on auto and see if it runs a bit cooler yet.

That's what I was thinking. But contrary to that, I thought someone had discovered that Z97 boards are over-volting these processors at default settings, or that they actually had to be manually reduced.

Still, it would only make sense that starting with a Turbo overclock to only the spec 4.4 on all cores should not give much of a voltage boost under "auto" settings.

Put it another way. Whatever the collective experience accumulates before I were to try it, and IF I were overclocking a DC chip, I'd spend a lot of time up front anyway to see if the stock "for all cores" turbo speed could be had for lower than the default volts.

That brings us back to the tests. Someone mentioned OCCT as excessive, but they really meant the LinPack OCCT test. He should try the OCCT:CPU test.

One also has the option to turn off AVX2 extensions.

Also, doesn't the XTU tool come out of the Intel shop? If so, I'd say use it.

 

[Tsavo]

Yeah, I'll be happy with 4C @4.4GHz for now...with the lowest possible volts I can get away with.

My 3570K ran for years at 4.2 GHz (1.2v), then recently wouldn't even do stock speed at stock volts. It turned into a crashfest. Sent it to a friend to practice delidding.

Everything is fast, smooth as glass and stable, which is what I want. The 3570k ran poorly in comparison to the 4790K that it feels like I've got a starship now vs a paper airplane.

 

[crashtech]

4.4 is nothing to sneeze at. I played around with mine a lot and finally just left it at 4.4 also. It's super fast like that, I don't miss the extra 400MHz that my max 24/7 OC gave me.

 

[deanx0r]

My 4790K maxed out at 4.7GHz, but I was really hard pressed to tell the difference in performance with its stock speed (4.4GHz). The difference in heat output and fan noise was however very noticeable.

 

[BonzaiDuck]

My 4790K maxed out at 4.7GHz, but I was really hard pressed to tell the difference in performance with its stock speed (4.4GHz). The difference in heat output and fan noise was however very noticeable.

That's the second person to come forward suggesting that OC'ing the chip beyond it's stock turbo speed offers little extra gain. Second -- that I've read on these forums, anyway.

 

[Tsavo]

That's the second person to come forward suggesting that OC'ing the chip beyond it's stock turbo speed offers little extra gain. Second -- that I've read on these forums, anyway.

Sure, 4.7 GHz is fast...but..4.4 GHz on all cores is fast enough for me. I prize stability over benchmark scores.

Although my 3570K died an early death, my assertion that lower volts = longer stable life is true.

 

[Dave3000]

My 4790K maxed out at 4.7GHz, but I was really hard pressed to tell the difference in performance with its stock speed (4.4GHz). The difference in heat output and fan noise was however very noticeable.

4.4 GHz is not the stock speed of the 4790k. You should be comparing 4.7 GHz to 4.2 GHz, not 4.4 GHz, assuming that 4.7 GHz is the all-core turbo.

 

[Dave3000]

Sure, 4.7 GHz is fast...but..4.4 GHz on all cores is fast enough for me. I prize stability over benchmark scores.

Although my 3570K died an early death, my assertion that lower volts = longer stable life is true.

If you prefer stability over benchmark scores then why don't you run your 4790k at default settings? 4.2 GHz on all cores is not much of a difference from 4.7 GHz. GPU is more important in most games and 4.2 GHz should not be bottlenecking a high end GPU.

 

[crashtech]

Stability issues at 4.4GHz? Not any 4790K I have heard of.

 

[Dave3000]

I haven't either but it's still running out of specifications. If you want guaranteed reliability, especially if you use your PC for work as well, you don't overclock it to even 4.4 GHz on all cores, even if it's the highest turbo bin of the 4790k for up to 2 cores, that doesn't mean it will be stable with all cores at 4.4 GHz at the same time.

 

[BonzaiDuck]

I haven't either but it's still running out of specifications. If you want guaranteed reliability, especially if you use your PC for work as well, you don't overclock it to even 4.4 GHz on all cores, even if it's the highest turbo bin of the 4790k for up to 2 cores, that doesn't mean it will be stable with all cores at 4.4 GHz at the same time.

Well . . . I think we (forum members in general) were having this dispute in some other concurrent thread, and it's a point of contention that has a long continuous history on this forum.

The point I already put forward: Intel tests the processors and chooses the specs for running them based on their own cost-accounting and litigation history. With the cost-accounting, you would choose a warranty-period that makes sense, and attempt to assure little or no RMA costs. Sometime back in the '90s, Intel produced a Pentium-Pro or similar processor that had some sort of floating-point error, and it would propagate errors in linear algebra used for business optimization problems. Some construction company in Florida (IIRC) was building some office-complex or skyscraper, and the errors cost them a bundle, so they sued.

If you think you "know what you're doing," raising the clock-speed of the processor while volting it subject to longevity and temperature concerns might be done to assure the same sort of reliability. In that case, though, you don't have Intel's experience or testing facilities. Instead of relying on Intel, you are relying on yourself and your best judgment.

So there's a specific meaning to "guaranteed reliability."

I advise folks who use their machines for work that they shouldn't overclock them. On the other hand, some say that they have to overclock for the work that they do. And I say -- you're welcome to do that and guarantee your own safety anyway. Maybe you can? Maybe you can't!

But simply choosing "for all cores" in the BIOS for the spec maximum turbo speed should only require a minor voltage adjustment. It may be "overclocking," but it's not "major leap."

The specs Intel choose for their processors is based on statistical distributions: they are points chosen on a continuous curve. Since different motherboards and chipsets may volt the processor with some slight difference, there's always been a voltage tolerance, even if they stopped publishing one with the Nehalem 32nm cores. I'm pretty sure that "for all cores" -- at most -- isn't likely to exceed those tolerance limits, and some folks suggested to me that they simply left the BIOS voltage settings on "Auto" when they did it.

 

[ShintaiDK]

Dont use AI style OC. It uses the singlecore turboboost VID for all 4.

 

[BonzaiDuck]

Dont use AI style OC. It uses the singlecore turboboost VID for all 4.

I assume you're referring to the "built-in" OC features of motherboards, like ASUS "AI Tweaker" or whatever it's called?

I need to be modest about my understanding of these things, but unless something has changed, those "quick-and-easy" options always over-volted the processor. I'm still using 4-year-old processor technology, but my experience with that was what I consider a "necessary first step:" Find out what that feature gives you to get a basic baseline of information. I was rather stunned by how much less voltage you needed just to reach the same speed and "prove" -- however thorough -- the CPU's stability.

 

[superstition]

Are you using LLC (load line calibration)? Most decent boards have multiple levels and they can wreck havoc on temps if they're too aggressive. Also, higher levels sometimes do a worse job of combating droop, which is counterintuitive.

Look at the OCCT charts here for an example of that: http://www.modders-inc.com/gigabyte-990fxa-ud3-rev-4-0-motherboard-review/4/

As you can see, with that board and chip, the medium setting is the best followed by the low setting. Everything else is worse, including standard, off, and anything higher than medium.

 

[BonzaiDuck]

Are you using LLC (load line calibration)? Most decent boards have multiple levels and they can wreck havoc on temps if they're too aggressive. Also, higher levels sometimes do a worse job of combating droop, which is counterintuitive.

Look at the OCCT charts here for an example of that: http://www.modders-inc.com/gigabyte-990fxa-ud3-rev-4-0-motherboard-review/4/

As you can see, with that board and chip, the medium setting is the best followed by the low setting. Everything else is worse, including standard, off, and anything higher than medium.

Was this question directed to me? I looked back and forth, but couldn't find a reference to the OP's own motherboard.

I'm still using Z68 boards. I think some of us back in 2011 had put our heads together to compare overclock settings on 2600K's.

I also keep in my mind the residual intuitive wisdom based on a 2007/2008 Anandtech article about OC'ing the high-end C2Q "X" based on Conroe cores. There is a good reason to keep actual VCORE below VID in an overclock regime. Out of that, I and somebody with a Maximus IV Extreme board, had concluded that you can boost LLC to keep vDroop at around 2 to 3 mV. That's the way my two SB rigs are overclocked.

On the BIOS LLC scale, I think it was a notch above Medium but third from highest. I would think, however, that choosing "which notch" to use would depend on the motherboard and chipset.

Some folks don't think that LLC is in any way useful. And of course, the old Anandtech article offered up a caution based on its very thorough analysis of all voltage features: VID, VCORE, vOffset, vDroop, etc.

 

[Tsavo]

I think he might be asking me.

My board is an MSI Gaming 5.

There are no settings for LLC, though there is one for Vdroop control. 17.5% to 100%. I'm assuming Auto defaults to 100%

 

[Ramses]

I had a similar heat/stress test experience when I first got my 4790K. I found it odd that some stress testing software was able to overheat, or at least make them run really hot, it seems counter intuitive to say "don't run this stress test because it stresses it too much" when that is sort of the point. Absolutely you'll never see loads like that in real use, but when has that ever stopped us before? Very strange.

I also had an Asus Z97-WS with everything set to auto that was jacking the voltage under turbo or turbo'ing all cores or something like that (it's been awhile) and it took me awhile to figure it out as I expected "auto" to run it as Intel intended. That was another part of my initial high temps. Also strange.

No problems in use now, just an odd experience that wasn't like any other CPU I've ever had going back to 386's. It's a hot running CPU when loaded and volt'd though.