Dual channel works when each channel has the same amount of memory installed.
Hybrid dual - different size, but it depends on the implementation of the memory controller. It may not turn on at all.
Practical position of the RAM sticks in specific slots - until you get the inscription "dual" and get stability.

Were it my system I would check the SPD to see exactly which 8GB DDR3 Crucial Ballistix they are, find another two to on such as eBay that are identical and upgrade the system to 32GB. I just looked it would likely cost around GBP 20 for 2 DIMMs.

I wonder what speed/timings the system is currently running at, have you checked? I just looked in a SIV save file from a Gigabyte Z97-HD3 and they were as below, assuming XMP 1.3 is enabled your system should have lower timings.

Thanks so much for the great info shevalier and red-ray! I definitely have some homework to do. Once I get it back together I'll check the timings and play around with the order. That is if the board still works, I'm pretty confident that it does but I have an ASROCK Z97M that I bought hoping to do a small form factor pentium g2358, 750ti build on XP but I can use it in the meantime if need be. I'm just not looking forward to having to chose between my wifi card and my soundblaster 🤣. The Zalman cooler it also makes me a little concerned it seems like there's not much clearance for ram sticks. I was planning on running four gigs on that build but it may be a tight squeeze for 24 or 32, I just worry about the hot heatsink so close to the ram chips.

red-ray wrote on 2025-04-16, 06:37:

assuming XMP 1.3 is enabled your system should have lower timings.

Before the quantum entanglement memory bus was invented 😀, frequencies looked something like this
Per channel
1 single sided stick = base frequency + overclock
1 double sided (or 2 single side) = base freq
1 doubleside + 1 singleside = base - 1 speed bin
2 singleside = base - 1(2) speed bin.

You shouldn't count on low timings and high frequencies with a large number of memory modules
Although sometimes there are successful combinations of processor, motherboard and memory

shevalier wrote on 2025-04-17, 04:32:

You shouldn't count on low timings and high frequencies with a large number of memory modules

It all depends on the DIMMs and in this case he probably should. Op specified me has Crucial Ballistix and I know they will be faster than the ones in the system I posted the timings for. I don't know exactly which DIMMs he has, but the SPD from the the above timings are as below as are the Crucial Ballistix timings from one of my systems. Note the actual timings on my system are the same as the XMP 1.3 #4 timings with all 8 DIMMs installed.

The general rule of thumb is that you want all the channels to be mirrored, so in Channel 0 insert one 8GB and one 4GB, then in channel 1 insert one 8GB and one 4GB. Each channel will have 12GB and thus be mirrored and as a result the CPU will run in dual channel mode. Another example would be if you had one 4GB stick and two 2GB sticks. Put the 4GB stick in channel 0 and then put both 2GB sticks in channel 1, in this way channel 0 will have 4GB and channel 1 will also have 4GB and it will run in dual channel mode.

The GA-Z97-HD3 has colour coded slots to make it easier, so put the 8GB sticks in the black slots and the 4GB sticks in the grey slots (or vice-versa, it doesn't matter). Just don't put the 8GB sticks in a black and a grey slot or the 4GB sticks in a black and a grey slot. If you do this, it will still work and you'll still see the full 24GB, but the CPU will be running in single channel mode and you'll half the memory bandwidth.

red-ray wrote on 2025-04-17, 06:48:

It all depends on the DIMMs and in this case he probably should.

Nope.
All electronic components have capacitance, and it needs to be recharged.
Memory controller buffers into CPU can easily recharge one memory input at high frequency. Two is more difficult, three is even worse. Four is not a given.

The resulting operating frequency in a real system with several modules depends not only on the quality of the memory itself, but also on the processor/chipset.

red-ray wrote on 2025-04-17, 06:48:

Note the actual timings on my system are the same as the XMP 1.3 #4 timings with all 8 DIMMs installed.

Lucky

shevalier wrote on 2025-04-17, 14:34:

Lucky

Nope, it's working as expected. I suspect all we can do is agree to disagree on your incorrect post and let others decide who to believe.

Here are two more systems working as expected, I probably have hundreds of SIV save files that would also show working as expected. Post some information to prove your statement.

red-ray wrote on 2025-04-17, 15:27:

shevalier wrote on 2025-04-17, 14:34:

Lucky

Nope, it's working as expected. I suspect all we can do is agree to disagree on your incorrect post and let others decide who to believe.

Here are two more systems working as expected, I probably have hundreds of SIV save files that would also show working as expected. Post some information to prove your statement.

Sorry, I don't have any other electricity for you. 🙁
Quantum entangled data buses have not yet been invented.
But for some reason, fully buffered memory modules have been invented.

red-ray wrote on 2025-04-17, 15:27:

working as expected.

Sorry, I didn't look at your screenshots.
You have very modest "expectations" which confirm what I said.
PC4-21300(1333) is "speed bin -2", i.e. PC4-25600 (1600) -> PC4-23466 (1466)
PC3-1866 is "speed bin -2", i.e. PC3-2400 -> PC3-2133
Ivy Bridge can handle DDR3-2400.

shevalier wrote on 2025-04-18, 03:57:

Ivy Bridge can handle DDR3-2400.

Wrong again, the HP DL580 Gen8 speed is low as there are 3 DIMMs on each of the 32 channels, were 64 rather than 96 installed they would run faster.

AFAIK there are no Ivy Bridge desktop systems that support 3 DIMMs per channel.

If the 8's are 1600MHz and the 4's are 1333, I'd just run the pair of 8's for 1600
if the 8's are the slower ones, then I'd take the hit down to 1333, and definitely same order in each bank

red-ray wrote on 2025-04-18, 09:39:

speed is low as there are 3 DIMMs on each of the 32 channels, were 64 rather than 96 installed they would run faster.

Thanks again for confirming my point.
More ranks per channel - lower speed.
Manual completely agrees with you.

Matth79 wrote on 2025-04-18, 13:36:

4's are 1333

DDR3 is very overclocking friendly.
As far as I remember, back then "RAM nanometers" overtook memory controller frequencies.
A good 1333 MHz DDR3 often runs at 1866 and very often at 1600. Even without increasing the voltage.
Feel free to try 1600 MHz

red-ray wrote on 2025-04-18, 09:39:

AFAIK there are no Ivy Bridge desktop systems that support 3 DIMMs per channel.

shevalier wrote on 2025-04-18, 13:45:

Thanks again for confirming my point.
More ranks per channel - lower speed.

Op's desktop system is specified for 2 DIMMs per channel and he will only have 2 DIMMs per channel so this should not happen for his system.

red-ray wrote on 2025-04-18, 14:11:

Op's desktop system is specified for 2 DIMMs per channel and he will only have 2 DIMMs per channel so this should not happen for his system.

in manual we trust 😀

shevalier wrote on 2025-04-18, 14:28:

in manual we trust 😀

I don't trust anyone who posts such an image without a link to where it came from.

It is unclear which manual this is within, given it's talking about 3 DIMMs per channel clearly it's not op's GA-Z97-HD3 so is irrelevant. This thread is meant to be helping turbohydramatic, it's not a general discussion about DIMMs in servers!

Update: I don't wish to mess this thread up any more so I will not respond to posts about my HP DL590 Gen8 as they will not help turbohydramatic

red-ray wrote on 2025-04-18, 14:46:

It is unclear which manual this is within,

Dude, I thought you read your server manual.....

red-ray wrote on 2025-04-18, 09:39:

HP DL580 Gen8

HP ProLiant DL580 Gen8 Server
User Guide
page 47.

https://support.hpe.com/hpesc/public/docDispl … docId=c04086387