Asus X99-E WS USB3.1 with 128 GB

M

mikeblas

[H]ard|DCer of the Month - May 2006
Joined
Jun 26, 2004
Messages
12,777
I've done a build around an Asus X99-E WS USB/3.1 motherboard with 128 gigs of memory. I've got an Intel i7-5960x processor and eight Crucial CT16G4RFD4213 parts. I've got an EVGA GTX-970 card, plus an Areca ARC-1882 RAID controller.

I'm not over clocking, and everything in the BIOS is at the default setting (as far as I can tell.)

With 128 gigs installed, the machine isn't stable. It might run Windows for a few minutes, or it might hang while the RAID card is POSTing. If I remove two sticks, dropping to 96 gigs, I'm fine and the machine is very stable.

The manual says the motherboard can run with 128 gigs. Should I have bought different memory?
 
According to intel even the 5960X only supports 64GB of memory. Only Xeons support more.
 
Yeah, I ws confused -- I was also looking at one of the Z10 boards for my build, and I might have conflated the specs. Thing is, it works fine with 96 gigs; I've been beating it up pretty good.
 
Not sure what your needs are but with that amount of ram you should be looking at server grade components.
 
Supported or not, sounds like some of that memory is bad, run a stick at a time until you find the culprit.
 
You may need to raise the ring and SA/IO voltages because mem controller has to work a bit harder.
Maybe even slop up the timings a notch or bump the DRAM voltage a little.
 
On a related note, I would recommend using ECC DRAM on the X-99 boards. That's what I've put in my ASRock X-99WS. That would help with identifying and protecting against any failures on the DIMMs. The performance and cost overhead seems negligible.

Still, I'm looking for some definitive information on the advisable memory configurations for the Quad Channel arrangement. From reading other threads it seems multi-channel memory controllers will run with sub optimal arrangements (perhaps why 96Gb works in this case even when it is not expected to) ... I will ask my q on another thread.

Personally I wouldn't run the 96 in there if the CPU says it only supports 64Gb.
 
It's the NewEgg page for the product that led me to believe it supported 128GB.

and the product manual says this:

8 x DIMM, Max. 128GB, DDR4 3200(OC) / 3000 (OC) / 2800 (OC) / 2666 (OC) / 2400 (OC) / 2133 MHz Non-ECC, Un-buffered Memory
(when installing Intel Socket 2011-v3 Core TM i7 Processors)
Click to expand...
 
Last edited:
Is your RAM on the HCL for the board? Is there RAM on their HCL that's tested in 128GB capacity?
 
CMG said:
On a related note, I would recommend using ECC DRAM on the X-99 boards. That's what I've put in my ASRock X-99WS. That would help with identifying and protecting against any failures on the DIMMs. The performance and cost overhead seems negligible.
Click to expand...
Why do you make such a recommendation? If an ECC part has a correctable error, isn't it transaparent? When it has an error that's not correctable, it causes an MCE just like regular parity memory. How is that fault mapped back to a physical part? (Or did you mean something else by "locating"?)

I normally do run ECC memory, but didn't for this build. At least, not yet -- maybe I'll return the memory I bought . Workstation-class boards are sometimes nerfed server-class boards, or sometimes upscale desktop boards. In this case, I think I have the latter, and I'm starting to regret not getting one of the dual-socket boards I was considering.
 
0ptional said:
Is your RAM on the HCL for the board? Is there RAM on their HCL that's tested in 128GB capacity?
Click to expand...

The sticks I have are not on the HCL. There are three 16 GB parts on the HCL, but they're ECC/RDIMMs. (So that would be one rational reason to use ECC.) The configuration isn't specified; that is, the HCL lists the part, but doesn't specify the total memory configuration tested.

The manual implies that the I7 part doesn't support ECC and that the Xeon part is required to use ECC. This is substantiated by the Intel ARK page linked previously.
 
It's quite weird that your chip posts with both ECC and greater than 64GB then...
 
I know, right? I'm bummed I don't have 128 GB of memory, but it seems like I'm lucky anything works. I haven't screwed up a build this bad in ... well, never, I don't think.
 
something something unicorn chip*

Sounds like you just need to but a Xeon and get what you really want. Probably capable of doing a trade in FS/FT.
 
mikeblas said:
I know, right? I'm bummed I don't have 128 GB of memory, but it seems like I'm lucky anything works. I haven't screwed up a build this bad in ... well, never, I don't think.
Click to expand...

You should be able to swap in a Xeon and be good to go. I'd bet you'll end up with 256GB support eventually as well.
 
mikeblas said:
Why do you make such a recommendation? If an ECC part has a correctable error, isn't it transaparent? When it has an error that's not correctable, it causes an MCE just like regular parity memory. How is that fault mapped back to a physical part? (Or did you mean something else by "locating"?)

I normally do run ECC memory, but didn't for this build. At least, not yet -- maybe I'll return the memory I bought . Workstation-class boards are sometimes nerfed server-class boards, or sometimes upscale desktop boards. In this case, I think I have the latter, and I'm starting to regret not getting one of the dual-socket boards I was considering.
Click to expand...
OK, I guess only Xeon CPUs support ECC. If a parity failure is detected it will fail on POST and show that the failed DIMM needs replacing,
 
CMG said:
OK, I guess only Xeon CPUs support ECC. If a parity failure is detected it will fail on POST and show that the failed DIMM needs replacing,
Click to expand...

What? POST on the next boot? That doesn't make any sense. A parity error can happen any time. If it's just a parity error (one bit), then ECC should able to correct it. If it can't, then there's an ECC failure and an MCE is raised immediately -- otherwise, the machine would continue running with bad data floating around.
 
mikeblas said:
Yeah, I ws confused -- I was also looking at one of the Z10 boards for my build, and I might have conflated the specs. Thing is, it works fine with 96 gigs; I've been beating it up pretty good.
Click to expand...
You'll have to check the motherboard manual to make sure you are plugging in the dimms to the right slots based on how many dimms you have, what channel config your using and if they are dual or quad ranked dimms (it matters when you're using that much memory).
 
bigdogchris said:
You'll have to check the motherboard manual to make sure you are plugging in the dimms to the right slots based on how many dimms you have, what channel config your using and if they are dual or quad ranked dimms (it matters when you're using that much memory).
Click to expand...

The 32GB sticks I've seen are quad rank, the 16GB in use here are only dual, with quad rank your stuck with 2 dimms per channel but X99 only has 2 dimms per channel anyway. I think the main issue here is the 5960X only supports ECC ram, not ECC Registered.
 
Have you tried with slightly increased voltage yet? And, have you checked the socket for bent pins?
 
Last edited:
It wasn't a socket problem. I used 6 sticks in any config and got 96 GB. With 128 GB, I'd get errors and go nuts. I didn't try fooling with voltage. I actually returned the extra two sticks. I guess I'm lucky it's stable, but I had to do this build under duress and didn't have time to vet my shopping cart before I had to have the rig rebuilt and back online.
 
You must log in or register to reply here.
Back
Top