SuperPI - "Not Convergent in SQR05" with waza?

[Guest Alex Bese]

Hey guys!

I tried the copy-waza tweak with SuperPi today for the first time, as GtiJason explained in this: https://community.hwbot.org/topic/167622-spi32m-help-needed/ thread. However I couldn't get the benchmark to finish with the tweak, because it always says "Not Convergent in SQR05". I first assumed it demands very high stability, so I lowered the bclk from 195 to 173, and the issue isn't solved, but it finishes without waza at 206 bclk.  

Have anybody experienced this before? If so, what was the solution? 

[Noxinite]

What platform, CPU and memory? Any screenshots?

[Guest Alex Bese]

1 minute ago, Noxinite said:

What platform, CPU and memory? Any screenshots?

LGA1366: X58A-UD5, not good memory kit and an X5550. I could do this on windows 7, but because of the much lower bclk, I can't imagine it's not enough stability for the tweak. 

What screenshots would be necessary to determine the problem? 

[Noxinite]

Erm, not sure about X58. Not a platform I have lots of experience on.

Stability for wazza specifically can be either memory (not likely in this case it seems) or due to some cache or read/write instability - as wazza is copying lots of files in a short spare of time.

[Mr.Scott]

1.58v on air? And turbo'ing to 4.7?

Even on 2 cores I bet you're running hot. Heat creates random instability.

It finished at 206 bclk how many times?

Just once might have been just a fluke.

[Guest Alex Bese]

9 hours ago, Noxinite said:

Erm, not sure about X58. Not a platform I have lots of experience on.

Stability for wazza specifically can be either memory (not likely in this case it seems) or due to some cache or read/write instability - as wazza is copying lots of files in a short spare of time.

And how could I eleminate that cache or read-write instability?

8 hours ago, Mr.Scott said:

1.58v on air? And turbo'ing to 4.7?

Even on 2 cores I bet you're running hot. Heat creates random instability.

It finished at 206 bclk how many times?

Just once might have been just a fluke.

Yeah, if it would be all core load, I wouldn't be able to cool it, but with this on 2 cores, it goes up to like 75 celsius max (even though it's a soldered chip, this one is factory delidded, because it came from a Mac Pro... Maybe that's why I can cool it better)

206 finished a lot of times without waza, yes. 

[Noxinite]

1 hour ago, Alex Bese said:

And how could I eleminate that cache or read-write instability?

Yeah, if it would be all core load, I wouldn't be able to cool it, but with this on 2 cores, it goes up to like 75 celsius max (even though it's a soldered chip, this one is factory delidded, because it came from a Mac Pro... Maybe that's why I can cool it better)

206 finished a lot of times without waza, yes. 

That's the thing, I don't know. I heard on 775 that Wazza put extra strain on the motherboard NB as that handles file transfers or something, but as I haven't used X58 I don't know if that is helpful.

[Guest Alex Bese]

1 hour ago, Noxinite said:

That's the thing, I don't know. I heard on 775 that Wazza put extra strain on the motherboard NB as that handles file transfers or something, but as I haven't used X58 I don't know if that is helpful.

That's unfortunate. I'll try to reinstall the system, maybe that'll solve it. If not, I'm very confused. 

[Guest Alex Bese]

43 minutes ago, Alex Bese said:

That's unfortunate. I'll try to reinstall the system, maybe that'll solve it. If not, I'm very confused. 

Allright, it does the same thing. Even at stock.

Maybe it doesn't like tinyxp? 

[TerraRaptor]

Do you have enough space on disk? Not convergent is common when you run into disk space issue. With waza, writing files on disk, that may be an issue:)

Edited March 6, 2020 by TerraRaptor

[Splave]

Maybe its QPI related, can try lower that or playing with the voltage for it

[Guest Alex Bese]

2 hours ago, TerraRaptor said:

Do you have enough space on disk? Not convergent is common when you run into disk space issue. With waza, writing files on disk, that may be an issue:)

Well, about 109GB is free. I assume it's enough for this. 

1 hour ago, Splave said:

Maybe its QPI related, can try lower that or playing with the voltage for it

QPI was at slow mode, so I can't imagine that's the problem.  

 

Also I tried with another XP install. This is not TinyXP, it's just a normal XP iso that I edited in nLite. On this, I can't get ERAM to work, because even though it creates the Z drive, I can't open it. It says it's not formatted.  

On TinyXP, ERAM worked, but I couldn't get it to work with Memory Source -> External, because it didn't create the Z drive so I used OS: PagedPool instead. Could this be the error's cause? 

Edited March 6, 2020 by Alex Bese

[Splave]

you can try this 

https://www.dropbox.com/s/k33wkuc5atjnz8p/My modPISetup.rar?dl=0

From Fredyama

 

Edited March 6, 2020 by Splave

[TerraRaptor]

47 minutes ago, Alex Bese said:

Well, about 109GB is free. I assume it's enough for this. 

I mean the drive you are running superpi from - if you run it from eram then its size may be an issue. I would run benchmark without eram to isolate software issues from hardware issues.

[Guest Alex Bese]

44 minutes ago, Splave said:

you can try this 

https://www.dropbox.com/s/k33wkuc5atjnz8p/My modPISetup.rar?dl=0

From Fredyama

 

I'll check it out, but I think I got it to work now.

3 hours ago, TerraRaptor said:

Do you have enough space on disk? Not convergent is common when you run into disk space issue. With waza, writing files on disk, that may be an issue:)

This was a really great advice actually, because I gave eram 1gb and now it works, so I'm 100% sure I gave it a really small disk.  

[Guest Alex Bese]

I'm testing different modes and sizes for doing waza on AM3 now, and it's very interesting. 

OCX needs an increased stability, so because of this I have to clock the cpu lower for the same voltage or increase voltage for the same freq and that's equals to lower overall score.

Also after doing the same waza mode multiple times, I experienced that it's not really consistent. For example using only a manual c to eram copy with an 1200mb waza size, there's a 0.390s difference between the scores which isn't bad, but I tried a manual d-c, c-d, d-c waza too, and there the gap is around 2 seconds between the slowest and fastest run. 

Did you guys also experience the same stuff?  

Edited March 23, 2020 by Alex Bese