Overclocking 5820K. A personal experience.

[RejZoR]

Here you go @heky . I have it set to 1.175V in BIOS, but CPU-Z reads 1.185V. Still... This is my stable overclock that I use daily for games, heavy file compression and and occasional video encoding.

 

[heky]

Here you go @heky . I have it set to 1.175V in BIOS, but CPU-Z reads 1.185V. Still... This is my stable overclock that I use daily for games, heavy file compression and and occasional video encoding.

View attachment 75817

Sorry to burst your bubble, but running Realbench is far from being stable at full load.(with only the video encoding option selected at that). But yeah if it serves your purpose...

 

[RejZoR]

I've used RealBench just to ramp up the clocks so you see what I see at full load plus it shows some system specs. I've tested this OC in various stability programs. I used to have 1.150V but it wasn't stable under heavy load. This appears to be.

 

[heky]

I've used RealBench just to ramp up the clocks so you see what I see at full load plus it shows some system specs. I've tested this OC in various stability programs. I used to have 1.150V but it wasn't stable under heavy load. This appears to be.

Show me a 30 minute run of Aida64 stability test, with just the FPU option ticked. We will talk again after...oh, and do show your temps at that.

 

[RejZoR]

That's like setting a V8 car to run just 2 cylinders at 50.000 RPM to evaluate if the engine is reliable...

EDIT:
I guess we'll be running it anyway...

 

[RejZoR]

 

[xkm1948]

Why just stress FPU instead of all system?

 

[R-T-B]

Why just stress FPU instead of all system?

Cause the guy asked above.

 

[cadaveca]

Why just stress FPU instead of all system?

FPU pushes far higher power draw and increases temps fairly good too. Given his load temps, and his voltages, he's got a really decent CPU, and might get 5 GHz under high-end watercooling. His clocking is better than one of my 5930Ks...but my 5390Ks do not have gimped PCIe .

Same as with running the encode in RealBench... I get more crashes on the last test, passing encode is pretty easy (relatively speaking).

Then I also see his very low cache speed, 3 GHz. Then it's like...oh, OK then.

If ya wanna hammer the memory performance, you gotta increase dat cache, yo. Even the 6950X I'm playing with now need decent cache speed to get decent bandwidth @ 3200 MHz... otherwise the bandwidth isn't much better than the stock 2400 MHz; only copy perforamcne increases, and latency drops. Unfortuantely, I can't get much more than 3600 MHz on the Broadwell-E, while both my 5930K's can do 4.2 GHz with ease.

But dem 4 added cores..... mmmm....

So, 5820K can still be a very good chip to buy, if you can get a good one!

 

[RejZoR]

My CPU seems to be super sweet up to 4.5 GHz and then it's a total no go from then on. I've tried everything, 4.6, 4.7, 4.8, 4.9 and it wouldn't work regardless of voltage. 5 GHz wouldn't work even on one core. I'd have to really invest more time fiddling with it, but at 4.5 GHz on all 6 cores, I just can't be bothered to be honest. As for the cache, considering all the horror stories about degradation, I'm keeping it stock. For what I do, I don't really notice a difference. I might clock it higher later on when CPU will be a bit outdated and needing some boost.

The "gimped" PCIe is a non-issue since I only have 1 graphic card and 1 soundcard. I've had a M.2 SSD, but I sold it for the 2TB Samsung SSD.

As for the temperatures, I have my system setup for silent operation. If fans weren't set so conservative, the temps would be a lot lower I think. For short heavy loads, perfectly fine and it's so quiet most of the time I have to look at the lights if it's even working (if screen is in standby), especially now that I don't have any HDD in it.

 

[heky]

Nice one @RejZoR, you really have a golden sample. Thanks for sharing.

I need 1.254v to be stable at same clocks under full load( my cache is at 3.5ghz@1.15v though) . But i do have lower temps thanks to EK.

 

[xkm1948]

With my 128GB RAM my OC limit is down from ~4.4GHz to 4.25GHz. Running max RAM with DDR4-3000 speed seems to drag down OC headroom severely.

 

[xkm1948]

FPU pushes far higher power draw and increases temps fairly good too. Given his load temps, and his voltages, he's got a really decent CPU, and might get 5 GHz under high-end watercooling. His clocking is better than one of my 5930Ks...but my 5390Ks do not have gimped PCIe .

Same as with running the encode in RealBench... I get more crashes on the last test, passing encode is pretty easy (relatively speaking).

Then I also see his very low cache speed, 3 GHz. Then it's like...oh, OK then.

If ya wanna hammer the memory performance, you gotta increase dat cache, yo. Even the 6950X I'm playing with now need decent cache speed to get decent bandwidth @ 3200 MHz... otherwise the bandwidth isn't much better than the stock 2400 MHz; only copy perforamcne increases, and latency drops. Unfortuantely, I can't get much more than 3600 MHz on the Broadwell-E, while both my 5930K's can do 4.2 GHz with ease.

But dem 4 added cores..... mmmm....

So, 5820K can still be a very good chip to buy, if you can get a good one!

Hey I can take that 6950X off your hand if you want. I have use for 4 moar cores!

 

[cadaveca]

Hey I can take that 6950X off your hand if you want. I have use for 4 moar cores!

It's going back to Gigabyte in Taiwan this evening.

 

[xkm1948]

I assume a 6950X full review coming up soon.

 

[cadaveca]

I assume a 6950X full review coming up soon.

Wasn't provided for a CPU review and Intel declined to provide sample for said purpose (jerks! ). You will see a mobo review with it for sure, but I do have a few reviews now waiting to be published.

I am expecting another Broadwell-E sample some time in the future though, so maybe that chip will get reviewed. I'm not sure which chip I'll get next though. Intel's marketing department is in shambles since the lay-offs, so they need to pull their heads out of their collective butt and get me chips for review or I can't be bothered to do one properly. I wasn't very impressed with the CPU I got and is pictured above, TBH.

 

[avatar_raq]

I wasn't very impressed with the CPU I got and is pictured above, TBH.

Every review I read about Broadwell-E mentions the limited OC headroom, and because I was hoping the 5820K broadwell-e counterpart has the full 40 PCI-e lanes or 2 more cores I was very disappointed and went a head and bought a 5820K as it is cheaper, available now and hopefully overclocks better.

 

[avatar_raq]

I did some more overclocking and so far I got a stable 4.6 Ghz @ 1.3v core, 1.95v input voltage (without LLC, drops to 1.85v under load). I stopped using Prime95 and LinX and used ASUS realbench and Aida64 stability test instead. Also I connected the fans of H110i GTX to the motherboard directly and set a custom fan curve, as I found it easier this way because when the fans were connected to the CPU block, the Corsair link software used the water temperature instead of CPU temp. Pump profile was left at quiet.

The next step is to find if I can lower the core voltage without sacrificing stability, or probably see if I can achieve a stable 4.5Ghz with 1.2v or 1.1v. Fingers crossed.

 

[cadaveca]

even 4.6 @ 1.3V is pretty good. Your package temps are just a bit high though; try to stay under 75C. Lowering cache(ring) voltage and multi can have dramatic effects sometimes.

 

[RejZoR]

What's a stock voltage for cache anyway on 5820K ? The one showed in BIOS?

 

[avatar_raq]

even 4.6 @ 1.3V is pretty good. Your package temps are just a bit high though; try to stay under 75C. Lowering cache(ring) voltage and multi can have dramatic effects sometimes.

I reduced the Vcore to 1.25v and it looks to be stable with better temps. Good news. I left the cache multiplier and voltage at default (x30, 0.9v respectively), by selecting an offset of 0.001 and not even leaving it at auto. I do not know why the package temp is so much higher than core temps.

Now I have few questions:
1- Did I make the right decision by ditching the brutal stability tests (Prime95, LinX)?
2- Do I leave it at this OC (4.6Ghz@1.25v) or do I go back to my previous one (4Ghz@1v)? I am concerned about the longevity of the CPU and plan to keep it long as possible.
3- With the new OC, the package temp is in low 70s with occasional spikes to 80 C. Is this OK?
4- If the answer to question 2 is 'keep the 4.6 OC', do my bios settings look ok? Did I miss anything?

Here are my settings:
Bclock: 100
Core multiplier: x46
Vcore: Adaptive, offset +0.001, Total Turbo Voltage: 1.249v
Cache Multiplier: Auto
Cache Voltage: Adaptive, offset +0.001, Total Turbo Voltage: Auto
Input Voltage" 1.9v (Drops to ~1.8v under full load)
LLC: Level 1 (0%)
System Agent Voltage: Adaptive, offset +0.001 (result ~0.8v)

What's a stock voltage for cache anyway on 5820K ? The one showed in BIOS?

For me it is ~0.737v in idle (core multiplier x12) and ~0.9v under full load (core multiplier x46). Like you I did not OC or overvolt the cache.

 

[cadaveca]

What's a stock voltage for cache anyway on 5820K ? The one showed in BIOS?

There is no set "Stock"; it varies from chip to chip. So to find your "stock", on an ASUS board, clear CMOS, boot into BIOS after clear, and check monitoring page in BIOS. Voltage listed is your "stock".(note this doesn't include turbo boost).

Now I have few questions:
1- Did I make the right decision by ditching the brutal stability tests (Prime95, LinX)?
2- Do I leave it at this OC (4.6Ghz@1.25v) or do I go back to my previous one (4Ghz@1v)? I am concerned about the longevity of the CPU and plan to keep it long as possible.
3- If the answer to question 2 is 'keep the 4.6 OC', do my bios settings look ok? Did I miss anything?

1: Yes. There are many other tests that are good for testing stability for more users, such as ASUS Realbench. P95 and LinX are extreme load examples, and are useful for testing the capability of your cooling, or stability under those specific loads. For memory-specific or cache testing, I might suggest something else, but ASUS Realbench ATM is fairly good.
2: 1.25V is conservative, and should last you as long time, especially since you are leaving cache at default. what I would do is stop using offset voltage, since offset is relative to the multi in use, and as such, does not guarantee your voltage is at any set level for the domains that are set to offset.
3: Please upload screenshots of your BIOS for proper suggestions. You can do this using USB drive and pressing "F12" in BIOS to capture the screenshot.

 

[RejZoR]

Considering RealBench H.264 loads all cores to max and uses AVX, I think it's a very reasonable load.

 

[avatar_raq]

Here are the bios screenshots. I tried to be thorough

 

[avatar_raq]

.... what I would do is stop using offset voltage, since offset is relative to the multi in use, and as such, does not guarantee your voltage is at any set level for the domains that are set to offset.

I did not get this. I do not use offset, I use adaptive with an offset of 0.001 as it is the lowest value and I do not want to leave the offset option under adaptive to auto in fear of the motherboard adding more voltage. The goal is to keep the voltage at default when the multiplier is 33 or less.