i7 7700k @ 5.2GHz/3600 RAM

T

techguymaxc

Limp Gawd
Joined
Jul 6, 2016
Messages
139
Delidded the old fashioned way (with a razor). Cool laboratory Liquid Pro beneath the IHS, Phobya Nanogrease Extreme above it.

Specs:
Custom water cooling (single D5 pump w/reservoir top, XSPC Razor CPU block, EK 1070 GPU block, RX360 radiator with Corsair SP120s in push/pull, EX140 radiator with Corsair AF140s in push/pull).
Asus Strix Z270h
Corsair Vengeance LPX 16GB 3600 (18-19-19-39-2T) running at rated speeds
Gigabyte GTX 1070 (Founder's Edition) EK block & backplate @ 2062MHz GPU/8800MHz RAM
PC Power & Cooling Silencer III 1200W PSU
Fractal Design Define S

Here's the chip delidded:
delid_zpsgsiewmga.jpg


Overclocking Results:
IMG_1059_zpsqwamruvv.jpg


Can pass basic stress testing with 1.32 set in BIOS, needs about 1.37-1.38 for longer runs. This is where I have it set for daily use. With an AVX offset of 500MHz it goes like this all day.

Here are a couple benchmarks. I can run others if anyone is terribly interested.
cinebench5200_zpsdj7zqpgx.jpg

3dmarkphysics_zpshnruorzx.jpg


And a couple pics of Furmark stress testing @ 4k to give you an idea how much cooling capacity is in the water loop:
download4_zpsdod1nun2.jpg

download5_zpsazknn9yb.jpg


And a couple shots of the system to round it off:
download2_zpsorx0tdzg.jpg

download3_zpsodg9qs9h.jpg
 
Nice chip! How much of an improvement in temperatures did you see from delidding?
 
blade52x said:
Nice chip! How much of an improvement in temperatures did you see from delidding?
Click to expand...

I didn't test that. I've delidded 4 previous chips I've owned, and 4 more for friends/customers over the years (since Ivy Bridge launched). I don't bother to check anymore because I know the difference is massive. The first chip I ever delidded, a 3770k, saw over 30C drop @ 4.9GHz compared to stock TIM.
 
Zarathustra[H] said:
That's the highest single threaded Cinebench score I've seen.

Pretty awesome!
Click to expand...
Thanks. This PC is primarly used for flight sim, a program which is notorious for its dependency on single-thread performance. I'm seeing higher FPS in FSX than I've ever seen in my life. Highest I can find anywhere on the net either.
 
techguymaxc said:
Thanks. This PC is primarly used for flight sim, a program which is notorious for its dependency on single-thread performance. I'm seeing higher FPS in FSX than I've ever seen in my life. Highest I can find anywhere on the net either.
Click to expand...


Nice. Yeah, your multi-threaded Cinebench score of 1133 is only slightly behind that of my hexacore Sandy-E at 4.8Ghz, which is pretty damned amazing considering the 50% core difference.
 
Zarathustra[H] said:
Nice. Yeah, your multi-threaded Cinebench score of 1133 is only slightly behind that of my hexacore Sandy-E at 4.8Ghz, which is pretty damned amazing considering the 50% core difference.
Click to expand...
It's amazing what newer chips can do. Even though each year only brings incremental gains, go back a few years and it adds up. If you go back far enough (say Nehalem days) you can compare this amount of rendering power to dual processor servers. I've got a couple decomissioned Dell servers from work, each with dual 6 core X5660s and those are only slightly faster than the system you see at the top of the list in that screenshot. To think that 4 cores/8 threads could almost match 12 cores/24 threads is impressive no matter how old the chips. I've got a 5820k @ 4.4GHz in my home media server (poor clocker) and it scores about 1230, IIRC. Had a 5960x that only did 1290 at stock a couple years ago. Of course that's a very low stock clock, OC'd to 4.5 it did 1733.
 
Pretty high score congrats.

but sub 2 mins testing and no avx? Have you tried any longer/harder stability test ?
 
SvenBent said:
Pretty high score congrats.

but sub 2 mins testing and no avx? Have you tried any longer/harder stability test ?
Click to expand...

Yeah, that was just an early run to find the minimum threshold for some level of stability. I've done longer runs since but they require more voltage as I mentioned. That version of IBT does use AVX, also.
 
techguymaxc said:
It's amazing what newer chips can do. Even though each year only brings incremental gains, go back a few years and it adds up. If you go back far enough (say Nehalem days) you can compare this amount of rendering power to dual processor servers. I've got a couple decomissioned Dell servers from work, each with dual 6 core X5660s and those are only slightly faster than the system you see at the top of the list in that screenshot. To think that 4 cores/8 threads could almost match 12 cores/24 threads is impressive no matter how old the chips. I've got a 5820k @ 4.4GHz in my home media server (poor clocker) and it scores about 1230, IIRC. Had a 5960x that only did 1290 at stock a couple years ago. Of course that's a very low stock clock, OC'd to 4.5 it did 1733.
Click to expand...


Yep, my server in my basement is a dual Westmere hexacore L5640. It works fine for me, and my NAS and virtualization needs so I have no need to upgrade it, but if it were to shit them bed, and I were to replace it, I could probably get away with a server class Atom board today, and save a lot of power, if they only accepted registered RAM modules, and more RAM.

My server is currently rocking 192GB RAM, and I'd rather not downgrade that.
 
techguymaxc said:
Yeah, that was just an early run to find the minimum threshold for some level of stability. I've done longer runs since but they require more voltage as I mentioned. That version of IBT does use AVX, also.
Click to expand...

Are you sure on IBT with ABX?
I believe the project went dead before Linpack had AVX in it. Which was also the reason IBT fared worsen then prime 95 (w/avx) on my stress testing test i did some years back. IBT was harder thna prime95 on core 2 duo and amd system. but he AVX capeable I7 ha Prime95 as the harder test (temp and wattage pull)
Linx on the other hand continue on and got updated with a Linpack thah had AVX.

offause this is just from memory so i might be wrong. but i am curios.
 
SvenBent said:
Are you sure on IBT with ABX?
I believe the project went dead before Linpack had AVX in it. Which was also the reason IBT fared worsen then prime 95 (w/avx) on my stress testing test i did some years back. IBT was harder thna prime95 on core 2 duo and amd system. but he AVX capeable I7 ha Prime95 as the harder test (temp and wattage pull)
Linx on the other hand continue on and got updated with a Linpack thah had AVX.

offause this is just from memory so i might be wrong. but i am curios.
Click to expand...
If it doesn't then the new AVX offset feature isn't working correctly. When I apply an AVX offset and run IBT it downclocks the chip by that amount.
 
techguymaxc said:
If it doesn't then the new AVX offset feature isn't working correctly. When I apply an AVX offset and run IBT it downclocks the chip by that amount.
Click to expand...
Thank you for the info.
It might be possible i was testing with an older version but im pretty sure it was the same GUI.
i just looked into documentation and it didnt mention avx so i thought it hadnt been updated since my testing.
now i have to add in IBT in my stress testing testing
 
SvenBent said:
Thank you for the info.
It might be possible i was testing with an older version but im pretty sure it was the same GUI.
i just looked into documentation and it didnt mention avx so i thought it hadnt been updated since my testing.
now i have to add in IBT in my stress testing testing
Click to expand...

Probably.. you can find IBT with AVX on the IBT 2.54 version. but it may act weird with some Windows 7 system and some AVX2 capable processors, this doesn't occur with Win 8.1 or Win10.
 
Well GAWD DAYAAAAAAAAAAAAAAAAAAAMN
Really not much else to say other than that. Your system is beyond sick. Love the dual-rad setup too. Mad props on the achieved clockspeed and the successful delid. Geezus lawdy.
 
Nice !!

Very clean PC I have to admit !


The thing is, will it hold some severe hour long stress test ? I can push mine to the same, even keep the oc n the DDR4 but it will fail certain tests then, it will still do Passmark ( 3043 IPC points @ 5.2-3866 ), CineBench and a few others but fail on RealBench somewhere down the loop and fail IMMEDEATLY on prime95_v28.10b1 ( the newest one, has AVX hardcore like enabled ).

I could maybe raise the Volts to get it stable but I need to delid first, right now temps are OK if I stay where I am as in sig.


I found the hardest part was to match the OC of the CPU with XMP and then once I had that under control to drive the RAM further, it booted till 4133 on stock CPU settings. So I think the RAM would do even more but at the cost of even higher Voltage on VSA and VIO ( already at the limit ) and raise the CL from 16 to 18 or 19, both I think is contra-productive.

After many dozen stress tests with multiple apps I have found this setup to be one of the trickiest to oc I have personally ever had, compared to my golden Sandy this is a Maze like operation if you want to combine 5G with high DDR4 clocking. The thing with VCCIO is that, that once you are close to stable errors take long to occur, 1-4h like. A 1 hour test may go OK 2 times, fail another one, then be OK for another 3 runs each 1h. Raising VCCIO from 1.2000 onward in verysmall increments brought me to the stablest setting I could yet achieve. Knowing the CPU passes all tests with 5G and the RAM passes all tests at 3866, each one alone, it drove me almost nuts to find the right parameters for vDIMM VCCIO and VCCSA on even XMP. XMP on 5G did not run stable, it needed tweaking. Actually the settings are all the same on XMP3600 and 3866. Further pushing needs CL18 or worse for the Samsung "B" die. I dont think my IMC wants any higher Voltage either. You have to make a balanced decision here as you cant have both with just water m( everyday usage, quiet too ). The KabyLake just wont deliver that, I accepted that. High CPU at 5G or above and high RAM settings are a thing of its own if you only consider a 4h run as sufficient proof of stability, 8h even better.

For my combo, it will run at VCCIO at 1.2500, but fail soon in prime, ..1.275 may hold for a whole 30min, maybe 1h but it will fail, sooner or later...with 1.300VCCIO it holds it for 4-5h ( my longest I have done yet, played WoWarships while I primed it :p ). vDimm I had to raise also, if I lower it by 0.025 to 1.375 it will also fail...sooner or later, both XMP and 3866 at 5G. VCCSA has been an issue since Z170 Skylake,
my MSI M9 board pushed 1.35 on Auto through my IMC for 3200 XMP till it smoked the CPU very likely, at least the CPU went RMA and got replaced, along with mobo back then, Asus M9E then and new 6700k, same XMP3200...same VCCIO..Asus also pushed the SA to 1.35 and even 1.4 in OC for the same Corsair 3200. MAde me think. I gave it all back then after those bricked 2 times and got me the rig in sig. That hasnt bricked yet and never had an issue with rebooting and oc...Qcodes and what not else...1/3 of the price to the before boards and more solid, more bang for the buck.

I dont think a higher Asus board would make that easier I have to say, my 2 Z170 boards were such and as described, also troublesome when you oc'ed the CPU and the RAM. I never got lucky with those boards due to bricked FW's but I worked them for 2 month. VCCIO and VCCSA are the Zx70's problem with high overclocking. I dunno what killed my 6700k but it may well have been that.


Thinking about the following:

30,- bucks for Intel Insurance for overclockers

------------------------OR------------------------

30,- for a delid tool


it's an either or, you really cant have both here

I would like to see how a stock DDR4-2400 module would behave in the system. Wondering how much simpler overclocking the CPU alone would be and what difference it would make in hardcore games that rely on IPC.
 
Dannotech said:
Do you think it could survive running my newest stress test in Statuscore? http://www.dannotech.com/statuscore
Click to expand...

I've downloaded and installed it, I can see that it uses AVX since my CPU is downclocking to 4.7GHz due to the 500MHz AVX offset I specified in the BIOS. I'm sure the chip could run at that speed with no problem. Are you asking for a 5.2GHz AVX pass? In my experience it's really only stress tests that make use of AVX so I'm not too concerned about the stability of this instruction set. So far in real world use this week I've had zero crashes with plenty of benchmarking and gaming.

If you want to see the results @ 4.7 I'm fine with that. What length of time do you consider necessary to determine stability with your stress test?
 
Doing a run on 5-3866....will see when I come back if it still runs @ 1.35 adapt. vCore.

From what I see now it is much more conservative with AVX than p95_v28.10 and draws less power than even p95_v26.6 if I read that right. Temps are ok, highest yet was 71°C. The machine has been priming all day and is warmed up btw.

How does it show that it failed ?
 

Attachments

  • Unbenannt.JPG
    Unbenannt.JPG
    278.6 KB · Views: 63
Nice system... reminds me of when I had my Q9650 and broke 4.2ghz with decent voltages.
 
techguymaxc said:
I've downloaded and installed it, I can see that it uses AVX since my CPU is downclocking to 4.7GHz due to the 500MHz AVX offset I specified in the BIOS. I'm sure the chip could run at that speed with no problem. Are you asking for a 5.2GHz AVX pass? In my experience it's really only stress tests that make use of AVX so I'm not too concerned about the stability of this instruction set. So far in real world use this week I've had zero crashes with plenty of benchmarking and gaming.

If you want to see the results @ 4.7 I'm fine with that. What length of time do you consider necessary to determine stability with your stress test?
Click to expand...


It uses a good mix of the X86 instruction set. So yea its doing AVX as well as standard FP and integer math, loads and stores, branch prediction and macro op fusion, hardware prefetching, and hyper-threading specific tasks, i.e. it's really trying to do a bit of everything at once.. I was really just curious if it could do the test at 5.2GHz for no more than 5 minutes, since 4.7GHz is pretty much a no-brainer for a Skybylake.

:)
 
Dannotech said:
It uses a good mix of the X86 instruction set. So yea its doing AVX as well as standard FP and integer math, loads and stores, branch prediction and macro op fusion, hardware prefetching, and hyper-threading specific tasks, i.e. it's really trying to do a bit of everything at once.. I was really just curious if it could do the test at 5.2GHz for no more than 5 minutes, since 4.7GHz is pretty much a no-brainer for a Skybylake.

:)
Click to expand...

I'll turn off the offset and see how it does for a few minutes.
 
I let it run for about 20-25 minutes with no offset, here's a screenshot:

statuscore_zpsyc2tv5i3.jpg


BTW, feature recommendation: you should create a log each time the software is run and capture the results. I like your test methodology, seems like a mix of instructions would be more likely to
1) put the whole chip to work
2) simulate a real world stressful workload (or even multiple simultaneous workloads)

Question: is the MIPS rating inclusive of the load/store instructions or do you not count those? Also, I'm guessing no HT support?
 
techguymaxc said:
I let it run for about 20-25 minutes with no offset, here's a screenshot:

statuscore_zpsyc2tv5i3.jpg


BTW, feature recommendation: you should create a log each time the software is run and capture the results. I like your test methodology, seems like a mix of instructions would be more likely to
1) put the whole chip to work
2) simulate a real world stressful workload (or even multiple simultaneous workloads)

Question: is the MIPS rating inclusive of the load/store instructions or do you not count those? Also, I'm guessing no HT support?
Click to expand...


That is a damn impressive chip.

So yes Statuscore does have HT support, actually, its just that it combines the scores from both threads and shows you the statistics on a per code basis The reason is because it is running 2 different (asymmetrical) programs on each thread; thread 0 is doing the ALU/FPU/AVX instruction mix while the 2nd thread is doing the load/stores. The reason for this to make sure the load stores do not cause dependencies in the pipeline that would otherwise stall the IPC from reaching its full potential. Thread0 is pretty close to 4 IPC anyway and I think Thread1 is much lower (in isolation).

Thanks for the feedback. I am definitely interested in making it more usable for data recording because I'd love to see it used in CPU reviews. And with Ryzen coming I think the IPC is totally relevant to the conversation. I want to see who's got the better IPC without respect to any clockspeed variations =) And I am definitely interested in hearing more feedback!


[EDIT] P.S. Yea the mips is inclusive of ALL instructions that execute on the processor core, even in the operating system.
 
SvenBent said:
to bad its win10 only i would have lot to put in in my stress test testing.
I do have core damage but coredamage is only doing so so.
Click to expand...

I've compiled it for Windows 7 but as soon as it gets installed BSOD. And Microsoft is providing no support for developing drivers on Windows 7 so I'm on my own. I need a Virtual Machine with x86 remote kernel debugging and probably an old version of the WDK because I guess Microsoft doesn't want people using Windows 7 anymore.
 
You must log in or register to reply here.
Back
Top