AMD Ryzen 5000 Builders Thread

[Justinus]

I just installed a 5900x bought today.
I have post.
But, on reboot after F10 saving settings in BIOS ...
... goes right back into BIOS screen
( loop )
Never leaves BIOS ... wtf

Tried the last 3 BIOS releases.
( surprised I could flash back )

Temps are good.
All devices present.

any ideas ?

I assume you've checked your boot menu to ensure the boot devices have priority set correctly?

Only time I've seen this is if no boot devices are present or detected, or they are all disabled (boot order set to nothing).

Edit: Actually, one time on X99 when I had CSM settings all weird it would detect my boot devices but fail to boot and enter bios directly. I assume you've done a cmos reset? (I'd think flashing the uefi would also do that for you, but you can't be too sure..)

 

[Noid]

its the chip ...

put my 3900xt back in and it boots.

a couple BIOS screen shots with the 5900x
I turned off the BIOS slash logo
Got the following messages on my screen
before going back to BIOS

I dont see the M2 detected

 

[repoman0]

For the lulz I set my 5950x to 4.2 ghz all core and 1.144v (reported under load). On my (hampered) custom loop I'm seeing around 55C averages with peaks closer to 60C.

I think that falls in line with what you'd expect from an AIO... I think. It's honestly hard to tell.

As for score, locked at 4.2 I see 28,225, which would be slightly higher prorated to 12 cores at 21,169. Probably a function of me locking my clocks and your boost fluctuating a bit.

Thanks for the test. I got mine to ~21100. Put my ram at 3466 with the "fast" timing calc preset values for Micron E-die (it won't do 3600), and -5 all core in curve optimizer with default PBO limits. Very silence-oriented fan curve as well. I'm happy with that for now, need to just use the computer for a couple weeks to see if it's stable, then mess with the Auto OC additional boost. I haven't actually tweaked a system or run benchmarks in a long time ..

 

[JoeRambo]

its the chip ...

put my 3900xt back in and it boots.

Several things to try:

1) I'd disconnect the SATA devices
2) I'd wonder around in BIOS and chance M1 PCIE speed from auto to fixed 3 or 2
3) Switch between M1/M2 slots, maybe new CPU has trouble with direct connected M1 slot PCIE pins on CPU, that would help to diag it
4) An important test if BIOS/CPU can see the drive is the following -> BIOS usually has some BIOS FLASH menu, look if your drive is selectable there

 

[Noid]

grr sry

 

[Noid]

I changed my 3900xt boot to UEFI only
( from both LEGACY/UEFI )
It would not post .... so
" LEGACY " has something to do with my M2 drive

This option is GREYED OUT - NO SELECTION
when using my 5900X in BIOS

... wtf ... 6 months wait for this lol

 

[repoman0]

I changed my 3900xt boot to UEFI only
( from both LEGACY/UEFI )
It would not post .... so
" LEGACY " has something to do with my M2 drive

This option is GREYED OUT - NO SELECTION
when using my 5900X in BIOS

... wtf ... 6 months wait for this lol

Maybe the compatibility support module that enables legacy operation is broken in your board's BIOS? Support was not removed with Zen 3. Looks like there is a new beta bios for your board, maybe give that a try.

 

[Noid]

Last 2 BIOS releases have no hard disk support for the 5900x
( 5 BIOS flashes later ... )

AMD AGESA ComboAm4v2PI 1.0.8.1 -- boots

However, it seems I have unsupported chip information
CPU-Z has alot of missing info.

Looks like MSI dropped the ball on mine.

If I can't use the last 2 BIOS's ...
... this chip will never reach it's full potential.

I wonder is it will work in my ASUS Prime x470 now ...

Currently at the settings in attachment.
I need to clean and re-seat the chip and the Noctua now.
( be back later to see if I'm wasting my time tuning this MB now )

 

[reqq]

and my previously WHEA errors are gone with latest MSI 1.1.9.0 beta bios Also fast curve optimzer test..some cores bossting 5075.

 

[Justinus]

and my previously WHEA errors are gone with latest MSI 1.1.9.0 beta bios Also fast curve optimzer test..some cores bossting 5075.

What fabric and memory clocks and timings? And what board? And what memory config (dimms)?

 

[amrnuke]

Updating my previous post, I moved to AGESA 1.1.0.0 Patch D. PBO +200, CO with no offset for fastest, -5 for 2nd fastest, and -10 for other 4 cores. RAM at 3666 16-16-16-32 and IF at 1833.
PBO is limited by my board to +200 on my ASRock B450M Pro4.

First score is stock with PBO, 3200CL16 RAM.
Second score is PBO +100, 3600CL14 RAM.
Third score is PBO +200, 3666CL16 RAM.

GB5 ST: 1598 -> 1643 (2.8%) -> 1676 (2.0%)
GB5 MT: 8572 -> 9200 (7.3%) -> 9428 (2.5%)
Speedometer 2.0: 174.4 -> 178.2 (2.1%) -> 182.2 (2.2%)
CB20 ST -- 600 -> 611 (1.8%) -> 623 (1.9%)
CB20 MT -- 4609 -> 4660 (1.1%) -> 4662 (<0.1%)
CB23 ST -- 1538 -> 1575 (2.4%) -> 1609 (2.1%)
CB23 MT -- 11929 -> 12011 (0.7%) -> 12028 (0.1%)
CPU-Z ST -- xxx.x -> 653.1 -> 662.9 (1.5%)
CPU-Z MT -- xxx.x -> 5151.4 -> 5149 (<-0.1%)

Really seems like the curve optimizer isn't doing much here for sustained boost / multithreaded tasks, which is something I was hoping it would help on.

Any tips on how I can further tinker? I'm absolutely going to work on tightening up the RAM and/or trying 3733 with IF at 1866. I'm a bit unsure where to head next with the curve optimizer though.

 

[repoman0]

Anyone have any suggestions or tricks for figuring out which cores can't handle negative curve optimizer offsets? I tried -10 all core and almost immediately got a WHEA uncorrectable error BSOD, and finally got one at -5 offset as well after a few hours leaving the computer alone at idle. I know AMD has mixed messaging on undervolting with newer Zen chips -- they tune the voltage vs frequency vs temperature lookup table to within an inch of stable life from the factory and don't recommend undervolting, but elsewhere their marketing touts curve optimizer as a way to improve boost behavior by reducing power consumption.

 

[Justinus]

Anyone have any suggestions or tricks for figuring out which cores can't handle negative curve optimizer offsets? I tried -10 all core and almost immediately got a WHEA uncorrectable error BSOD, and finally got one at -5 offset as well after a few hours leaving the computer alone at idle. I know AMD has mixed messaging on undervolting with newer Zen chips -- they tune the voltage vs frequency vs temperature lookup table to within an inch of stable life from the factory and don't recommend undervolting, but elsewhere their marketing touts curve optimizer as a way to improve boost behavior by reducing power consumption.

One thing to try would be to ID your best cores in each CCX per ryzen master then apply a 0 offset to them. Go ahead and apply -10 to every other core and see how stable it is.

It seems like the best cores typically require the least offset, likely because they boost the highest with a thin margin and need the most voltage.

Otherwise your option is a very slow and methodical "one core at a time" approach, which will give you the best and most accurate result while taking way, way, way, way more time than I think anyone wants to spend doing this.

 

[Midwayman]

I just installed a 5900x bought today.
I have post.
But, on reboot after F10 saving settings in BIOS ...
... goes right back into BIOS screen
( loop )
Never leaves BIOS ... wtf

Tried the last 3 BIOS releases.
( surprised I could flash back )

Temps are good.
All devices present.

any ideas ?

I had the bios boot loop on a MSI board with my 5900x. Drove me nuts. Eventually I took all my drives out except for the m.2 on cpu lanes. It finally detected it as a boot drive and got past bios. Also if you're running XMP, boot on JEDEC default first. Once I had got through the first boot I was able to add everything back in. I'm on the 1.1.0.0 patch D bios from early Dec.

 

[Noid]

" cpu lanes " --> ???

I am thoroughly discouraged with MSI.
It takes 3 BIOS steps to reach IF 1900.
( 3600 / 1800 ---> 3800/1900 ---> mem tweaking )
If I get a boot failure ... my steps start all over again.
Loading good BIOS profiles ... fail on re-boot reset.

My AIDA scores look great.
Although I have a significant loss in L3.
( from my 3900XT )

BTW, AIDA is broken - no IF displayed for this chip -- Pro version

 

[Midwayman]

" cpu lanes " --> ???

Usually the m.2 closest to the socket. on x570 you'll have one with pcie lanes direct to the cpu and one or two that go through the chipset. Not sure it actually matters, but that's the one that worked for me.

 

[Noid]

I'm 7 hours P95 stable at the timings above.
CPU temp maxed at 71C over 7 hours.

Max CPU temp with gaming was 78C.
( on AIR with ambient temp of 80F )

Knowing my DDR settings from 2 other chips in this MB helped.
Too many BIOS flashes yesterday for me...
.. so I wont be flashing forward for awhile.
( to try disconnecting my SATA drives on boot )

So far, this chip scores same/lower than my 3900XT in gaming benchmarks.
AIDA boost hovering at 4.9 MHz with voltage on AUTO.
( AIDA static voltage 1.33v boosts at 4.5 MHz -- IIRC )

Positives:
Very low temps.
Low Latency.

Negatives.
Appears to have same strength as my 3900XT ( so far )
BIOS issues

[ Edit ]

Attached CB23 side by side result compare to my 3900XT
3000+ point increase in Multi-core
( did not expect such a jump in numbers ! )

... going to get the "feel" of this chip in my day-to-day stuff.

 

[reqq]

What fabric and memory clocks and timings? And what board? And what memory config (dimms)?

5900x
MSI x570 gaming pro carbon
2x8 g skill b die @ 3733 1:1 14-15-14-14-30-288 + dram calc fast subtimings(those at 3600)

 

[repoman0]

Think I might have successfully stabilized my 2x16 GB 3200CL16 kit at 3600 CL16, still stock 1.35v. I used the E-die "bad bin" preset in Ryzen Timing Calculator which suggested loosening a few subtimings. Before, Memtest for Windows would catch errors right away, now 10 instances testing 24 GB have been going for 15 mins.

Why do I care? No idea, other than 3600 MHz is a more round number than 3466

 

[Justinus]

Think I might have successfully stabilized my 2x16 GB 3200CL16 kit at 3600 CL16, still stock 1.35v. I used the E-die "bad bin" preset in Ryzen Timing Calculator which suggested loosening a few subtimings. Before, Memtest for Windows would catch errors right away, now 10 instances testing 24 GB have been going for 15 mins.

Why do I care? No idea, other than 3600 MHz is a more round number than 3466

Higher memory speeds, higher fabric speeds, higher performance in general. It's pretty straightforward.

I'd highly recommend using memtest86 on a bootable USB to test instead of in-windows testing though. It ensures every hardware memory address gets tested so there are no picky or weak modules on your DIMMs. It's also a more thorough and reliable memory test in general.

 

[repoman0]

Higher memory speeds, higher fabric speeds, higher performance in general. It's pretty straightforward.

I'd highly recommend using memtest86 on a bootable USB to test instead of in-windows testing though. It ensures every hardware memory address gets tested so there are no picky or weak modules on your DIMMs. It's also a more thorough and reliable memory test in general.

Of course, the point was I will never notice in a million years outside of bandwidth and latency benchmarks. I know it's just for the fun of tinkering. I usually do memtest86 after running the Windows one for a while. In the past for me memtest for windows has usually caught ram instability faster, or caught instability that the OG has missed. In fact I just bought the Pro version earlier this afternoon because I realized it's only $5 and I got tired of opening and starting 12 instances. Now at ~360% coverage of 28GB out of my 32 even after tightening my timings to the "safe" preset aside from Trfc which I left pretty high. 1.35 vdimm and 1.0 vsoc. Guessing Trfc was my problem all along. I wish I knew exactly what all these timings mean -- I believe Trfc is something to do with the regular interval at which the capacitors are recharged in order to preserve data. Last time I messed with subtimings was Athlon64 days, before I learned about logic circuits and basic chip design in undergrad.

edit: tRFC is just the number of clock cycles that the capacitors spend recharging each refresh interval. Pretty simple. Makes sense then that having that set too low in earlier testing might have been the single culprit disallowing a 3600 MHz ram clock if the leakage current for some of the ICs was higher. I have it set to 630 clock cycles now, equivalent to 560 cycles at 3200 MHz. In earlier testing raising the voltage up to 1.40 and then 1.45 slightly increased the amount of time I could run memtest for windows without an error .. capacitors inside the ICs would recharge slightly faster at a higher voltage and therefore current so that makes some sense.

 

[Kenmitch]

I guess if this winds up being true then there's a possibility of 5600x's with some decent overclocking headroom.

Dual-CCD Ryzen 5 5600X and Ryzen 7 5800X In the Wild | TechPowerUp

I might have to shut down my 5900x rig and boot up the backup 5600x rig and do some playing around.

 

[mooncancook]

Just finished building my 5800x with ASRock B550M Steel Legend. Now how do I install Windows 10 from USB? I've always done it from dvd drive and haven't done it from usb. I tried using the MS utility and it got stuck on blue windows logo. I tried using Rufus and it got stuck on ASRock logo. For windows utility I used 32gb partition with FAT32, and for Rufus I don't have option other than NTFS with full 80gb capacity. I have Secure Boot disabled.

 

[mooncancook]

Just realized Rufus created boot drive show up as two UEFI boot drive in BIOS. I changed the order and now it is showing some boot messages, and stuck on "launching 'efi\boot\bootx64.efi'... message

Edit: never mind, it is working after I updated the BIOS. The pre-installed BIOS didn't properly support ryzen 5000 CPUs

 

[Ajay]

Just realized Rufus created boot drive show up as two UEFI boot drive in BIOS. I changed the order and now it is showing some boot messages, and stuck on "launching 'efi\boot\bootx64.efi'... message

Edit: never mind, it is working after I updated the BIOS. The pre-installed BIOS didn't properly support ryzen 5000 CPUs

Building computers is FUN!