First Ryzen build (5600x) frustrations(?)

Skull_Angel

[H]ard|Gawd
Joined
May 31, 2010
Messages
1,664
Hey guys, just needed to vent a bit because this is the first build I've put together in over 20 years that's been so simple to tune that it's actually frustrating; I swear it's going to give me anxiety due to the lack of issues throughout stability testing.

I went into this build fully expecting to encounter some issues I've read about in various forums, but have been pretty lucky with nothing showing up, which is great. The only tuning problems I've had so far have been from manual RAM overclocking, but that's been ironed out with great success on settings I'm content with.

The CPU overclocking through PBO, auto-oc, and curve offsets has been what's driving me up the wall though; it's simple (too simple?? I've only ever done manual OCs before), but mostly undocumented and seemingly works even with horrible settings. I spent too long looking for info only to find users experiences, so decided to experiment myself until reaching the performance limits of what my chip and setup could do... In all that time testing and benching the only settings to cause any issues were curve optimizer and it's got me paranoid as to whether the damn thing is actually stable! It's run the gauntlet of tests from 10 cycles of y-cruncher, various OCCT tests and settings, 20+ hrs CoreCycler, TM5, 20+ hrs P95, ect. Is FIT just too aggressive at protecting the CPU? If it is, can this really be considered stable? Am I overthinking this??(probably)

For all intents and purposes tuning should be considered finished and all that's left to finalize it is a clean install. There's just that little voice that keeps saying "It can't be this easy, right??".

/rantoff

System specs for anyone interested:
AMD 5600x (aOC +150, PPT 114, TDC 72, EDC 104, minor curve offsets 0, -3, -12, -12, 0, -3)(NH-D15)
MSI x570 Tomahawk (1.2.0.3b)
G.skill FlareX 2x16gb (3800/1900 16-16-16-16-24-40 @1.43v)
Sapphire Nitro+ 6700xt (stock OC)
BeQuiet Straight Power 11 750w
 
Last edited:
Relax and play some games or something entertaining. Good choices on the components makes it very simple enjoy it now as that’s a sweet little rig.

Haha, thanks. It's likely boredom; I'm use to spending at least a month tuning a new rig, but after all that was said and done, this only took a little over a week worth of actually doing anything.
 
I've built at least 6 similar rigs for friends and family and had only minor issues with a couple of them. Most were built on the B450 Tomahawk which isn't as refined as your board but performance was always really good regardless of using the cheap solution.
 
I've built at least 6 similar rigs for friends and family and had only minor issues with a couple of them. Most were built on the B450 Tomahawk which isn't as refined as your board but performance was always really good regardless of using the cheap solution.

Nice, it's been a bit since I've put together builds for buds, kinda miss the free booze (y)

The x570 Tomahawk was mostly about the VRM, price (got it before the price hike), and feature set. Still seems like the bios/uefi is going to need more maturity, but it's enough for it's primary use in gaming for now. The only weak part has been the RAM so far; it's older RC type B-die on rc rev B1, so there have been issues with temperature stability past 42C when primary timings are tighter than 16 (tried everything from loose secondaries/tertiaries, voltage meddling, and cad_bus).

yes its that simple now. go game.

Haha, thanks. It feels wrong being that easy, but it's not like there was a lot of headroom for gains. The multi-core boost is definitely better than stock though; I just wish there was a bit more granularity to curve optimizer to keep max single/low-core boost VID under 1.45v (throttle) easier.
 
Nice, it's been a bit since I've put together builds for buds, kinda miss the free booze (y)

The x570 Tomahawk was mostly about the VRM, price (got it before the price hike), and feature set. Still seems like the bios/uefi is going to need more maturity, but it's enough for it's primary use in gaming for now. The only weak part has been the RAM so far; it's older RC type B-die on rc rev B1, so there have been issues with temperature stability past 42C when primary timings are tighter than 16 (tried everything from loose secondaries/tertiaries, voltage meddling, and cad_bus).



Haha, thanks. It feels wrong being that easy, but it's not like there was a lot of headroom for gains. The multi-core boost is definitely better than stock though; I just wish there was a bit more granularity to curve optimizer to keep max single/low-core boost VID under 1.45v (throttle) easier.
have you tried just using a voltage offset with pbo +200? everyone ones gonna say thats not the way, but its is the simple way...
 
have you tried just using a voltage offset with pbo +200? everyone ones gonna say thats not the way, but its is the simple way...

vCore offset instead of curve optimizer offset? I haven't thought about it, does it not disable PBO? I've got 2 weak cores that don't hit 4850 boost, so limited it to 4800; I got a funky CPU, 2 ultra efficient cores (don't boost well), 2 normal cores, and 2 leaky cores that get hotter than the sun, but boost to the heavens (the heat is problematic for FIT boost algorithm), lol.
 
There is no such thing as overclocking anymore. Cpus are built to run to thier highest frequency sqfely and naturally now.

A good metaphor is akin to driving a Bugatti to 190mph. Is it overclocking the car? No. It just capable of that much speed.

Same with modern CPUs no one overclocks anymore. That died a decade ago.
 
yeah. i have a -.075v offset, pbo on, fmax enabled and +200, i get all core 4850 and temps seem fairly even across cores.

Thanks for the heads up, I'll give it a try when I've finished tweaking RAM again. Anta777 started posting on the ddr4 thread at overclock.net with some good info, so I'm trying to see if refining my timings with his info gets me better efficiency.
 
There is no such thing as overclocking anymore. Cpus are built to run to thier highest frequency sqfely and naturally now.

A good metaphor is akin to driving a Bugatti to 190mph. Is it overclocking the car? No. It just capable of that much speed.

Same with modern CPUs no one overclocks anymore. That died a decade ago.

The analogy is close, but I'm not sure if it's quite correct. Using PBO is more akin to using an ECU or plug-in module that has some basic map presets you can select, but not modify (normal, sport, eco); the CPU is able to take care of it's personal safety due to FIT, but you can refine what it's fed [to a degree] for prolonged boost performance under certain circumstances (multi-core loads).

Manual overclocks sill have value in certain applications, but due to heat density [and wide variations in core quality on the same chip] on 7nm you're unlikely to reach the same clocks under high multi-core loads as you would under single/low-core loads.You would optimally want multiple profiles and the ability to swap them on-the-fly for a general use manually overclocked machine. But, even at that point you'd have to have a good way of assigning core affinity because Windows and AMDs solutions aren't always the best at figuring it out, and something like process lasso is sill going to add latency which may entirely negate any benefits...

Generally with AMD PBO*, the best performance will come from limiting voltage (PPT, off-sets) so that it doesn't reach the 1.45v VID limit [but really only for multi-core], and current (combination of PPT, EDC, off-sets, cooling solution) to reduce temperature based boost-throttling aggressiveness.

*edit: added for clarity
 
Last edited:
Haha, thanks. It's likely boredom; I'm use to spending at least a month tuning a new rig, but after all that was said and done, this only took a little over a week worth of actually doing anything.
My first ryzen build annoyed me the same way. With all the ram compatibility issues and struggles to hit 3200mhz I was expecting a week to tune out the best ram timings.

Put all the pieces together, boot, enable XMP, it worked!? /cry.
 
There is no such thing as overclocking anymore. Cpus are built to run to thier highest frequency sqfely and naturally now.

A good metaphor is akin to driving a Bugatti to 190mph. Is it overclocking the car? No. It just capable of that much speed.

Same with modern CPUs no one overclocks anymore. That died a decade ago.
and overclocking is removing the restrictor plate; aint exactly legal, better keep that on the down low.
 
My first ryzen build annoyed me the same way. With all the ram compatibility issues and struggles to hit 3200mhz I was expecting a week to tune out the best ram timings.

Put all the pieces together, boot, enable XMP, it worked!? /cry.

It's the primary reason I started RAM tuning! Played with PBO and got frustrated/bored, looked into more in-depth RAM tuning... And here I am, around 2 weeks later (mostly reading between testing), still messing with timings that already work to see if I can get better results :ROFLMAO:

Apparently got "shitty" early revision b-die, but getting good results so it doesn't bother me much. At least the whole process is leading me to accumulate even more knowledge I'll rarely use, hahaha.
 
Same with modern CPUs no one overclocks anymore. That died a decade ago.

Um... err... ok. Not particularly true. But kind of true sort of...

These days the need, or desire, to overclock is not nearly as strong. There are experimenters (like me) who will overclock for the fun of it. There's still a lot of us out there.

Rather you might want to modify the statement to: "Very few people run their systems overclocked". Which may be more accurate.

Back in the day there was real motivation for overclocking. You could get a AMD Thunderbird or an Intel Coppermine and gain 20-50% performance boost on air. Build a custom water loop and even a bad chip might give you a 50% overclock (definitely true for the Thunderbirds).

Today... erhm... I'm not going to spend days tweaking a chip and tweaking the RAM for a 5% performance boost. Which is a far cry from achieving a 1.2Ghz overclock, on an 800Mhz chip, on air, or for less than $300.00 in hardware store parts. This is boring and not impressive.

I get a new chip, overclock it with stock cooling (for fun), and go back to stock. But run overclocked? Nah.
 
Um... err... ok. Not particularly true. But kind of true sort of...

These days the need, or desire, to overclock is not nearly as strong. There are experimenters (like me) who will overclock for the fun of it. There's still a lot of us out there.

Rather you might want to modify the statement to: "Very few people run their systems overclocked". Which may be more accurate.

Back in the day there was real motivation for overclocking. You could get a AMD Thunderbird or an Intel Coppermine and gain 20-50% performance boost on air. Build a custom water loop and even a bad chip might give you a 50% overclock (definitely true for the Thunderbirds).

Today... erhm... I'm not going to spend days tweaking a chip and tweaking the RAM for a 5% performance boost. Which is a far cry from achieving a 1.2Ghz overclock, on an 800Mhz chip, on air, or for less than $300.00 in hardware store parts. This is boring and not impressive.

I get a new chip, overclock it with stock cooling (for fun), and go back to stock. But run overclocked? Nah.
This. My old Tbird shaved load times on UT2k3 from 40 seconds to 15 seconds, which was VERY worth the effort. 9500 Soft mod to 9700 saved several hundred dollars. Etc.

Now? It's on water, auto-boost is set, and it's fast enough for anything I want to do. Pushing things gets me... an extra 2-3%? Meh, not worth the effort and the random crashes till you get it fully dialed in - unless you enjoy doing that as your hobby, which is totally cool (I used to play with autoexec/config.sys to get an extra k of memory free, even though I already had 604k!).
 
This. My old Tbird shaved load times on UT2k3 from 40 seconds to 15 seconds, which was VERY worth the effort. 9500 Soft mod to 9700 saved several hundred dollars. Etc.

Now? It's on water, auto-boost is set, and it's fast enough for anything I want to do. Pushing things gets me... an extra 2-3%? Meh, not worth the effort and the random crashes till you get it fully dialed in - unless you enjoy doing that as your hobby, which is totally cool (I used to play with autoexec/config.sys to get an extra k of memory free, even though I already had 604k!).

Gosh. I'm old. You're old too.

Not many remember the innovation put forth by the original modders between 1995 and 2005. Every innovation we have now in cooling, case design, overclocking, and GPU hacking is relatively off the shelf technology now. It's Built in.

Blows my Mind. I have so much uselss knowledge in my head now.
 
Gosh. I'm old. You're old too.

Not many remember the innovation put forth by the original modders between 1995 and 2005. Every innovation we have now in cooling, case design, overclocking, and GPU hacking is relatively off the shelf technology now. It's Built in.

Blows my Mind. I have so much uselss knowledge in my head now.
Oh yes. So much ancient history that doesn’t do anything useful anymore. But it also means that we understand what the automatics are doing now and can adjust / troubleshoot too
 
Oc'ing memory is where it's at these days. And still diminishing returns.

Pretty much. Current RAM tune I'm in the process of finalizing will look like it's on the weaker side, but latency is good and seems to put less stress on the imc, so I'm seeing a slight gain in boost frequency duration when benching. Fingers crossed there are no issues, haha
 
You know your old if your first PC storage media was a cassette tape. 5.25 floppy disks were just beginning to appear back in 83 when I was a senior in HS.
 
Final settings for RAM for now. Probably took more work than it was worth, but it was a fun journey.

Final.png
 
did you happen to do any before/after tests to compare, see if its worth it?
Compared to stock, XMP? I don't think I even benched stock settings, XMP 3200 14-14-14-14-34 was ~58ns, iirc. Early 3600 stable averaged ~57ns, while earlier 3800 revisions were closer to ~55ns. Like I mentioned earlier the kit is an older b-die revision (type RC) on raw card revision B1, so it's got it's issues; what hurt me the most was heat instability when lowering primaries at higher frequencies, may revisit if I ever upgrade cooling solution, but it'll be a bit of work to tighten up.

test.png
test5.png


Later testing moved to ABSOLUT@anta777 config on TM5. A bit more brutal and took about 50min longer with suggested .cfg changes for 12 thread CPUs, but it may also need more than the standard 3 cycles as well. When nearing stable it always seemed to pass the first round of testing, but not a second round after start up (failures occurred near the end of cycle 3).

The toughest settings to work with were cad_bus since there are some standard combos that work well (24-20-24-24, 40-20-30-20, 60-20-30-20, etc.), but the closer you get to the edge of stable frequency, timings, and voltage the more you have to hone in on looking for settings unique to your individual setup which is a bit too time consuming for me.

Best not to, it ruins the fun learning your hours of tweaking produced a 1% increase in performance. Best to just assume it's awesome.

I'm not sure how a 8~9% latency improvement translates to realworld performance, but I'm going to assume it's about as good as upgrading from a 120hz monitor to 144hz :ROFLMAO:

edit: footnote about the TM5 ABSOLUT profile - it doesn't seem to pick up on imc related instability, so make sure to do a full suite of RAM related testing when finalizing results.
 
Last edited:
Back
Top