"silicon lottery" i7-6700k 4.7 cpus?

[markm75]

Anyone purchased a 4.7 cpu from silicon lottery before and had success? They list it as being 1.40 or so or below in terms of the voltage (in the bios?) needed to be stable, however their stability is just 1 hour of realbench, which isnt enough.

I've had some testing with the 6700k and the other one i have and tested, i can only really achieve stability for long periods of time at around 4.6 and usually around 1.360-1.375 (under load, while maybe 1.345 + 0.025 adaptive in the bios).

Wondering if these would fair better, any experiences out there, particularly with the 4.7 and how close the voltage required came to what they list these at?

Thanks in advance

 

[thesmokingman]

I don't think it's worth it. I grabbed one of the shelf at Microcenter, threw good cooling at it and managed to run [email protected] with the igpu on. It passes 15 hour sessions of handbrake.

Skylake clocks very well and the percentage of good overclocks from what I've seen is quite high. It gets even better after delidding. Btw my chip is stock and comparatively is not a gold chip. That's why I think buying a binned chip that isn't even well binned is a waste.

 

[mls1995]

I actually had a bad 6700k and returned it. The one I got after that has been a real winner. I have a 4.7 6700k that can also handle all 4 dimms filled and ram running at 3200. It's the best chip I've ever had. In total I've had 3 6700ks and only one can hit 4.7.

But, I still doubt I'd buy a binned chip. I'm just not sure it's worth it other than bragging rights.

 

[i7Baby]

Depends a lot on the cooling you're using. Water cooling will help.

 

[markm75]

Update..

So i purchased the 4.7 rated at 1.408 or lower vcore SiliconLotteryCPU for $339 + $50 delidded.. Its rated this way using a specific configuration plus 1 hour of realbench (only this test).

I'm still choosing to meet old school results of 12-24 hours with p95 blend 27.9 (no mod for avx though timer set to 10 instead of 15 for quicker fails).. i wish there was a way to make this all fail quicker.. i can pass small data checks of OCCT for 1.5 hours, IBT for 5-10 rounds, but P95 blend is always the clear fail here, but just takes a lonnnnng while to get to it. Yes you'll see i probably should have set ram to 2133 till i figured the vcore out, but i was going for the kill in one shot, because after my fixed voltage test at 3200, i thought i had found the right setting to begin with.

I have to say the delidding was done quite well.. about 15c on average less than I was getting with my 6700k from amazon (which appeared to need 1.375 volts or better to do 4.7 and i hadnt fully tested it with p95 blend, so likely it needed more).

For the $50, the cost of buying a delidder, it was worth it to not worry about screwing the chip up.. since i wont be delidding more than one it was also cost effective.

So onto my testing..
All this is now done on a Thermaltake 3.0 Ultime Riing 360mm water cooler, which after testing 3 different models i found to be the coolest/quietest for the value (tested the h115i, the h100i, the regular TT 3.0 ultimate).. i have stats on all these compared as well.. another story.



Fixed voltage:

I found that 1.40 fixed in the bios was a bit of a winner.. though i had had vccio and system agent all the way up to 1.25. (this with ram already at 3200mhz).. with LLC6 vcore under p95 blend load was hitting 1.45 volts actually.. (so when people say they are at 1.35 and 4.7 etc, i sometimes wonder if they mean the bios setting or the vcore load, because they could be 1.35 adaptive with + 0.10 etc).. at any rate, i made it several hours (see below on adaptive i failed at 6-8 hours on p95 blend, so unsure if this was truly the winner).. 1hour of realbench succeeded here too. Many tests did, just not p95 blend.




For the tests below i switched to P95 without the local.txt mod that is referenced out there with this value (CpuSupportsFMA3=0).. i'm still very unclear on whether p95 blend should be run with this set in the local.txt, i found i had fewer errors (coincidence?) if it was NOT included.. its related to a skylake freeze bug with avx instructions.. even with the line in the text file, it still says its testing with AVX, so this seems strange.. so these tests below do NOT include the line:



Adaptive testing:

i started off NOT lowering LLC from the value of 6 out of 7.. i figure, if it needs so much vcore.. whatever value.. then lowering the LLC will only make me have to offset it more to get it to where it "needs" to be anyway? Or is this not the right way to think about it?

I tried initially setting adaptive to say 1.390 and maybe +0.008.. this would result in say 1.408 or 1.395 under load (keep in mind i already had found that manual volts of around 1.40 were the ticket (with the higher vccio 1.25 not 1.20)
So, testing 1.390 + 0.008, with vccio/system agent around 1.20 (instead of 1.25 like my manual tests).. would result in bsod reboots or errors (see below for the examples)

Strangely at some point here, i think i was having success with realbench for 1 hour where prime95 blend would fail within 3 hours.
I then tried 1.395 and either + 008 or -005 i forget.. or both.. but those too failed with p95 after so many hours, in this case 6 hours, despite seeing mostly 1.424 under load of P95 blend (drops to 1.408 etc)..

So i just assumed the core voltage set in the bios must definitely need to be 1.4 or better..

I now have vccio at 1.20 and agent 1.20 and the volts set to 1.402 - 0.008.. this results in 1.440 volts or 1.424 with the p95 blend tests.. I also changed the blend setting from 15 to 10 minutes intervals.. (i was able to catch errors quicker this way).. at this point i'm at 10hrs + and going. Max temp 86C, ambient 24C (delidded/thermal take 360 riing)

**side note, i did set the Fclk to 1000 instead of auto, per a bug others mentioned.


Some of the previous blue screens included:

ATTEMPTED_EXECUTE_OF_NOEXECUTE_MEMORY
DRIVER_IRQL_NOT_LESS_OR_EQUAL

I'm pretty sure these were memory related (clock set to 3200 this whole testing time)..
So far it appears to be the case.. needing say 1.424 minimal under load maybe, unless i'm missing something, seems pretty obvious now. (off chance maybe increasing vccio and agent would have helped, unclear)


Also i would think trying for 4.8 is out, i made a few attempts with fixed voltages up to 1.43, usually resulting in spontaneous reboots. I realize i can go as high as 1.5 volts and my temps at 1.44 are "ok" at 86c or lower, but i think i'm at my own limit so to speak.

 

[Streiw]

I actually had a bad 6700k and returned it. The one I got after that has been a real winner. I have a 4.7 6700k that can also handle all 4 dimms filled and ram running at 3200. It's the best chip I've ever had. In total I've had 3 6700ks and only one can hit 4.7.

But, I still doubt I'd buy a binned chip. I'm just not sure it's worth it other than bragging rights.

Nice admission of fraud there.

 

[mls1995]

Nice admission of fraud there.

fraud my ass. I bought CPU from newegg, and it couldn't pass stress tests at STOCK speeds with 4 sticks of ram installed. That's called defective. I exchanged it with newegg and I got a good chip. End of story. I didn't buy from silicone lottery.

I've had 3 total 6700ks because I have two systems.

 

[Streiw]

fraud my ass. I bought CPU from newegg, and it couldn't pass stress tests at STOCK speeds with 4 sticks of ram installed. That's called defective. I exchanged it with newegg and I got a good chip. End of story. I didn't buy from silicone lottery.

I've had 3 total 6700ks because I have two systems.

Bullshit. It didn't clock as high as you wanted it to, so you RMA'd it.

 

[Furious_Styles]

Bullshit. It didn't clock as high as you wanted it to, so you RMA'd it.

Dude just told you it did not meet his expectations at stock. As far as I know you can return anything if you're dissatisfied with it and that it is 100% not fraud.

And to the guy doing the OCing: You need to either severely underclock the memory or run the cpu at stock and make sure your memory is 100% stable @ 3200. Otherwise you really don't know what the limitation/problem is. Also you can't really compare delids vs stock on different CPUs.