Overclocking

I'm looking to get opinions on AVX instructions with regards to overclocking and stability.

My recent venture into overclocking my new 8700k lead to somewhat disappointing results at only 4.8Ghz stable, that was Prime95 stable for about an hour and with AVX instructions on.

I see many people now using AVX offsets in the BIOS to reduce the core speed by 200-300Mhz when AVX instructions are detected since they seem to add about 15 degrees to temps under Prime95. Some people are just avoiding benching with AVX based tools all together.

My new build is a 8700k on a MSI Gaming Carbon Pro AC mobo cooled with a Noctua DH15. it's not delidded and I don't intend to.

So I'm trying to wrap my head around this, would it be best to bench under both AVX and non-AVX use and find the most stable settings for both and then use an AVX offset to achieve those settings?

Or is it not worth being a purist and just benching with it turned of as long as muh gaymes are stable? Or is there better "real world" benchmarks worth running?

This kinda reminds me of the "power virus" debacle with Furmark...

Just to bump and update, looks like I've found a stable clock at 5Ghz benching Prime95 with the AVX flag set to 0 and 4.8Ghz with it set to 1

So I've set 5.0Ghz in the BIOS with a -2 AVX offset and that seems to work OK.

79 degrees at 5Ghz @1.35 vCore ain't bad.

Whats the point of throttling the cpu in only synthetics?

That doesn't sound too bad, I could only get mine to 4.5 stable

Have you delided?

>Buys CPU with toothpaste TIM
>Hur dur I'm not delidding

I think he's a dumb cunt...

He wants to push a few extra Mhz but he's not willing to get free performance from a delid. Alright mate...

>would it be best to bench under both AVX and non-AVX

How are you to stupid to figure this out? find what freq is stable with AVX enabled and find what freq is stable with AVX disabled and then you have your overclock and offset for AVX freq. it's not rocket science.

>Or is there better "real world" benchmarks

Rendering x264/x265

It's not specific to synthetics, In fact it looks like Prime95 might be the only synthetic stress test to use AVX. It's specific to that instruction set and what apps use it.

By all accounts because of the specific nature of the instruction set, only a handful of specialist applications like say the adobe suite, MATLAB and that kind of stuff use this. And of that small subset an even smaller subset use them in such a way to max out your CPU, like video encoding.

Most motherboards have an "Auto" vCore which will be very high if you leave it Auto and run a high multi, you can manually tease that down if you set your own voltages.

I've not delidded and it's primarily because I'm not experienced with doing that and because you're messing about right on the chip to clear the TIM a small mistake can brick a £350 CPU.

It's a bit of an over reaction to say that I want to push a "few" extra Mhz, but won't de-lid for free performance. Going from the stock Turbo of 4.3Ghz to 5Ghz is a 700Mhz increase, where as de-lidding from everything I've read will maybe push you an extra 200Mhz if you're lucky with the binning of your CPU.

>it's not rocket science.
No but there seems to be a lot of different conflicting ideas out there about what really constitutes "stability" and whether that should be context sensitive to your usage, and some purists who think AVX offset is cheating etc. So just wanted to gather some different perspectives.

>Most motherboards have an "Auto" vCore which will be very high if you leave it Auto and run a high multi, you can manually tease that down if you set your own voltages.

So I set my own voltage using adaptive, but it keeps going above the limit anyway, and it seems that the temps are linked to the core speed, not the voltage which seems backwards. As in, at 4.5 GHz temps are fine, with 4.6 GHz, but same voltage settings the temps go above 90. Is this just ASUS being shit?

Delidding is required if you want to overclock 8700k.

I hate to hijack this thread but I'd also say making a new one on the same topic for just another CPU is ridiculous.
I have a (what you'd call) poverty-tier AMD Ryzen 3 1300x with the stock cooler in pic related. It's running at 3.7GHz at stock voltage. I'm trying to juice more CPU performance since that seems to be what's making the system slow paired with an RX 480. What happens when you bump the voltage up on a CPU? How do I check for warning signs if I do? Can it reach 4GHz?

Check temps don't get too high. That's really the main limit for OCing. I think Ryzen can just reach 4.0 GHz

So just install a temperature monitor software, check it's not going above 80*c or so and put the voltage up? I'm guessing around 1.4v for 4GHz?

ryzen > coffee kek

Yeah, I use HWinfo. I think there are recommended voltage caps as well, typically about 1.4V, regardless of temperature. At least that's what I'm trying to do with my 8700k, but I'm having some difficulties

overclock with stock voltage.
overvoltage is not efficient.

Depends on the CPU, if you are unlucky it might not even reach 4 GHz stable.
The sweet spot for Ryzen CPU is around 1.3 Volt, going higher is not really worth it.

This is generally because of the heat that's generated at high overclocks, heat alters resistance and that alters voltage/power so the demands change under load, this is known as vdroop. That's why LLC was invented, all good motherboards have this option in the bios and allows you to control for vdroop, play with those settings to help maintain stable vCore between light and heavy loads.

I understand that to hit the highest overclocks de-lidding is required. But the question I'm asking is relevant to both regular and de-lidded CPUs as both suffer more vCore when AVX instructions are used. All de-lidding does is allow you to replace the stock TIM with something better and achieve better cooling.

Worth it in what way? Temperature or power draw?
Not concerned about the latter, not concerned about the former as long as it isn't reaching unsafe levels.

Not worth it in the sense that going from 1.3 to 1.4 Volt would probably get you 100 MHz at most while temperature and power draw would be much higher.

overcocking with stock voltage is literally free performance. temps and power will be the same.

ryzen's ceiling is 4GHz on a good binned chip anyway. 4.1GHz on a special snowflake reviewer chip.
try overcocking with stock voltage first and see where it goes.

Right, I've got it stable at 3.625v at 3.9ghz and it's a huge improvement. However i can only set this with Ryzen Master and it doesn't persist through reboots. If i try to set it via bios these are the only power options i have. If i just set the clock it just reboots and gets stuck. These options have auto, normal and +-x.xxxv settings but i have no clue how to set it to 3.625 from here.

Sorry, seems i mean 1.3625 not 3volts, jeezus

Nevermind, I figured it out. Seems to be a stupid Gigabyte thing. I needed to know the absolute base Vcore setting (in this case 1.225v which you can ONLY find from Ryzen Master) and add increments from that. I've given it +0.120v on the DVID Vcore and a +0.050v on the SOC DVID for good luck. Got me up to 3.9. Would it be unsafe to push the DVID Vcore up to +0.180 to try for 4GHz? Temperature seems the same.

S

fucking 4chanx didn't let me comment.

Do these temps seem kinda high for 1.25 V under intel burn test at stock voltage?

I just reapplied thermal paste because I was having issues getting to 4.6 GHz under a Noctua D9l, it seems to have helped a bit, previously the heat spreader wasn't completely covered.

8600k at 4.8ghz comfy, 1.3v (was 1.28 until bios update )
Have plenty of room for 5.0ghz. Just being conservative.
Manual Volts.
140% power settings.
Line limit 3

Oh at 66degrees C average ..... so comfy
, noctu a dh15s