X10QBI and v3/v4 cpus (e.g. supermicro sys-4048b-trft)
- Thread starter gmaxwheel
- Start date
-
- Tags
- 2011-1 quad socket x10qbi
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
Depends on whether the QRx4 are LRDIMMs or RDIMMs.
X10QBi memory boards are 3 SPC (Slots per Channel), so the right part of the table is relevant.
If you use Quad Ranked RDIMMs, you can only use 4 per memory board, while with Quad Ranked LRDIMMs, you can fill the whole memory board (12 pcs.)
Another note: The table in the X10QBi manual is wrong, you should use the table in the 8048B-TRFT manual instead (I learnt that the hard way...)
X10QBi memory boards are 3 SPC (Slots per Channel), so the right part of the table is relevant.
If you use Quad Ranked RDIMMs, you can only use 4 per memory board, while with Quad Ranked LRDIMMs, you can fill the whole memory board (12 pcs.)
Another note: The table in the X10QBi manual is wrong, you should use the table in the 8048B-TRFT manual instead (I learnt that the hard way...)
tks for sharing! I was about to learn in the hard way too!!
Indeed I was a bit confused because 2SPC thing! but it became crystal clear after your mention! never crossed my mind that this was a generic table (2 SPC does not apply to x10qbi)
Yeah just keep in mind that the speeds in the X10QBi table are wrong...
This is the right table:
Note the biggest difference: The X10QBi does not support 1600MHz in 2:1 Performance mode with QRx4 LRDIMMs (like shown in the X10QBi) table. It only supports 1333MHz, like shown in the 8048B-TRFT table I posted above
This is the right table:
Note the biggest difference: The X10QBi does not support 1600MHz in 2:1 Performance mode with QRx4 LRDIMMs (like shown in the X10QBi) table. It only supports 1333MHz, like shown in the 8048B-TRFT table I posted above
This is all a bit confusing to me.
I currently have in the X10QBi 4x Intel Xeon E7-8893 v2 6-Core 3.4GHz CPUs installed.
I have 4 Mem cards installed (X10QBI-MEM1 Rev 1.01) in SMI SLOT P1M2, P2M1, P3M2, and P4M1. I run this slot configuration to have all PCI-e slots available to me for running my Nvidia Quadros.
Each Mem card is populated with two 16GB Dimms (SK Hynix HMT42GR7AFR4C-RD DDR3 PC-14900R 1866MHz Memory) but I am only able to run on 1333MHz in either Lockstep mode. What memory modules would I need to get 1600MHz? Do you have a part number?
Thanks!
Yeah, you need LRDIMMs. Your Hynix Modules are just RDIMMs.
With LRDIMMs (e.g. Samsung M386B4G70DM0-CMA4, I'm also looking for those), you can run the sticks at 1600MHz in 1:1 Lockstep mode, but it is NOT possible to achieve more than 1333MHz in 2:1 Performance mode with DDR3 DIMMs.
Also keep in mind that 1:1 Lockstep mode has *much* worse performance compared to 2:1 Performance mode.
Oh, and another hint: The Supermicro board is a diva. If you have 1:1 Lockstep mode and not populated 32 or 64 or 96 DIMMs (but 31 instead, for example), the RAM will only run at 1333MHz even if it could do 1600MHz (learned that the hard way).
With LRDIMMs (e.g. Samsung M386B4G70DM0-CMA4, I'm also looking for those), you can run the sticks at 1600MHz in 1:1 Lockstep mode, but it is NOT possible to achieve more than 1333MHz in 2:1 Performance mode with DDR3 DIMMs.
Also keep in mind that 1:1 Lockstep mode has *much* worse performance compared to 2:1 Performance mode.
Oh, and another hint: The Supermicro board is a diva. If you have 1:1 Lockstep mode and not populated 32 or 64 or 96 DIMMs (but 31 instead, for example), the RAM will only run at 1333MHz even if it could do 1600MHz (learned that the hard way).
I am running programs which love faster memory but don't care if there is more than 64GB available. Do you think if I go to LR Dimms and run on 1600 that I would have less performance then sticking with 1333?
Yes, Lockstep Mode at 1600MHz has a lower performance than Performance Mode at 1333MHz, see here: https://lenovopress.com/lp0048.pdf
OK. Thanks for the explanation! It is hard to find any good info about this board.
I have another question you may be able to have an answer to. It is regarding lockup on Boot code A9.
The following occurs: When I reset the Bios and remove all graphic cards I can enter the Bios setup and do not get A9 or boot issues.
The BIOS sometimes fails to save the new settings but after two or three times repeating the setup, it finally saves it.
I then shut down and re-install the graphic cards but I have to connect a separate monitor to the VGA of the
Supermicro AOM-X10QBi-A card to see the boot screen. It does not matter if I set the VGA in the Bios to Auto, Offboard, or Onboard. It seems I can't get it to post on anything but the VGA. Ideally I would like to disable the VGA port on the AOM module and boot offboard but I can't get this to work at all. My board came with Bios 3.2a and I wish I could try an older one (the one before the Meltdown microcode update) but I can't find that either. Any ideas? Could this be memory type related?
I have another question you may be able to have an answer to. It is regarding lockup on Boot code A9.
The following occurs: When I reset the Bios and remove all graphic cards I can enter the Bios setup and do not get A9 or boot issues.
The BIOS sometimes fails to save the new settings but after two or three times repeating the setup, it finally saves it.
I then shut down and re-install the graphic cards but I have to connect a separate monitor to the VGA of the
Supermicro AOM-X10QBi-A card to see the boot screen. It does not matter if I set the VGA in the Bios to Auto, Offboard, or Onboard. It seems I can't get it to post on anything but the VGA. Ideally I would like to disable the VGA port on the AOM module and boot offboard but I can't get this to work at all. My board came with Bios 3.2a and I wish I could try an older one (the one before the Meltdown microcode update) but I can't find that either. Any ideas? Could this be memory type related?
Last edited:
Apologies in advance that this isn't a solution, but I would like to comment that the correct behaviour of the machine with video card installed is as follows:
1. BIOS VGA setting changed by user
2. Reboot and post, initialisation using previous settings - it should seem like the change has not registered.
3. However, when supermicro splash screen is reached, auto shut-off occurs.
4. Automatically reboots, BIOS VGA setting is activated.
In summary, two post events to preserve your settings.
This means that if you change from Auto/Onboard to Offboard AND with a GPU installed, first post will be like every other one before - shows up in the AOM-X10QBi-A VGA port. Then second post will be completely blank until it reaches the Supermicro splash screen - this will show up in your GPU. There is nothing you can do about this because the initialisation is done at a different level to the GPU and the AOM-X10QBi-A VGA port is forcibly disabled when offboard graphics are used.
Thanks for your comment on the subject. I think I may have posted my problem incorrectly and I didn't want to get off topic. I would be fine with the behavior you have stated, but the real problem is that I can not enter the BIOS settings after saving the changes, regardless if a Graphics card is installed or not. I have created a new thread which hopefully brings my problem better across.
Hi everyone, maybe you know what PLL (clock generator chip) is used on X10Qbi? It could be fun to overclock FSB slightly by modifying PLL chip registers.
slightly related: if you have V3 (haswell) cpus, you could potentially do the 4-way EFI hack to get full-speed all-core turbo. haven't tried it though but im interested to see whether someone has done that before on the x10qbi.
Does anybody tested config with mixed CPUs? like 8890v3 and 8891v3? does the core count should be the same?
This has since been worked through and this information is no longer accurate. but im keeping this for archive purposes.
[resource] RE: 64GB DDR3 DIMMs and mixing of RAM.
I just upgraded to 24 x 64GB = 1536GB ram. Unfortunately, the process was not smooth.
The official Supermicro compatibility website states that only "Samsung M386B8G70DE0-YH93" is compatible with the X10QBi. There is a current shortage of these RAM sticks.
I bought 24 sticks of this from CloudNinjas on Ebay after explicitly asking for this exact model number but they just arrived today with only 6x M386B8G70DE0-YH93 and the remaining 18 sticks were M386B8G70DE0-YH93Q, which I did not ask for. The Q version seems identical to the non-Q version and there is no official documentation describing how the Q version may be different to the non-Q version, so hopefully the data I provide here will prove useful to someone in the future.
It took me an entire afternoon, most of my brain cells, and really questionable RAM topology to get the system past POST and I'm running jobs in OS right now. It seems stable so far, but I'm not sure of long-term stability.
Current POSTable RAM configuration: Total 4 memboards in the "configuration to utilise all PCI-E slots". All DIMMA and DIMMB slots are populated. For CPU1 and CPU3 memboards, DIMMA and DIMMB all contain YH-93Q. However for CPU2 and CPU4 memboards, DIMMA all contain YH-93 and DIMMB all contain YH-93Q. On all boards, DIMMC and DIMMD are empty.
What I learned:
All for now. I'm contemplating a refund or return, but I hope this information helps someone.
I just upgraded to 24 x 64GB = 1536GB ram. Unfortunately, the process was not smooth.
The official Supermicro compatibility website states that only "Samsung M386B8G70DE0-YH93" is compatible with the X10QBi. There is a current shortage of these RAM sticks.
I bought 24 sticks of this from CloudNinjas on Ebay after explicitly asking for this exact model number but they just arrived today with only 6x M386B8G70DE0-YH93 and the remaining 18 sticks were M386B8G70DE0-YH93Q, which I did not ask for. The Q version seems identical to the non-Q version and there is no official documentation describing how the Q version may be different to the non-Q version, so hopefully the data I provide here will prove useful to someone in the future.
It took me an entire afternoon, most of my brain cells, and really questionable RAM topology to get the system past POST and I'm running jobs in OS right now. It seems stable so far, but I'm not sure of long-term stability.
Current POSTable RAM configuration: Total 4 memboards in the "configuration to utilise all PCI-E slots". All DIMMA and DIMMB slots are populated. For CPU1 and CPU3 memboards, DIMMA and DIMMB all contain YH-93Q. However for CPU2 and CPU4 memboards, DIMMA all contain YH-93 and DIMMB all contain YH-93Q. On all boards, DIMMC and DIMMD are empty.
What I learned:
Systems with any M386B8G70DE0-YH93Q sticks seem to only work when ALL the RAM is in slots DIMM A1/A2/A3 and DIMM B1/B2/B3. If you attempt to populate the DIMMC or DIMMD slots, you will be met with "DC detect failure" or "VMSE failure" or "DDR training failure" on POST. This means you can only use half of your RAM slots.This behaviour is very consistent, and I have reason to believe that this is because the M386B8G70DE0-YH93Q cannot operate within the constraints of the memory controller of the X10QBi for quad-channel configuration.Mixing of memory types may very well be possible as a general feature. To maximise likelihood of the mixing working, CPU1/3 memboards must have the same configuration and CPU2/4 may have a different configuration. Furthermore, each channel (e.g. DIMMA slots) must have the same configuration between each pair of CPU1/3 or CPU2/4 memboards.YH93Q sticks are fine with 3 slots per channel when using 4 memboards, which means fully filling DIMMA/DIMMB and leaving DIMMC and DIMMD slots completely empty. I have not tested whether YH93Q sticks tolerate being distributed over all 8 memboards. I only know 4 memboards are POST-able.By just eyeballing it, RAM speed does not seem to be adversely affected by only using 2 channels in a real life workload, but performance will surely take a hit in synthetic benchmarks.
Last edited:
Interesting - I've never had an issue mixing Q and non-Q variants, albeit for 32G DIMMs (M386B4G70DM0-CMA4). I always assumed the "Q" just means that the DIMM was an OEM variant, but I've not contacted Samsung and asked.
Another interesting point: I've always made sure to get CMA4 DIMMs, mainly because I already had CMA4 DIMMs. YH93 uses a different memory buffer (3 = Inphi iMB02-GS02B, 4 = Montage MC CI)
Another interesting point: I've always made sure to get CMA4 DIMMs, mainly because I already had CMA4 DIMMs. YH93 uses a different memory buffer (3 = Inphi iMB02-GS02B, 4 = Montage MC CI)
the seller claims that Q variants just mean they're made elsewhere.
it's so weird. i tried with minimal DIMMA1, DIMMB1 with all Q variants and it works. but as soon as i put anything in DIMMC1 or DIMMD1 it errors when it POSTs.
I've tried with and without CMOS reset so BIOS settings shouldn't matter.
I absolutely know that all slots are working, because I've upgraded from a 64x 8GB sticks configuration.
you know what? i should contact samsung!
it's so weird. i tried with minimal DIMMA1, DIMMB1 with all Q variants and it works. but as soon as i put anything in DIMMC1 or DIMMD1 it errors when it POSTs.
I've tried with and without CMOS reset so BIOS settings shouldn't matter.
I absolutely know that all slots are working, because I've upgraded from a 64x 8GB sticks configuration.
you know what? i should contact samsung!
I assume you have tried the A1,B1,C1,D1 config with different DIMMs to rule out a faulty one?
Resetting CMOS would've been my next idea, there are some LRDIMM specific settings which might not have mattered when you used RDIMMs before.
Which BIOS version are you using?
The X10QBi is a nasty diva, I always pray that I don't get and crude VM memory errors when rebooting the machine
Resetting CMOS would've been my next idea, there are some LRDIMM specific settings which might not have mattered when you used RDIMMs before.
Which BIOS version are you using?
The X10QBi is a nasty diva, I always pray that I don't get and crude VM memory errors when rebooting the machine