AMD Ryzen 9 5950X Spike Temps Question

[AManWithPlan]

My PC: https://pcpartpicker.com/list/t9dXQD

 

Dear LTT,

 

I am newly back on the AMD platform, been off since the Bulldozer; was on team Intel with a 5960x. Just put together a new build linked above.

 

I am now on a completely custom water loop with over 1,000 watts of cooling power. The EKWB waterblock on the CPU is using Conductonaut (which I correctly applied) as the thermal "paste".

 

Is this processor known for running hot? Because I'm around 60C at idle (was on 50C back on my Intel processor which had a 140W design compared to my new 105W one; and I was using a Corsair AIO cooler at the time so had no where near this custom loop I have now).

 

I am also getting spikes to 75C when doing simple things like downloading a large file. The pump ramps up per my bios profile as soon as the processor runs hotter, but these spikes are odd, I've never seen them on a processor (usually temperature rises up slowly by 2 degrees or so).

 

My coolant temperature at equilibrium is around 36C, so that is the above-ambient threshold to keep in mind.

 

I've had the processor run as hot as 87C when benchmarking, something I couldn't even get higher than 65C on my old Intel with a weaker cooling solution.

 

I am currently undervolting it to 1.378 with an aggressive PBO2 profile that takes me to around 4,875mhz on the best cores (all achieved with undervolting, not overvolting).

 

Let me know your experience so I can compare, cause reviewers are sitting pretty around 70C or so under load with much weaker cooling solutions (like air only).

 

Aman

 

[Pikatchu]

I don't have an 5950x, running a 5800x.

It is running very hot overall, web browsing can push it to 60C. This seems to be an issue with 5000 series AMD; they are hot commodities! (literally)

I am currently have it at a -0.0875 offset with PBO turned completely off.

[AManWithPlan]

1 minute ago, Pikatchu said:

I don't have an 5950x, running a 5800x.

It is running very hot overall, web browsing can push it to 60C. This seems to be an issue with 5000 series AMD; they are hot commodities! (literally)

I am currently have it at a -0.0875 offset with PBO turned completely off.

How do you think these benchmarkers are doing it? Just insanely low voltage, like 1.31V or something? PBO wouldn't even clear 4.1ghz at that level. Hell, multi-core dropped down to 587mhz, haha.

 

If the heat is okay (meaning chip won't die as long as its under 90C), I'm fine, I just want to make sure I am doing all I can. I am used to far lower temps with Intel on a higher wattage chip.

[Guest]

EDIT:  I mean curve not core.         The specs for the 5800X say 105W... yet Precision Boost by default puts it to 142W.. lowering this with a negative curve offset can improve temperatures a lot.

 

I'm current running 115W with -30 core offset in PB advanced settings - 76c full core load vs. 85+ with defaults, performance within 1-2%.

[AManWithPlan]

1 minute ago, si1enze said:

The specs for the 5800X say 105W... yet Precision Boost by default puts it to 142W.. lowering this with a negative core offset can improve temperatures a lot.

 

I'm current running 115W with -30 core offset in PB advanced settings - 76c full core load vs. 85+ with defaults, performance within 1-2%.

I get doing -30 in PBO, but that will mean you're not going to achieve anywhere close to 5ghz single-core performance; which is what I am shooting for. Also, the 16 core chip that I have won't let you do -30 on every core, you'll get WHEA errors and crashes. I'm currently -15 all cores, -7 to -8 on the fastest cores.

I'll never be doing super-intensive all core workloads, so I don't need high voltage overall, but I do want aggressive single core performance (preferably under 60C, which I am currently above at around 66C-68C). The more I offset my voltage (down), the less single-core performance I get while testing.

[Guest]

EDIT: curve not core, my bad.       I get the same 4.85Ghz boost clocks with this configuration compared to default.  I've noticed that too, with the voltage, I leave mine on auto.  With this config a full core load sits at about 4.5Ghz at 1.15v, I don't think I'd want it lower but that's from the core offset.  I might raise that a tad.

[Chris Pratt]

4 minutes ago, AManWithPlan said:

How do you think these benchmarkers are doing it? Just insanely low voltage, like 1.31V or something? PBO wouldn't even clear 4.1ghz at that level. Hell, multi-core dropped down to 587mhz, haha.

 

If the heat is okay (meaning chip won't die as long as its under 90C), I'm fine, I just want to make sure I am doing all I can. I am used to far lower temps with Intel on a higher wattage chip.

That's actually not that amazing. Most Zen 3 can do a full -30 step undervolt with the curve optimizer, which amounts to a max -150mV undervolt. They perform much better that way too. Not uncommon at at all to hit 4.6 all core and 5GHz single core. That's what my 5900X hits. The key is using the curve optimizer and not just applying a constant undervolt or even offset, though. Ryzen has never undervolted well that way.

[Guest]

I meant curve, not core.  Edited my posts.  Whoops.

[AManWithPlan]

10 minutes ago, si1enze said:

I get the same 4.85Ghz boost clocks with this configuration compared to default.  I've noticed that too, with the voltage, I leave mine on auto.  With this config a full core load sits at about 4.5Ghz at 1.15v, I don't think I'd want it lower but that's from the core offset.  I might raise that a tad.

I'm seeing something a bit different on my chip. Leaving core voltage on auto makes the board run it at 1.45-1.5v. This gives me spikes into the 80's temperature-wise.

 

I get about the same PBO performance on single-core (maybe a spike to 5.1ghz here or there, but ultimately within 1% Cinebench R20 scores), at the cost of my higher spikes in temperature from simple things like downloading a file, running an intensive website (like Coinbase), or simply copying some files around. The spike is quickly cooled by my pump speeding up and it comes back down into the high 60C's, but still, I don't want it to spike to above 90 (already had this happen while testing, even though I manually set it to throttle at 90C, it spiked to 97C with Vcore on auto making me quickly kill the benchmark).

[AManWithPlan]

8 minutes ago, Chris Pratt said:

That's actually not that amazing. Most Zen 3 can do a full -30 step undervolt with the curve optimizer, which amounts to a max -150mV undervolt. They perform much better that way too. Not uncommon at at all to hit 4.6 all core and 5GHz single core. That's what my 5900X hits. The key is using the curve optimizer and not just applying a constant undervolt or even offset, though. Ryzen has never undervolted well that way.

I don't think the 16-core 5950x is capable of the same PBO undervolts as the 12-core 5900x. Even Ali (Optimum Tech) couldn't do better than -12 all core on a 5950x (though he didn't play with the fastest cores to tweak like I did, which is why I can do -15).

 

 

[Chris Pratt]

1 minute ago, AManWithPlan said:

I don't think the 16-core 5950x is capable of the same PBO undervolts as the 12-core 5900x. Even Ali (Optimum Tech) couldn't do better than -12 all core on a 5950x (though he didn't play with the fastest cores to tweak like I did, which is why I can do -15).

 

 

Well, it does have more and more tightly packed cores, which means both less power and thermal headroom. I would still expect better.

[AManWithPlan]

6 minutes ago, Chris Pratt said:

Well, it does have more and more tightly packed cores, which means both less power and thermal headroom. I would still expect better.

As I said, instability/WHEA errors start for me right around -20 on all the non-fastest cores. So I keep them on -15.

 

I am also undervolting by 0.085v, so that is a factor, but I tested -20 all-cores at Auto voltage.