My i7 7700k Has Arrived - Insights, Benchies & Overclocks Inside (Now With Delid!)

[RichUK]

So, first stability checkpoint / baseline reached. The headlines:

• 6 hours Prime95 Small FFTs stable - using Prime95 v28.7
• 4.7 Ghz @ 1.264v (memory stock - 2133) - note, HWinfo doesn't report volts correctly - see CPU-z.
• Average temp 83 degrees
• Corsair H110i set to quiet mode (I'd never leave it in performance mode due to the increased noise - I don't want to sit next to a vacuum cleaner!)
• A-Tuning was used to adjust the offset - 0.02v and 0.03v failed in under an hour, 0.05v achieved stability at 4.7Ghz.

For me, Prime95 Small FFTs is the gold standard in testing overclock stability, as OCCT and IBT pass at 5Ghz 1.36v where as P95 does not. Prime95 also produces the most heat - for these reasons, i feel it better tests the max load conditions (BF1 only gets temps to around 65 degrees).

Given i'll need more voltage for higher frequencies resulting in 99 Degrees temps on the current setup (where it will also continually throttle), I'll be waiting until I delid and install the Kraken X62 on Thursday/Friday before going for Prime 95 Small FFT stability at 4.9Ghz which should be achievable with less than 1.4v (i may even attempt 5Ghz again, with better VRM cooling and CPU temps).

Benchmark sanity test:

• https://browser.geekbench.com/v4/cpu/1263060
  
• https://s4.postimg.org/if8o998il/CR15_4_7_Ghz.jpg

The below should be a useful comparison against 6700k overclocks:

 

[buklau]

Have you try to delid your chip and see if you can hit 5ghz stable?

 

[RichUK]

Have you try to delid your chip and see if you can hit 5ghz stable?

That's Thursday / Friday this week, along with the newly released Kraken X62.

 

[buklau]

Thanks for the heads up. I'm itching to get a 7700k if it does 5ghz stable delided.

 

[UncleCrusty]

Are you able to set a different max clock for AVX workloads? I believe I read that would be a Kaby feature somewhere, but perhaps only on z270 boards. Otherwise, are you able to just disable AVX/AVX2? It seems to me that the critical path is in the AVX2 execution pipe (induced by heat density or just raw logic delay), I wonder where you'd be prime stable if it was only allowed to use SSE.

 

[RichUK]

Are you able to set a different max clock for AVX workloads? I believe I read that would be a Kaby feature somewhere, but perhaps only on z270 boards. Otherwise, are you able to just disable AVX/AVX2? It seems to me that the critical path is in the AVX2 execution pipe (induced by heat density or just raw logic delay), I wonder where you'd be prime stable if it was only allowed to use SSE.

Hmmmm. I could test with a version of prime that doesn't support AVX (from what I've read, before version 28.5). I suppose this would be valid if my intended use won't require AVX.

What's interesting is IBT and OCCT doesn't fail at 5ghz, and in fact passes with 1.36v but temps are high (which I can resolve).

 

[UncleCrusty]

Hmmmm. I could test with a version of prime that doesn't support AVX (from what I've read, before version 28.5). I suppose this would be valid if my intended use won't require AVX.

What's interesting is IBT and OCCT doesn't fail at 5ghz, and in fact passes with 1.36v but temps are high (which I can resolve).

Would be interesting to see.
This is where I got that AVX clock information from:

http://www.fudzilla.com/news/processors/41048-kaby-lake-has-learned-two-overclocking-tricks

Old and lackluster article, so who knows. It's "AVX negative ratio offset" that's of interest; I think BDW-E supports something like this as well. Might be useful if you could set an aggressive non-AVX clock, and an AVX clock which is a couple 100 MHz lower (lower voltage as well would be good).

 

[RichUK]

Would be interesting to see.
This is where I got that AVX clock information from:

http://www.fudzilla.com/news/processors/41048-kaby-lake-has-learned-two-overclocking-tricks

Old and lackluster article, so who knows. It's "AVX negative ratio offset" that's of interest; I think BDW-E supports something like this as well. Might be useful if you could set an aggressive non-AVX clock, and an AVX clock which is a couple 100 MHz lower (lower voltage as well would be good).

How very interesting. This is what happens when you've been out of the overclocking game for so long, i assumed using the latest version would be fine.

At 4.7 Ghz, I am observing the following:

• P95 v28.7 - 83 degree average (peaked at 89 degrees)
• P95 v27.9 - Peaked at 81 degrees (didn't capture the average)
• P95 v26.6 - 73 degree average

OCCT with AVX Linpack ticked - 81 degrees.

 

[UncleCrusty]

How very interesting. This is what happens when you've been out of the overclocking game for so long, i assumed using the latest version would be fine.

At 4.7 Ghz, I am observing the following:

• P95 v28.7 - 83 degree average (peaked at 89 degrees)
• P95 v27.9 - Peaked at 81 degrees (didn't capture the average)
• P95 v26.6 - 73 degree average

OCCT with AVX Linpack ticked - 81 degrees.

Quite the difference. I wonder how it translates to stable voltages as well.

 

[RichUK]

Interesting indeed. 5Ghz at 1.34 stable for 15 minutes under Prime95 v27.9 (OCCT temps are still less and doesn't fail):

 

[RichUK]

If this is anything to go by, using Prime95 v27.9 will suffice: http://www.overclock.net/t/1570313/skylake-overclocking-guide-with-statistics

 

[UncleCrusty]

Interesting indeed. 5Ghz at 1.34 stable for 15 minutes under Prime95 v27.9 (OCCT temps are still less and doesn't fail):

Thanks for testing this. Seems to support the notion of a separate AVX clock being a very useful overclocking feature, especially when AVX is hardly used.

 

[RichUK]

Thanks for testing this. Seems to support the notion of a separate AVX clock being a very useful overclocking feature, especially when AVX is hardly used.

Thank you for pointing it out in the first place! I recall doing some reading a while back that suggested using an older version of Prime95 for this very reason.

 

[Burpo]

http://overclocking.guide/download/prime-95-v27-7-64-bit/

 

[RichUK]

http://overclocking.guide/download/prime-95-v27-7-64-bit/

What's the difference between this version and 27.9?

 

[Majcric]

The stress testing and how far ones chip will overclock is obviously subjective depending upon his/her kind of testing . Unfortunately I find myself in the crowd that has to run Prime95 small FFTs before I am satisfied. And preferably the latter version. My i7 6700k will probably do more than 4.6 if I'd be willing to let Prime95 (considered a power virus to some akin to furmark for GPUs) go but at this point I'm not willing to do so.

Anyway, It appears your chip will do 4.7 to me. Again the findings are nice info.

 

[.vodka]

What's the difference between this version and 27.9?

P95 has a "whatsnew.txt" included, that's the changelog with explanations you're looking for.

Basically:

New features in Version 28.9 of prime95.exe
-------------------------------------------
4) Method for choosing the best FFT implementation changed. In previous versions, the FFT
implementation that resulted in the fastest single worker timing was used. In this version
the FFT implementation that had the best throughput was selected. For FMA3 FFTs I used a
4-core Skylake to measure best throughput. For AVX FFTs I used a 4-core Sandy Bridge
to measure best throughput. Not many FFTs were affected, but you may see a few percent
variation in throughput with this version.


New features in Version 28.6 of prime95.exe
-------------------------------------------
2) AVX2 support for trial factoring. About a 50% speedup over the previous version. However,
all trial factoring should still be done on a GPU. A GPU is on the order of 100 times more
efficient at trial factoring than a CPU!!!


New features in Version 28.3 of prime95.exe
-------------------------------------------
1) Supports Intel's new for fused multiply add instruction introduced with the Haswell CPU.
This results in faster FFTs. Note that performance on many Haswell systems is memory-bandwidth
limited. This means that when running workers on all cores performance gains will be small.
2) Some minor optimizations may give a very small performance boost for AVX CPUs.
3) All new test torture test data for AVX CPUs. The new data runs more iterations, thus more time is spent torturing the CPU rather than initializing the FFT routines.


New features in Version 27.9 of prime95.exe
-------------------------------------------
1) None, only minor bug fixes.

New features in Version 27.7 of prime95.exe
-------------------------------------------
1) When an error occurs reading a save file it is renamed with a .bad extension.
On rare occasions the file can be read successfully at a later time.


New features in Version 27.1 of prime95.exe
-------------------------------------------
1) 32-bit FFTs optimized for AVX-capable computers. Intel Sandy Bridge computers should
see a 25% speed increase.

27.1 introduced AVX support (Sandy Bridge)
28.3 indroduced FMA support (Haswell) with more stressful torture testing
28.6 uses AVX2


That's why you're 5GHz stable for 27.9, yet not for later versions which are more stressful per se and make use of extra hardware/instructions introduced with Haswell back in 2013.

Your current 5GHz settings will probably fail with software that makes use of such extensions. If I'm not mistaken an example of a day to day task using them is video encoding with Handbrake.



Hopefully delidding + liquid pro/ultra and shaving >10°C off load temps will get you stability on versions later than 27.7-27.9

 

[RichUK]

Delid done - seeing 65 degrees full load with Prime v28.7 which was previously failing but now it isn't (well, it doesn't insta-fail, i stopped it after 1 min)!!!

 

[RichUK]

Before and after delid with Liquid Ultra - 5Ghz, 1.344v, Kraken X62 @ 50% fan and 65% pump (i.e. silent), Prime95 v27.9 - 26 degrees difference (~30 degrees max temp difference)!

 

[Arachnotronic]

Before and after delid with Liquid Ultra - 5Ghz, 1.344v, Kraken X62 @ 50% fan and 65% pump (i.e. silent), Prime95 v27.9 - 26 degrees difference (~30 degrees max temp difference)!

It is a crying shame that Intel ships these things with the poor TIM. Great results!

 

[[DHT]Osiris]

Before and after delid with Liquid Ultra - 5Ghz, 1.344v, Kraken X62 @ 50% fan and 65% pump (i.e. silent), Prime95 v27.9 - 26 degrees difference (~30 degrees max temp difference)!

That's an impressive delta. I wish Intel would just do away with their TIM/lid system, it's f'ing terrible if you can get a 30c difference under load (albeit overclocked).

 

[vissarix]

Before and after delid with Liquid Ultra - 5Ghz, 1.344v, Kraken X62 @ 50% fan and 65% pump (i.e. silent), Prime95 v27.9 - 26 degrees difference (~30 degrees max temp difference)!

Since you went so far why dont you put your cooler directly on the die without ihs?

 

[LTC8K6]

Since you went so far why dont you put your cooler directly on the die without ihs?

Well, you can crack the die easier that way.

 

[.vodka]

That's an impressive delta. I wish Intel would just do away with their TIM/lid system, it's f'ing terrible if you can get a 30c difference under load (albeit overclocked).

It is a good system when they solder the chip to the IHS, this makes up for the extra height the black adhesive adds, that TIM can't effectively work around. Not soldering K parts in the mainstream 115x socket anymore is a brain dead, bean counter move. I suspect part of Sandy's great OC headroom when above 4.5GHz is thanks to it being soldered and avoiding the heat transfer bottleneck we see on Ivy and later.

Yeah, yeah, more steps to get to the final product, extra cost for a multi billion dollar worth company... yet look at the results. 30 freaking less degrees when pushing the chip to the limits! In line, if not better than we usually see on delidded + liquid metal'd Ivy/Haswell (makes sense, smaller die each generation...) Isn't that worth it for the very reason the K parts exist?

It boggles the mind.

 

[HilbertSpace]

Thanks for the pictures. I believe those are the first pictures of a Kaby Lake de-lidded anywhere. I spent a lot of time trying to find some. Looks 'almost' identical to Skylake, but hard to tell about die size... my quick image analysis with counting pixels results in ~124-125 mm^2 versus 122 mm^2 for Skylake. Location seems a bit different too, die looks a little closer in one direction and further in the other to the gold marking triangle than for Skylake, so it looks like this product (https://shop.aquacomputer.de/product_info.php?products_id=3378) might not fit Kaby Lake too.