Ryzen 5800 owners complain about very high MT load temps

[Zach_01]

my 3700X pulls way more than 88W try double it's TDP at 135.8W during CBr20 multi mind you it's also pulling an all core of 4450MHz and reaches 77c

That is interesting...
Can you show to us a screenshot of HWiNFO sensors mode, full window, during R20 MultiT?

 

[8BitZ80]

I think you are wrong. The 5900X uses 6+6 and that is why the price is so competitive comparing to the 5800X because this one has to use 8 core chiplet. So the 5800X and 5950X are the only chips using 8core chiplets. Also, the 5950X uses the top quality silicon while the 5800X uses any 8core chiplet which passed testing process.
That is why there is a discrepancy in the voltage that the 5800X uses (all 5000 series chips use around 1.2v while the 5800x uses 1.35v and that is a noticeable difference). The voltage is higher than 3800X and that is the only chip in the 5000 series family which has temps and voltage higher than the previous gen counterpart.

Then we'll wait until someone with a 5900X can use Ryzen Master to verify the CCX configuration. Let's actually prove that I'm wrong, rather than just saying it.

 

[R0H1T]

Well AMD can use any combination of 4+8 or 6+6 IIRC even on zen2, there's no technical limitation in there. So it could simply be that they're making it with a combination depending on the chiplets they have as well as demand of such ships.

 

[ratirt]

Then we'll wait until someone with a 5900X can use Ryzen Master to verify the CCX configuration. Let's actually prove that I'm wrong, rather than just saying it.

You can watch reviews and you will have that answer. There might be a possibility to use defective chiplets 4 core but these (as you guys probably know) are rare (80mm2 size) on a matured 7nm node.
Besides, AMD will not make 4 core processors with chiplets rather monolithic CPU with iGPU. At the beginning of shipping, AMD will surely focus on the chiplets and bin them accordingly. 6 cores per chiplet for 5600X and high quality for 5900X same goes with 5800X and 5950X. You can see it with the 5800X higher voltage (lower quality) than 5950X (higher quality). It's fairly simple I think.

 

[Vayra86]

Yeah well the CPU knows its own silicone better than people so...

Stop talking about my tits will you

You know that scheduler on an OS, that also takes control away from the user ~ I bet most people don't like that either.

No thats why they disable page file

And then complain why things don't work

 

[oobymach]

No thats why they disable page file

And then complain why things don't work

Off topic but even I use invisible tasks in task scheduler, I have an inaudible tone play every 5 mins to keep my T50's from turning off. I actually wrote the damn script for it, but more and more software is getting invasive to the point where it's disruptive. Ads for upgrades in your software, multiple useless tasks to run a single device, if I weren't a hobbyist and didn't know how to disable such crap I might be put off by the invasiveness.

Also you need pagefile but not for windows to have control of it, mine averages 16-60mb thanks to custom size setting. In these days of ssd's where writes need to be minimal taking control of your pc isn't a bad idea. I re-routed all my temp folders to an hdd and have a m2 ssd just for my pagefile thus keeping writes to my m2 c: drive to a minimum.

 

[birdie]

my 3700X pulls way more than 88W try double it's TDP at 135.8W during CBr20 multi mind you it's also pulling an all core of 4450MHz and reaches 77c

The 3700X pulls at most 93W, period, unless you've enabled PBO/OC/increased its PPT.

Then we'll wait until someone with a 5900X can use Ryzen Master to verify the CCX configuration. Let's actually prove that I'm wrong, rather than just saying it.

2 six-core CCX'es:

 

[8BitZ80]

Cool, then I stand corrected.

 

[birdie]

The plot thickens:

I will post the thermals on the new one. Ended up cleaning and repasting the damn thing four times before I decided it was the damn cpu and not my cooler or paste application technic. Called my companies AMD tech group (servers) and had them take a look at it remotely. They have a very handy diagnostic suite that interrogates the hell out of the chip on the hardware level (tried to snag a copy and got my hand smacked (damn); two hours later and they said it was the TIM under the heat spreader. AMD flagged the chip's serial number and are going to RMA it back to the QC lab to figure out what happened and to see if more also got dodgy TIM (the 5800x and 5900s get soldered; not sure about the rest of the series) application in that batch; I doubt enough got messed up to warrant a recall.... at least I hope not. Due to the temperature spikes those cores are now damaged. So I returned it and got a store credit and a promise of a phone call when they get their next shipment. Statistically I was the one to get a dud chip; they randomly test chips after the head spreader is attached before they package them for QC. If they had to test everyone then they would never get them to market. When they test to bin them it is before they place the head spreader and test stability right on the core. They have a solid copper block cooler with a liquid metal pad thingy that they use to cool the core when binning them (have a picture somewhere I got from someone). I have my 3900x till then so I am not sitting with a pile of dismantled computer parts pining for a round of whatever game tickles my fancy at the time.

Hopefully it's an isolated issue and it's not what people are dealing with.

 

[Glaceon]

I suddenly don't regret buying a 3900X in August...

 

[Frick]

The plot thickens:

Hopefully it's an isolated issue and it's not what people are dealing with.

How hot did that one get?

 

[8BitZ80]

Just did some testing to see how the thermals were doing on my 5600X:

Idle: 45C
Gaming: 72C
Cinebench R20 Multi: 89C

Granted I'm running a 92mm tower cooler but at least it's a point of comparison. Not throttling in Cinebench R20 and getting 4.4GHz on all cores constantly. There's no way I'd be able to run a 5800X without a major cooling upgrade and so it's little surprise that the 5800X club members are having issues.

 

[Sybaris_Caesar]

Stop talking about my tits will you

Just as point of reference, how big are they?

 

[birdie]

Just did some testing to see how the thermals were doing on my 5600X:

Idle: 45C
Gaming: 72C
Cinebench R20 Multi: 89C

Granted I'm running a 92mm tower cooler but at least it's a point of comparison. Not throttling in Cinebench R20 and getting 4.4GHz on all cores constantly. There's no way I'd be able to run a 5800X without a major cooling upgrade and so it's little surprise that the 5800X club members are having issues.

89C is just one degree away before the CPU starts throttling (90C for the Ryzen 5000 series). Your cooling solution is clearly insufficient even for your 5600X CPU.

 

[Chrispy_]

These temperature spikes are no different to what we've had with the XT Zen2 models. It's all part of the boost alogorithm and if you cool it harder it will just boost higher and get hotter again. I have a 3600XT (gift from AMD, not something I wasted money on) and it's exactly the same behaviour as the 5600X. Later-model 3950X in our renderfarms behave much the same way, though I don't use PBO on those - just 100% plain stock and with motherboard/BIOS combos that don't 'cheat' with power reporting.

If you give it headroom, the CPU will use it. If you don't want such aggressive boost you can reduce the TDP or set custom PBO settings that are actually lower than the default PBO.

 

[8BitZ80]

89C is just one degree away before the CPU starts throttling (90C for the Ryzen 5000 series). Your cooling solution is clearly insufficient even for your 5600X CPU.

Eh it's fine. Cinebench isnt really that critical beyond testing and benchmarking. I dont do any rendering workloads.

 

[tabascosauz]

Eh it's fine. Cinebench isnt really that critical beyond testing and benchmarking. I dont do any rendering workloads.

Honestly, it looks like the firmware still has a long way to go. Most reviewers pegged the 5600X as similarly cool-running if not more so than the 3600, so looking at the scattered user reports about how the chips are behaving suggests that the 5600X isn't a hot-running chip, just the firmware is still in such a primitive state that the new Vermeer chips don't understand when to "go" and when to "stop". The 3600 would never give the U9S a run for its money at stock settings; from the reviews, neither should the 5600X.

So essentially, launch firmware is just wack, in classic Ryzen fashion, just like pre-1003 AGESA. Good news is Hallock already promised improved firmware coming up soon.

If AMD reins back these chips' Vcore a bit and board vendors do a better job of calling for more consistent stock Vcore, the 5800X may get a little bit better.

Optimumtech's 5950X is both bizarrely underperforming in all-core and also boosting to a ridiculously hot extent in mere productivity and web browsing, which points to the firmware needing work. Unsurprisingly reminiscent of how Ryzen 3000 behaved before AGESA 1003ABBA's boost filter. The 5950X frequencies during Heaven (a GPU benchmark, of all things) look a lot like the spikes you see if you graph Tctl/Tdie idle temps in HWInfo over time.

 

[R0H1T]

Just lower the voltage on them, it's possible to get'em beyond 5GHz mobo makers are pumping way too much.

Unless I see some decent sample set with voltages, cooling, clock speeds & what application was running I wouldn't call this an alarming thing just yet.

 

[Vayra86]

Just as point of reference, how big are they?

5nm

 

[EarthDog]

5nm

Smaller nodes FTW!

 

[Deleted member 24505]

Honestly, it looks like the firmware still has a long way to go. Most reviewers pegged the 5600X as similarly cool-running if not more so than the 3600, so looking at the scattered user reports about how the chips are behaving suggests that the 5600X isn't a hot-running chip, just the firmware is still in such a primitive state that the new Vermeer chips don't understand when to "go" and when to "stop". The 3600 would never give the U9S a run for its money at stock settings; from the reviews, neither should the 5600X.

So essentially, launch firmware is just wack, in classic Ryzen fashion, just like pre-1003 AGESA. Good news is Hallock already promised improved firmware coming up soon.

If AMD reins back these chips' Vcore a bit and board vendors do a better job of calling for more consistent stock Vcore, the 5800X may get a little bit better.

Optimumtech's 5950X is both bizarrely underperforming in all-core and also boosting to a ridiculously hot extent in mere productivity and web browsing, which points to the firmware needing work. Unsurprisingly reminiscent of how Ryzen 3000 behaved before AGESA 1003ABBA's boost filter. The 5950X frequencies during Heaven (a GPU benchmark, of all things) look a lot like the spikes you see if you graph Tctl/Tdie idle temps in HWInfo over time.

Did he say he is using a 240mm rad for the cpu and a 3090? no wonder temps are so crappy.

 

[Zach_01]

Just as point of reference, how big are they?

5nm

Oh... you couldnt be more in to the discussion! We are talking about squished nodes... apparently...

Seriously now, I really would like to see a view of HWiNFO during R20 MT run on these 5000 CPUs.

 

[EarthDog]

Seriously now, I really would like to see a view of HWiNFO during R20 MT run on these 5000 CPUs.

I deleted a post in this thread, but regretting it. This thread really feels like a witch hunt (considering the source).

Anyway, I'll have a 5950x landing today so I can put something up late tonight.

 

[Octopuss]

In these days of ssd's where writes need to be minimal

Excuse me, have you timewarped here from 2005 or something?

 

[Athlonite]

That is interesting...
Can you show to us a screenshot of HWiNFO sensors mode, full window, during R20 MultiT?

sure hold on few moments and I get it up and running and here ya go

as you can see from the screenshot CPU temps and Package power are 77°c and 130W