7920x on air?

[Ziphnor42]

I have been holding back upgrading my 5820k. I was considering the 9900k or waiting for 7nm Ryzen. However, i have now stumbled across a possible good deal on a used 7920X CPU. I use my computer for both VR gaming and software development, so i like high turbo boosts, but also lots of cores for Visual Studio/Resharper.

I think i will be able to get it for a price similar to a new 9900K (maybe a bit more), but i am unsure how hard it is to cool? I am thinking to send it by a delid service, but i am unsure how it would OC on air (with a premium air cooler in a vertical case (Silverstone FT05)). I don't need all-core 5ghz or anything like that, but it would be nice with a good 24/7 OC on say 4 cores.

 

[mackerel]

Only have 6 cores of the stuff myself, but I'd guess if you stick to lower voltages you could still hit low to mid 4.x GHz and still have reasonable thermals under sustained workloads. Also from reading around, these CPUs seem to get less of a difference from delid compared to consumer ones, so maybe try it first without delid and see how you get on.

 

[Ziphnor42]

Only have 6 cores of the stuff myself, but I'd guess if you stick to lower voltages you could still hit low to mid 4.x GHz and still have reasonable thermals under sustained workloads. Also from reading around, these CPUs seem to get less of a difference from delid compared to consumer ones, so maybe try it first without delid and see how you get on.

I guess the real question is whether i can OC by active core count and get something like 4.7 for say up to 4 cores? Currently my 5820k runs at 4.2 all cores with a power limit that throttles it under AVX and similar. I would like to gain significant single threaded performance with an upgrade to 7920x.

 

[mackerel]

From memory, my 7800X runs 4.3 all cores non-AVX at stock voltage (can't recall 1.10 or 1.20?). Clocking beyond that does mean hitting the voltage, so it'll be a tradeoff you'll have to make as it will significantly increase power. I think I'm at 1.35v+ to hit 4.8 and that is thermally questionable on custom water. Personally I think it is a poor trade off to increase voltages to get fewer core higher clocks, than optimise for all cores.

 

[Woomack]

For 7920X you will need top air cooling what means it will be big. Count something like Noctua NH-D15, Cryorig R1 or something similar. If you decide to get this CPU then I also recommend to set voltage manually at 1.10V or 1.15V. It's enough to overclock most of them by 200-300MHz and in the same time it's less than voltage at auto settings so under full load it will have lower temps.
My 7900X runs at 4.7GHz 1.20V on all cores but on water cooling. I wasn't testing it on larger air cooler but something more standard like ~150W cooler couldn't handle it above stock clocks.

 

[mackerel]

Correction to my earlier post. The 4.3 is my AVX OC (but not AVX-512) at stock voltage - it is Prime95 stable. I don't recall at all what the non-AVX clock limit was as that is less interesting to me. Not sure I ever tested that. I did need high voltage to hit 4.8+ non-AVX. 4.7 might not be at the wall yet.

I'm going to connect the system to a chiller this weekend so will be retesting it then.

 

[Ziphnor42]

For 7920X you will need top air cooling what means it will be big. Count something like Noctua NH-D15, Cryorig R1 or something similar. If you decide to get this CPU then I also recommend to set voltage manually at 1.10V or 1.15V. It's enough to overclock most of them by 200-300MHz and in the same time it's less than voltage at auto settings so under full load it will have lower temps.
My 7900X runs at 4.7GHz 1.20V on all cores but on water cooling. I wasn't testing it on larger air cooler but something more standard like ~150W cooler couldn't handle it above stock clocks.

I already have a Noctua NH-D15 actually, and its in a vertifical airflow case (Silverstone FT05) which also helps i guess. Everybody seems very focused on all core overclocks, but what about OC by #active cores? I mean, most games cannot use 12 cores anyway (but VS+R# can) so as long as ~4-6 active cores can hit a high clock, then it is fine if all core clocks are a bit lower.

- - - Auto-Merged Double Post - - -

Correction to my earlier post. The 4.3 is my AVX OC (but not AVX-512) at stock voltage - it is Prime95 stable. I don't recall at all what the non-AVX clock limit was as that is less interesting to me. Not sure I ever tested that. I did need high voltage to hit 4.8+ non-AVX. 4.7 might not be at the wall yet..

Good to know that we agree on what stable means I test GPU with Furmark and CPU with Prime95

 

[EarthDog]

I test GPU with Furmark

Not a great idea. Furmark is considered to be, by both AMD and NVIDIA, a 'power virus'. While it will not damage the card (when leaving power limits untouched), it doesn't do a particularly good job testing your clock speeds. If you notice, when you game and such, your clock speed will be, for example 2000 MHz. When you test with Furmark, typically the card will throttle to protect itself and run at MUCH lower clockspeeds. So in essence, you are not testing the clock speed you are running.

 

[Ziphnor42]

Not a great idea. Furmark is considered to be, by both AMD and NVIDIA, a 'power virus'. While it will not damage the card (when leaving power limits untouched), it doesn't do a particularly good job testing your clock speeds. If you notice, when you game and such, your clock speed will be, for example 2000 MHz. When you test with Furmark, typically the card will throttle to protect itself and run at MUCH lower clockspeeds. So in essence, you are not testing the clock speed you are running.

My 1080TI boosts to around 1950mhz in Furmark reaching a max of 73C. The clock (GPU Boost) is set to 1736 in Auros Engine and power target to 130% (voltage at default). Furmark seems to do a pretty good job at crashing if i increase clock or go too high on memory, while running perfectly stable at "stock" (which is pretty OC'ed on the Auros Extreme). What test should i use instead then?

 

[EarthDog]

You can run any of the UL 3DMark benchmarks that are free. Free game demos like world of tanks encore, final fantasy CV (loop).

WhWhat does it boost to in games? Over 2000 I'd imagine.

 

[Ziphnor42]

You can run any of the UL 3DMark benchmarks that are free. Free game demos like world of tanks encore, final fantasy CV (loop).

What does it boost to in games? Over 2000 I'd imagine.

In 3dmark benchmarks it runs at about the same 1950-1980mhz pretty constant it seems

 

[EarthDog]

I'm absolutely surprised to see that. In all testing I've completed in multiple systems and cards, running furmark drops boost bins immediately, running notably lower clocks than what is seen in gaming.

 

[Ziphnor42]

I'm absolutely surprised to see that. In all testing I've completed in multiple systems and cards, running furmark drops boost bins immediately, running notably lower clocks than what is seen in gaming.

MSI Kombustor seems to show this effect somewhat at least, causing some fluctuations in clock down to 1800mhz. Maybe i am using an older furmark version, or some settings that were less stressful?