Winchester 90nm and Prime95

why do you guys always trust only memtest+?
only because it passes hours with flying colors at e.g. 267 means nothing when it comes to prime stability. memtest+ test ONLY memory and memory controller. it DOES NOT test your CPU.
so when you succesfully memtest at 267x9 it doesn't mean that you will pass 12-24hrs of priming. this is absolutely no-brainer ;)
with memtest+ there no maximum heat-up on the CPU.
i suppose when the CPU heats up it WILL affect memory's controller as well since it is on the die.

Quote:

---

Originally Posted by bachus_anonym

why do you guys always trust only memtest+?
only because it passes hours with flying colors at e.g. 267 means nothing when it comes to prime stability. memtest+ test ONLY memory and memory controller. it DOES NOT test your CPU.
so when you succesfully memtest at 267x9 it doesn't mean that you will pass 12-24hrs of priming. this is absolutely no-brainer ;)
with memtest+ there no maximum heat-up on the CPU.
i suppose when the CPU heats up it WILL affect memory's controller as well since it is on the die.

---

I was simply confirming that the CPU was stable at a given speed for over 24 hours, 2.4ghz but only with memory async (218.5). Now, one would think that the memory I am using which was tested using the same motherboard over 24 hours with my brothers 130nm NC while priming with 267x9 1:1 3/4/4/8 1t, would be stable with the same settings with my 90nm 3000+. Unfortunately it wasn't. The message here being that memtest is less of a good indicator of on-die memory controller stability and it could falsely lead you into believing the memory is stable, and I am not talking peanuts in terms of HTT, there is a huge gap as to what my memory is capable of and what memtest reported as stable with this CPU. Bottom line, use memtest as a very very rough indicator of what stability is. What I find surprising is that this is contrary to my previous experience with memtest when overclocking different platforms, such as prescott, Athlon XP, P4, etc where the gap between actual stability and memtest stability was much lower.

Uhm, don't got a Winchester myself, but a Clawhammer instead. Reason I post is because, I have similiar problems on my Clawhammer, that has been reported by some on their Winchesters.

At 12 x 250mhz / 1ghz HTL - I can make endless passes of M32 with no problem, can even run lots of other CPU benches at the same time, and it all works. 3DMark and other 3D benches, same thing - all works. But, Prime95 at same settings, with or without memory divider - fails within seconds. I can hold the Blend torture for hours at 12 x 235mhz, but it will fail the custom test, at its default settings. Not only does it fail, it sometimes even gives blue screen to boot. The RAM can pass memory86 tests(usually only use nr 5) at 245 - 250mhz CL2.2.2.5 T1 @ 3.5v, but is unstable at same settings in WinXP. Never done any extensive burn in of the RAM.

Now, here is the really "fun" thing with my Clawhammer:

At 12x200mhz 1:1 CL2.2.2.5 T1 3.5v, I fail a custom 800mb test... Windows gives "error window" and program shuts down. Same results, if I use 1x512mb or usual 2x512mb setup. Basicly, I can bench all day to 3ghz and some beyond, but Prime95 really hates my system. Isn't that what you Winchesters users basicly experience?

Just wanted to add all of this to frey, just to point out that problem could be more than memory controller related, as I believe the 3700+ has the best single channel controller out there. From results above, my controller should suck, but it doesn't as far as anything but Prime95 is related.

I think, at least on my system, that the fault lies beyond the controller. I have a very, very messed WinXP install and the whole PC could use some lovin. I am almost sure(I hope), that Prime95 will run well, when everything is cleaned up. If not, well, I never liked Prime95 much anyways. 32M and 3DMark01 running at the same time, is stability enough for me, as I don't run OCed 24/7 - never will and don't have need for it, as I have enough of stock power for my needs, OCing is just for fun.

Results with my 6800GT(410-420mhz/1100-1120mhz): 29.3k - '01, 13.6k - '03, 5.6k '05, and 30s Pi M1 at 12 x 242mhz. So it doesn't at least, have bad performance. I hope VX RAM will improve the above just a notch to 30k/14k/6k and 29s respectively :)

Here's my response from the Prime 95 author.




Hi,

At 07:22 PM 12/6/2004, alvin hayek wrote:
>I am a member of a computer
>enthusiast forum called xtremesystems.org. We at xs
>strongly believe there is a major compatiblity flaw/
>or hardware flaw on the aforementioned. When running
>Prime95 torture tests on the A64 Winchester (90nm) We
>are experiencing errors at stock cpu speeds. Is there
>a way to revise the software for the new processor, or
>is this definitely a hardware flaw on AMD's part.

There are no known problems with prime95 and AMD64 parts.
You could have bad CPUs, but more likely it is memory. Bad
motherboards,
power supplies, or overheating could also be the culprit.

Good luck,
George

Quote:

---

Originally Posted by trans am

Here's my response from the Prime 95 author.




Hi,

At 07:22 PM 12/6/2004, alvin hayek wrote:
>I am a member of a computer
>enthusiast forum called xtremesystems.org. We at xs
>strongly believe there is a major compatiblity flaw/
>or hardware flaw on the aforementioned. When running
>Prime95 torture tests on the A64 Winchester (90nm) We
>are experiencing errors at stock cpu speeds. Is there
>a way to revise the software for the new processor, or
>is this definitely a hardware flaw on AMD's part.

There are no known problems with prime95 and AMD64 parts.
You could have bad CPUs, but more likely it is memory. Bad
motherboards,
power supplies, or overheating could also be the culprit.

Good luck,
George

---

Thats what the sensible people have been saying all along.
Nice to get some confirmation :)

Quote:

---

Originally Posted by dnottis

Thats what the sensible people have been saying all along.
Nice to get some confirmation :)

---

That doesn't confirm anything! He said there are no ~KNOWN~ problems.

:stick:

OMG , people get off Prime's back!
there's nothing wrong with Prime95 that won't let Winnies pass Blend. period.
just look for your problems elsewhere, but not in Prime's code.

This is just silly.

Just because you overclock, run super low latencies or do some other tricks and fail prime95 stress test, it doesn't mean prime95 is the culprit.

Look at the most logical explanation: the hardware is unstable, because it was used over the specified limits (or near it).

Scientific calculation programs work in a deterministic manner. If they fail, they will fail in the same situation always. Not chaoticly here and there.

Hardware that is run out of spec does not behave deterministically. It may run ok for 24 hours and then fail a second run just after the first 15 minutes.

If you want your system to be proven 100% stable, then it neeeds to pass prime95 test through several runs of several hours.

No buts or ifs about it.

If you can present a case, where ALL CPUs of certain type at stock speed ALWAYS fail in the SAME manner (same time/test/calculation instance), then there is a cause to belive that the software could be to blame.

Otherwise, youd'be better off looking at the most logical explanation: it's that instability of the hardware, regardless of how unwilling you are to accept it.

Another issue altogether is to consider whether you think prime95 stability is important for you or not.

But it is a separate issue of WHETHER the system is stable to begin with.

regards,
halcyon

Curious...I installed 2x512Mb VX sticks in my NEO2 (3500+ winny), and I'm having the same problem. Prime fails at stock settings even though I'm Memtest stable at 260x11. Previously I had no problem running 2x256MB Mushkin L2 BH5 at 265x10 or 253x11 (my cpu maxes out at 2.8GHz Prime stable).

Well, these are all good arguments, but no ones knows what the issue really is. When Prescotts (90nm) first came out P95 blend would not run on them. A new revision of P95 came out and guess what? It ras fine. Same goes for P95 and 1gb of memory. It would not run reliably with 1gb and a new revision fixed that.

My point is that we don't know what the issues is. Bottom line is that Winchester 90nm will not run P95 blend relaibly.

I have had it running for 8 hours and it as had no errors. It loads the CPU 100%, so I would say this is a good test.

My setup is MSI Neo2, 3500+ 90nm, OCZ PC3200 Platinum Rev 2 @ 2-2-2-10, 200fsb and stock multi. I will report back on Prime95 blend at these speeds.

don't you think that it would be kind of odd that Prime95 would run on overclocked machine but not run reliably on stock machine? i find it really weird :stick:
that's why i would advise to everyone that has this problem to poke around with a multimeter and see maybe at stock settings (after LOAD DEFAULTS) vcore set is just not high enough to run Prime95...

as we speak, i'm on another ~10hr run of Prime95 Blend with vcore lowered to 1.75v REAL and vdimm lowered to 3.76v REAL @ 260x10 6-2-2-2.0 - just about to finish one full loop...

Quote:

---

Originally Posted by andyOCZ

Well, these are all good arguments, but no ones knows what the issue really is. When Prescotts (90nm) first came out P95 blend would not run on them. A new revision of P95 came out and guess what? It ras fine. Same goes for P95 and 1gb of memory. It would not run reliably with 1gb and a new revision fixed that.

My point is that we don't know what the issue is. Bottom line is that Winchester 90nm will not run P95 blend relaibly.

I have had it running for 8 hours and it as had no errors. It loads the CPU 100%, so I would say this is a good test.

My setup is MSI Neo2, 3500+ 90nm, OCZ PC3200 Platinum Rev 2 @ 2-2-2-10, 200fsb and stock multi. I will report back on Prime95 blend at these speeds.

---

Ok, Prime95 blends failed in 4 minutes with the PC3200 Plat Rev2 . This is a proven DC kit and works perfectly in P95 on my DFI 754 rig running both sticks at 255fsb or at 200fsb with the same timings/voltage.

The P95 results are identical to the same RAM/CPU on the GB K8NS Ultra-939 board.

Quote:

---

Originally Posted by andyOCZ

Ok, Prime95 blends failed in 4 minutes with the PC3200 Plat Rev2 . This is a proven DC kit and works perfectly in P95 on my DFI 754 rig running both sticks at 255fsb or at 200fsb with the same timings/voltage.

The P95 results are identical to the same RAM/CPU on the GB K8NS Ultra-939 board.

---

if you're not using DDR Booster checck vdimm with a multimeter. i had this odd issue that without DDR Booster even if i set vdimm in BIOS to 2.85v it would NOT be 2.80v but 2.65v when i run DDR400 6-2-2-2.0!!! it was just fine with 7-3-3-2.0 !!!

andy, what is the date on that winnie?

i've tried with one stick of 512PQI memory but still prime fails in no time.
Said that because some people said that prime95 can fine on winchesters with less than 512mb memory...

Quote:

---

Originally Posted by Beenthere

The latest version of Prime 95 works fine.

http://www.houseofhelp.com/forums/sh...ad.php?t=22249

---

So basically, this would mean I'm an idiot a few posts back? Right. :(

Sorry trans, I forgot about the total memory usage! You'd think with the clattering of the heads of my hard drive I would have remembered this... Doh!

blend=bad. custom=good.

Quote:

---

Originally Posted by halcyon

If you can present a case, where ALL CPUs of certain type at stock speed ALWAYS fail in the SAME manner (same time/test/calculation instance), then there is a cause to belive that the software could be to blame.

---

Exactly what the logical people in this thread keep saying. If the majority of people can run their Winchester Prime stable at stock for a few hours, you have already disproved the software conflict. If the majority only get errors when overclocking then the problem is... from overclocking! :stick:

Quote:

---

Originally Posted by andyOCZ

Ok, Prime95 blends failed in 4 minutes with the PC3200 Plat Rev2 . This is a proven DC kit and works perfectly in P95 on my DFI 754 rig running both sticks at 255fsb or at 200fsb with the same timings/voltage.

---

So the mem controller running in dual channel can't keep up. :shrug: That's not a big jump in logic.

Quote:

---

Originally Posted by Byron

i've tried with one stick of 512PQI memory but still prime fails in no time.
Said that because some people said that prime95 can fine on winchesters with less than 512mb memory...

---

Board conflict, bad mem controller or bad stick? I have been having issues with the PQI myself. XP platform would not run it stock stable. A64 CH would not either. Switched the A64 chip and, no surprise the problem fixed itself - so it was a bad memory controller on the first and probably a board conflict on the XP system.

Quote:

---

Originally Posted by Tek

At 12x200mhz 1:1 CL2.2.2.5 T1 3.5v "Prime fails"

---

So your memory can't handle 3.5v or just maybe it's damaged from running it at 3.5v You've already said it's unstable in XP.

Quote:

---

Originally Posted by bachus_anonym

if you're not using DDR Booster checck vdimm with a multimeter. i had this odd issue that without DDR Booster even if i set vdimm in BIOS to 2.85v it would NOT be 2.80v but 2.65v when i run DDR400 6-2-2-2.0!!! it was just fine with 7-3-3-2.0 !!!

---

its weird but when i set it to 2.85 volts in bios, msi corecentre would record it as only 2.80v!!

strange, whacked it up to 2.85v in coretester and will retest stability now :)

I can prime stable with my 3200 @ 240 x 10 and 267 x 9 at stock volts for 12+ hours. However over 2.5G I can prime for 6 hours, then crap out at 10 mins. If it's the memory controller why can I consistently prime at 267 x 9? As long as my CPU doesn't get over 2.5G it seems to Prime just fine, it has more to do with overall clock speed than the HT. It's once I hit 2.5G that things become unstable, (and it doesn't matter how 250 x 10, 278 x 9) it has nothing to do with the memory controller, the CPU despite what people are saying, is just generating errors once it's clocked too high.

..or it's the chipset generating errors, voltage dips, power issues, memory issues, timings too tight, etc, etc if you are stock clocked and experiencing issues with Prime - or believe it or not maybe it's just a bad CPU. Of the 4 winchester CPUs I've had one had issues priming at stock speeds - the other 3 have not. So was it Primes fault that one CPU had problems, hell no, it was a bad CPU that got RMA'd and the replacement worked fine.

WiCKeD,
my mem is fine and my board, can run dual channel at same speeds no problems.
Must be prime problem, because system is rock solid, can run all benchmark programs etc, or bad mem controller... but if the mem controller is bad, then why it my system is stable?!

Quote:

---

Originally Posted by dnottis

I can prime stable with my 3200 @ 240 x 10 and 267 x 9 at stock volts for 12+ hours. However over 2.5G I can prime for 6 hours, then crap out at 10 mins. If it's the memory controller why can I consistently prime at 267 x 9? As long as my CPU doesn't get over 2.5G it seems to Prime just fine, it has more to do with overall clock speed than the HT. It's once I hit 2.5G that things become unstable, (and it doesn't matter how 250 x 10, 278 x 9) it has nothing to do with the memory controller, the CPU despite what people are saying, is just generating errors once it's clocked too high.

..or it's the chipset generating errors, voltage dips, power issues, memory issues, timings too tight, etc, etc if you are stock clocked and experiencing issues with Prime - or believe it or not maybe it's just a bad CPU. Of the 4 winchester CPUs I've had one had issues priming at stock speeds - the other 3 have not. So was it Primes fault that one CPU had problems, hell no, it was a bad CPU that got RMA'd and the replacement worked fine.

---

Yes that may well be, but remember running at 2.4GHz is not the same as running it at 2.5GHz even if both is set to 200MHz HT as the higher overall frequency pushes memory controller harder.

I agree it's most likely a hardware problem with either the chips and/or the motherboard/RAM combo that just can't hack it above certain speeds.

We should get a clearer picture when people start selling their chips for the newer revision winchester cores come Jan/Feb... :banana:

Quote:

---

Originally Posted by trans am

andy, what is the date on that winnie?

---

0437 3500+

Remember I m running TCCD at this point. My VX got shipped to my Idaho address and I am in California. OOPS :banana4:

andyOCZ
Will it not pass blend @ stock speeds?

Quote:

---

Originally Posted by IvanAndreevich

andyOCZ
Will it not pass blend @ stock speeds?

---

No it won't run P95 blend and this is the 3rd motherboard I have had this same processor in an EPOX, GByte and now MSI. I have tried at 200fsb 2-2-2, 200fsb 2.5-3-2, 250fsb 2.5-3-3 with and without the booster, I tried different RAM too. This has to be a CPU/Prime95 issue.

i think this is going to be my last episode of "Prime95 & A64 S939 90nm Winchester" gallery :)
i've been "primed" out of my head recently :D i gave my setup over 45hrs of Prime95 torture within the last two days and i think that this should really convince you guys about one thing...
there is no issue with at least latest batches of Winchester 3200+ on MSI Neo2 Platinum concerning Prime95 stability...

21 hours of Prime95 torture - 2 full loops of Blend and a bit more ;)

260x10=2605MHz @ 1.76v REAL LOAD, DDR520 6-2-2-2.0 @ 3.372v REAL LOAD;OCZ DDR Booster

MSI K8N Neo2 Platinum - Q47353.1 0440A2 S Taiwan
A64 3200+ 90nm "Winchester" - CBBFD 0441UPCW
Thermalright XP-90 w/ YTech 120mm

CLICK for Full Size Screenshot - 2048x1536 397KB

http://www.xtremesystems.org/forums/...chmentid=20767

Quote:

---

Originally Posted by andyOCZ

No it won't run P95 blend and this is the 3rd motherboard I have had this same processor in an EPOX, GByte and now MSI. I have tried at 200fsb 2-2-2, 200fsb 2.5-3-2, 250fsb 2.5-3-3 with and without the booster, I tried different RAM too. This has to be a CPU/Prime95 issue.

---

Andy
I've had 4 Winchesters and one just wouldn't prime at stock either. I RMA'D the CPU for a new one which didn't have the same issue. Maybe you just have a bad CPU - it can happen. These AMD64 parts are so flakey.