AMD Ryzen 5000 Builders Thread

[B-Riz]

	Cores Threads	Base Freq	Turbo Freq	L3 Cache	TDP	MSRP
Ryzen 9 5950X	16c/32t	3400	4900	64 MB	105 W	$799
Ryzen 9 5900X	12c/24t	3700	4800	64 MB	105 W	$549
Ryzen 7 5800X	8c/16t	3800	4700	32 MB	105 W	$449
Ryzen 5 5600X	6c/12t	3700	4600	32 MB	65 W	$299

5600X comes with cooler, the rest do not.

above from https://www.anandtech.com/show/1614...en-3-on-nov-5th-19-ipc-claims-best-gaming-cpu

 

[Leeea]

I haven't gotten to the end of it, but it looks like a microcode issue to me.

It could be, but seems unlikely. Microcode is held in the mainboard anyway, and loaded to the CPU on boot up.

What he is seeing is a memory error, which is usually associated with memory timings. He has overclocked the memory and infinity fabric (it overclocks with the memory).

XMP profiles do some crazy . . . stuff. Anyone who can commit the time should really use the Ryzen Timing Calculator instead.

It most certainly does crazy stuff. However, it is also what the manufacturer tested/claims the memory can do, so typically offers the most trouble free experience when overclocking memory.

 

[DrMrLordX]

PBO is no different ...
.. they are both subject to BIOS programming.

PBO's quite a bit different. It changes TDC and EDC but it won't change vSoC, VDDG, or VDDP. XMP/DOCP can do those things. Which it really doesn't advertise and can be difficult to monitor, since VDDP and VDDG might not even be accessible except through fairly-obscure UEFI menus like . . . AMD CBS or what have you.

I found the " Dram Calculator " is good for getting you ball park settings.

Sure beats doing it the old-fashioned way: strictly by hand with guesswork.

It could be, but seems unlikely. Microcode is held in the mainboard anyway, and loaded to the CPU on boot up.

There were people in the thread I linked getting the same error, bone stock. I realize that Vermeer is a different creature than anything that's come before it, but there are lots of ways that unstable memory settings manifest, and that isn't necessarily one of them.

It most certainly does crazy stuff. However, it is also what the manufacturer tested/claims the memory can do, so typically offers the most trouble free experience when overclocking memory.

Not in my experience. Maybe on Intel rigs.

 

[Leeea]

There were people in the thread I linked getting the same error, bone stock. I realize that Vermeer is a different creature than anything that's come before it, but there are lots of ways that unstable memory settings manifest, and that isn't necessarily one of them.

The weird thing is that thread ends with them claiming HWINFO software program was causing the issue by activating some compute feature somewhere.

Not saying they are not having problems, but not sure if I agree with any conclusions coming out of that thread.


Regardless cytg111 has another CPU on hand, so he can do the swap and rule that out rather quickly.

Not in my experience. Maybe on Intel rigs.

I have only built three Ryzen systems, and on all three I just set the profile and forgot about it.

The settings your referring to, VDDP, VDDG, and etc, those are being specified by the engineers of the manufacturer who made the memory. They are going to know what their memory needs to overclock consistently and stable. They are going to have the tools that are far more finely grained then ours to test with.

While I will certainly agree they leave performance on the table, I think it is a bit of a push to claim they sacrifice stability by setting obscure settings.

 

[Noid]

I need your visual advice the next time you look at your 5900xt 's

Mine has a visible gap between the CPU circuit board and the metal " cap ".
Big enough to slide a corner of a piece of paper under ...
( or anything else for that matter )

Lucky I have no thermal paste in the separation ( yet ).
There is clearly an epoxy running around 3 sides ...

And, now that I looked at a 3900XT I have,,,, It also has a " gap "
( on the side that has the "R" for " Ryzen " )

Why would they leave holes for shyt to get into ...?

Also, if the epoxy is relevant to the dissipation of heat, then my chip is clearly defective.

 

[nurturedhate]

I need your visual advice the next time you look at your 5900xt 's

Mine has a visible gap between the CPU circuit board and the metal " cap ".
Big enough to slide a corner of a piece of paper under ...
( or anything else for that matter )

Lucky I have no thermal paste in the separation ( yet ).
There is clearly an epoxy running around 3 sides ...

And, now that I looked at a 3900XT I have,,,, It also has a " gap "
( on the side that has the "R" for " Ryzen " )

Why would they leave holes for shyt to get into ...?

Also, if the epoxy is relevant to the dissipation of heat, then my chip is clearly defective.

Seems to show a gap on the left side.

 

[Noid]

Mine has a significant " gap ".
I think at least 1 mm.

 

[Leeea]

Mine has a significant " gap ".
I think at least 1 mm.

If it works I would not worry about it.

To elaborate, my understanding is there are three sub chips underneath that metal cover, with the cover bonded down to them with some sort of solder.

See:
https://www.google.com/search?q=ryzen+delidding&tbm=isch&sa=X

There is a black seal / rubberish type material that runs around the border a 2-3 mm in from the edge.

As long as the heat spreader is bonded to the chiplets and the heat spreader is flat with itself so the heatsink can make good contact it should not matter if it is a bit off angle.

Getting thermal paste in along that edge will have no effect.

 

[blckgrffn]

I just wanted to share that my 5800x's both are runaway hot with Gigabyte boards.

I can easily temp throttle on my noctua in the giant HAF case with lots of airflow when leaving the CPU with it's "stock" settings. It was really drawing a lot of power according to Ryzen Master.

My work PC with a AMD RGB cooler and PBO off and ecomode on at least doesn't crank the fans up so quickly after some aggressive fan curve tuning but temps can still change very dynamically.

Even with eco mode enabled on my bigger PC it is crazy beast, my 3600 was much easier to keep temps under control. It's actually really annoying because the 12cm fan will be really peaky doing normal things like logging me into windows.

 

[Geegeeoh]

You just can't compare 3000 and 5000 temps.
Very different "design" behind.

 

[Noid]

I finally found a way to limit PPT.
In " Digi Power " menu settings.
Parameter called " CPU VDD Full Scale Current " ( amps )

I was watching HWINFO numbers and tried to target
" CPU Core Current ( SVI2 TFN) "

It does not represent the same number.
However, I put 118A in there and my screenshot is my results.
As you can see , PPT was NOT my heat issue.

Crazy spikes !!. ( 110C )
( remember - I am a STATIC oc )

50.9 C average ... for 2 hours of P95 .. hmmm

[ BIG EDIT - corrections + BIOS screenshot ]

Being that the Noctua max fan speed is 1500rpm, I removed one and placed the fan from my dead 750PSU that is much more powerful and silent. It's also bigger than those 140mm ( with more blades ).

Now that I know PPT is not my problem ...
I'm going to fiddle with Core voltage ( 1.33v ) vs. LLC ( Mode 3 ) now.
( looking at individual core temps also )

Despite being P95 stable at 4.6Ghz in winter, I can't hold it in summer.
I may have to drop to 4Ghz for heat issues.

 

[DrMrLordX]

The weird thing is that thread ends with them claiming HWINFO software program was causing the issue by activating some compute feature somewhere.

Yup. Though in the end the only fixes I've seen people come to that cover all use cases have either been:

a). Different motherboard firmware/UEFI revision
b). Different CPU

Others have speculated that it's an issue with individual cores boosting beyond the point of stability, but that isn't a consistent conclusion either. Check out this thread I dug up on a 5900x.

The settings your referring to, VDDP, VDDG, and etc, those are being specified by the engineers of the manufacturer who made the memory. They are going to know what their memory needs to overclock consistently and stable. They are going to have the tools that are far more finely grained then ours to test with.

Ehhhhh we're just gonna have to agree to disagree with that one. The extent to which they push vSoC (for example) gets completely insane. Anything pushing vSoC to 1.2v or higher is really suspicious. It's one of the reasons why I won't touch XMP/DOCP on an AMD rig.

 

[Leeea]

Ehhhhh we're just gonna have to agree to disagree with that one. The extent to which they push vSoC (for example) gets completely insane. Anything pushing vSoC to 1.2v or higher is really suspicious. It's one of the reasons why I won't touch XMP/DOCP on an AMD rig.

I have seen 1.35v common for years with that. Never had an issue in my limited circle. Used to be 1.35 was jedec.

More interestingly, according to the spec sheet for my chips 1.35v is the low end, with anything between 1.35v to 1.5v being acceptable. There is a reason they are putting heatsinks on ram these days.

1.2v is the jedec timing, but running higher speeds at that voltage is out of spec according to the manufacturer. Real overclocking with all the instability and other annoying issues.


The high voltage on the spec sheet lets them drive the chips harder (most likely the latches in the ram circuits), which likely yields more reliable and consistent results at overclocked speeds. As for the voltage damaging the chips? That seems very unlikely. If the manufacturer rated it accordingly, I am happy to trust the manufacturer of my chips.

If you do not trust the manufacturer to give you a valid spec sheet, or produce ram that will last when in spec, buy from a different manufacturer.

 

[Dave3000]

What RPMs are your fans spinning at on your NH-U12A? Those fans go up to 2000 RPM, so maybe you need to increase your fan curve in your UEFI (or set it to performance).

Outside of the NH-D15, your cooler is the highest performing Noctua cooler, so something's not right if you're seeing 90c (with good case airflow and good ambient temperatures).

When I did a manual fan curve. I set it to spin both fans at around 1200-1250 rpm (54% duty cycle) for 20C to 60C CPU temperatures (not CPU Package temperature) and 100% duty cycle for CPU temperatures of 70C and up. If I set the fan profile to standard or turbo, the CPU fans will rev up and down frequently even during light usage or some background tasks happening and I find that annoying. On my 5800x, a CPU temperature of 70C is equivalent to around a CPU package temperature of 81C, so at around 81C CPU package temperature, my CPU fans are set to hit full blast. Also today I moved my stock rear case fan to the front of the case as a 2nd intake fan and use a spare 120mm Coolermaster Silent that I had lying around as a rear exhaust fan. CPU temperatures didn't change, but I think my 5800x boosts slightly higher now at full load, but it's still what I would consider within margin and not really a big deal. Would a 360mm AIO mounted in the front of the case as intake be worse for CPU cooling than a 240mm AIO mounted at the top of the case as exhaust. Also my GPU is runs hot enough as it is during heavy GPU load, and I don't want to bring hot air into my case and I heard the front mounting an AIO brings hot air into a case because of the radiator and it may be better for CPU temperatures than a top mounted AIO but worse for other components such as video card, is there any truth to that?

 

[DrMrLordX]

I have seen 1.35v common for years with that. Never had an issue in my limited circle. Used to be 1.35 was jedec.

vSoC, not vDIMM. Big difference.

 

[Noid]

Scraped 4C off that spike I get @ 2hr 23 minutes into the Blend test.
My max was 99.6C until then.
Currently 38C with 44C avg at the same place as yesterday. ( 50C )
Vcore @ 1.285 ( 1.29 BIOS ) x 45 multiplier.

If your curious when my CPU shutsdown --- its's at 113C lol

I wish I could determine the test that produces that spike.

Still not happy ... dropping Vcore to 1.25v lol

 

[cytg111]

Hmm. Try reading through this:

5950x Cache Hierarchy Error

I have a new system built and running seemingly well, apart from this "fatal hardware error" which has popped-up at least twice (maybe more). Admittedly, I have another PSU coming from Corsair that will hopefully fix a 5% to 10% reproducible startup issue where the PSU has to turned off and...

community.amd.com

I haven't gotten to the end of it, but it looks like a microcode issue to me.



XMP profiles do some crazy . . . stuff. Anyone who can commit the time should really use the Ryzen Timing Calculator instead.

Yea, went through that a dozen or so other threads on the same subject. Doesnt sound too good.

I have not swapped the CPU yet but I do seem to have gotten rid of it for the past two days, and I have been trying to replicate it.

What did I do? Open the case and put a big a* ventilator on it.. we are talking a one meter fan that should have been operating in a wind tunnel, this thing moves AIR.

So maybe something heat after all?

 

[DrMrLordX]

So maybe something heat after all?

Maybe. My experience with Vermeer is nil, but AMD's implementations of N7 seem pretty temp-sensitive. You get notable improvements to the v/f curve, usually in 20C increments. You can get much smaller ones lowering operating temps by 5-10C. If you are getting a hotspot temp drop and IF you have a funky CPU that is not boosting properly, then maybe - maybe! - a 5C drop could bring it into the edge of stability. That's just reaching though.

 

[Ajay]

Yea, went through that a dozen or so other threads on the same subject. Doesnt sound too good.

I have not swapped the CPU yet but I do seem to have gotten rid of it for the past two days, and I have been trying to replicate it.

What did I do? Open the case and put a big a* ventilator on it.. we are talking a one meter fan that should have been operating in a wind tunnel, this thing moves AIR.

So maybe something heat after all?

I'd be inclined to send that CPU back, if you can. Then I'd put a good AIO on it or a Noctua NH-D15. This would be too much bother for me.

 

[LikeLinus]

I'd be inclined to send that CPU back, if you can. Then I'd put a good AIO on it or a Noctua NH-D15. This would be too much bother for me.

Had an odd situation with my 5950x and RAM. Did you try removing the RAM and running MemTest86 with only 1 stick in ? Then swap out the sticks of ram and only having the other stick of RAM in? Try both slots for each. I had mine both pass, but then found out that one stick would not pass in a single slot when the other was populated, but would pass in a certain slot. So I had to figure out which stick worked best in which slot. Before, I was having random crashes and thought I had a bad CPU! Turns out, it was just the ram in slot orientation that was causing the problem. Not sure why, but I finally got it to work.

 

[bigboxes]

Not in my experience. Maybe on Intel rigs.

I use XMP settings on my Ryzen 3 CPU and on my previous Haswell build. Kind of set it and forget it. I value stability.

 

[DrMrLordX]

I use XMP settings on my Ryzen 3 CPU and on my previous Haswell build. Kind of set it and forget it. I value stability.

XMP 2.0 is still an Intel standard though. You can tell just from looking at the tRFC values it'll pick. If it works great, but the things that have to be done to get those timings to work on AMD CPUs are not always the best.

 

[bigboxes]

XMP 2.0 is still an Intel standard though. You can tell just from looking at the tRFC values it'll pick. If it works great, but the things that have to be done to get those timings to work on AMD CPUs are not always the best.

DOCP. Right after I booted into the bios for the first time, I set it for XMP profile #1. Never had a problem. Even more stable than my last build. My main rig has now been up 50 days! Of course, I made sure the ram was on the QVL. It was hard to find though and expensive.

 

[CentroX]

Was able to tweak my 5800x to 76c max with cinebench scores of 16K

 

[Makaveli]

Was able to tweak my 5800x to 76c max with cinebench scores of 16K

detail your tweak?

 

[DrMrLordX]

DOCP. Right after I booted into the bios for the first time, I set it for XMP profile #1. Never had a problem. Even more stable than my last build. My main rig has now been up 50 days! Of course, I made sure the ram was on the QVL. It was hard to find though and expensive.

What tRFC settings did you get that way?