• Please review our updated Terms and Rules here

Buffing up my old main i7, thoughts and advice

The memory controller has been on the CPU since 2009, the motherboard has little to no say how the CPU treats memory. It's not like the old days where the memory controller was on the north bridge and there were rigid limits. Unless you had a VIA chipset, those things didn't care what kind of memory you had, as long as it looked like memory.
 
Even on the older stuff, a lot depended on the chipset firmware. I've got an old HP Vectra here with P3 (goosed up with a slocket) that doesn't seem to care if I install "naked" SDRAM or the buffered, registered stuff, so long as it's all the same. Chipset is 440BX.
Similarly, I've got a 440GX P3 that takes either type. But I also have other 440BX systems that demand unbuffered SDRAM.
 
Well it just arrived in the country, so we'll see soon enough. ECC is going to take a week or so.
Truthfully it all depends on the fan noise level. The Xeon supermicro workstations I ran at work were standard big gamer type fans, I can't say I noticed increased noise.

I have a Quadro, when you put that thing in a normal PC, it's airport levels of noise. The card is meant to be put in a rack and to be cooled by a horizonal ventilation system in servers. The cooling on it is a fan for 486, which goes wild if something isn't cooling down the entire machine. When you replace the stock vent with something from GTX series, the card isn't slim any more but it's able to cool down itself normally with low rpm on fans.

So, having in memory what heat blocks look like on 1U and 2U rack servers, and considering that block in my computer is 3U high just in itself, able to keep a 3GHz base clock CPU on 40(c) while fan is barely audible, I wonder what mileage I get from this, by advertisement, same TDP processor.
 
Posting this from Xeon. Good news - temperature, bad news - memory. In reverse than expected ;)

Idle temp is ~42 which is same as i7. On normal load temp is about 55, again like i7. Where the load is recompiling FreeBSD with 44 jobs. It's not a massively parallel build system so no core ever goes above 15% usage.
ECC RAM is not working, the board is stuck in POST loop, passes 19, hangs a bit in there, then loops back over couple of codes to reach 19 again and so on...my old ram seems to work, at least the 16GB stick, I'll put others in too and see.

Btw it takes a whole lot longer in POST, about three times more to get the video out and BIOS banners. I'm thinking about downgrading the BIOS to the one recommended for the CPU.
 
P.S. forgot about the M.2, 1TB Samsung SSD 980 in a most bought Amazon 16x adapter card just works, plug and boot.
 
So the BIOS Setup on latest (2019) version in combination with Xeon has an issue, doesn't reboot to flash utility, doesn't reboot at all, hangs on exit. I think post displays A6 at that moment. Plus the extremely prolonged POST.

I'll pop in the i7 back tomorrow, and then reflash the board to the recommended 2016 BIOS for this CPU.
 
The extended POST time with the huge amount of memory is because of RAM training. If the system is unstable with that RAM, I'd suspect some of it is bad.

If you can get the machine far enough, you may try running Memtest.
 
More detailed explanation: I have a bunch of this Hynix ECC RAM and I have my old non-ECC RAM, from which I picked a 16GB Kingston module for testing.
The Xeon works normally with a single non-ecc 16GB stick, albeit the prolonged POST. The memory is working because I've filled it up (and then some more in swap) when test-recompiling the FreeBSD userland. I'm just operating the PC normally since yesterday on 16 GB RAM. Did also Userbench, Cinebench on Windows and so on.

No ECC stick works in any position. Per manual POST code 19 is south bridge init, phase before memory stuff.

So, with Xeon and 2019 BIOS;

- ECC doesn't work, classic RAM works
- Prolonged POST
- Setup utility save/reboot issue
- Otherwise the computer and booted OS'es are fine.

Unfortunately I don't have a server board to test this Hynix ECC RAM against, but considering that all sticks hang in the same POST sequence, they seem to be consistent in that regard. Of course the POST is very short and halts on a different code if RAM is not properly connected. I guess the purchased Hynix are ok (seller packaging was nice too).

I have to play with BIOS versions now, I do have two BIOS slots on the board - can keep the latest on A and try out the earlier Xeon-labeled version on B.
I don't know what's on slot B and if it's even populated, I hope it's populated with the version whose Setup works without issues, so I can enter flash mode. If no, I have to switch CPU to preform flashing.
 
I wonder if it's a matter of buffered or unbuffered memory. Some systems are extremely picky about that.
 
Last edited:
^ that's the last straw here, Xeon-recommended BIOS works without glitches in Setup utility, but no cigar with ECC RAM.

As a current final resort I've manually modified that BIOS image. No time tonight to switch RAM around, but the modified BIOS is running - used AMIBCP 5.02 to explicitly enable ECC under CPU category.
Gonna try it out tomorrow as it's late night here already. If it fails, well I'll try trading it for unregistered ECC.
 
You must log in or register to reply here.
Back
Top