George's dream build

[Madpoo]

Originally Posted by Prime95

Never tried that. For some reason, I don't think that will provide any benefit over DDR4-2400. I don't remember where I got that impression -- perhaps from the Asrock web site which suggests an 8-ish% bandwidth increase, not a 50% possible increase.

What's the max speed that board supports? If it doesn't support 3200 then installing those modules is just adding a little bit more latency without a speed boost to make up for it.

If it did support speeds up to 3200 then it should be faster even with the tick higher CAS latency. Whether it's enough speed boost for the price difference, that's probably more subjective.

[Mark Rose]

Originally Posted by Madpoo

What's the max speed that board supports? If it doesn't support 3200 then installing those modules is just adding a little bit more latency without a speed boost to make up for it.

If it did support speeds up to 3200 then it should be faster even with the tick higher CAS latency. Whether it's enough speed boost for the price difference, that's probably more subjective.

The board officially supports DDR4-2133. It also has a couple of DDR4-2400 modules listed as supported, which is an overclock. I haven't been able to find the maximum speed it will attempt to use. I would buy 3200 if it will work, since that's a big boost in bandwidth if it works. If it only does 2800 or 2400 there is no sense in spending more money. If 3200 works there is also an argument for getting an i5-6600 instead of an i5-6500.

[endless mike]

Oh George, what have you started?



Boards one, two, and three are in the case. Board four is waiting to go in. Board five, six, and seven should be arriving early next week. All will be running Core i5-6500.

As to the RAM discussion, I've got Crucial DDR4-2400 (this stuff) on the first four boards with BIOS 1.50, and the nonZ-OC does not show up as an option.

I'm going to try the RAM George linked earlier with the next three boards, and maybe try downgrading the BIOS on one of the first four boards to see if that changes anything.

I figure I have the summer to get it all straightened out. Then, once it starts getting cold outside again, run full out.

[Mark Rose]

So according to this news release that speeds up to DDR4-2800 may be possible with the H110m-ITX.

Now to find out which chips are used in Kingston and Samsung memory.

[Fred]

Originally Posted by Mark Rose

So according to this news release that speeds up to DDR4-2800 may be possible with the H110m-ITX.

Now to find out which chips are used in Kingston and Samsung memory.

I had really good luck with the first 4 or 5 boards I put in my build (same mobo as George), using the cheaper memory instead of specifically Kingston and Samsung brand. The OC function always appeared.

Then, out of the blue, for the last few boards I couldn't seem to get a pair that worked. I finally just bit the bullet and bought http://www.newegg.com/Product/Produc...82E16820104532
which worked great for just a couple dollars more (at the time of purchase).

[Mark Rose]

Originally Posted by Fred

I had really good luck with the first 4 or 5 boards I put in my build (same mobo as George), using the cheaper memory instead of specifically Kingston and Samsung brand. The OC function always appeared.

Then, out of the blue, for the last few boards I couldn't seem to get a pair that worked. I finally just bit the bullet and bought http://www.newegg.com/Product/Produc...82E16820104532
which worked great for just a couple dollars more (at the time of purchase).

Yeah, I'm not opposed to spending a few more bucks for the brand name. Apparently the inexpensive G.Skill Ripjaws V Series is based on Samsung modules. I'm thinking if I go ahead with this I'll just buy a bunch of different RAM and see what works and eat the restocking fees.

[Prime95]

Mild weirdness. Check out this double-check result:

http://www.mersenne.org/report_ll/?e...exfactor=1&B1=

36 unreproducible roundoff errors with a good result. What happened was resuming mprime after a reboot would immediately give a roundoff = 0.5 error, resume from last save file, and repeat endlessly.

I rebooted the system and this time mprime started without problem. This is the second or third time (out of roughly 30 reboots and I think only after a power on/off rather than a reset) I've seen this behavior and only on this one CPU. Any ideas what might cause such strange behavior? The only oddity on this CPU is it is undervolted by 60mV whereas all others are undervolted a max of 40mV. The CPU has produced dozens of good double-checks at this voltage.

This isn't a major issue, but it does mean one CPU might stop producing results if there is a power failure while I'm on a vacation.

Does anyone think adding a pause before starting mprime after a reboot would help?? I'm thinking it would not.

[Mark Rose]

Why not increase its voltage to match the others?

[kladner]

Will it pass torture testing at its current settings? (Joke intended, but really, isn't it worth checking?)

EDIT: With all respect, could it be possible that the undervolt is borderline?

[Prime95]

Originally Posted by kladner

Will it pass torture testing at its current settings? (Joke intended, but really, isn't it worth checking?)

Actually, it passes torture testing at -70mV. My rule-of-thumb when overclocking or undervolting is to find the settings at which the torture test passes and then add a little more voltage for good measure.

@Mark: I can increase the voltage. I was merely curious if anyone had a working theory as well as to warn other cluster builder's of a possible weirdness.

[endless mike]

Boards one thru seven installed. One thru four have an OS installed; I haven't thought about trying network booting yet. It might have been cheaper to do that instead of separate drives for each board. I did try USB flash drives at first, but with each shutdown and reboot, I ended up with file system errors.



I think with some cable management, I can get an eighth board in place above the power supply. There might end up being a heat issue though.

All three boards with the RAM George linked to above have the nonZ-OC option, even with the latest BIOS 1.50 installed. The four boards with Crucial RAM that I linked to above don't have that option, no matter which BIOS I tried.