Memory expansion

[marcus]

Hi.

My X1000 came with 4GB or RAM in the form of 2 2GB DIMMs. Yesterday I bought a second set of 2GB DIMMs to bring the memory up to 8GB (useful for Linux).

Now, the memory is recognized and Linux boots up with 8GB of RAM, but I'm getting random reboots which I don't get with just the original 4GB.

The DIMM in slot1 is identical to the one in slot3 (these were delivered with the machine), and the DIMM in slot2 is identical to the one in slot4 (these are the new ones). So that should satisfy the requirement that slot1+slot2 is populated identically as slot3+slot4.

The new DIMMs (from Corsair) are marked "800MHz 2048MB 5-5-5-18 1.80V ver4.5".

Anyway, I wanted to check how CFE had configured the modules (CAS latency and such), so I tried the show dramcfg command, but this did not print anything at all! Do I need to upgrade CFE? Ideally I'd also like to be able to modify the configuration, but the dramcfg command doesn't seem to do anything either...

[trevordick]

Marcus,

Do I need to upgrade CFE?

You should already have the latest version of CFE installed on your machine. The presence of the early boot menu will confirm this. Note: the memory test does not work under CFE, this is a CFE related bug.

I presume you are refeering to random reboots in Linux? What kernel/distro are you using?

Trevor

[marcus]

I presume you are refeering to random reboots in Linux? What kernel/distro are you using?

The most frequent case is a reboot during early Linux kernel startup, even before the fbdev starts displaying anything on the screen. This happens maybe 50% of the time, both with 3.12.1 which I use for Gentoo, and with an older kernel used for Wheezy (don't remeber the exact version right now). Both kernels are quite stable with 4GB. And once I managed to boot Gentoo up with 8GB, I could run it for several hours before the system rebooted. But triggering the problem at early boot is quite easy.

Another quite weird thing is that EHCI seems to lock up on some boots when i have 8GB installed. OHCI is not affected, so I can use the keyboard which is connected direcly to a root hub port (which is there purely for the benefit of CFE, which can't handle my KVM), but not the one connected via an external USB2 hub...

But let's get back to the DRAM config. Is there no way at all to check it? I would like to compare it against the config used when running on 4GB, as well as against the datasheet of the new RAM...

[mechanic]

Just wondering what happens if you set the cpu speed to 900MHz for example.

[javierdlr]

Hi.

My X1000 came with 4GB or RAM in the form of 2 2GB DIMMs. Yesterday I bought a second set of 2GB DIMMs to bring the memory up to 8GB (useful for Linux).

Now, the memory is recognized and Linux boots up with 8GB of RAM, but I'm getting random reboots which I don't get with just the original 4GB.
..

I don't own a X1000, just a SAM460ex, but:
1)What if you use only the new DIMMs (2x2GB) either in slot1 & slot3? OR in slot2 & slot4?
2)If you change your original DIMMs in slot2 and slot4?

Is just to check if all RAM/DIMM slots are ok.

[marcus]

1)What if you use only the new DIMMs (2x2GB) either in slot1 & slot3? OR in slot2 & slot4?
2)If you change your original DIMMs in slot2 and slot4?

Hm, well, the problem is that the DIMMs in slot 1 and 3 are glued into place (not my fault, the machine was delivered that way)...

[marcus]

Ok, I'm back. It turns out that the "show drawmcfg" command indeed does something, but only for people who pay £4.80 for the serial cable, because it prints on the serial port only...

Anyway, here is the original config (with just the old 4GB):

CFE> show dramcfg
SDRAM: ECC off, Non-ECC DIMM used on channel 0.
SDRAM: ECC off, Non-ECC DIMM used on channel 1.
DRAM runtime configuration:
Total memory: 4096MB
Channel interleaved
XAUI

Channel 0:
2048MB of RAM, Address Range 0 - 6144MB
Memory clock = 400MHz (DDR2-800), tCK = 2500ps, CL = 5
Low Latency Config Tuned for Fcore = 2000 MHz
High performance mode
MMC advanced features
DRAM Rtt: 150 Ohms
MMC Rtt0: 75 Ohms
MMC Rtt1: 75 Ohms
MC & DRAM ODT
Addr/Cmd 2T Timing
Ctrl Output Phase wrt CLK: 0
Addr/Cmd Output Phase wrt CLK: 0
Rank interleaved
UDIMM

Channel 1:
2048MB of RAM, Address Range 0 - 6144MB
Memory clock = 400MHz (DDR2-800), tCK = 2500ps, CL = 5
Low Latency Config Tuned for Fcore = 2000 MHz
High performance mode
MMC advanced features
DRAM Rtt: 150 Ohms
MMC Rtt0: 75 Ohms
MMC Rtt1: 75 Ohms
MC & DRAM ODT
Addr/Cmd 2T Timing
Ctrl Output Phase wrt CLK: 0
Addr/Cmd Output Phase wrt CLK: 0
Rank interleaved
UDIMM

*** command status = 0
CFE>

and here is the new one (with all the DIMMs):

CFE> show dramcfg
SDRAM: ECC off, Non-ECC DIMM used on channel 0.
SDRAM: ECC off, Non-ECC DIMM used on channel 1.
DRAM runtime configuration:
Total memory: 8192MB
Channel interleaved
XAUI

Channel 0:
4096MB of RAM, Address Range 0 - 10240MB
Memory clock = 400MHz (DDR2-800), tCK = 2500ps, CL = 5
Low Latency Config Tuned for Fcore = 2000 MHz
High performance mode
MMC advanced features
DRAM Rtt: 50 Ohms
MMC Rtt0: 150 Ohms
MMC Rtt1: 150 Ohms
MC & DRAM ODT
Addr/Cmd 2T Timing
Ctrl Output Phase wrt CLK: 0
Addr/Cmd Output Phase wrt CLK: 0
Rank interleaved
UDIMM

Channel 1:
4096MB of RAM, Address Range 0 - 10240MB
Memory clock = 400MHz (DDR2-800), tCK = 2500ps, CL = 5
Low Latency Config Tuned for Fcore = 2000 MHz
High performance mode
MMC advanced features
DRAM Rtt: 50 Ohms
MMC Rtt0: 150 Ohms
MMC Rtt1: 150 Ohms
MC & DRAM ODT
Addr/Cmd 2T Timing
Ctrl Output Phase wrt CLK: 0
Addr/Cmd Output Phase wrt CLK: 0
Rank interleaved
UDIMM

*** command status = 0
CFE>

Well, the only difference, about from the amount of memory per channel, is the impedances. If it could be those that cause the problem, I guess a potential solution might be to put both the new DIMMs in the same channel, allowing the impedances to be set differently for old and new DIMMs. (Athough that stupid glue is still preventing me from doing that...) Since the speed and size are the same, it should still be possible to use interleaving.

Or am I thinking this completely wrong, and the change in impedance is just a natural compensation for the fact that the fan-out changed when the total DIMM count changed?