mersenneforum.org  

Go Back   mersenneforum.org > Extra Stuff > Blogorrhea > jvang

Reply
 
Thread Tools
Old 2020-10-03, 22:08   #1
jvang
veganjoy
 
jvang's Avatar
 
"Joey"
Nov 2015
Middle of Nowhere,AR

1101100112 Posts
Default General Hardware Discussion

Basically I'm just gonna post some sorta hardware stuff here every day or so to speculate/converse/whatever upon. Might be something widely discussed like all the upcoming launches, might be some weird thing I'm messing with. First off... I'm messing with some stuff!

I got a great deal on a 3900x a few weeks ago, planned on testing it compared to my 3600 and messing with it a bit, then selling it. But it's still sitting in the box cause I haven't been bored enough to take my d15s off lol. So I plan on doing that in the next day or so. I've got a few questions:

A) I'd like to see how high the FCLK will go on this chip. IIRC anything past 1600 is technically an overclock, but 1800 is common and that's what I've got my 3600 sitting at. How would I verify the stability of a higher FCLK? Should I test it with my MCLK set at 1:1 or decouple it and run at JEDEC or something?

B) Overclocking RAM is nice and all that, basically I've just plugged my dimms into 1usmus's calculator and copied everything into BIOS. For now I've got 2x16gb B-die running at 3600c14 @1.45v. Degradation is a given with any usage at all but how high would be a safe voltage to use 24/7? If I can get 3733mhz or 3800mhz on the 3900x I'd probably have to give the memory a bit more voltage to keep up with the same timings, which would put me near 1.5v which might be too high for "safe" constant use. Not sure there. They're G.Skill Trident Z RGB sticks, mobo is an Asus X570-E.

C) Anyone messed with 1usmus's new automatic overclocking thing that was recently released? Would be interesting to see some results for that software compared to stock, PBO, and manual OCs.
jvang is offline   Reply With Quote
Old 2020-10-04, 12:25   #2
Nick
 
Nick's Avatar
 
Dec 2012
The Netherlands

52·61 Posts
Default

I haven't played with hardware directly for a few years, so I won't try to answer those questions directly - I hope someone else will!

But it may be interesting to think about how you could start testing and measuring hardware-level signals yourself.
What sort of kit would you need to build up for that?
For example, are there accessible JTAG interfaces or something similar?

Last fiddled with by Nick on 2020-10-04 at 12:53
Nick is offline   Reply With Quote
Old 2020-10-04, 17:11   #3
storm5510
Random Account
 
storm5510's Avatar
 
Aug 2009
U.S.A.

6A016 Posts
Default

I would not recommend anyone overclock anything. I do not do it myself...

storm5510 is offline   Reply With Quote
Old 2020-10-04, 17:23   #4
paulunderwood
 
paulunderwood's Avatar
 
Sep 2002
Database er0rr

5×701 Posts
Default

Quote:
Originally Posted by storm5510 View Post
I would not recommend anyone overclock anything. I do not do it myself...

When you see iteration times of gpuOwl on a Radeon VII drop dramatically with memory overclocking without result degradation, you have to ask yourself "why not?".
paulunderwood is offline   Reply With Quote
Old 2020-10-04, 22:41   #5
storm5510
Random Account
 
storm5510's Avatar
 
Aug 2009
U.S.A.

25×53 Posts
Default

Quote:
Originally Posted by paulunderwood View Post
When you see iteration times of gpuOwl on a Radeon VII drop dramatically with memory overclocking without result degradation, you have to ask yourself "why not?".
I do not know from Radeon VII, so I will simply take your word for it. IIRC, these are better for PRP's and the like than Nvidia. Each has their own area of usability in this process.
storm5510 is offline   Reply With Quote
Old 2020-10-04, 23:23   #6
jvang
veganjoy
 
jvang's Avatar
 
"Joey"
Nov 2015
Middle of Nowhere,AR

1B316 Posts
Default

Quote:
Originally Posted by Nick View Post
But it may be interesting to think about how you could start testing and measuring hardware-level signals yourself.
What sort of kit would you need to build up for that?
For example, are there accessible JTAG interfaces or something similar?
oh jeez, some googling has confused me more than i expected. what kinda uses does this JTAG stuff have? this article i found has a lot of information that is basically incomprehensible to me lol https://www.xjtag.com/about-jtag/what-is-jtag/


Quote:
Originally Posted by storm5510 View Post
I would not recommend anyone overclock anything. I do not do it myself...

aside from scenarios where absolute stability is a necessity, i dont see why not if youve got the time. my current ram overclock alone nearly doubles its bandwidth for PRPs with my 3600 compared to some 3000c15 sticks i had lying around, the extra performance for free is nice
jvang is offline   Reply With Quote
Old 2020-10-05, 01:49   #7
retina
Undefined
 
retina's Avatar
 
"The unspeakable one"
Jun 2006
My evil lair

134108 Posts
Default

Quote:
Originally Posted by jvang View Post
... extra performance for free is nice ...
It sure is. But it isn't really for free, right? Extra power usage. Extra wear and tear. Extra possibility of early failure. More retries when bad data is returned. etc.

There is always a trade-off. You just have to pick the trade-off point you are comfortable with.
retina is offline   Reply With Quote
Old 2020-10-05, 02:39   #8
jvang
veganjoy
 
jvang's Avatar
 
"Joey"
Nov 2015
Middle of Nowhere,AR

3·5·29 Posts
Default

Quote:
Originally Posted by retina View Post
It sure is. But it isn't really for free, right? Extra power usage. Extra wear and tear. Extra possibility of early failure. More retries when bad data is returned. etc.

There is always a trade-off. You just have to pick the trade-off point you are comfortable with.
barring an undervolt + overclock, sure. hence my asking about safe voltages for b die since my trade-off point is longevity, and id prefer my stuff to not degrade significantly within a year or so. i'd figure theres still a good bit of headroom on many components to use the same or less power for the same or more performance, while being as stable as any other system. though i'd also figure that that's the ideal point at which a product is released to consumers, hence, for example, the new ampere cards having extremely low overclocking headroom (though this time they may have even overshot their auto-boost with all the crashing issues and such).
jvang is offline   Reply With Quote
Old 2020-10-05, 02:54   #9
retina
Undefined
 
retina's Avatar
 
"The unspeakable one"
Jun 2006
My evil lair

23×11×67 Posts
Default

Safe voltages depends upon the die you have, and the process it was made with. No one can give you a definitive answer because no one knows about your particular die and every one of those billions of transistors in it. The process it is made from will suggest a certain "safe" voltage, but there will always be some transistors at the extremes of the range that will be the weakest, and those will fail first.

If you want utmost longevity then undervolt. If you want utmost performance then overvolt like crazy until it melts and burns. Or pick your point of risk somewhere within that range.

Last fiddled with by retina on 2020-10-05 at 04:30
retina is offline   Reply With Quote
Old 2020-10-05, 07:57   #10
Nick
 
Nick's Avatar
 
Dec 2012
The Netherlands

52×61 Posts
Default

Quote:
Originally Posted by jvang View Post
oh jeez, some googling has confused me more than i expected. what kinda uses does this JTAG stuff have? this article i found has a lot of information that is basically incomprehensible to me lol https://www.xjtag.com/about-jtag/what-is-jtag/
You're right, it's confusing! That's because what it gets used for has changed so quickly.
This Intel document is a few years old now but still gives a good introduction:
http://www.intel.com/content/dam/www...149x-paper.pdf

I've mainly seen it used for debugging embedded systems
(processors in non-computers - anything from a settop box for satellite TV to a washing machine).
But, to start with, you just need to understand enough to work out whether this is a useful interface for you
to get deeper into the hardware, or whether there is something else more suitable available.
Nick is offline   Reply With Quote
Old 2020-10-05, 08:34   #11
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

8,963 Posts
Default

Almost all devices that include a not-the-lowest-performance processor, say they have at least one 32-bit or 16-bit "brain" (like CPU, MCU, chipset, FPGA, all kind of controllers, etc) have a JTAG or similar (SWD, SWIM), interface. But unless the documentation is provided, nobody know how to use it, except the designers/manufacturers.

JTAG per se, is a very simple thing, just a VERY-VERY big serial shift register (google: shift register). Imagine all bits inside of your CPU, all registers, all internal settings, input and output pin values, etc., placed in a very-very long string/queue. You can clock them out bit by bit and read them, or you can clock them in bit by bit and set them into the CPU, just by using three lines, one "data input", one "data output" and one "clock" line. Every pulse you give on the clock line will shift in a bit on DI, and shift out a bit on DO. You can shift in a number of bits till they "align" with internal registers, then set them all in the same time. Or read all internal registers at once, them shift them out one by one and see what they contain. That's all.

But the alignment, you have no way to guess, until the manufacturer provides you with the datasheet.

We use JTAG interfaces in our daily work, and could not do half of out job without them. They are used to program and debug almost all the hardware we make. Very few we make public.

We don't think jvang needs JTAG access to his toys, unless he kills the BIOS and needs flash rewrite by external artillery (programmer/debugger, etc). The BIOS of the mobo itself, and high-level utilities (i.e. run from OS) to play with that BIOS, should provide all the tools needed to do what he wants.

Last fiddled with by LaurV on 2020-10-05 at 08:41
LaurV is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Mandela effect, general discussion and how would you study it scientifically? jasong Lounge 12 2018-04-15 05:13
Is this solvable in general. jwaltos Homework Help 34 2017-09-21 14:12
S9 and general sieving discussion Lennart Conjectures 'R Us 31 2014-09-14 15:14
General Unix job-scheduling discussion pinhodecarlos Software 19 2014-04-17 15:04
General Status??? R.D. Silverman NFSNET Discussion 4 2007-07-19 18:43

All times are UTC. The time now is 00:27.

Sun Dec 6 00:27:28 UTC 2020 up 2 days, 20:38, 0 users, load averages: 1.44, 1.70, 1.90

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.