Winchester 90nm and Prime95

More and more I am leaning toward recommending newcastle over winchester if you plan to do 3500+ or higher.
I just got another newcastle and it did 2550 priming @ stock voltage on air. Don't remember any winchester I had doing this feat on stock voltage. Will try to get it up a little at a time and see where air ceiling of this will be with stock cooling. This might be the perfect replacement for my deceased newcastle that benched at 3.15ghz on mach 2. I always thought that chip should do air around 2.7ghz or so and if this one gets there I might be in luck again :)

Prepare urself for an unedjucated guess at whats the problem.

Maybe since it's 90nm there is of course more of a chance for crosstalk. At higher vdimm the voltage goes through the cpu as well, it is more prone to leak than the 130nm. So maybe thats why higher vdimm causes problems. The same thing can be causing problems since the same thing happens, to my knowledge with increased clock frequencies (and I don't know much about this so I could be competely wrong). So the whole problem might just be because it is 90nm and it is more prone to crosstalk than 130nm. And that is what happens after you get to 250fsb (and in some cases 200fsb which could be caused with 2.9 or more vdimm).

I'm not sure if the terminology I used is right, so sorry if it's hard to understand. For people who know what I'm talking about, does this sound like a logical cause for the problems?

edit: haha... I just noticed I spelled educated wrong when andy made fun of me there :p: ... I'll leave it cause it just makes me sound like I know what I'm talking about.

Quote:

---

Originally Posted by kryptobs2000

Prepare urself for an unedjucated guess at whats the problem.

Maybe since it's 90nm there is of course more of a chance for crosstalk. At higher vdimm the voltage goes through the cpu as well, it is more prone to leak than the 130nm. So maybe thats why higher vdimm causes problems. The same thing can be causing problems since the same thing happens, to my knowledge with increased clock frequencies (and I don't know much about this so I could be competely wrong). So the whole problem might just be because it is 90nm and it is more prone to crosstalk than 130nm. And that is what happens after you get to 250fsb (and in some cases 200fsb which could be caused with 2.9 or more vdimm).

I'm not sure if the terminology I used is right, so sorry if it's hard to understand. For people who know what I'm talking about, does this sound like a logical cause for the problems?

---

Yor unedumacated guess is a fairly sound theory. :) This is more than likey exactly what is going on. Some of the CPU's just leak or crosstalk more than others.

Quote:

---

Originally Posted by kryptobs2000

hey, I'm just wondering. What is the benefit of even having the memory controller on die? I can see the obvious benefits of less latency or something (I don't know how it all works, but just guessing). But other than that.. I just don't see any benefits. I mean, on crappy low end boards, sure they would have equaly crappy memory controllers, but who cares? On high end boards, say if the 754 dfi had it's own memory controller. I'm sure it'd have a really good memory controller easily capable of 300mhz+ 2-2-2-5 (dosn't mean the ram can do it I'm just saying). And you could have dual channel on the 754 cpu's. So whats the point? Is it just to cut costs or a promotional gimmick, or is there actually a real benefit?

---

A K8 without an ondie memory controller is called an Athlon XP. (well technically the K8 in 32 bit mode also has SSE2, slightly tweaked cache and TLBs, and a little bit longer pipeline, but none of this massively affects per clock scaler performance, the difference is the memory controller)

Try one out sometime if you want to see what difference it makes.

You mean try an Athlon XP out? I had a 2500+ and I had better memory bandwidth than I do now, but I think I was running it in dual channel too I'm not sure. I know I was running it in dc, but not sure if I every benched it in dc because if I did, my score only went up 200mbps or so which dosn't seem right.

>>I know I was running it in dc, but not sure if I every benched it in dc because if I did, my score only went up 200mbps or so which dosn't seem right.

That's because nForce2 is not a true dual channel chipset and the Athlon XP is limited by the FSB. Which is offtopic and now history.

Hello all,

I have a CBBHD 0444 Winnie, K8N Neo2 platinium, and went cheap on the RAM side cause I wanted a lot of it (2Gb) and couldn't afford the Crucial Ballistix 1Gb PC4000 sticks ... ended up with 4 x 512Mb MarkVision ValueRAM (88 Cdn$ a stick here in Ottawa)

Didn't run MemTest86 at first, directly installed Windows XP, started a quick overclocking to see what this puppy was able to, reached 2.5Ghz with not much trouble (DRAM max = 133), then started Prime95 and ... FATAL ERROR after 3 minutes.

Up-ed voltage 5% then 8.3% ... same pb.

Went back to 2.4GHz, ran 3h @+8.3% then same pb

Back to 1.8HHz, all defaults ... same pb !!!

Then I used MemTest86 at stock defaults, and found 1 memory location giving 5 to 10% memory errors - tested all 4 sticks separatly, only 1 exhibits this 1 memory location errors at 5 to 10% rate (even clocked down to DRAM max = 100MHz)

This is probably my Prime95 problem at stock defaults - will return this stick tomorrow, and update you if I can Prime stable afterwards - at stock speeds to start with.

Cheers
Dom

So you were running 2Gb(4x512) of ram?

dom993
Sounds like you have a screwed up stick?

The Small FFTs don´t fail, but Blend and, large fails, which sounds logical.
Superpi 32M also fails after 15min.

The best is Memtest: It doesnt give me any errors at all, while D00m3 is crashing after 10 min. ahha this prog realy sucks, i will never use it again on A64 Systems.

n0w4i4
Which test in memtest are you using?

The default, dont now exactly, but it ran about 5-6 hours over night.

I am not a memtest user, but people have been saying something about using test 5. Try it! Apparently it is much better.

I'll try it when i am back from work ! :upset:

To summarize my investigation so far:

At stock speeds and voltage both SuperPI and Prime95 fail but memtest86+ extended test (7.5hrs per pass) passes OK.

At 2-2-2-5 1T (stock) and 2.6Vdimm both SuperPI and Prime95 fail.

At 3-4-4-8 2T and 2.6Vdimm both SuperPI 32M and Prime95 (14hrs) pass!

I might try leaving my settings as they are and bumping the voltage to 2.7Vdimm (stock) to see what happens. I've also got a 3500 Newcastle coming so I'll be able to check that it is the chip that is the problem.

If you have tested your NC please post your results. I can't wait to know why higher Vdimm causes to crash Prime and superpi but not Memtest !?

My guess: 10% AV8 (Bios) related
89% Winchster related
1% PSU

Just tried 2.7Vdimm, that's all I changed and SuperPI 32M passed fine. Dunno if that disproves your idea. Later I'll try changing one setting at a time and see where it'll fail, I'll also Prime95 to check it's still stable. Not sure when my Newcastle is coming, it's coming by courier but I haven't had an email to confirm it's been dispatched.

Edit: just to confirm - this is 3-4-4-8 2T but at 2.7Vdimm (stock).

Mhm ok .. i think things will change as soon you try to oc HTT ~ +20 Mhz !

i just got mine to pass 8.5 hours in blend mode with the settings in my sig... i'm a happy camper!

btw its a week 37 cpu... :P

Quote:

---

Originally Posted by spacey

i just got mine to pass 8.5 hours in blend mode with the settings in my sig... i'm a happy camper!

btw its a week 37 cpu... :P

---

Nice going, is the reason you are using a devider for your ram is because they cant handle running 1:1?

week 41 90nm 3500
over 24 hours prime in blend mode
epox EP-9NDA3+ 1gb ocz plat rev 2 o/c to 2420 mhz and memory is at 1t 2-3-3-5
air cooled

This will be my test rig until my water stuff arrives so if you want me to try some settings LMK

BeerCan
Memory 1:1?

spacey
I also think that you should try running MAX 1:1 your memory can handle. See if it passes then ;)

Does anything other than Prime crash with these faulty winchesters?

Are they still ok for gaming????

as far as I know, mostly yeah. But they obviously have some problems hardware based so they are not 100% stable.

I just wanna know if it's still possible to play games like ut04 and doom 3 while running overclocked for a day without it crashing.

That's all i would need it for. :P