AMDFX-8350+ASUSM5A97R2 MemoryCompatibility(SolvedNotClosed)

Message

nlsa8z6zoz7lyih3ap · Post by **nlsa8z6zoz7lyih3ap** » Wed Apr 03, 2013 5:33 pm

I have recently built A PC based on

(1) AMD FX-8350 cpu
(2) ASUS M5A97 R2.0 mob
(3) CORSAIR Vengeance 32GB (4 x 8GB) 240-Pin DDR3 SDRAM DDR3 1866 Desktop Memory Model CMZ32GX3M4X1866C10
(memory)

To the best of my knowledge, after considerable reading, these should be compatible.

However the memory fails memtest after a few hours and the PC always crashes after a few hours of compiling.
Heat is not the issue as I have set the make_opt j=4 and this keeps the temp below 50degC.

Does anyone have any experience with this cpu and this mob? If so can they tell me what memory has worked successfully for them?

A curious point: The crashes stop if I disable cores 3-8 in the bios. I can not imagine what this has to do with
the problem . It is a clue, but obviously not a solution.

DONAHUE · Post by **DONAHUE** » Wed Apr 03, 2013 5:46 pm

cooling method is??
overclocking??
heating of the cpu and heating of memory each/both depend on how many cores are working
is memtest telling you anything consistent about which bank is failing?
"always crashes after a few hours of compiling." what kind of crash? anything on monitor? oops? panic? power goes off? ???

nlsa8z6zoz7lyih3ap · Post by **nlsa8z6zoz7lyih3ap** » Wed Apr 03, 2013 7:25 pm

"always crashes after a few hours of compiling." what kind of crash? anything on monitor? oops? panic? power goes off

Nothing on the monitor. Sometimes the power goes off, but more frequently it just freezes and doesn't respond to network pings.

is memtest telling you anything consistent about which bank is failin

I'm afraid that I can't answer this as I have already RMA'd the memory.

The cooling was the stock cooler, but I did keep the temp below 50degC max while compiling by setting j<=4.

I am not currently interested in trouble shooting the situation but am interested in finding out the following:

(1) Does anyone have a working machine with this cpu,mob and memory?
(2) If anyone has a working machine with this cpu and mob, what memory are they using?

I am really preparing myself for the eventuality that the replacement memory doesn't work either (Pessimistic, aren't I?)
and wondering what should I then try next?

Many thanks for you reply and suggestions.

nlsa8z6zoz7lyih3ap · Post by **nlsa8z6zoz7lyih3ap** » Wed Apr 03, 2013 7:49 pm

PS

is memtest telling you anything consistent about which bank is failing

While I didn't try memtest on the separate memory modules, I did try them separately while compiling.

Each of the modules by itself caused a crash.

depontius · Post by **depontius** » Wed Apr 03, 2013 8:01 pm

I'll repeat DONOHUE's question... Are you overclocking anything? I don't know how to equate speeds between Crucial, Newegg, etc.

Newegg indicates that things are being overclocked at 2133, but not 1866 or 1600.
Crucial recommends 1600 and offers 1866, though not as the front-line recommendation.

It would be worth double-checking, perhaps worth an experiment under-clocking, to see if that stabilizes things.

By the way, I've been interested in this board, but am afraid of UEFI. What instructions did you use?

krinn · Post by **krinn** » Wed Apr 03, 2013 8:03 pm

I don't remember any asus motherboard i have brought that didn't include memory module compatibility list in its manual.

nlsa8z6zoz7lyih3ap · Post by **nlsa8z6zoz7lyih3ap** » Wed Apr 03, 2013 8:21 pm

I'll repeat DONOHUE's question... Are you overclocking anything? I don't know how to equate speeds between Crucial, Newegg, etc.

No. I generally used stock settings --which yielded 1333 for memory, even 'tho it is sold as 1866. I also tested it at 1600 and 1866, with the same results.

I did no overclocking on the cpu.

By the way, I've been interested in this board, but am afraid of UEFI. What instructions did you use

The UEF| is merely an option that can be used. I chose not to use it, so from my point of view it was just like standard bios.
My pre-upgrade grub2 still worked just fine with it. I used to always use Gigabyte mobs, but a bit of googling caused me to worry about he "hybrid" UEFI-BIOS on their new ones.

As too the cooling method. I have an order placed for a liquid corsair cooler. I am not interested in overclocking,
but as a retired mathematician, I certainly would like to have 8 cores running on some of my research programmes.
The hardware has advanced so much in the last 20 years, that I am hoping to push the programmes to investigate cases that got bogged down in earlier times.

Thanks for your comments and interest.

nlsa8z6zoz7lyih3ap · Post by **nlsa8z6zoz7lyih3ap** » Wed Apr 03, 2013 8:26 pm

I don't remember any asus motherboard i have brought that didn't include memory module compatibility list in its manual

Mine did too and the memory in question is on its list.
However I am asking because searches with google yield some people who claim that not all memory compatible with an mob
will work well with all cpu's that are compatible with that mob. I know that this sounds strange, but it is for that reason
that I asked if anyone with my cpu and my mob had had success with my memory.

depontius · Post by **depontius** » Wed Apr 03, 2013 10:43 pm

That's a much more sensible question to ask now than it would have, say more than 5 years ago. Beginning with the Opteron and Athlon-64 families AMD moved the memory interface onto the CPU chip. So while the motherboard is not completely divorced from memory performance, the CPU is much more the determinant than in "the old days."

In case you're not familiar with the terms, "in the old days" the CPU talked to the northbridge, and then the northbridge talked to the memory, graphics card, and PCI slots. These days the CPU talks to the northbridge and the memory, and the northbridge talks to the PCI-express bus. (including the graphics card(s)) Conversely, "yesterday" the southbrige talked to the rest of the peripheral chips on the motherboard, these days it pretty much sweeps them all up into one piece of silicon. If this is stuff you already know, I apologize for being pedantic.

nlsa8z6zoz7lyih3ap · Post by **nlsa8z6zoz7lyih3ap** » Wed Apr 03, 2013 11:12 pm

If this is stuff you already know, I apologize for being pedantic.

No apologies required. In fact I enjoyed (and needed) your clear explanation.

However, if the replacement also does not work, I will be faced with the difficult question:
Which is defective, mob, memory or cpu? Or are they all OK by themselves, but just don't fit together in some subtle way.

Thanks again for your explanation.

DONAHUE · Post by **DONAHUE** » Thu Apr 04, 2013 12:01 am

waiting to fail and failing under load screams overheating to me.
while waiting for replacement memory and/or water cooler it might be worthwhile to reseat the stock cooler with a fresh thin coat of arctic silver or shin etsu thermal grease.

nlsa8z6zoz7lyih3ap · Post by **nlsa8z6zoz7lyih3ap** » Thu Apr 04, 2013 12:44 am

waiting to fail and failing under load screams overheating to me

I certainly understand that feeling, but gkrellm (via lm_sensors) never showed temperatures over 50C when compiling, and usually they were in the high 30's or low 40's. I may be missing something, but would the load really be high when "j=4" in /etc/make.conf? (while running emerge -eq @world)
In contrast, when running my research programmes on all 8 cores ,the temp rose to 60C within a minute, at which point I killed the process.'s. This never resulted in any crashes, but the heat generated is the reason that I have ordered the liquid cooler.

while waiting for replacement memory and/or water cooler it might be worthwhile to reseat the stock cooler with a fresh thin coat of arctic silver or shin etsu thermal grease

Good suggestion, but I will have the replacement memory and the water cooler before turning it on again. Until then, I will have no memory at all in the house.

Thanks again for you comments and suggestions.

depontius · Post by **depontius** » Thu Apr 04, 2013 1:00 am

I'm following this with quite a bit of interest, because I'm in silicon design by day, and in about 6 months I'm planning to build a system very similar to what you're describing. I've never had overheating problems since my old K6-3, and for that I wrote a "CPU throttler" that would read the temperature and throttle the CPU if it got too hot. I've never gotten terribly exotic about my cooling, either. I think that these days Linux has built-in thermal regulation - I know that the kernel has had "thermal zones" for quite a while, but don't really know how to use them. I would suggest a little time with google and 'linux thermal zones" would give you some peace of mind. I'm probably going to do the same.

Anon-E-moose · Post by **Anon-E-moose** » Thu Apr 04, 2013 12:02 pm

nlsa8z6zoz7lyih3ap wrote:I have recently built A PC based on

(1) AMD FX-8350 cpu
(2) ASUS M5A97 R2.0 mob
(3) CORSAIR Vengeance 32GB (4 x 8GB) 240-Pin DDR3 SDRAM DDR3 1866 Desktop Memory Model CMZ32GX3M4X1866C10
(memory)

To the best of my knowledge, after considerable reading, these should be compatible.

However the memory fails memtest after a few hours and the PC always crashes after a few hours of compiling.
Heat is not the issue as I have set the make_opt j=4 and this keeps the temp below 50degC.

It could be something as simple as the memory not getting enough voltage.
That's something to check in the bios.
1866 is an overclocked memory, which needs higher voltage.
The packaging or the website should tell you what the "minimum" voltage needed is.
And it may need to be bumped up a little above minimum or recommended.

Good luck.

nlsa8z6zoz7lyih3ap · Post by **nlsa8z6zoz7lyih3ap** » Fri Apr 05, 2013 5:02 am

. I would suggest a little time with google and 'linux thermal zones" would give you some peace of mind.

Thanks for the suggestion I have already learnt lots of interesting stuff that I will use in the future.

It could be something as simple as the memory not getting enough voltage.
That's something to check in the bios.
1866 is an overclocked memory, which needs higher voltage.

This certainly gives me an "aha" moment. As I am not an overclocker, I had overlooked the fact that this is promoted
as an "overclocked" memory and done nothing at all in this regard in the bios settings. The manual shows that the bios does indeed offer such features, including a "memory tuning" feature for overclocked memories which it says takes about 5 minutes to run. I had not not done this. At the moment I have no memory whatsoever, but when I do, in about 2 weeks, I will report back with the results.

Thanks again to everyone for their good comments and suggestions.

mattty · Post by **mattty** » Fri Apr 19, 2013 12:39 pm

nlsa8z6zoz7lyih3ap wrote:I have recently built A PC based on

(1) AMD FX-8350 cpu
(2) ASUS M5A97 R2.0 mob
(3) CORSAIR Vengeance 32GB (4 x 8GB) 240-Pin DDR3 SDRAM DDR3 1866 Desktop Memory Model CMZ32GX3M4X1866C10
(memory)

Hi, I have exactly the same problem and partial solution. I bought the same hardware just only two RAM modules(2x8GB) the CPU and motherboard are the same. I had some talk to the seller about this problem, but they refused it saying that this CPU cannot handle such big RAM modules at the speed. I played with the system for a few days and I discovered partial solution. Setting DRAM voltage to the 1.35V(instead the 1.5V default) removed the memory failures with memtest and made the system stable. The downside of this setting is that it won't let you set RAM to more that 1600Mhz.

Hope this helps.

nlsa8z6zoz7lyih3ap · Post by **nlsa8z6zoz7lyih3ap** » Sat Apr 20, 2013 1:00 am

The replacement to the memory that I RMA'd arrived today.
Results: (1) The first of the 4 chips tested by itself showed errors on test 6 of memtest (the orignal memory failed test 5) but the second did not.
(2) I am now running this system with the second chip alone. The mob only gives me 1333 Mghz with this chip. However my PC still crashes when compiling something like gcc, unless I disable all but the first 2 cores, in bios. However other programmes such as x264 encoding and my own mathematics programmes run just fine on all 8 cores, with the new memory.
(3) Heat : My Corsaire H80 liquid cooler has arrived and is now installed. While running all 8 cores at 100%, it keeps the cpu temp at around 45C with x264 and around 50C with my own Math programmes. I also have a small fan blowing on the NorthBridge, which keeps it cool enough to touch.

Thank you, Matty, for sharing your experiences with us.

Setting DRAM voltage to the 1.35V(instead the 1.5V default) removed the memory failures with memtest and made the system stable.

I was able to do this with only one module installed, but with two, I was unable to boot and had to clear CMOS in bios to get up and running again. However even with only the one module at 1.35V, I was still unable to compile gcc or chromium without disabling all but cores 1 and 2. Increasing memory voltage makes things worse.

So the system works, sort of, but not well enough to run "emerge" with all cores enabled, even when I set j to 2 in /etc/make.conf. In this sense it is worse than my 4 core amd system which it is supposed to replace.

I had some talk to the seller about this problem, but they refused it saying that this CPU cannot handle such big RAM modules at the speed.

Did the seller mean that the 8GB modules are too big for the FX-8350 (I could not get them to run stably at ANY speed, no matter how slow (800mghz for ex))
If so do you know what size of modules it would support. Do you know how I could get information on what memory would run stably with this mob cpu combination.

My current Goal: Get some advice on a 16GB memory set that would run stably with this cpu and mob. Can anyone advise me in this regard. I had thought that choosing memory off of the mob memory compatibility list would be enough, but apparently that is not so when the CPU is the FX-8350. If I can't succeed in this, I will revert to my previous system.

Thanks again to everyone for their help and comments regarding my very frustrating hardware problems.

PS If anyone thinks that the problem is a faulty mob or cpu, I would be interested in hearing that too.

ADDED IN EDIT I no longer have confidence in my power supply. I noticed that I could smell it a bit after a full load test. Further investigation reveals that its fan no longer turns. Before investigating memory any further, I shall replace the power supply. I should be able to do so in less than a week.

Anon-E-moose · Post by **Anon-E-moose** » Sat Apr 20, 2013 9:43 am

The power supply could certainly be your problem.
If the supply wasn't enough for the motherboard including memory, cpu, etc then you might get some of the problems you have experienced.

I was running an MSI board and was having intermittent problems that I thought were due to the age of the board
but I started having similar problems with a new board so swapped out the power supply.
Fixed all the problems, and I've tested out the old MSI board with another power supply and it works fine.

It was supplying power, but evidently not enough in certain circumstances.
The fan dying is certainly a bad sign.

Good luck.

mattty · Post by **mattty** » Sat Apr 20, 2013 2:49 pm

Hi,

Did the seller mean that the 8GB modules are too big for the FX-8350 (I could not get them to run stably at ANY speed, no matter how slow (800mghz for ex))
If so do you know what size of modules it would support. Do you know how I could get information on what memory would run stably with this mob cpu combination.

actually the seller's customer service was quite annoying. They just kept repeating that those AMD CPUs are unreliable and problems are common. They told me to use 2x4G modules at 1333MHz, but it is not acceptable for me. From technical point of view I do not see a reason why "big" 8G modules should be the problem. On the other hand more modules can make the situation worse.

I also tried to find some relevant supported memory list, but there are only the motherboard<->memory compatibility lists, but we would need the CPU<->memory list as the CPU talks directly to the memory.

A few more tips:
You can try setting memory timing by hand (no leaving it on auto) even when the values are the same as auto-detected. You can also check is memory timings match the XMP/SPD profile. You can display them in the BIOS/tools section.

Maybe you already now, but when the system hangs at start with red LED next to the memory lighting you can power it off by long-pressing the power button. At next start it will let you correct the settings. It remembers all the values you entered. I managed to test a lot of configurations without clearing the CMOS.

Kollin · Post by **Kollin** » Sat Apr 20, 2013 6:29 pm

Yes! Asus boards are picky when comes to memory modules!
When i bought my board ~5 years ago (m2n32ws pro) i put 8GB OCZ inside, the result was exactly the same as yours

However the memory fails memtest after a few hours and the PC always crashes after a few hours of compiling.

I spent countless hours fiddling with bios settings and memtest and finally i gave up and bought G.Skill modules.
Since then everything works perfectly and OCZ modules are collecting dust somewhere in the drawer

My advice is: Don't bother, jest get some other modules from different manufacturer

depontius · Post by **depontius** » Mon Apr 22, 2013 12:10 pm

re: "big memory modules"

As previously mentioned, I've been spec'ing a similar system. I tend to like to buy from Crucial. I've been in the memory/silicon/design business for over 30 years, and I like the fact that Crucial is selling their own (Micron) memory, and can control test from wafer to DIMM.

Crucial's web site has a memory configuration tool, where you can punch in the hardware by brand/type/product and they will recommend a memory product.

What was interesting is that for this category of motherboard, "max 32G", when I tried to get ECC DIMMs they would not recommend 32G, but only 24G. I'm presuming that from a sheer numeric point of view, 32G might be right at the top end in terms of bus loading, and when you start adding the extra chips for ECC you go over the limit.

That doesn't necessarily factor into the current discussion, other than that perhaps stuffing 32G onto one of these boards is hovering right at the edge of reliable operation.

In terms of my own plans, I don't need this machine right away, and don't want to spend the money right now. I've been figuring that in the fall the 2nd generation Piledriver/Vishera will be out, and that sounds like a good time to pounce. Steamroller is due next year, but I don't want to wait that long.

nlsa8z6zoz7lyih3ap · Post by **nlsa8z6zoz7lyih3ap** » Thu Apr 25, 2013 11:45 pm

I have had the new PSU since Tuesday and can now report the results of testing since then.

(1) One of the 4 memory chips. run by itself still fails memtest at the default settings whereas the other 3 (by themselves) pass.
Therefore I am now running the PC and testing with the 3 that passed, for a total of 24GB.
In this regard note depontius comment:

What was interesting is that for this category of motherboard, "max 32G", when I tried to get ECC DIMMs they would not recommend 32G, but only 24G. I'm presuming that from a sheer numeric point of view, 32G might be right at the top end in terms of bus loading, and when you start adding the extra chips for ECC you go over the limit.

(2) I currently have this running at 1733mhz by first using the mob',s memory overclock tuner, second choosing 1733 for the DRAM speed, and third setting the memory voltage to 1.4000 volts..
The Dram voltage change was necessary for success, and this was inspired by mattty's comment:

Setting DRAM voltage to the 1.35V(instead the 1.5V default) removed the memory failures with memtest and made the system stable. The downside of this setting is that it won't let you set RAM to more that 1600Mhz.

However I am fortunate in getting the 1733 speed. I haven't tried 1866 yet with the new PSU, but that did pass memtest with the old PSU, so perhaps it will
with the new too. I haven't tried it yet. Of course with the old psu everything crashed under load, as mentioned earlier, so the 1866 was of no help.

(3) The mob defaults this memory do 1333mghz although it is rated by corsaire at 1866.

(4) Regarding this main question of this thread:

....... that not all memory compatible with an mob
will work well with all cpu's that are compatible with that mob?

I now believe that the answer is that the cpu does matter. The mob manual states that

When overclocking, some amd cpu models may not support DDR3 1600 or higher frequency dimms

The manual has other such comments as well. While I don't overclock my cpu, the mob defaults this memory to 1333 and claims that I have overclocked it when I set it to anything higher. All of this makes it very difficult to choose parts and design a system from part specifications. For that reason I am very glad to have depontius's helpful comment

Crucial's web site has a memory configuration tool, where you can punch in the hardware by brand/type/product and they will recommend a memory product

So thanks to the help of everyone who responded in this topic, I now have a fully functioning system that has not crashed in 3 days.

While my hardware problems are solved, I think that many interesting questions remain concerning memory, mob's and cpu's. Any further comments in this regard would be received with great interest. Thus I am going to mark this topic as solved but not closed.

I do still have a question pertaining to my new PC ('tho it is a bit off of the memory topic.)

Question: What is the correct kernel setting for cpu type. ? Opteron/Athlon64/Hammer/K8 comes with the advice Select this for an AMD Opteron or Athlon64 Hammer-family processor. │
│ Enables use of some extended instructions, and passes appropriate │
│ optimization flags to GCC Of course the FX-8350 doesn't look to me like and opteron or Hammer family cpu. It doesn't even support all of the instruction set from my previous 4 core.
For that reason I have chosen Generic-x86-64.. Does anyone know if Opteron/Athlon64/Hammer/K8 would be a reasonable or better choice?

trismo · Post by **trismo** » Fri Apr 26, 2013 1:47 am

CMZ32GX3M4X1866C10(Ver3.23)(XMP)
M5A97_R20-Memory-QVL says 4 DIMM not supportet only with 2

depontius · Post by **depontius** » Fri Apr 26, 2013 1:50 am

nlsa8z6zoz7lyih3ap wrote: Question: What is the correct kernel setting for cpu type. ? Opteron/Athlon64/Hammer/K8 comes with the advice Select this for an AMD Opteron or Athlon64 Hammer-family processor. │
│ Enables use of some extended instructions, and passes appropriate │
│ optimization flags to GCC Of course the FX-8350 doesn't look to me like and opteron or Hammer family cpu. It doesn't even support all of the instruction set from my previous 4 core.
For that reason I have chosen Generic-x86-64.. Does anyone know if Opteron/Athlon64/Hammer/K8 would be a reasonable or better choice?

Sadly, I know you asked for the kernel configuration, not the CFLAGS. I suspect that for now you're probably better off with generic, as you have done. I'm pretty sure that gcc-4.6.x won't support "-march=native" for piledriver, so here are some other references.

From the forums...
http://forums.gentoo.org/viewtopic-t-95 ... art-0.html
From the Gentoo wiki - note that this is for bulldozer, not piledriver, but it's probably closer than anything else...
http://en.gentoo-wiki.com/wiki/Safe_Cfl ... lldozer.29
From elsewhere, incidentally I believe piledriver is bdzv3...
http://www.linuxquestions.org/questions ... 175444876/

nlsa8z6zoz7lyih3ap · Post by **nlsa8z6zoz7lyih3ap** » Fri Apr 26, 2013 2:00 am

I'm pretty sure that gcc-4.6.x won't support "-march=native" for piledriver, so here are some other references.

Thanks.
That is my understanding too. I am using gcc-4.7.2-r1. I have made complete backups of
my installation :
(a) compiled using -mtune=generic, and
(b) compiled using -mtune=native.

Both of these seemed to run stably for the day that I tried them.
I am now recompiling everything using "-march-native". I'll let you know how this works out.

AMDFX-8350+ASUSM5A97R2 MemoryCompatibility(SolvedNotClosed)

AMDFX-8350+ASUSM5A97R2 MemoryCompatibility(SolvedNotClosed)

Re: AMD FX-8350 +ASUS M5A97 R2.0+Memory Compatibility

Re: AMD FX-8350 +ASUS M5A97 R2.0+Memory Compatibility