Ryzen - Return of the Jedi

[savagebunny]

Thanks to chew* to motivate me, and well.. no one loves the GT7 like I do. I'm just a pleb in the wind, can edit if I made a mistake.

[chew*]

Thanks for posting it

[AMD_OCer]

I have exhausted every option except newer bios which i do not want to use...i know its slower backflashing is a pain and some microcode is not leaving...im seeing this on taichi right now...annoying..

My llc is set the way its set so when i set 1.40 bios i see 1.40 socket with dmm. Same with soc...

I set manual voltages to ensure board is not playing dirty tricks...Auto at mercy of bios..

Fan is ocing...= run cooler

basically by setting the LLC you are disabling vdroop completely.. keeping constant and set voltage to the CPU under load at 1.40.

Did you try increasing the SOC with the 2x16gb and CH6 just a tad little bit... to 1.10 - 1.15.. just for the fun of trying..

[chew*]

I tried everything.

Some users are running 4x16g in it at 3000 though..

None at 3200 however.

[lowdog]

Chew

Any chance you could throw up a pick of the pad behind the cpu socket for the Taich and highlight the measurement points for Vcore and SOC pleeeeeease

I'm looking at my board's pad and can't for the life of me figure out what points to measure with my DMM

[Formula350]

T_T Damn. Running the latest BIOS through 32M while at BCLK 102, and something is JUST barely out of whack and causing it to not make it through a full run :\ It was stable enough for a Cache Stress test yesterday for over 15mins though :| Reason I'm bummed is I was heading for a time of (by my estimate) around 8min33sec, based on my previous best of 8min53sec. This next attempt, so far, will bring it down to roughly a 8m31s run at the rate it's going. That's pretty impressive I think, for running stock everything, and on Auto (volts, timings, etc) at DDR4-3200, with only adjustment being BCLK +2MHz lol (Making for DDR4-3264 and a boost of 3970MHz, as I'm in Balanced profile).

Go baby go! Loop 23 finished! 7m 55.771s lol FFFFFFFFFFFFFU....... *sigh* Spoke too soon. "NOT EXACT IN ROUND" SO close!!! >_<

[chew*]

Well needs to be win 7. You will most likely find it being slower rtc in w10 is buggy. Gives false good times when you mess with bclk.

[chew*]

Chew

Any chance you could throw up a pick of the pad behind the cpu socket for the Taich and highlight the measurement points for Vcore and SOC pleeeeeease

I'm looking at my board's pad and can't for the life of me figure out what points to measure with my DMM

I just start measuring caps. I try to keep vddp down till i can sort out soc...easy to mix them up when both are 1.00 in bios.

Test with vddr default @ 1.20. If testing near stock cpu volts...can get a false positive on that to if vddr and cpu are @ 1.35 lol...

[Formula350]

Well needs to be win 7. You will most likely find it being slower rtc in w10 is buggy. Gives false good times when you mess with bclk.

Wouldn't that result in the Checksum not matching up with the time, causing for it to not pass validation check?

[ZenEffect]

Wouldn't that result in the Checksum not matching up with the time, causing for it to not pass validation check?

No the checksum is really... Not as useful as you would think.

For cas + speed + cpu speed the run looks bugged sorry but I guess even under "ideal" circumstances it can still occur, I caught my fair share of heat running Windows 10 for 1 sub. Try it in Windows 7. For the most part what chew* said it will be slower, but you can tune most of it out and get over the "actually faster than" hump by spending a ton of time tuning the board to bench well, which is basically the polar opposite of 24/7 stability. I haven't even bothered with a DMM because in my opinion the values *set* in bios vs actual is inconsequential except for llc. It's a nice to know for things like llc but ultimately the setting in bios dictates what the end result will be regardless of actual voltage. I'm not after trying to find that golden chip which does XXX freq at XXX volts. To me it just needs to do XXX freq and that's it.

Just my 2c. DMM is a more correct way of doing things, but when you boil it down to action vs reaction, you can get away with a lot short cutting. The down side is you have no real sharable data as standards were not defined.

That would have been around 8:20 if it passed BTW.

[Formula350]

No the checksum is really... Not as useful as you would think.

For cas + speed + cpu speed the run looks bugged sorry but I guess even under "ideal" circumstances it can still occur, I caught my fair share of heat running Windows 10 for 1 sub. Try it in Windows 7. For the most part what chew* said it will be slower, but you can tune most of it out and get over the "actually faster than" hump by spending a ton of time tuning the board to bench well, which is basically the polar opposite of 24/7 stability. I haven't even bothered with a DMM because in my opinion the values *set* in bios vs actual is inconsequential except for llc. It's a nice to know for things like llc but ultimately the setting in bios dictates what the end result will be regardless of actual voltage. I'm not after trying to find that golden chip which does XXX freq at XXX volts. To me it just needs to do XXX freq and that's it.

Just my 2c. DMM is a more correct way of doing things, but when you boil it down to action vs reaction, you can get away with a lot short cutting. The down side is you have no real sharable data as standards were not defined.

That would have been around 8:20 if it passed BTW.

Thanks for the info.
Yea, in reflection, I have a feeling it indeed was likely a buggy run. The 1.0.0.4a/b/whatever were supposed to completely fix the Sleep Bug, but you know... that run WAS post-sleep. So, I think there are still some kinks to work out. I can say that the main bug I was experiencing of having the temps lock to not being accurate/responsive is indeed fixed, but perhaps that's where I've been having "stability" issues with AIDA64 runs...? I'm one who loves Sleep, particularly because I've found it'll really tell you if a system is stable or not, being that I've had game stable configs but fail to wake up from sleep. Usually those ones were BClk in nature, as it was when AMD first went with that on FM1 APUs where everything tied together, resulting in primarily issues with the Storage Controller flaking out under AHCI (needing to drop back to IDE mode).

I'm really tempted to try and get Win 7 running though, but I'm so on the fence about it *sigh* I'm not a bencher, so this all is just "something to do" and comparative really. However, Win10 (at least v1607) is really irritating me with some of the crap that's been changed :\ I have v1511 on this laptop and more-or-less am content with everything, I mean a few niggles here and there but overall it's to me. The MAIN issue I'm having with v1607 (and I wish I knew when specifically this was implemented) is they made it so UAC being disabled prevents built-in-apps from being able to run:

Which at first blush a person might think "Who the hell uses Win10 apps?", and I'm with you, except things like Photo Viewer are now 'apps', so any time I want to look at an image I have to open it in Paint Actually, now it's acting weirder than normal... Before, Edge wouldn't work either, giving the same 'error', but now that worked. Calc, and other programs, would similarly error, but now they kinda fake-open (you see the window's border of the application but it vanishes), no error. *scratches head* lol NEVERTHELESS, in v1511 that's on my laptop here it doesn't happen and I have UAC Disabled. Whether that comes down to versioning or the difference from Home on the laptop and Pro on the desktop... I dunno. Tis very annoying, though! I can appreciate the underlying point of UAC, but if I want it off, it should be off like Win 7. I don't need to click OK on a dialogue box when running anything (bad enough I have to deal with the Firewall one).
[/rant] heh Sorry

Regardless, I had thought that the 32M run being was down to a couple things: #1 AGESA being updated, as I finally flashed a BIOS other than the shipping ver. #2 BCLK @ 102, despite being an ever-so-small increase. #3 That I had killed Explorer and other background tasks, so I was mainly hoping that was the biggest reason for the improvement. Also, fun fact! If you use Process Monitor and 'suspend' MsMpEng (the Antimalware service), it basically knackers the entire system, preventing it from being able to do, well, anything. Open a program... just kinda stalls. Un-minimizing something on the taskbar... pops up with totally blank box other than the header bar. Close something... sits there unresponsive. For that matter pretty much anything that wasn't already open that you try and use becomes "Not Responding" heh KINDA problematic given I can't un-suspend it now. Whodathunk?!
...... and 8mins later, now things start to show up (load) LOL Guess they timed out waiting for whatever it was (probably to scan the program) :P

[AMD_OCer]

k so I decided to go with 2x16GB and aim for the 3100 CL14 DDR4 kit with the CH6. Its more futureproof, and makes more sense than going 4x8gb.

I will start OCing when I get all the parts.. and post some hopefully nice results. I will try to mess with the mem as much as I can too to get it running at 100% spec speed for the ENTIRE 32gb.

Thx chew for all the info and help, in the mean time I will stick around to bug you with more stuff lol

[chew*]

Lol formula.

Win 10 is meh imo...my daily is still on 7.

Maybe a stripped win 10 would be tolerable but currently i prefer 7 even more so after using 10.

Win7 install is easy by the way. Ive loaded multiple versions from a usb stick.

As far as 32m we run it hardcore all the time so we know exactly what timings/ memspeed/ cpu clock is needed for xx time with xx tweaks.

Preety much just a glance and we can tell if something is "off"

In fact im supposed to be working on my 32m times...somehow got sidetracked working on 32g instead to help end users lol.

[aGeoM]

@Formula350

Nice 32M try, if you mod Bios do show HPET and then disabled it in your BIOS settings, you'll be exposed to the Sleep bug.

Anyway, have you tested the E7A31AMS.151 bios?

[Formula350]

@Formula350

Nice 32M try, if you mod Bios do show HPET and then disabled it in your BIOS settings, you'll be exposed to the Sleep bug.

Anyway, have you tested the E7A31AMS.151 bios?

I have not. Have they been releasing Betas on the forum again?
And yea I noticed they hid HPET now, thought that was odd. Disabling HPET will prevent Sleep Bug, though?



And Chew, I considered trying to make a "stripper" model (to coin a car phrase lol) of Win10, but I still suspect it won't grant me what I'm after since I suspect that crap is baked into 10. I'll probably end up copying over this older HP version off my laptop and installing that instead. I only hope it's not afflicted with the same bug that RTM had which crashed the system every few minutes. Couple people had it too and installing the newer 10 fixed their's just like it did for me, so there's definitely something not Ryzen friendly about the original Win10 (installed it says v10.0, not the newer 'build' style version of v1511 etc).

[AlleyViper]

On a side note, there are laptops and MBs such as my DFI 790FX that will show a sleep resume bug under W10 when HPET is enabled on bios (a peculiar BSOD that can be identified later with bluescreenview, but at the time it crashes on resume no image on monitor will show), but only when the system clock isn't forced to it. For some reason, IIRC previously W8.1 and early W10 builds were automatically set to HPETs 14.3Mhz instead of the default 3.xMHz on that particular MB without me touching anything (while it didn't on my Asus 890GX that behaves fine on either setting). But since 1607, I must set bcdedit /set useplatformclock true if I wish to be able to resume from S3 (S1 is OK) at default HTT, but it'll still fail at my daily HTT 240MHz setting no matter what.

[Hell Hound]

On a side note, there are laptops and MBs such as my DFI 790FX that will show a sleep resume bug under W10 when HPET is enabled on bios (a peculiar BSOD that can be identified later with bluescreenview, but at the time it crashes on resume no image on monitor will show), but only when the system clock isn't forced to it. For some reason, IIRC previously W8.1 and early W10 builds were automatically set to HPETs 14.3Mhz instead of the default 3.xMHz on that particular MB without me touching anything (while it didn't on my Asus 890GX that behaves fine on either setting). But since 1607, I must set bcdedit /set useplatformclock true if I wish to be able to resume from S3 (S1 is OK) at default HTT, but it'll still fail at my daily HTT 240MHz setting no matter what.

Get this let me know if its fixed, I'm using the free version w/ 8.1 though


http://www.lucashale.com/timer-resolution/



TimerResolution is an application to change the resolution of the default windows timer. The standard timer on Windows XP can vary between 10 and 25 milliseconds. Therefore if your code uses a timer or sleep value less than the timer resolution on your system you won?t be getting the results you expect.

The timer resolution will be globally changed while the application is running and will revert to the previous value when it is closed.

There is a difference for me

GTA = NO MORE STUTTER

Windows responds a lot faster

Consistent min fps when benching

[aGeoM]

I have not. Have they been releasing Betas on the forum again?
And yea I noticed they hid HPET now, thought that was odd. Disabling HPET will prevent Sleep Bug, though?

Not on English Forum. I got it from the German forum, just for studying Modules. I can post the link if you will. I have same feeling, that they hide HPET for prevent Sleep Bug, letting it enable on BIOS, away from "users" to change it.

[chew*]

was fooling around earlier......I really need a good set of B die for benching...this set is meh...

Quick reference point run


Hardware optimized run....no clue how to optimize OS for this benchmark.....

[I.nfraR.ed]

What volts for that last screen?

I got FlareX 3200C14. No problemt at XMP settings even with lower voltage (1.325V), now testing 3600 16-16-16-36 1.35V, simulating a higher binned kit.
Will post results when done, but it seems on par with my other 2 B-Die kits.
The strange thing is I have TRC at 61 with exact same bclk and divider with this FlareX on 0082 CH6 bios, while you're at 63.

[chew*]

that screen requires 1.55v

Like I said your ram and my ram......far from equal......

1.35.....maybe I can get into bios at c19......

My flare x? have not over volted but they are not hearing it @ stock volts and optimal timings at 3600.

maybe its my board, maybe its my imc, maybe your chips cold.

I'm using this bios cuz i'm planning on working on 32m and not sure if you tested but it got really slow on newer microcode......

I'm also running a stock win 10 unstripped and no max mem which allows me to push much easier in 7 btw.........

[I.nfraR.ed]

Maybe my IMC, don't know, but CPU is currently on Megahalems just put on top with 1 unmounted fan. Have no other chip to test. Had 1700 for few days, but sold it.
Plus my chip is happy to run even 3800+ on chiller, no IMC coldbug at these temps, while 1700 was coldbugged. Still have to test on LN2.
Maxmem is needed if you're running high mem clock and high vdimm (all my 3800+ C12 screens are with maxmem, but that's also required on Intel). I'm testing "daily" settings now, so running at full memory capacity, although I won't advice running 135 bclk for 24/7.
I noticed slowdown in Pi, yes.

For 3600C12 you definitely need above 1.6V on the mem.

[chew*]

Cant run over 1.55 vdimm on this chip.

1700 was 1.575 max.

Im doing your little test right now with 16-16-16 3600 @ 1.35. Only thing it looks like i will accomplish is hosing my OS...

[I.nfraR.ed]

3600C16 test done. Booted 3600 15-15-15-35 1T 1.35V same settings, will run tests.
Maybe I'm just lucky.

PS: Forgot CAS15 is rounded to 16.

[AMD_OCer]

Those BCLK though... oufff