Do you feel sinful overclocking an ECC RAM ?

[BravoV]

Just bought bunch of DDR4 ECC for SWMBO and also refresh my current desktop, both on Ryzens.


The evil side  in me says ...

C'mon, its Samsung B-Die anyway ! Even they're not cherry picked, your 1st try without even any efforts, without any bump on the voltage at BIOS, its already reaching 2933Mhz (2400Mhz rated), passed Memtest86+ for few iterations. C'mon ... squeeze more !!! Once it passed the memory test for overnight or at least 24 hours, that already exceeded your normal average usage time.


While the angel inside says ...

ECC is made from the 1st place is for a reason, and overclocking the RAM defies it's purpose, just DO NOT do it.


What to do ? What to do ? 


... just pointless rant ... 


PS :
Btw, just fyi, the freeware bootable Memtest86+ thru USB/CD, just recently updated to version 5.31b after a long hiatus years ago -> https://www.memtest.org/

[wraper]

ECC is made from the 1st place is for a reason, and overclocking the RAM defies it's purpose, just DO NOT do it.

Actually overclock and ECC go together very well. If there is little instability, ECC will fix that and you will be able to check if ECC ever intervened.
Overclocking without ECC is a crapshoot, you never know if system is actually stable and if any bit flips ever happened.

[chriva]

ECC memories are awesome to overclock! 
They're are not perfect tho. Push them hard enough or be unlucky and they'll still not detect and fix errors. There's only so much they can detect with one extra bit


Stay away from the buffered ones tho. You're just putting another link in the chain instead of making it stronger.

[wraper]

Analogy:
Crisps are 30 pence, for a normal packet.
You are paying 40 pence, for the special, reduced salt version, for health reasons.
Then adding salt to it, to make it taste nicer.

Pretty crappy (non) analogy. It's like buying a car (RAM) + insurance (ECC). Thinking that insurance becomes redundant if you like to drive fast is quite ridiculous IMHO.

[wraper]

Will ECC, stop an overclocked cpu, from being less stable, but more crashing, and more likely to have (potentially silent) data corruption ?

Hold on. How RAM ECC has anything to do with overclocked CPU stability? 

Will the car insurance, stop you getting accidents/injured/killed/arrested/imprisoned if you drive like an idiot ?

What insurance will do is buy you a new car (correct the data) if you crash. ECC log will show that you run RAM a bit too fast without any BSOD happening or silent data corruption. Without ECC you can only guess what's actually happening. Running memory tests for 24h does not mean that computer is stable. It just means it is not unstable enough to show the issue while testing. My experience with ECC shows that bit flips may happen once in a week or even month with overclocked RAM.

[magic]

Actually overclock and ECC go together very well.

Slight overclocking is how I verify if ECC works

And yes, I have had a problem once with some particular application sometimes producing different results on the same input data, which started after upgrading RAM. The new RAM passed memtest just fine and there were never any other issues with it

[rdl]

Is overclocking actually useful these days? Seems like the gains are pretty small. The last time I bothered to overclock anything was back in the 90s. It was a Celeron 300A that ran at something like 500MHz, which was a pretty significant increase.

[wraper]

Even for Zen 2 RAM speed makes a big difference in gaming. Though there is very little gain going above 3200 MHz with fairly low timings. Also I prefer manually overclocking ECC ram rather than buying "3200 MHz" consumer RAM with 2133 MHz rated chips. In both cases it's overclocked.

[BravoV]

Thanks for the suggestions & replies, appreciate your feedback no matter which side you are. 

To both wrapper & MK14 (even you've deleted all your posts  , which is not needed imo), just to say thanks, and you two are exactly like the voices in my brain. 

Just fyi, the RAM's stock timings, which is set in the SPD is so loose.

[chriva]

Have fun

I miss the days where you could buy workstation processors and overclock the living heck out of them.
Especially Opterons. Holy heck you could push them 

Ran a dual core Opteron 165 (1.8 Ghz stock) at 3.4 and pc3200 ram at some way too high frequency and CL2 back in the day.
I literally spent weeks fine tuning all the timings on that thing. These days I'm just lazy

[BravoV]

These days I'm just lazy

Thats my problem as well.

1st lazy attempt just plain set at CL14@2933Mhz, without any volt tweaking, manual set stock volt @1.2V,  they passed >24 hrs memtest without error at all. 

I'm not new in OC-ing scenes, since at early era where Intel CPU was replaced with NEC V20 to gain more speed, or mobo xtal mod and etc.

I guess I am too old for this. 

[MK14]

MK14 (even you've deleted all your posts  , which is not needed imo), just to say thanks, and you two are exactly like the sounds in my brain. 

A disagreement, which (at least on my part, I think I was having a bad day, as well), was occurring. So, to avoid, even a possibility of distracting the moderators.
I removed myself from the thread (deleted all my posts), and it appears, the other chap (which wasn't Wraper), also removed the post(s), which started the situation.

In all fairness, I was at least partly (or more, opinion dependent), to blame.

Also, I'd MISUNDERSTOOD your opening post.  I thought (from my too quick skimming of it), you meant overclocking, as in CPU overclocking. Which caused lots of the exchanges of posts, between myself and Wraper.
Which was my fault for not properly reading it in the first place.

tl;dr
My correction(s), because I had misunderstood the original post, seemed to annoy someone else, in the thread. Who (aggressively) complained about it. They wanted me to stop doing multiple posts, and just edit the original post(s).
So, for peace, quiet and no-moderator involvement, I decided to exit the thread.

Back on topic:
I'm NOT a fan of overclocking. Either the cpu, graphics cards or memory. In the same way, I'm not a fan of exceeding 1 amp, for a 1 amp rated transistor, whose datasheet says, recommended maximum 1 amp.
Especially, if the computer is for serious/important uses.

But, if the computer is only used for gaming, and the game save files are NOT vitally important to you. Then that can be a different situation.

Basically my opinion, is if the reason you bought the ECC ram, was for stability/data-integrity, then overclocking it is counter-productive.
E.g. Adding salt, to the reduced/zero salt, crisps you bought, for health reasons.

But, if you got the ECC ram, purely for an overclocking machine (e.g. gaming only), and the ECC is just being used as an aid, to help you successfully overclock the ram (and/or rest of the computer). Then, that is also a different matter.

[BravoV]

Query of memory status in Windows 10 I got this.




Photo of the physical RAM sticks.



The "memoryerrorcorrection" type = 6 means it is "(0x6) Multi-bit ECC" according to Microsoft reference.

While at the internet, most Ryzens with ECC memory reported has the DataWidth=64 and TotalWidth=72 queried through Windows WMIC command, while my TotalWidth=128, what is this mean ? 


Also about the ECC, that the single-bit type supposed to be capable of correcting single bit flip, how about multi-bit type ?


Fyi, mobo is X370 Taichi (officially it supports ECC memory) and a 1700.

[wraper]

Query of memory status in Windows 10 I got this.

That's what you should expect IME. I got the same on 2 motherboards and was getting "WHEA corrected hardware error" events when pushing RAM clocks.

Also about the ECC, that the single-bit type supposed to be capable of correcting single bit flip, how about multi-bit type ?

Single bit ECC cannot correct anything. It can only detect single bit error. Multi-bit ECC can correct single bit errors and detect multiple bit errors.

[BravoV]

Query of memory status in Windows 10 I got this.

That's what you should expect IME. I got the same on 2 motherboards and was getting "WHEA corrected hardware error" events when pushing RAM clocks.

Noted and thanks. 

Any tips how to trigger this intentionally and "quite" consistently if possible, while the OS still capable of booting ?

As I'm aware Ryzen desktop is not capable of error injection in the ECC memory, right ?

Also about the ECC, that the single-bit type supposed to be capable of correcting single bit flip, how about multi-bit type ?

Single bit ECC cannot correct anything. It can only detect single bit error. Multi-bit ECC can correct single bit errors and detect multiple bit errors.

Great, good to know that.

Btw, on others with ECC memory installed, but show "DataWidth=64 and TotalWidth=72", does this mean it can only detect bit flip, but no correction capability at all ?

[wraper]

Btw, on others with ECC memory installed, but show "DataWidth=64 and TotalWidth=72", does this mean it can only detect bit flip, but no correction capability at all ?

AFAIK it's just what windows reports. OS does not have anything to do with error correction. If you get corrected error warning, means error was corrected.

Any tips how to trigger this intentionally and "quite" consistently if possible, while the OS still capable of booting ?

IMHO the easiest would be finding max RAM frequency computer boots at and if you don't see any WHEA warnings within a few hours trying lowering timings.

[Bassman59]

The evil side  in me says ...

C'mon, its Samsung B-Die anyway ! Even they're not cherry picked, your 1st try without even any efforts, without any bump on the voltage at BIOS, its already reaching 2933Mhz (2400Mhz rated), passed Memtest86+ for few iterations. C'mon ... squeeze more !!! Once it passed the memory test for overnight or at least 24 hours, that already exceeded your normal average usage time.

The digital logic design engineer in me, for whom static timing analysis is a vital part of the job, says, "don't overclock."

The whole point of speed grading parts is to guarantee that they work under the defined worst-case conditions. That doesn't mean parts won't work when run with faster clock speeds, or reduced (or increased) supply voltages, or above temperature specs. Very often the parts will work. The questions are always, "For how long will this work?" and "What are the symptoms of failure?" and "do you feel lucky?"

[wraper]

The whole point of speed grading parts is to guarantee that they work under the defined worst-case conditions. That doesn't mean parts won't work when run with faster clock speeds, or reduced (or increased) supply voltages, or above temperature specs. Very often the parts will work. The questions are always, "For how long will this work?" and "What are the symptoms of failure?" and "do you feel lucky?"

FWIW ICs with different speed grades have exactly the same silicon die. They might be binned to lower grade because they really were worse or just because manufacturer needed to "produce" a certain quantity of them. As long as you don't over-volt them above spec, you are safe to not fry them. And ICs unlikely to degrade and start glitching. If you run them on borderline of maximum possible, you may have long-term troubles though. Most of higher speed  consumer DDR4 RAM is rated for 1.35V instead of standard 1.2V anyway and use slower speed grade chips than on a sticker.

[magic]

The digital logic design engineer in me, for whom static timing analysis is a vital part of the job, says, "don't overclock."

The OCer in me says that you guys employ wide safety margins to avoid putting as much time and effort into testing your products as we do

Seriously, I run OCed ECC RAM for a few years with zero errors. Yes, I verified that ECC works by OCing it even further and backing off.