So lets do the final checks of the "project". The goal was to build up 10x stronger PC as I had before.
Pentium III-S 1400 @ 1638Mhz VS Xeon x3360 2830 @ 3400Mhz
465 gflops VS 1280 gflops on single core * 3,65 Multicore effectivity 4672
156Mhz SDR VS 400Mhz QDR
1248mb/s VS 12 800mb/s
156mhz SDR VS 800Mhz DDR dual channel
1248mb/s VS 25 600mb/s
Real memory reads
1202mb/s vs 9500mb/s
Not Completed = See notes
Graphic slot bandwidth:
AGP 4x VS PCI-E 2.0 16x
1064mb/s vs 8000mb/s
Completed See notes
Harddisk read rates
SCSI Seagate Cheetah 15k RPM VS OCZ Revodrive x2
70mb/s VS 690mb/s
Completed = See notes
Single core performance should be better, but overall total CPU cooperation hit the total limit. Real memory rates are not as effective as I expected from synchronized overclocking. This seems to be mainly problem of CAS latencies. Old SDRs I had and chipsed matched very well resulting in 98% effectivity in read. New system barely hit 75%. Also comparing AGP 4x and PCI-E 2.0 16x ... Since I was able to run old system with agp 2x without any impact on system performance it does not matter as much. And since old drive was very very fast, but still classic HDD the new SSD RevoX2 still wins.
Conclusion on Dfi X48-t3rs after 2 years of usage:
CPU VRM - Excellent.8 Volterra chips do their job as expected. CPU voltage is stable even without CPU Droop control.
Memory slots - Terrible. The connection between board and DIMM module is bad.
PCI-E/PCI slots - Are fine. There is nothing to complain about.
X48 chipset - On this board it is able to fulfill its specifications, with one exception - 8gb ram on 1600Mhz should be possible even on 1,265v vNB.
Onboard accesories - Jmicron ATA controller is junk. Other devices seems fine.
It is true that this board does not have big potential on Core 2 quad processors. Testing on this week shows that when most slots on the board are occupied the Northbridge suffers on frequency and voltage overshoots. Increasing level of signal strenghts causes it to fail. Only cure for this is higher vNB. And even when it sounds strange its possibly caused by the high quality of components which were used for the manufacturing.
The heat what X48 produces is incredible. In most cases its much more hot than CPU.
After numerous torture tests I was unable to destroy CPU or memory. Presumably dead WD disk drive has to be re-tested. When Elpida MNHs survived on 1,9v and Linx testing it means that i can expect quite a long-term life of this system.
According to the real performance of this system it seems to be better as a workhorse system with some overclocking potential, but not as a pure-bred overclocking machine. I just wonder why the X48 was not shrinked down to 65nm... But if the shrink should have some power supply issues its quite certain that high voltages could kill it easier than this 90nm furnace.
Test with clock generator and PLL voltage shows why the CPU was not able to be clocked higher. Signalling from CPU to NB was quite strong even on default. Higher Clock generator and PLL voltages improved the signalling further and decreased system stability... In combination with diginal PWM of CPU the northbridge was unable to handle such strong signal under heavy load of 2 separate CPUs and 4 dimm modules. Only correction - higher vNB voltage - was after all the limiting factor.
This kind of power management was just too good for the northbridge. When problems with DDR3 implementation accumulated on this the board was doomed (in eyes of many overclockers). This is one of the few boards which might suffer from components overvoltage. High voltage is not the solution, but the right voltage is.
Most I had to test system like "lower voltage until the boot fails, set higher voltage until the boot fails = right voltage is in the middle". And this takes a lot of time or really good overclocking method (and even mine is not as good as I wanted).
When I replaced PSU last time i completely forget to connect additional power for PCI-E. Will test it, today, but I fear it will not have much impact on system, but I have to take into account that that power lines were designed for multiple graphics. Both PCI-E slots are occupied. Only one of them is graphic card, but this possibility has to be checked.
The cables are connected. No change observed. Everything that happened afterwards was related to some bios settings. Some of them were not saved correctly and when they were reloaded, some values were different.