[mattliston]

That's true, but the binaries that are packaged with IBT are ancient. If you run the latest binaries, it's harder to pass than Prime 95.

And some versions of Prime95 are notorious for killing 2011-3 chips. Not sure on exact details, but it has claimed 2 of mine. Spiking upwards of 320-350 watts is insane to see from a 140 TDP chip, though we know TDP doesnt directly correlate to actual power draw..

While the intelburntest is old, it still does a great job of testing cpu, cache, and memory all at once.

If you do the "very hard" 4gig test with the xtreme setting (right clicking on the start button), it will make an overclock fail in 5-20 minutes what the regular setting of "very high" 4gigs can run for hours, in my experience. The program overall is a great way to "walk" into your overclock. Pass a normal test, then fail a "high" test. Add some voltage, passes the high test, but fails the very high test. Add some voltage, still fails the very high test, add a bit more, it passes. Try teh extreme mode "high" test, fails. For fun, tried the very high under extreme, lock up and forced reboot.

I like the "ladder" steps this program enables for testing.

2011-3 and AM3+ both react favorably to the Intelburntest. If it said it was stable, I generally never had random BS happening.

 

[JSHamlet234]

And some versions of Prime95 are notorious for killing 2011-3 chips. Not sure on exact details, but it has claimed 2 of mine. Spiking upwards of 320-350 watts is insane to see from a 140 TDP chip, though we know TDP doesnt directly correlate to actual power draw..

Yeah, you really don't want to make a habit of running Haswell-E at more than 2X TDP. If it ends up higher than that, lower the clocks a bit or don't run certain things at certain speeds. I boot up at 4.2, and I can play with the clocks/voltages later if I want to. 4.5 at 1.27 runs cool and stable in 99% of applications, but in all honesty there isn't a practical difference between 4.2 and 4.5 for me, so I'm always at 4.2 unless I'm running a benchmark. This CPU is still going strong after 7 years, so I must be doing something right.

 

[rewease]

Another thing I found interesting:

Using Power Limits for AVX Stability on Haswell/Broadwell OC

Not the pure teaching but solves certain problems.

 

[mattliston]

But that is for Broadwell/Haswell, not the "-E" variants.

 

[JSHamlet234]

Another thing I found interesting:

Using Power Limits for AVX Stability on Haswell/Broadwell OC

Not the pure teaching but solves certain problems.

I've played around with this a lot, but in the end it doesn't help on Haswell-E because the load on the processor varies from microsecond to microsecond, and the CPU will eventually clock up at the wrong time and error out or crash, especially in something like OCCT Large or Prime 95 large FFT, which don't pull that much continuous power but are very hard to pass. The quickest way to break this method is to run Prime 95 with something like 6 threads. It'll run at full speed and crash in a hurry.

 

[rewease]

Ok, that's a problem. But isn't this much like the normal operation of Haswell-E Xeons constantly bouncing off their TDP limit therefore clocking cores up and down? Like a HCC E5-2699 v3 with 18 cores, 3.6 GHz Turbo but only 145W TDP.

 

[JSHamlet234]

Ok, that's a problem. But isn't this much like the normal operation of Haswell-E Xeons constantly bouncing off their TDP limit therefore clocking cores up and down? Like a HCC E5-2699 v3 with 18 cores, 3,6 GHz Turbo but only 145W TDP.

Yes, they do and stock Haswell-E and Haswell-EP also run extra voltage when executing AVX2 code if they are run completely stock. The separate VF curve for AVX2 is disabled as soon as the multiplier is changed (including running all cores at the max turbo ratio). The only way to keep it while overclocking meaningfully is to overclock using BCLK only with offset voltage and the 125MHz strap. Unfortunately, the stock multipliers are too low on most chips for this to be useful. However, I've always wanted to try it on a 1680V3, which has a stock all-core multiplier of 35. 35x125 would get you to 4.375, and 35x127.5 would get you to 4.48, while keeping the stock AVX behavior. Unfortunately these chips are still too expensive on the used market for me to roll the dice on one just to satisfy a curiosity.

 

[mattliston]

You dont have to bother with anything AVX related. my E5-1660 V3 simply downlocks if it hits a power limit or thermal limit. Else it runs modified turbo multiplier for 4.5GHz. If I run a stress test that generates far too much power usage, it simply downclocks and pops back up.

This is on static voltages, SA offset, and cache is static as well at 3600mhz

Setting a decent power limit is key to keeping it alive. I have mine set to 300 watt long duration and 325 short, with a 32second window.

Proof of it working goes into testing and looking at average power usage and average temperatures at the same test for the same duration.


I use hwINFO64, and I clear the min/max/average before a test, hit it for 10 minutes, and view the results.


I use an odler version of XTU to change power limtis within windows. If I leave them at 300 L 325 S, I get lower average power and temps than if I change things to 400/450 limits. At the same clocks and voltages.


So it is critical for longer 2011-3 chip life to set reasonable power limits.

If I see a sustained active power draw of 287 watts, but it shows a spike of 345, that is an indication to me to make the short term power limit something like 300 watts and long term somethign like 280 or 275 watts.

Things just happen so fast. hwINFO64 luckily has settings for how often it checks for data, and whether it needs to check embedded chips based on cycles (example, i have my cycles set to check every 3 cycles, rather than every 1. This makes the system more responsive to data like temps and pwoer, while things like SATA devices and GPU's update at a rate of 3 times slower.

 

[Blameless]

If you're going for a HW-E or EP, make sure it's a J batch; the others are much more hit or miss in overclocking capability.

 

[rewease]

However, I've always wanted to try it on a 1680V3, which has a stock all-core multiplier of 35.

I might try that on my current 1660v3 with its all-core multiplier of 33. But I guess 4,125 GHz is out of its league. However I gave the silicon lottery another roll on a different 1660v3 that should arrive next week. Hope it will clock better.

 

[mattliston]

None of the Xeon 16xx chips that are unlocked are particularily terrible at overclocking. Just seems some, like my example, suck for cache overclocks. My limit is pretty much jsut under 3700mhz cache when running 4 sticks of ram faster than 2400. my current ECC ram is 4x8gb of 2666 and 3600mhz cache is perfectly fine, but it darn near refuses to boot at or above 3700mhz cache. I have to overclock via Intel XTU to go above 3700, and that is with a hefty cache voltage. It runs 3600mhz at 1.19v, but it wants over 1.3v to run 3800mhz properly.

 

[JSHamlet234]

None of the Xeon 16xx chips that are unlocked are particularily terrible at overclocking. Just seems some, like my example, suck for cache overclocks. My limit is pretty much jsut under 3700mhz cache when running 4 sticks of ram faster than 2400. my current ECC ram is 4x8gb of 2666 and 3600mhz cache is perfectly fine, but it darn near refuses to boot at or above 3700mhz cache. I have to overclock via Intel XTU to go above 3700, and that is with a hefty cache voltage. It runs 3600mhz at 1.19v, but it wants over 1.3v to run 3800mhz properly.

That's normal if you don't have an OC socket,

 

[mattliston]

That's normal if you don't have an OC socket,

The OC socket was more for extreme overclcocking and does nothing for normal overclocks.

My Rampage V Extreme had the OC socket, and these Asrock X99M killer boards did around the same core and cache clocks for 2x 5930k, 3x 5820ks, and this E5 1660 V3. Just didnt have as much memory overclocking, due to actually being able to lock down the SA voltage. On the Asrocks, it is an offset.


I tested all these before I decided I didnt mind the downsize in motherboard size and features. I never plan on LN2 cooling anyways.

 

[JSHamlet234]

The OC socket was more for extreme overclcocking and does nothing for normal overclocks.

My Rampage V Extreme had the OC socket, and these Asrock X99M killer boards did around the same core and cache clocks for 2x 5930k, 3x 5820ks, and this E5 1660 V3. Just didnt have as much memory overclocking, due to actually being able to lock down the SA voltage. On the Asrocks, it is an offset.


I tested all these before I decided I didnt mind the downsize in motherboard size and features. I never plan on LN2 cooling anyways.

Is this your board? ASRock > Fatal1ty X99M Killer/3.1

It says that it has an OC socket. I didn't realize that it did.

 

[mattliston]

well shoot, I guess that means my point is irrelivant. I assumed only the Asus boards had them.

That means I cannot determine if the OC socket makes a difference or not.

But I was under the impression that the OC socket simply meant it could handle far more power draw, aka, during LN2 overclocking or such.

I dunno now...

 

[D-EJ915]

I think Asus was the only 1st gen boards with oc socket, everyone copied it for 2nd gen x99 boards. My 1st-gen X99S xpower definitely didn't oc as well as my 2nd gen boards, sold it off a while ago but I can't really compare since I just had 1. My work pc has the X99M fatality original version but again it's not a board I can spend a lot of time testing on since it's for work.

 

[JSHamlet234]

well shoot, I guess that means my point is irrelivant. I assumed only the Asus boards had them.

That means I cannot determine if the OC socket makes a difference or not.

But I was under the impression that the OC socket simply meant it could handle far more power draw, aka, during LN2 overclocking or such.

I dunno now...

It mainly helps with cache OC and debatably with memory OC. Every single one of my Haswell-E chips got over 4GHz cache on my ASUS board with voltages of under 1.2. My current chip can do 4.2@1.17. Without an OC socket, anything above 3.8 would be impossible.

I do believe ASRock's implementation of the OC socket is less effective than ASUS's. I once bought a 5960X from someone who used it in an X99 OC Formula. I couldn't match the 4.7GHz core clocks he claimed to get with it, but I got about 500MHz more out of the cache than he did.

 

[D68D]

First post on the forum, forgive me if its in the wrong spot.
I've a E5-2699 V3 running in a Asus X99-Deluxe motherboard. I've done the BIOS turbo unlock as per...
Unlock Xeon E5-2678 V3 Turbo Boost forever on Asus protected Bios - YouTube
Which appeared to work just fine. However I can't get the system to post multipliers higher than 30 even though it reports it can go to 36 and 36 is entered in BIOS. What stupidly obvious thing am I missing?
Running Cinibench (13,691 multi core) it stays pegged at x30, so I don't think its at the TDP limit.

The behavior I was expecting was that it'd post to 36 and then throttle back under load due to TDP at which point I'd need to stuff about with other settings to minimize the throttling or is that wrong too?

Cheers.

 

[mattliston]

First post on the forum, forgive me if its in the wrong spot.
I've a E5-2699 V3 running in a Asus X99-Deluxe motherboard. I've done the BIOS turbo unlock as per...
Unlock Xeon E5-2678 V3 Turbo Boost forever on Asus protected Bios - YouTube
Which appeared to work just fine. However I can't get the system to post multipliers higher than 30 even though it reports it can go to 36 and 36 is entered in BIOS. What stupidly obvious thing am I missing?
Running Cinibench (13,691 multi core) it stays pegged at x30, so I don't think its at the TDP limit.

The behavior I was expecting was that it'd post to 36 and then throttle back under load due to TDP at which point I'd need to stuff about with other settings to minimize the throttling or is that wrong too?
View attachment 2552904

Cheers.

This is indeed the wrong place, and you knew it.

If you followed that youtube video exactly, you gave yourself an undervolt and a 120 watt limit

I highly suggest re-watching that video and taking some notes. It seems to go over things too fast, and important information is skipped over fast.

 

[rewease]

Have you tried disabling power state C-6 in BIOS? Seems to be a common thing with this mod.

Btw: My second 1660v3 just came in and this sample seems to be in the normal range of OC capability . Both are J-batch Malayans, the 'bad' one is 2015 production, the current one is 2016.
Guess I will never know if it's just a bottom bin part or has seen some abuse previously. Noticed however that it has 50+ mV higher idle VID than its replacement.