Sempron 64 3000+ AM2 w/ DDR2-800 = Instability?

[sheh]

I have a socket AM2 Sempron 3000+ (1.6GHz, 90nm) on a Geforce6100 + nForce410 mobo (MSI K9NGM). I'm trying to run it with DDR2-800. Memtest86 checks okay. I also ran a CPU/memory benchmark under a Linux distro without a problem. But Windows XP gets stuck occasionally, particularly when opening the nVidia control panel, when starting AIDA64, or when running the AIDA64 memory benchmark.

The memory is identified correctly by the BIOS as 400MHz but in fact runs at 320MHz with a divider of 5 off the CPU speed (1600/5 = 320). I'm not sure it's because Semprons don't support more than DDR2-667 or because the divider can't be lower than 5. Anyway, whether I run one DDR2-800 memory module, two as single-channel, or two as dual-channel, the problem persists. But if I mix a DDR2-533 module with a DDR2-800 module (only tried one), in which case both run as DDR2-533 single-channel, Windows becomes stable.

I'm using the IGP, if it may matter. Another thing that I noticed while running the DDR2-800 is that there's some extra delay while navigating a bootable collection of diagnosis tools (UBCD).

Any idea if the problem may be due to the CPU, the chipset, the memory, or maybe even Windows drivers?

 

[BUnit1701]

Considering that it is stable under Memtest and Linux, I would lean towards a Windows driver issue.

 

[Arkaign]

In that config, I'd just go ahead and keep the stable configuration with the 533 chip in there. The performance difference will be nearly nil anyway.

I have a drawer full of AM2 stuff, and I bet tons of people do. I'm sure you could post a local CL ad and get something better for $10 or so (Single Core or maybe even dual core Athlon 64 that will be MIIIIILES better than the Sempron).

 

[DrMrLordX]

About the memory clock speed: You're bringing up an issue that I had years ago with my Brisbane-core X2-3600+, which is the same generation of processors as your Sempron. It consistently ran memory below expected speeds. There is probably an old post on this board where I exhaustively documented the issue . . . oh yeah, here it is. Check out the third post and later.

But, to summarize, your CPU multiplier is going to affect memory speed. You are getting exactly the drop in memory clock speed that I experienced on Brisbane. Your CPU multi on that Sempron is 8x, so you are getting 80% of the expected memory clockspeed. The only way to avoid this entirely is to use the 1:1 ratio and run your HTT speed up to get the memory clocks that you want. Of course, you will have to lower your CPU multi and Hypertransport multi to avoid overclocking those, and your board may not want to raise HTT by much.

Not sure what's causing AIDA64 or your Nvidia utility to glitch, but it may not be the memory doing it.

 

[sheh]

BUnit, the extra delays in UBCD when using DDR2-800 do suggest something going on outside of Windows. Maybe Memtest/Linux just don't exercise or enable whatever part is leading to the instability. It is a strange problem.

Arkaign, but I wonder if dual-channel might worth the trouble, beyond the clocks.

Yeah, I might look for another AM2 CPU and see if it helps. More for lower power consumption than anything else; this Sempron is a 59W TDP part, some newer and faster chips are 45W or less (I'm curious about those 22W TDP parts, but they're rare). I also think newer chips might have more aggressive CnQ options.

DrMrLordX, BIOS-wise at least there's nothing tweakable. I'm not concerned so much with having 320MHz instead of 400MHz, but it would be nicer to have dual-channel 320MHz rather than single-channel 266MHz.

If not the memory leading to instability, what else? It's fine when the DDR2-800 runs at 533. The only differences seem to be the memory clocks and the divider.

 

[Arkaign]

Less than $10 shipped, use guest checkout if you don't have an Ebay account :

http://www.ebay.com/itm/AMD-Athlon-...GAD-/151267716443?pt=CPUs&hash=item2338422d5b

Worth it

 

[Arkaign]

And another one for less than $6 shipped!

http://www.ebay.com/itm/AMD-Athlon-...4CN-/111302357575?pt=CPUs&hash=item19ea233247

 

[Arkaign]

And this sounds exactly up your alley for $12!

http://www.ebay.com/itm/AMD-ATHLON-...rgy-/330763713633?pt=CPUs&hash=item4d030dbc61

2.8Ghz Dual Core 45W low power version.

 

[DrMrLordX]

DrMrLordX, BIOS-wise at least there's nothing tweakable. I'm not concerned so much with having 320MHz instead of 400MHz, but it would be nicer to have dual-channel 320MHz rather than single-channel 266MHz.

Alas, your BIOS will not save the day. Oh well! Such is life.

Re-reading your original post, I have to say that you may have problems with the on-die memory controller. I haven't used AIDA64 before. Maybe I should give it a whirl . . . hey, since AIDA64 is crashing, have you tried good ol' Prime95 Blend? Sometimes it digs up IMC/NB related problems on AMD chips.

If not the memory leading to instability, what else? It's fine when the DDR2-800 runs at 533. The only differences seem to be the memory clocks and the divider.

The IMC. I've had IMC instability issues on my Brisbane before, after forcing too much voltage across the chip. After awhile, it started hating any ratio other than the DDR2-533 and DDR2-400 ones.

 

[CuriousMike]

And another one for less than $6 shipped!

http://www.ebay.com/itm/AMD-Athlon-...4CN-/111302357575?pt=CPUs&hash=item19ea233247

LOL - $6 shipped.

That is one damned cheap upgrade.

 

[sheh]

Arkaign, thanks. Yeah, quite a few interesting pieces of old gear on eBay. I'm not in the US so availability and prices aren't as good, but it's still my default go-to place for this sort of thing.

LordX, this CPU hasn't been mistreated. But I'll see how a different one behaves.

But anyway, seeing those new Kabini results of 20W at Windows idle... Maybe I should try to find a 35W chip rather than 45W. This AM2 computer currently idles at 70W. The computer it will replace idles at about 50W. I'm not sure a CPU change can shave 20W (and that's just to break even), but we'll see. There's also CnQ to test. Though the chipset is probably wasteful as well.

 

[sheh]

Same instability with an Athlon 1640B (45W TDP, 2.7GHz, 65nm). It's probably the memory, somehow. The DDR2-800 modules are unstable also when running at 533MHz, whether as single or dual channel. I didn't find a software to change AMD K8 memory timings under Windows, nor does the BIOS provide any tweakability of this sort, so I edited the SPD to try various speeds. Though, I didn't dig so far into the detailed timing parameters, just the clock rate.

As for power consumption, it didn't help as much as I vaguely hoped. At least it's a little better, with 65W at idle (sometimes lower, as low as 62W). That's with CnQ, so idle is 1GHz at around 1.1V.