Nehalem and 4 Gbit DDR3

While discussing Intel desktops with DDR2 memory using 2 Gbit technology (4 GB UDIMMs), the question of Intel’s next generation and 4 Gbit DDR3 (8 GB UDIMMs) came up. It’s more or less the next iteration of exactly the same problem.

I decided to focus on the LGA1366 platform which was the first post-Core 2 generation and Intel’s first platform that used exclusively DDR3 memory (recap: later Core 2 chipsets supported both DDR2 and DDR3). Reviewing Intel’s documentation turned out to be… confusing.

The datasheet for the first generation Bloomfield Core i7 (45nm Core i7 LGA1366, Nehalem microarchitecture) says that DDR3 "512Mb, 1Gb, 2Gb, technologies/densities are supported", 24 GB RAM maximum. The memory controller in the CPUs could handle 6 DIMMs, and with 2 Gbit technology/4 GB UDIMMs, that obviously results in 24 GB maximum.

Can this 2008 desktop CPU support 48 GB RAM? Intel says no…

The datasheet for the follow-up Gulftown processors (32nm Core i7 LGA1366, Westmere microarchitecture) does not state what technologies or memory limits are supported. It was not hard to find reports of people using Westmere generation LGA1366 CPUs with 48 GB RAM so I started digging further.

I only have one desktop LGA1366 board that could possibly take 48 GB RAM, Intel’s own DX58SO2 (Smack Over 2). That’s where it gets interesting. The DX58SO2 Technical Product Specification clearly says 24 GB maximum. But the DX58SO2 Product Brief just as clearly says “Maximum system memory up to 48 GB using 8 GB double-sided DIMMs.” Who do you trust, Intel or Intel? Could they be both wrong?

No, the DX58SO2 product brief is right and the board indeed supports 48 GB RAM.

First of all, the BIOS release notes include an interesting tidbit for BIOS Version 0876: “Fixed the memory Maximum Capacity value reported in SMBIOS Type 16 to support 48GB.” That wouldn’t affect what memory sizes were actually supported but it implies that the BIOS was originally meant to handle only up to 24 GB.

My initial assumption was that Westmere CPUs would support 8 GB DIMMs and the older Nehalem generation would probably not. That assumption turned out to be wrong.

My DX58SO2 happened to have a Xeon W3690 in it. That CPU is, depending on who you ask, either Gulftown or Westmere EP (Intel’s ark is obviously very inconsistent in that Xeon W3680 is supposedly Westmere EP and W3690 is Gulftown). Some people also call it Westmere-WS, perhaps just to make the point that the code names are meaningless.

At any rate, the DX58SO2 with a W3690 does support 8 GB DIMMs, and although it is picky and initializing newly added memory modules can take rather long time, I was able to stuff the board with 8 GB DIMMs:

48 GB RAM with Xeon W3690

The memory works just fine, I had no trouble booting Linux and using all the memory. In fact with 48 GB RAM and six cores, the now antique platform (LGA1366 came out in 2008!) is surprisingly usable.

Okay, so Intel’s product brief was not lying and the board does support 48 GB RAM. But does it actually work with other CPUs? The next test was a Nehalem generation Xeon W3580. That processor might be a Nehalem-WS or a Bloomfield… thank goodness Intel reduced the insane number of code names when Sandy Bridge came out.

And lo and behold, the W3580 which supposedly supports up to 24 GB physical memory, does this:

48 GB RAM with Xeon W3580

So the Intel datasheets are wrong and the claim on Intel’s ark that this CPU supports up to 24 GB physical memory is wrong too. The memory is not just detected, it really works:

memtest86-ing 48 GB RAM in a Nehalem desktop

I did not run a full test but I let memtest86 run long enough that it should have failed if it couldn’t address all the memory. And yes, I have a funny mix of memory modules there.

The next test was an i7-965 (Core i7-965 Processor Extreme Edition, which people generally agree is a Bloomfield CPU). The CPU I have is from late 2008, one of the first batch of LGA1366 CPUs.

At first I was convinced that it can’t work with 8 GB modules. The board would not boot and kept getting stuck at various POST codes. But apparently the real problem is that this Intel board, like some other Intel boards (looking at you, DX79SR) is not very good at changing CPUs.

That is to say, if you plug in a different CPU, the results depend on what was in the board before. It also matters whether the BIOS jumper is in the “config” position or whether the B2B button on the back is pressed. It is… not helpful.

After some more CPU swapping, and messing with jumpers and buttons, the i7-965 somehow decided to boot up after all:

Core i7 965 with 48 GB RAM

And memtest86 was happy enough with the memory, too:

memtest86-ing 48 GB RAM in with another Nehalem desktop CPU

Again, I was able to boot into Linux and use 48 GB RAM without any apparent trouble. It is worth repeating that this is a 2008 desktop CPU, albeit a high-end one. When I bought a regular Sandy Bridge desktop in 2011, it did not support that much memory. And a Haswell based Intel DZ87KLT from 2013 also maxes out at 32 GB. It’s really unfair.

Just for the sake of sanity I checked that a Core i7-975 also works with 48 GB RAM (it does).

Next I wanted to test if the “real” E and X series Xeons also work. Those are well known to have no trouble with 8 GB registered DIMMs, so the chances were good they’d work with 8 GB unbuffered DIMMs too. First I tried a low-end Xeon E5504 (Nehalem), and found that it can handle 48 GB as well. Next I went to the very opposite end and tried a Xeon X5690 (Westmere), which of course worked too.

I should note that if the DX58SO2 board didn’t have the super useful POST code display, I would have almost certainly concluded that it can’t support 48 GB RAM. It can take surprisingly long time (perhaps 30 seconds) for anything to show up on the screen when the RAM is fully populated. But from the POST code display it’s apparent that the board is not, in fact, stuck.

Conclusion

After testing a broad selection of LGA1366 processors in a desktop board, it appears that these CPUs can generally handle 48 GB RAM using 8 GB UDIMMs, despite Intel’s documentation stating that they only support 24 GB RAM. This is true for Core i7, Xeon W, and Xeon E/X LGA1366 processors of the Nehalem and Westmere generations. A prerequisite is obviously a desktop board which has six memory slots and can properly configure six 8 GB UDIMMs, but such boards do exist.

This entry was posted in Intel, PC hardware, PC history. Bookmark the permalink.

12 Responses to Nehalem and 4 Gbit DDR3

  1. Zir Blazer says:

    I ended up googling about Lynnfield with 4 * 8 GiB since I said in the other blog post that I recall that some people had issues with it, and ended up in this Thread: https://hardforum.com/threads/core-i7-860-with-p55-chipset-32gb-ram-support.1845468/

    Where at least two users reports success. So I save you the need to test that since there is proof it works, at least Processor/IMC-side.

    Reminds me when in 2013 DDR3 prices were record low and I decided to purchase 4 * 8 GiB for my 2009 Athlon II X4 on a Motherboard that officially only supported 4 *4 GiB. After a few “Your Motherboard doesn’t support that much RAM” posts, I ended up with the last laught: http://www.xtremesystems.org/forums/showthread.php?286094-Trying-to-choose-DDR-III-Memory-Modules-for-32-GB-RAM-(4-*-8-GB)&p=5193882&viewfull=1#post5193882

  2. Yuhong Bao says:

    Did you test 16GB registered DIMMs?

  3. Michal Necasek says:

    I have not… because I don’t think I have any. Do you mean in the DX58SO2 board though, or in a board that officially supports registered memory?

    There are also at least in theory 16 GB unregistered DIMMs using 8Gbit technology. But I’ve never seen one, I just noticed they’re in the JEDEC DDR3 UDIMM specs. There aren’t any 16 GB UDIMMs listed in, say, Samsung’s 2014 catalog.

  4. Michal Necasek says:

    Good to know. Doesn’t surprise me too much since the Lynnfield CPUs are actually newer than Bloomfield, so if the original Bloomfields could support 8 GB UDIMMs, why not the follow-up.

    I remember there was a lot of trouble with the early Core i7 laptops in that they refused to work with 16 GB DDR3 RAM (2x8G). Very often the laptop recognized the RAM but was unstable to the point of being unusable, quite possibly due to the firmware. I’ve seen that happen with Dells and MacBook Pros at least.

  5. Yuhong Bao says:

    Yea, I believe that only Micron and Intelligent Memory made the 8Gbit DDR3 chips around 2015.

  6. Yuhong Bao says:

    And I believe that on the Intel side only Broadwell and Skylake ever supported 8Gbit DDR3.

  7. Nagel says:

    I can confirm that a HP Z400 workstation (newer mainboard version with 6 DIMM slots, BIOS 3.61, Xeon W3670 CPU, X58 chipset) does work with in a 6 x 8 GByte configuration, providing 48 GB RAM. The Intel ark database nevertheless states “formerly Westmere EP” and 24GB max memory size. As a lot of discussion can be found in the internet about a successful 48GB setup of this generation of the Z400 workstation, one can indeed deduce that certain generations of Socket 1366 CPUs do support 8 GByte DIMM despite Intel stating a 24GB limit. There also seems to be a lot of flexibility, as I have used Corsair CML16GX3M2A1600C10 modules for this setup. According to dmidecode my W3670 CPU is Type 0, Family 6, Model 44, Stepping 2.

    Also thanks for your very interesting website. Carry on!

  8. Michal Necasek says:

    Based on the testing I’ve done, I am 99% certain that all Socket 1366 CPUs support 8GB DIMMs, contrary to Intel’s specifications. I just couldn’t find any that wouldn’t work, going to the oldest 2008 models. Whether LGA1366 boards support such memory is probably a separate question, and especially the firmware might be a problem.

  9. Squall Leonhart says:

    why are we talking about this in 2021, when it was discussed in depth on other easily found blogs back in 2015?

  10. Michal Necasek says:

    I don’t know, why are you making that comment in 2024? πŸ™‚

  11. Prototyped says:

    The “High-End Desktop” (HEDT) parts like LGA1366 were really a different beast compared to the “ordinary” desktop parts like LGA 115x from this and successive generations. HEDT were effectively Core-branded uniprocessor Xeon parts, and it showed in terms of the amount of RAM and I/O they supported. This of course continued through at least the Cascade Lake-X (Core i9 109x0X) processors. I don’t recall having seen any newer Core-branded HEDT processors after thisβ€”I think Intel may have switched to the more honest Xeon W branding afterward.

  12. Prototyped says:

    (This is not to confuse them with the Xeon-branded desktop parts on LGA 115xβ€”the “Xeon E” parts, where the difference from the Core-branded desktop parts was typically just ECC support and features like SMT/HyperThreading available further down the offering stack. In other words, market segmentation rather than any real difference in capabilities. That Xeon E branding continues today, though it’s become much less meaningful, as ECC is available now in ordinary “Core” branded processors where paired with a W-series chipset. Previous to about Alder Lake-S, the options were to use either Xeon E series or, oddly, specifically designated Celeron and Pentium branded low-end parts, presumably so manufacturers could shift fully functional servers but for the buyer to replace the CPUs with Xeons they’d acquired in bulk elsewhere.)

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.