4800+ on chip memory controller problem

[tompson]

this is long... but bare with me.

ok, so I want lots of memory bandwidth (3DMark05 LOVES the bandwidth) so I'm running my CPU at 10x multiplier and I want to find my highest FSB / HTT overclock at this speed. So I seem to be hitting this limit of 264MHz which seems ok. Prime95 is stable for 6+ hours with two threads of it running simultaneously. However, Memtest fails!

So my first guess is that it's a memory problem. So I lower my multiplier to 8x and up my memory voltage to 3.5V... No good. So I eventually raise my CPU from 1.40 to 1.475 volts (big jump) and all of a sudden memtest is stable. Not a memory problem but a CPU problem? This doesn't make much sence since a blend test in prime95 is ok AND I've run the CPU at faster speeds using higher multipliers. I've been able to get 2676 = 12 x 223 vs 2640 x 10. So my best guess is that the on chip memory controller must be failing at 264MHz HTT (ok at 263) while the rest of the CPU can go faster.

Is this normal? Has anyone else found that the on chip memory controller is the limiting factor in their overclock??

I thought that the ALU and the internal cache were the biggest problems at high speed.

This makes me believe that the new revision E mem-controllers aren't very good at overclocking.
__________________
my specs:

eVGA e-Geforce 7800 GTX KO video card
Enermax Noisetaker 600W
OCZ EL DDR PC-4000 Dual Channel Gold VX (voltage extreme)
AMD Athlon 64 X2 4800+ CPU
2 X Western Digital 8MB buffer, 74GB Raptor in Raid 0 array
DFI LANParty nForce4 SLI-DR
Lian-Li PC-V1200 Plus case modified with window
Thermalright SI-120 CPU heatsink and Panaflo (Panasonic) 120mm M speed HydroWave fan
Dell UltraSharp 2405FPW 24-inch Wide Aspect Flat Panel
IBM Thinkvision L190 19" LCD Monitor
Sound Blaster X-Fi Fatal1ty FPS
Logitech Z-5500 digital 5:1 speaker system

 

[Ronin]

Stability isn't always measured by a program, per se. Put it where you're comfortable and GAME with it, and see if it's stable. If it is, who cares what memtest says?

 

[stevty2889]

I get similar results, with higher FSB, the memory controller doesn't seem to be as stable, even if the CPU is running at the same speed it was stable with at a lower FSB with no trouble. I can run fine at 238x11(2.618ghz) but at 260x10, I just reboot when trying to get in to windows, even though the cpu is running at a slightly lower speed, and the ram is running at about the same speed, and I lowered the HTT multi to 3x.

 

[imported_michaelpatrick33]

I have the same problems with my X2. I have Patriot dimms and they can go over 230 at 1:1 but if I used dividers on the memory (5:6 etc) I can't much go over 205 or I get bsod regardless of cpu speed.

 

[uOpt]

The memory controller on 1024 KB cache San Diegos and X2 4400 and 4800 is often reported to not drive RAM up to the speeds that the same RAM brings on 512 KB chips.

If you really want memory bandwidth you have the wrong chip. I offer you my 3800+ in trade 🙂

Real-world performance wise that doesn't matter in the slightest.