Discussion Ryzen 3950X Overclocking/undervolting benchmarking data: Temps and CB20 scores.

[tiberium87]

Hello Friends,

I am new to the overclocking world, and so far it's been a fun and educational ride.

I wanted to share some stats I collected on my Ryzen 3950X, and hear your thoughts on them. Maybe people can even use this as a very rough estimate / comparison of what they might be able to expect for their own 3950X in their OC adventures. I know I would have liked something like this a few days ago!

I think mine runs a tiny bit hot, but not too bad. Using a 360mm water cooler (120mmx3 fans), 3 bequiet chassis fans, and a bequiet Pure Base 500 case. Mobo is the X570 Taichi.

Here is the data:
spreadsheet_1
spreadsheet_2

Best,
tiberium

 

[zx128k]

Sounds about right...I'm running a 240mm AIO. A 3800x should get better performance than my 3700x, and being better binned at lower voltage, therefore run cooler. But even as it is I have to push things really hard just to get to 70C and only goes into the low-mid 70's in crazy stupid tests like prime95 small FFT's. Your custom loop probably benefits greatly from high-flow waterblock and pump, transferring heat a little faster off the heat spreader. But to get only 3 or 4 degrees better cooling, I again just don't see that the cost-benefit ratio is there to justify it. I'd need a new, big case (to fit the plumbing, pump and reservoir) and the requisite maintenance is just not something I'm interested in. I'll just toss my $79 AIO and get another one (at inflation adjusted price) in 5 years once the block becomes gunked up.

You think that because you are not pushing that AIO. Once you hit the limit your temps will rocket. Easy way to do that is a caseking overclock https://www.caseking.de/en/der8auer...30-ghz-mit-wraith-prism-kuehler-hpam-177.html 4.3GHz @ 1.4 volts. Should push 90c with an 240mm AIO.

Caseking sells the 3800x with an 360mm AIO https://www.caseking.de/en/amd-ryze...-all-in-one-liquid-cooler-360mm-cpbu-181.html It gives a high stable boost at a wid range of ambient temperatures.

I already had a water cooling setup, so all I had to buy was an AM4 block. So happy I did. ~40-50c in game temps ftw and 4.4-4.45GHz all core boost frequencies. This is why an all core 4.4GHz overclock which would net me 11700-11800 time spy cpu is not that good. It reduces ingame core frequency by 25-50MHz all cores and kills the 4.5GHz ST boost.

 

[drea.drechsler]

You think that because you are not pushing that AIO. Once you hit the limit your temps will rocket. Easy way to do that is a caseking overclock https://www.caseking.de/en/der8auer...30-ghz-mit-wraith-prism-kuehler-hpam-177.html 4.3GHz @ 1.4 volts. Should push 90c with an 240mm AIO.
...

Like I said, I've not managed to saturate the water in my AIO, even running Folding@Home 24/7 for several days in a row whether CPU only or CPU/GPU combined. It's sometimes pretty hard to get GPU work units.

But no, I'm totally not going to run a Caseking overclock now. Used to think the voltages they used had to be 'safe'... after all they do replace the (now lost) AMD warranty with their own. But after all the people who're now reporting processor degradation so soon with even lower overclocked voltages I'm simply not interested any more.

I did test my 3700X at 4.3Ghz using their recommended voltages though...I think I still have the overclocking profile even. While I do recall a nice uptick in CM20 multi thread, I remember at the time I thought the single thread wasn't that good. Now, since I've gotten PBO figured out and optimized, I don't think the multithread improvement will be nearly so nice and I'm doubtless going to be losing out on single thread since it rather eagerly boosts to 4425. I'm all about single thread since that's much more going to benefit the day-to-day sort of work. And of course...games

 

[zx128k]

Like I said, I've not managed to saturate the water in my AIO, even running Folding@Home 24/7 for several days in a row whether CPU only or CPU/GPU combined. It's sometimes pretty hard to get GPU work units.

But no, I'm totally not going to run a Caseking overclock now. Used to think the voltages they used had to be 'safe'... after all they do replace the (now lost) AMD warranty with their own. But after all the people who're now reporting processor degradation so soon with even lower overclocked voltages I'm simply not interested any more.

I did test my 3700X at 4.3Ghz using their recommended voltages though...I think I still have the overclocking profile even. While I do recall a nice uptick in CM20 multi thread, I remember at the time I thought the single thread wasn't that good. Now, since I've gotten PBO figured out and optimized, I don't think the multithread improvement will be nearly so nice and I'm doubtless going to be losing out on single thread since it rather eagerly boosts to 4425. I'm all about single thread since that's much more going to benefit the day-to-day sort of work. And of course...games

The main issue with ryzen overclocking is tightening the ram timings. As I have OCD when overclocking I naturally had to find the lowest settings posible. To get the lowest latency for games posible. With the higher clock speeds that also reduces latency. The RAM overclocking gives the biggest performance boost but also is a painful long process.

To get the most out of the cpu, a chiller is the best bet. I would bet a manual 4.5GHz all cores oc would be posible. PBO might be easy to set to 4.7GHz ST and motor along at 4.5GHz all core's in games.

Voltage wise my cpu runs prime95 small ffts @1.4 volts vcore and has done for 6 months. Frequency at 4.2GHz. This is with pbo off and LLC set to level 3 for vcore or medium. If ambient temp are low the frequency will increase to 4.25GHz.

With the RAM overclock, procODT 53ohms was what got it stable at 3800MT/'s IF 1900. Even with the EDC bug. With cooling getting that load tdie temp down is 100% the goal.

With the vcore voltages I am getting with PBO disabled in bios, I appear to get the same level with PBO and scalar x10. PBO scalar x1 does nothing at all, same with the EDC bug and anything other than scalar x10. No game performance increase. Cinebench r20 on a cooler day will increase from approx. 5200 to 5300 with the pbo bug.

So the pbo bug gives little or nothing in games even with +200MHz auto oc.

I think 4.4GHz all cores is posible and temp will be fine but I get that in game anyway.

 

[drea.drechsler]

The main issue with ryzen overclocking is tightening the ram timings. As I have OCD when overclocking I naturally had to find the lowest settings posible. To get the lowest latency for games posible. With the higher clock speeds that also reduces latency. The RAM overclocking gives the biggest performance boost but also is a painful long process.

To get the most out of the cpu, a chiller is the best bet. I would bet a manual 4.5GHz all cores oc would be posible. PBO might be easy to set to 4.7GHz ST and motor along at 4.5GHz all core's in games.

Voltage wise my cpu runs prime95 small ffts @1.4 volts vcore and has done for 6 months. Frequency at 4.2GHz. This is with pbo off and LLC set to level 3 for vcore or medium. If ambient temp are low the frequency will increase to 4.25GHz.

With the RAM overclock, procODT 53ohms was what got it stable at 3800MT/'s IF 1900. Even with the EDC bug. With cooling getting that load tdie temp down is 100% the goal.

With the vcore voltages I am getting with PBO disabled in bios, I appear to get the same level with PBO and scalar x10. PBO scalar x1 does nothing at all, same with the EDC bug and anything other than scalar x10. No game performance increase. Cinebench r20 on a cooler day will increase from approx. 5200 to 5300 with the pbo bug.

So the pbo bug gives little or nothing in games even with +200MHz auto oc.

I think 4.4GHz all cores is posible and temp will be fine but I get that in game anyway.

I do think PBO in general and the EDC=0 bug in particular gives highly variable results motherboard to motherboard and even processor to processor. I'm pretty sure different motherboard setups, mainly in VRM control, are affecting it. I get really good results from it probably because my motherboard isn't so well optimized in base 'default' settings. It's a low-end mATX B450 board...not low end for mATX but low end BECAUSE it's mATX. A B450 Mortar, pre-MAX. PBO scalar 10X does offer good results but 5x seems to be just as good and it's a tad cooler in Prime95 and the VCore averaging within 0.050V of each other.

I have been able to get my FlareX memory up to 3600mtps, at least. With CAS 14 timing and I used the DRAMCalculator which did instruct to change ProcODT as you note. I don't really want to try for 3800, not since this RAM was only rated 3200 anyway. I think pushing it to 1.41V is enough.