R6E: Any clue on what is the expected i9-7940X overclock?

[Inkar]

Hi,

Does anyone have a clue about what is the all core expected i9-7940X (non-avx) overclock range with the Rampage VI Extreme and an AIO?.
I know nothing is guaranteed other than the factory specs. I just can't find any reviews with this kind of data for the 7940X.


Thanks.

[Protocol]

Silent Scone wrote:
Leave phase profiles at defaults, simply increase the current limit. If the CPU has not been delidded, AVX range is limited. You're pulling 300W+ through the die, and you're not able to pull the heat away fast enough. You can set a small offset for cache and keep voltages low. Should be able to reach 3Ghz with 1.05v or less, this can help with temps. Lowering VCCIN whilst checking for stability can help, too.

Short of that, you'll need to delid the CPU.

Thanks for the info there. At the time I hadn't gotten around to the cache clocking but you are right - a small bump to 3 GHz needed only ~1.020v (offset mode). Stable at that voltage. Gained ~2% in graphics heavy benchmarks. Like 12% in memory-intensive benchmarks though.

I did set the CPU power phasing back to standard. Didn't appear to make a difference in temps or stability. Leaving it at standard.

Thinking about recording all settings and starting over with BCLK instead just to see what I can get out of it - we will see how my availability to tackle that is though. It's been a busy week so far.

[Kalorn]

Not bothered really pushing mine to see what I can get.

Settled myself on 4.5 GHz @ 1.2V - system is happy and in general under heavy (not extreme) load maxxes out around 65-70C. Extreme load (i.e benchmarking it can hit up to 95C (as shown on my realbench score).

May have to at some point delid it - but not fussed myself, I think I will be more than happy to sit at this for the foreseeable future, unless I get around to finally doing a full water cooled system rather than just using an AIO.

[Skunkfoo]

I had a Thermaltake 360 Floe AIO for bench testing. Anything above 4.5 couldn't pass Aida64 CPU System Stability Testing due to 90C+ temps then thermal lockup/shutdown. I even tried seating the CPU 2 times to make sure the TIM was evenly spread and the block made proper contact with the IHS.

I may get the CPU delidded down the road if I find that I can safely pump more volts into the Chip (like 1.3+). But for now i'm under custom water cooling and I'm now at 46/47/48 GHz. Using Adaptive gave best results and thermals once I figured it out. Doing OC By Specific Core really helped fine turn each core based on its characteristics.

For example:
12 cores are set to 4.6 GHz with max voltages between 1.208 and 1.247
1 core is set to 4.7 GHz with max voltage @ 1.251
1 core is set to 4.8 GHz with max voltage @ 1.254

Cache Frequency is currently set to a modest 2800 MHz with .01 + Offset (I have a goal of 3200 MHz and will take baby steps keeping all other OC settings the same)

DRAM Speed set to 3600 MHz on a 3733 MHz kit using Auto on all settings. Every time I try to set 3733 using all Auto my system cant ever stay stable for long no matter what Cache and CPU frequency I use. I haven't gotten around to playing with and trying to tighten the timings on the DRAM yet.

Side Note: Anyone been able to get Intel Turbo Boost Max Technology 3.0 to actually have a targeted application use the top performing core(s) yet? No matter how I configure the application I cant get my Test Subject Game (Heros of Newerth) to always run on the 4.8 GHz core.

[Silent_Scone]

Skunkfoo wrote:
I had a Thermaltake 360 Floe AIO for bench testing. Anything above 4.5 couldn't pass Aida64 CPU System Stability Testing due to 90C+ temps then thermal lockup/shutdown. I even tried seating the CPU 2 times to make sure the TIM was evenly spread and the block made proper contact with the IHS.

I may get the CPU delidded down the road if I find that I can safely pump more volts into the Chip (like 1.3+). But for now i'm under custom water cooling and I'm now at 46/47/48 GHz. Using Adaptive gave best results and thermals once I figured it out. Doing OC By Specific Core really helped fine turn each core based on its characteristics.

For example:
12 cores are set to 4.6 GHz with max voltages between 1.208 and 1.247
1 core is set to 4.7 GHz with max voltage @ 1.251
1 core is set to 4.8 GHz with max voltage @ 1.254

Cache Frequency is currently set to a modest 2800 MHz with .01 + Offset (I have a goal of 3200 MHz and will take baby steps keeping all other OC settings the same)

DRAM Speed set to 3600 MHz on a 3733 MHz kit using Auto on all settings. Every time I try to set 3733 using all Auto my system cant ever stay stable for long no matter what Cache and CPU frequency I use. I haven't gotten around to playing with and trying to tighten the timings on the DRAM yet.

Side Note: Anyone been able to get Intel Turbo Boost Max Technology 3.0 to actually have a targeted application use the top performing core(s) yet? No matter how I configure the application I cant get my Test Subject Game (Heros of Newerth) to always run on the 4.8 GHz core.

Assuming you've applied the profile, some games don't work well with IBT. You can always try handing it over to native control and see if that helps. Personally, have seen mixed results with that.

[Homewrecker]

Any updates as to who's running stable frequencies now and their associated settings. Just got my 7940X from SL and looking for a baseline to start with. The CPU has been binned at 4.4Ghz but I should be able to get more out of it on custom water.

[Silent_Scone]

Homewrecker wrote:
Any updates as to who's running stable frequencies now and their associated settings. Just got my 7940X from SL and looking for a baseline to start with. The CPU has been binned at 4.4Ghz but I should be able to get more out of it on custom water.

Non delidded
4.7Ghz *Prefered cores 4.8 1.25v Adaptive
4000Mhz CAS 16-17-16-41-1T 1.4v
Cache 3.0Ghz @ 1.05v
VCCIN 1.85v LLC6
AVX -3 / AVX512 -5

[Hot_Trout]

Current Build :

CPU : Intel Core i9 7940X (Non Delid)
Motherboard : ASUS Prime x299 Deluxe
RAM : TEAM GROUP XTREEM 32GB 3600MHZ (16-16-16-36)
Optane : Intel Optane 32GB
Graphics Card : Asus GTX 980ti 20th Anniversary Edition

Boot Drive : Intel Optane 900P 480GB AIC Add in Card PCIe SSD
App Drive : 1TB NVMe Samsung SM961 M.2
Game Drive : 1TB NVMe Samsung 960 Pro M2 48 Layer
Data Drive : WD Black 3TB WD3003FZEX
Cache Drive : 250GB OCZ Revo Drive 3
Storage Drive 1 : WD Gold 10TB Helium WD101KRYZ
Storage Drive 2 : WD Gold 10TB Helium WD101KRYZ

Case : Corsair Obsidian 700D + Mods
PSU : Seasonic 1250 watt GOLD Modular
Radiator : RASA RX360 Special Edition
Water Block : EK Supremacy Evo x299 Prime Water & VRM Block
Water Fittings : EK Compression Fittings
Jizz : Mayhems Pastel White Coolant
Thermal Conduit : LIQUID ULTRA THERMAL PASTE
Pump & Tank : Eheim 300 + Home Made Header Tank 4litre



Current BIOS Settings :

4.6Ghz on all cores except 4, 6 and 10 which are at 4.5GHz for cooling reasons only
3600Mhz CAS 16-16-16-36 1.1v as per XMP Default
Cache 3.0Ghz @ 1.05v
Cache Ratio : 30
AVX : 4 (4200Mhz)
AVX 512 : 5 (4100Mhz)
VCCIN : 1.85v
CPU VCCIO : 1.1
CPU System Agent : 1.05
Load Line Calibration (LLC) : 6




This configuration passes the following tests :

Intel Diagnostic Tool v4.1.0.24.W.MP Full Test Max Temp 82oc
Computer Pulls and pushes data all day 24/7 with full CPU loads under this build.

Could I ask if anyone could offer some further advice on the correct VCCIO and CPU System Agent settings. Do you use auto for yours?

[Thireus]

Hi guys,

I can read that several of you are reaching frequencies above 4.5*GHz on all cores. Congratulations.

I was just wondering one thing, are these frequencies stable under extreme stress testing conditions? And by extreme I mean the following:

- Running Prime95 v26.6 (non-avx version) blend test 24h
- Running Prime95 v26.6 (non-avx version) 8K FFT test 20min (to make sure temperatures stay below 94C)
- Running Prime95 recent version (with AVX) blend test custom 1h for AVX
- Running y-cruncher HNT stress testing for more than 100 passes to test AVX-512

I have a 7980XE (so a bit off topic) but I'm currently able to pass all the above Prime95 testing with Vcore offset mode and per-core overclocking, I have a few cores at 42x, some x43, most at 44x and one at 45x.

The y-cruncher HNT stress test for AVX-512 has been the trickiest one so far. I've had freezes when setting cores at 44x/45x, so I've had to increase LLC for it, for which I'm still trying to find the right value (as I'm writing these lines I have LLC set to 6 and it successfully did 62 passes). Also had to set an AVX-512 offset down to 8 for it...

Anyone trying to achieve extreme stability or are you guys running softer stability tests?

[Marko911]

Thireus wrote:
Hi guys,

I can read that several of you are reaching frequencies above 4.5*GHz on all cores. Congratulations.

I was just wondering one thing, are these frequencies stable under extreme stress testing conditions? And by extreme I mean the following:

- Running Prime95 v26.6 (non-avx version) blend test 24h
- Running Prime95 v26.6 (non-avx version) 8K FFT test 20min (to make sure temperatures stay below 94C)
- Running Prime95 recent version (with AVX) blend test custom 1h for AVX
- Running y-cruncher HNT stress testing for more than 100 passes to test AVX-512

I have a 7980XE (so a bit off topic) but I'm currently able to pass all the above Prime95 testing with Vcore offset mode and per-core overclocking, I have a few cores at 42x, some x43, most at 44x and one at 45x.

The y-cruncher HNT stress test for AVX-512 has been the trickiest one so far. I've had freezes when setting cores at 44x/45x, so I've had to increase LLC for it, for which I'm still trying to find the right value (as I'm writing these lines I have LLC set to 6 and it successfully did 62 passes). Also had to set an AVX-512 offset down to 8 for it...

Anyone trying to achieve extreme stability or are you guys running softer stability tests?

Hey,In my opinion,you have decent 7980XE chip.I'm amazed with your results with stock TIM under that chip.But higher core count bigger die makes better contact with IHS 🙂
If you didn't delid your CPU,try it.You will be surprised with results.and I don't think you would need to do avx-512 offset at whooping -8.

[Protocol]

Thireus wrote:
Hi guys,

I can read that several of you are reaching frequencies above 4.5*GHz on all cores. Congratulations.

I was just wondering one thing, are these frequencies stable under extreme stress testing conditions? And by extreme I mean the following:

- Running Prime95 v26.6 (non-avx version) blend test 24h
- Running Prime95 v26.6 (non-avx version) 8K FFT test 20min (to make sure temperatures stay below 94C)
- Running Prime95 recent version (with AVX) blend test custom 1h for AVX
- Running y-cruncher HNT stress testing for more than 100 passes to test AVX-512

I have a 7980XE (so a bit off topic) but I'm currently able to pass all the above Prime95 testing with Vcore offset mode and per-core overclocking, I have a few cores at 42x, some x43, most at 44x and one at 45x.

The y-cruncher HNT stress test for AVX-512 has been the trickiest one so far. I've had freezes when setting cores at 44x/45x, so I've had to increase LLC for it, for which I'm still trying to find the right value (as I'm writing these lines I have LLC set to 6 and it successfully did 62 passes). Also had to set an AVX-512 offset down to 8 for it...

Anyone trying to achieve extreme stability or are you guys running softer stability tests?

No "forgiving" testing for me. If it won't run any instruction set in any stress program/benchmark/non-synthetic workload at absolute 100 percent for at least 8 hours then I do not consider my system stable. (I only perform 8-hour testing if a smaller 2-hour test passes. Usually minor tweaks over a day or so gets everything dialed in tightly). Being lidded, I am hitting a thermal wall myself. Just can't get enough heat out and into the waterblock/coolant.

Skylake-X in general suffers from heat-transference issues due to the TIM compound that is used. I'm sure a hefty number of people are able to hit clocks with decent voltage overhead remaining but are reaching the TMAX instead (non-delidded of course).