XMP & how trustable it is (Linus video comments)

[whismerhill]

Hello,
Recently I discovered linus tech tips. I must say I'm really a fan of how you explain some stuff. I'm kind of a techie myself, been fixing computers for years...
 

Anyways, what I wanted to ask is an explanation or a more thorough video on this Linus random comment @14:04 in the video:

 

Apparently an ASUS rep said to Linus "it just doesn't work"

it made me ponder if we could really trust XMP, if the memory modules were tested enough ? where is the issue lies exactly ? or "issues" if there's several sources.
is it the different CPU integrated memory controllers ? is it the overclocking/overvolting of those due to XMP ?

So I made some research about it, and found a lot of people with problems on forums with XMP however ....
As we all know, 3% of people with problems can make a lot of noise whereas people satisfied generally say nothing. It's also an overclocking feature some people may be unable to understand that & not set it up correctly ?

So it could in my opinion go up to 6 or 10% people having issues with XMP easily, but that would NOT mean XMP is not trustable ?

My point here is :  that I can't find any thorough testing of XMP and the stability of this feature

I guess it all boils down to : if I were to advice a new computer for a gamer with lots of budget, should I advise to setup XMP, or use default "safe" values instead.
or is manually setting it up to stabilize a computer 100% the only way ....
Alternatively what makes XMP not being trusted  ?

sitenote : I also watched this video about RAM speeds :

I knew some of this already, but it doesn't really answer

Thanks for reading.

[TempestCatto]

I guess it varies by the RAM manufacture, the motherboard, the BIOS version, and the CPU for starters. I have XMP enabled on my system (specs on my profile) and the RAM is rated for 1866mhz and the XMP profile has it set to 2133mhz. I assume it also adjust things like clock speed, voltage, latency, and other things I don't have any knowledge of. For me, it's worked flawlessly since day one. I have had zero issues with my RAM and it's XMP profile. Some RAM kits paired with some CPU's take manual tweaking just to get it to rated speeds. Because of how "interchangeable" most pc parts are, this doesn't really take me by surprise. Somethings need more attention on some configs than others, just the way of finicky pc components I guess. But as I said, I haven't had any issues or complaints with my XMP enabled RAM. So as they say, your mileage may vary.

[Alex Atkin UK]

I find it odd ASUS would say such a thing as my board is ASUS and it works absolutely fine.  Almost all guides online tell you to use it, only really aggressive overclockers are going to bother manually tweaking timings.

[Cryptonite]

53 minutes ago, whismerhill said:

 

Hey man, not sure how correct I am but the way I've always seen XMP was like the RAM manufacturer would test timings and speeds on the ram, find a nice stable speed with good timings for each chip and mark the xmp profile.

 

like a saved state for the ram essentially. It's probably more like they apply an xmp profile and run it through automated tests in the factory and which tests they pass and which xmp profile they hit stability at would be the different speeds we get? I might be totally off though. but XMP seems to me like a guaranteed state for specific chips to overclock to.

 

I mean buying a 2400 MHz chip doesn't mean it can't run at a higher speed but it might have not hit the timings the company want the higher speed to run at so it got labled at 2400 MHz / it only passed a certain speed on certain memory controllers but not all of the simulated ones, so a higher speed can't be guaranteed.

 

That's what I would think goes on. How else would they be certain that a chip can handle whatever it is they put the XMP profile to.

 

 

Side note: I recently moved over to a AMD Ryzen 5 1600 (using the Asrock Taichi X470) because my NVME drive had a lot of trouble on my Asus Deluxe X99 board (paired with 5960X) . and it's the first time I'm able to get the xmp profile to run stable without any issues. I could get the deluxe board to run XMP on their latest bios but then NVME would not work at all. I think it's an Asus bios issue for most of their xmp issues. certain timings still picking up as auto or smth else than the xmp profile is supposed to override them to.

[whismerhill]

yeah ok, thank you all for your replies
@Alex Atkin UK remember ASUS rep <> ASUS official statement

 

In the video Ivan also says : "the overwhelming majority of the time" it does not work
which would mean something like 55 to 65% where it doesn't work if I want to put it in numbers ?
that sounds crazy high.

and NOBODY speaks about it apart from a high number of people accross forums which basically means very little in the way of statistics ???
just like if you search for hard drive issues, you'll find lots of answers, but the stats for hard drive failures are more or less constant (apart from a few models here & there)
sitting at around 1% to 10%  with the average being closer to 3% AFAIK

keep the comments flowing, I'd like to hear other's opinion too. Thank you very much
 

[Alex Atkin UK]

Yeah that sounds very unlikely.  If the majority of the time it didn't work it would be widely reported on all the sites that currently say "simply enable XMP" in their build guides.

[whismerhill]

now maybe what they meant is actually :
the overwhelming majority of the time it doesn't work IF you're also overclocking your CPU???

a clarification would be great if nothing else

[Gullerback]

Personally, I'm running 5ghz (8700k) with XMP (3466) on and have zero issues

[aezakmi]

for me XMP never worked, always had unstability problems (name it BSOD, hangs, crashes, etc) so I prefer to tune memory frequency and latencies manually.

After building, I dedicate enough time to configure all of the settings in the BIOS setup even at stock, can't trust these "Auto" settings too much... my CPU was overvolted to 1.43V at stock frequency, what a joke. 

[xg32]

On 1/19/2019 at 6:41 AM, whismerhill said:

.

i find gskill ram to be "close enough" and probably the top of the pile, the rest is matching a set of ram thats on the qvl (i used the z370 list for my z390), it's more important for ryzen.

 

Now where i believe it doesn't work, is when motherboard manufacturers claim they can run 4266 or whatever on their mid range board, buildzoid did a video on that, those boards do not work with high speed ram, nor would i try to make it work. For high end boards i still don't expect the ram to work just by switching to xmp without manual tuning, vccio/vccsa/imc voltages, secondary timings etc.

[whismerhill]

21 hours ago, xg32 said:

i find gskill ram to be "close enough" and probably the top of the pile, the rest is matching a set of ram thats on the qvl (i used the z370 list for my z390), it's more important for ryzen.

 

Now where i believe it doesn't work, is when motherboard manufacturers claim they can run 4266 or whatever on their mid range board, buildzoid did a video on that, those boards does not work with high speed ram, nor would i try to make it work. For high end boards i still don't expect the ram to work just by switching to xmp without manual tuning, vccio/vccsa/imc voltages, secondary timings etc.

that looks interesting, if you can find back the link, please do share. thank you very much
I found his channel "actually hardcore overclocking" but not what you're talking about

[xg32]

18 minutes ago, whismerhill said:

that looks interesting, if you can find back the link, please do share. thank you very much
I found his channel "actually hardcore overclocking" but not what you're talking about

it was his z390 review, he was talking about how the lower end gigabyte boards just doesn't clock higher than 4000 (they are known for problems with memory compatibility issues that got better with this gen) thats one major distinction with the elite and the boards above it.

 

Also i've done about 4 builds on zen+, 3 of them were mid range boards, and they hit a memory wall rather quick, while the x470 taichi just clocked to 4000 no problems, i'd say it's way more sensitive for amd compared to intel.