Ryzen Owners Zen Garden

[igralec84]

If you can run a full round or two of 32m on a single core or thread it should be ok..

Edit:

If not the computer will probably blink out and restart

I've been sticking to that so far, if it crashes i try it again just to make sure, core 8 (1st one of CCD2) was pretty clear 4 times in a row that it doesn't like -13

Why not ycruncher? This is my testing config for CO, just remember to set the individual core every time you run ycruncher. I would go default cfg corecycler first because it's a faster test. I wouldn't go in reverse from something else to corecycler (unless that something else is something worthless like cinebench).

View attachment 278379

Max clock on x core never indicates anything for what that core can achieve on CO. Counterproductive to start going off of assumptions like that. CO value is only representative of how distant the stock core V-F is set from what that core is really capable of.

I think corecycler that i have uses some 1usmus ycruncher setup and preset, i've running it to confirm the values from time to time if superPi doesn't crash on a thread. It hasn't crashed yet but i'm running 2 minutes per core currently. Also did around 30 mins of prime95 small FFTs and it didn't crash, but i'm sure 3Dmark CPU profile still would in at least 2/5 cases. Maybe i should run the RAM at XMP 5600 CL40 first and see what happens, but i haven't had an idle or random crash while doing nothing for a while now, so for my daily/gaming use it is stable it seems.
CCD2 seems to be not as good for CO, on CCD1 all cores are over -20 while on CCD2 one doesn't do -13 and another doesn't do -18, the other six are currently at -15.
If PBO +100 means x58.5 multiplier, i'm almost reaching it on one core with 5830mhz.
I did notice ASrock pulled the 1.14 BIOS i'm using due to people having BSOD issues, so hmm... might continue playing with all this on the next BIOS if they think something is messed up

 

[Veseleil]

but prime95 is just so unrealistic for casual users after 10-15 minutes

From my own experience with small FFTs, if it doesn't pop up with errors after a few tests/passes, it won't show errors even if it's running for hours.
The obvious hint that the system isn't stable is any of the newer Cinebenches. Mostly it crashes in the middle or the very end, or it won't start at all, followed with an error notification.

 

[Space Lynx]

From my own experience with small FFTs, if it doesn't pop up with errors after a few tests/passes, it won't show errors even if it's running for hours.
The obvious hint that the system isn't stable is any of the newer Cinebenches. Mostly it crashes in the middle or the very end, or it won't start at all, followed with an error notification.

if my games never crash I no longer care about the synthetic crap. I get it though, most people like that peace of mind, fair enough

 

[freeagent]

OCCT is a quick and dirty test to get you in the ball park. I wouldn't rely on superpi.

 

[Space Lynx]

OCCT is a quick and dirty test to get you in the ball park. I wouldn't rely on superpi.

I have always felt that if I can pass Aida64 stability for 10 minutes and Cinebench 10 minute run I am good to go. I don't bother with anything else, never used OCCT once in my life, I might give it a go though, thanks!

 

[freeagent]

AMD CPUs are kinda funny at first if you are coming from Intel.. They are sneaky, and tricky, and they will eff you up

Don't be afraid to be thorough

 

[igralec84]

OCCT is a quick and dirty test to get you in the ball park. I wouldn't rely on superpi.

Did one benchmark single thread SSE and it rebooted almost immediately Wish i knew which thread it was

EDIT: oh it actually has a last known status and after 8s cores 3 and 7 were at 68 and 63°C while others were 40-50, might be one of them

 

[3x0]

Did one benchmark single thread SSE and it rebooted almost immediately Wish i knew which thread it was

Go to Event viewer and look for whea 18 errors, APIC ID will tell you what thread crashed. If it's APIC ID 7 then core 3 crashed

 

[freeagent]

3 and 4 always gave me trouble. Always the first to go in core cycler.

 

[igralec84]

If i choose just one core to run the test, core 0 throws 30k errors in a couple of seconds, so seems to be at least that one as it's at -25 Will go over all of them now for a minute, better than 10 minutes i spent on each in superpi and thought it was stable as it didn't crash.

Surprise surprise, it's both cores that have -25 CO

 

[AusWolf]

AMD CPUs are kinda funny at first if you are coming from Intel.. They are sneaky, and tricky, and they will eff you up

Don't be afraid to be thorough

Intel is "plug and play". AMD is "plug and play with it".

 

[freeagent]

Intel is "plug and play". AMD is "plug and play with it".

Nah..

I came from Intel and had no problems with my setup. I was totally green. The guys here helped me pick my parts and then to tune it later on. Smooth sailing for me

 

[tabascosauz]

Nah..

I came from Intel and had no problems with my setup. I was totally green. The guys here helped me pick my parts and then to tune it later on. Smooth sailing for me

hey 3600XT is like, cheating you missed all the "fun" of 2019

 

[AusWolf]

hey 3600XT is like, cheating you missed all the "fun" of 2019

Nah. If you wanted a real plug and play experience on AMD back then, 3100 + A520 board all the way!

 

[freeagent]

hey 3600XT is like, cheating you missed all the "fun" of 2019

I really and truly did. I didn't even read about the troubles people were having. So for me coming in at the end of AM4, it was really a smooth ride, a pleasure even

And then you guys were making me feel stoked by telling me how good I was doing.. meanwhile it was pretty easy because the silicon did all the work for me

The real challenge for me came when I got 5900X and I was obsessed with getting it to boost to its limit

Good times

 

[tabascosauz]

I really and truly did. I didn't even read about the troubles people were having. So for me coming in at the end of AM4, it was really a smooth ride, a pleasure even

And then you guys were making me feel stoked by telling me how good I was doing.. meanwhile it was pretty easy because the silicon did all the work for me

The real challenge for me came when I got 5900X and I was obsessed with getting it to boost to its limit

Good times

AMD within 6 months of launch:

AMD after 6 months:

Too bad I forgot all about my own advice when dealing with Radeon Technologies Group, so I'm reliving purgatory all over again.

 

[AusWolf]

I really and truly did. I didn't even read about the troubles people were having. So for me coming in at the end of AM4, it was really a smooth ride, a pleasure even

And then you guys were making me feel stoked by telling me how good I was doing.. meanwhile it was pretty easy because the silicon did all the work for me

The real challenge for me came when I got 5900X and I was obsessed with getting it to boost to its limit

Good times

Or you could have just bought a 3100 that boosts to its 3.9 GHz limit all the time, in every scenario, straight out of the box.

 

[freeagent]

Or you could have just bought a 3100 that boosts to its 3.9 GHz limit all the time, in every scenario, straight out of the box.

But...

Why have 3900 when you can have 5150?

Besides.. XT was the same price as the vaginilla, and was always meant to be a placeholder for 5600X. I am totally fine with 5600X btw.. I wanted a GPU when I bought 5900X but couldn't get one..

And with the way I have my setup currently, I can run it fast and quiet.. not as quiet as I could run my X3D though.. sadface.

Maybe tomorrow I will make another attempt to fix it.

 

[AusWolf]

But...

Why have 3900 when you can have 5150?

Besides.. XT was the same price as the vaginilla, and was always meant to be a placeholder for 5600X. I am totally fine with 5600X btw.. I wanted a GPU when I bought 5900X but couldn't get one..

And with the way I have my setup currently, I can run it fast and quiet.. not as quiet as I could run my X3D though.. sadface.

Maybe tomorrow I will make another attempt to fix it.

I was only joking.

I only had the 3100 as a placeholder when I sold my 5950X (I realised I didn't need 16 cores after a couple of months). Then I kept it as a HTPC processor because it grew so close to me. It's like a puppy wagging its tail when she sees you. Even though you wanted a cat all along, you end up keeping her because she's too cute to give away. A friend of mine wants a mini indie gamer rig for his daughter, and I just upgraded my HTPC with the 11700 that came out of my main rig when I switched to AM5, so the trusty 3100 might find a new home eventually, after all.

 

[Mussels]

The last time I pulled X3D from my system the CPU stuck to the cooler, was not expecting that when I was removing it. Of course the CPU had to come out with it.. bending some pins in the process . and I didnt notice until yesterday when my CPU would not socket.

And of course now that I am old I cant see shit up close, and all of those pins just blend into a golden blur so I cant fix it until I go get glasses

Never once had this problem with AM4 since I started using it, but this latest batch of TF7 is sticky. I quit using it. FFS. This is one of the reasons why I left 939.

I got those old feelings once again

Good news, even broken pins can be replaced
Always try and twist to remove, and heat up before removing when possible - i'll power up a mobo with CPU but no ram or GPU to let it warm up slightly, if i have to

quite often you can sit the CPU over the socket, and gently 'spiral*' it around - the sockets themselves have a slight curve to them to help guide the pins in, and they can self-repair slight bends
(If the bends are all in one direction, you can of course have it slightly to that side and push that one direction with gentle downward pressure)



*Terrible word for what i'm explaining, but i've done this with 939 CPU's long ago that i bent. Once you get the worst ones lined up, the rest self guide in

 

[freeagent]

Good news, even broken pins can be replaced
Always try and twist to remove, and heat up before removing when possible - i'll power up a mobo with CPU but no ram or GPU to let it warm up slightly, if i have to

quite often you can sit the CPU over the socket, and gently 'spiral*' it around - the sockets themselves have a slight curve to them to help guide the pins in, and they can self-repair slight bends
(If the bends are all in one direction, you can of course have it slightly to that side and push that one direction with gentle downward pressure)



*Terrible word for what i'm explaining, but i've done this with 939 CPU's long ago that i bent. Once you get the worst ones lined up, the rest self guide in

Thank you sir I will try that tomorrow!

 

[Mussels]

Is SuperPi32M a reliable way to test Curve Optimizer settings?

You need to leave it running for a long time at low/zero load as well, like browsing the web and such

Most of my CO crashes came about from ending tests, as the CPU Switched between low and high power states

Thank you sir I will try that tomorrow!

I'll do one of my infmaous MS paint examples
I got SUPER lazy and the unimportant bits are not to scale in the slightest
So if the red pins are bent out, you line up the CPU so they're straight and get them JUST in the socket the tiniest amount - then move the CPU closer and closer to the correct orientation gently
small amounts of up and down left and right movement as you straighten it increase the odds of any bent pins finding their home and sliding in

Zero harm in getting the outside ones straightened, lifting the CPU out, and trying again starting slightly lifted from all four sides in turn - only lifted a very slight angle


Normal install:

Fixing pins bent to the 'left' in this image, you'd move the CPU to the left of the socket (Now i mean like 0.1mm off center, not a lot) and gently push to the right and down at the same time
You can then try this from all four directions and simply see if one direction will let the CPU sit in the socket naturally, where straight down fails

The scale of how angled to have the CPU is of course off, we're definitely talking a very small angle and very slightly off center - think of how little it'd take a CPU with straight pins to be shifted around before it lined up and fell down into the socket, if it wasnt sat down slightly rotated or something

Like if this was a stock CPU and you just had to twist that little bit for the pins to line up, that's the gentle level of effort here - these bent pins just have a new position that may be a little to one side or a little rotated before they fit in, and once its in once and latched in, they'll work normally (or very close to it) for every future install

 

[igralec84]

You need to leave it running for a long time at low/zero load as well, like browsing the web and such

Most of my CO crashes came about from ending tests, as the CPU Switched between low and high power states

I see, even the OCCT errors i can't really replicate, can be 3 times in a row after 1s but then 5 times nothing for 5 minutes, so really random. I did get one of those yeah, when the test ended it just rebooted.

I've settled for CCD1 all cores on -20 and CCD2 all cores on -15 for now, a bit better than all core -13 i ran for a month.

Did some tweaks to the RAM timings & voltages and have to run memtest & TM5 again to see if it TM5 still gets errors after 2 hours while karhu doesn't after 5 hours.

I guess as long as i don't get crashes while gaming or after gaming / while launching a game (Steam client seems suspicious), i'll consider it stable, like @Space Lynx said.

 

[Veseleil]

You need to leave it running for a long time at low/zero load as well, like browsing the web and such

Most of my CO crashes came about from ending tests, as the CPU Switched between low and high power states


I'll do one of my infmaous MS paint examples
I got SUPER lazy and the unimportant bits are not to scale in the slightest
So if the red pins are bent out, you line up the CPU so they're straight and get them JUST in the socket the tiniest amount - then move the CPU closer and closer to the correct orientation gently
small amounts of up and down left and right movement as you straighten it increase the odds of any bent pins finding their home and sliding in



View attachment 278537


Zero harm in getting the outside ones straightened, lifting the CPU out, and trying again starting slightly lifted from all four sides in turn - only lifted a very slight angle


Normal install:
View attachment 278538


Fixing pins bent to the 'left' in this image, you'd move the CPU to the left of the socket (Now i mean like 0.1mm off center, not a lot) and gently push to the right and down at the same time
You can then try this from all four directions and simply see if one direction will let the CPU sit in the socket naturally, where straight down fails

The scale of how angled to have the CPU is of course off, we're definitely talking a very small angle and very slightly off center - think of how little it'd take a CPU with straight pins to be shifted around before it lined up and fell down into the socket, if it wasnt sat down slightly rotated or something
View attachment 278539


Like if this was a stock CPU and you just had to twist that little bit for the pins to line up, that's the gentle level of effort here - these bent pins just have a new position that may be a little to one side or a little rotated before they fit in, and once its in once and latched in, they'll work normally (or very close to it) for every future install
View attachment 278541

Like for the effort.

 

[Arco]

From my own experience with small FFTs, if it doesn't pop up with errors after a few tests/passes, it won't show errors even if it's running for hours.
The obvious hint that the system isn't stable is any of the newer Cinebenches. Mostly it crashes in the middle or the very end, or it won't start at all, followed with an error notification.

My system can pass prime95. But cinebench23 went boom due to the power kernel.

It happens when it starts or when the power goes from nil to a lot.