most likely yes.
Printable View
most likely yes.
i came along this thread and issue yesterday...
Ok guys....is ANYONE being "scientific" here ?
What i want to say...please lets exclude making *assumptions* but instead bring some system in this debate - otherwise its worthless.
I have to admit i did NOT get any valuable information off this debate - because one *assumed* fact usually gets disproved in a following message.
One person seems to have a faulty A64 core, week 37...swaps a later core in which works. Which proves ?
"All early A64 winchesters up to week 44 are faulty " ? <-- of course NOT !!!
I see people claiming Week 37 chips are defective - just to read two messages later another person has a week 37 chip which overclocks to 2700Mhz and runs prime.
I see people saying cores from week 44 and up are ok...then i see some people having problems with these chips too !
ALSO:
I see TOO MANY people in this thread having the MSI K8 Neo2 board, i see too many people all having the OCZ PSU.
If you/we want "scientific" data whats really going on we need to look at this from a broader base with (god forbid !) various hardware, boards and PSUs and put this all together, sorted after core (week) and used hardware. Make a list, poll etc...
Also...it is not clear whether only 3000, 3200 or 3500 are affected...*if at all*.
HOWEVER - not being able to exclude it *is* a fault with prime95 (which i just dont KNOW) - i also strongly think that a CPU which fails Prime95 at stock is FAULTY.
I use(d) P95 for years to determine my overclocks - and the thought of a brandnew CPU wouldnt be able to run this (at stock) is , well, not cool.
---
>>"All early A64 winchesters up to week 44 are faulty " ? <-- of course NOT !!!
No. A BUNCH of people have prime failing @ stock with older weeks of Winchesters. "A bunch of older Winchesters may be faulty" <-- YES
>>I see TOO MANY people in this thread having the MSI K8 Neo2 board
Not all. Not only nForce3 either.
>>i see too many people all having the OCZ PSU.
Definitely not all!
the much MORE logical assumption would be that memtest86 is fine, so is your memory.Quote:
Originally Posted by n0w4i4
ALSO..you shouldnt use memtest to test your CPU....because then it would not be called, well, guess it, "MEMTEST".
On ther same level are the people who use Prime, Prime fails, and THEN wanting to RMA theyr memory because Prime had errors :) The same story but backwards.
Not without a reason i always say overclocking (and systems testing for that matter) is much more complex than what MANY people believe.
Some people dont even get the basics right, use the wrong tools etc...but then going online and making statements about faulty hardware, faulty tools etc.
Sorry, i dont want to sound harsh..but this happens a lot...this thread probably is the best example. (People calling OCZ because P95 fails etc..)
greetings !
just to add something to this (very interesting) thread:
You would also have to eliminate all people who (as it seems) use(d) wrong Bios settings, wrong memory timings. Another person even had a faulty stick of memory.
There are MANY reasons why Prime could fail.
Also, AndyOCZ cannot make an assumption that "a bunch" of A64 fails because of a "hardware flaw" - based on the fact that people called in for RMA and complained about problems with their system and Prime.
Because, also, "a bunch" of people probably also had their systems set up wrongly (wrong memory timings etc.)..we dont know for sure WHY they couldnt run Prime !
How many of these people do we KNOW that it was actually really their CPU which causes the problem - and not bios settings or EVRYTHING else ?
And then the number of people who swapped out a "faulty" core for another one (in a self test at home :) ) - is this number significant enough..or is it only a handful people doing some testing and then making a general statement ?
Dont get me wrong...i have no business to defend AMD...i have a week 37 core 3500+ myself - cant test since i still wait for my motherboard :)
I just think we need to be more organized here to get to the "core" (literally) of the problem...
ugh.. this is annoying. Sure some people it's their fault it's failing. This is not some random forum tho. Most people here know what they are talking about to say the least. We know how to overclock and we can determine to a logical degree what is causing problems. It's just annoying how so many people (amd being 'one' of them lol) act like we don't know what we are doing and just up the fsb to 350 and get mad when it dosn't work.
I mean, ok, so some people it is a user related fault. But that dosn't change the fact that there is a known and well confirmed hardware related issue here.
Oh well, ill probably buy one if i get the chance anyway. :) Do i really want to wait for the 'Venice' cores sheduled for late january? hell no
Hi, I understand your concern. Be assured that an OCZ PSU and MSI Neo2 motherboards are NOT causing P95 issues. I did VERY scientific testing, starting at 200fsb with several known good kits of high quality memory on the Gigabyte K8NS 939 Ultra and the MSI board. By verifying that my personal week 37 90nm A64 failed P95 blend at 200fsb I dupicated what my what my customers had experienced. I requested that they run P95 blend test at 200fsb to verify that they had a good CPU/Mobo before they asked for an RMA. I didn't try and convince anyone that they had a bad CPU.Quote:
Originally Posted by flexy
Prime95 was verified as good and NOT causing the issue. I went out with my own money to Fry's and purchased a 3200+ Winchester week 41 that runs P95 for as long as I like at 200fsb and at 250fsb. The week 37 CPU I have that failed P95 blend was a 3500+ model.
If one understands the way CPU's are manufactured and binned for sale, one would realize that the rated speed has little to do with this issue. The CPU's are assembled and then tested for voltage and speed before the are determined to be a 3500+. 3200+, 3000+, etc., making the final speed meaningless. The finished cores that are used in assembly for a given week are made prior and can span several weeks of production, making it impossible to know what "batch" the cores came from or exactly what week is causing an issue (at least for the end user).
As far as the PSU goes I doubt that using even a cheap overated model would cause errors at 200fsb. Once again I scientifically proved that the PSU was not an issue due to the fact that it was used with the failing and passing CPU's along with the same 2 boards.
This thread does require careful reading as not all of the posters did this kind of testing. There are those that did and they know they have a CPU that will not run P95. If we were to say that it was only 3 CPU's with this issue out of our sampling here (although many more are convinced their CPU's are faulty), and then factor that percentage over how many Winchesters AMD sold with these cores, we would I am sure be able to deduce that a large number are floating around on shelves and in computers.
Make of this thread what you wish, but rest assured that there are 90nm Winchester core CPU's that will not run P95 at 200fsb with loose memory timings and the agressive timings setting off. Once agian this was "scientifically" proven and not guesswork.
Read my last post. I do not assume, thank you.Quote:
Originally Posted by flexy
Can anyone answer this? :)Quote:
Originally Posted by ZeroSanity
I didn't test much more than Prime95 on my 3500+ before sending it back to Newegg for an RMA.
At this point the earlier weeks are showing up less often and we have pretty much decided that not all week 37 (or earlier) CPU's were bad. Some have posted that they were otherwise stable, even though P95 failed. Of course we all have different requirements and interpertations of stability. Given that I purchased a very good week 41 Winchester recently I would say that you should be safe. It's up to you
>>>
Some have posted that they were otherwise stable, even though P95 failed.
>>>
how did they determine this stability ? Which other program does validation-checks for FFP calculations etc....comparing calculations to a list like prime does ?
I
mine has been 100% stable in everything, stock or overclocked (obvioulsy to a point) in all things includeing p95. Its a 0444 chip
SuperPi I think, and umm... there's on other that calculates pi I forget what it's called cause I never use it.Quote:
Originally Posted by flexy123
After lots of SuperPI and Prime95 testing I've found I can get a stable system at 2-3-3-5 1T using stock speeds and stock volts. I can't however get a stable system at my rams rated 2-2-2-5 :(, I even tried uping the Vdimm to 2.85 and down to 2.6 but nothing helps.
I've got my Newcastle 3500 coming tomorrow so I'll be able to verify it's the chip and not something else.
BTW when SuperPI or Prime95 has failed I've been getting occasional blue screens and other crashes so I can't just ignore Prime95.
Verification of Prime95 comes through changing only the CPU with another "newer" Winchester and having it run for hours. I verified it.Quote:
Originally Posted by flexy123
hi,Quote:
Originally Posted by andyOCZ
yes..i meant what program they use to verify the stability. (Knowing that P95 compares to internal values). I dont know whether other programs do that, eg. SuperPI.....if these other programs only use the CPU and calculate, but do NOT compare the calculations to a list - how would these people know that their system is stable ?.
There are countless programs around which put extreme load on the CPU, but thats about it. You cannot make a statement about stability (also meaning: CPU calculates right) just by running a program without an obvious crash. (Like MANY do btw and then claim that their systems are stable :)
We (as overclockers) KNOW that of course :)
Thats why i asked: "Which otehr program does such validation checks like P95 does ?"
---
but regarding your comment...yes...its clear. You had a week 37 core and replaced it w/ a week 44 and it worked.
This itself only proves that YOUR week 37 core *possibly* was faulty - you cant take this to make a general assumption.
EXCEPT if we had TONS of people who did the same and can verify this. And not based on ONE (or a handful) persons's experience.
If we want a "scientific" approach we'd need a LOT of week XYZ core owners who did the same testing and can confirm that the errors were, indeed, solely based on the CPU revision they used.
Just to eliminate windows driver or stability issues and programs in the background, you might try the mprime 23.9 in a bootable disk. http://www.mersenneforum.org/zip/ download the latest version and either use rawrite to write that image to a disk, or burn a bootable cd (easily done under nero or cdrwin) with that boot image, if don't touch anything, a blend 3 test will run. Good luck all.
marek_steele: Gonna try that, thought it was on three floppys, but it's actually three diff versions.
Got it working, at stock settings it failed after 1 min ;). Gonna try 2-3-3-5 1T again which I consider stable.
Sorry,Quote:
Originally Posted by flexy
Some use Super PI as this calculates also. Most just game and run normal apps to check stability. 3dMark 2001 is good. You can run that in a loop for hours to check stabilty.
You are right about having needing many more CPU's to test. I wish I had a few trays. :) The reallity is that we have some that have this issue and some that don't. Anyway XS is not real scientific in general. We are all too passionate about our rigs. ;)
@flexy
You mean memtest just only tests my memory ? Ok then tell me how memtest "communicates" with system and how it tests RAM? YES it has to go thru the Memory Controller to test my Memory ? right ? And the Memory Controller sits in my CPU, so some parts of the CPU SHOULD be tested by Memtest, but it doesn´t !
n0w, it doents stress the cpu like other programs do. Yes it test the memory controller, but not very much at all... its just passing the data through it.. Remember that memtest just moves the data around on the memory, it doesnt do any real calculations like prime or super pi
Ok that sounds right, but I thought it tests the MC aswell. Would be a nice Option for a next version of memtest (enhanced) :) !?
There is no way to isolate just the memory in any test. If you have your CPU overclocked too high you will fail tests 1 and 2 in memtest86. You may also fail test 5. If your memory controller is on the ragged edge you will fail all RAM tests. The memory controller handles everything that the memory reads, writes or moves.Quote:
Originally Posted by n0w4i4
So what should i believe now ? At first i believed the version of andyOCZ, then MaxxxRacer's version and now again andyOCZ's ! :stick:
I also thought that everything must go thru the MC. :rolleyes:
Surely all the memory controller does is make sure the memory is read or written correctly, so memtest will thrash it more as it is spending all its time reading and writing to memory. Anything doing calculations will thrash it less as they are spending time doing the calculations and the data will be in the cache a lot of the time.
BTW this bootable CD for Prime95 worked really well, I had the same results as in Windows so it's not an issue with drivers etc.
Crunchy try one stick of memory at a time and see if it works.
hmm, Mine passes prime if it's stable and it doesn't if it is not
lol, well thats how prime95 works.
My new 0441 Winch passes prime overnight, but once I start using it with Prime in the background it freezes :) Bump vcore 0.025V more and all is good.
n0w4i4: sorry for the confusion. Let me try to explain. As andy said the memory controller handles all the movement of the data in the memory. But the cpu itself is not being stressed which makes it easier on the memory controller. Like my memory controlller works at higher speeds when my cpu is at lower clocks. when i turn up the mhz on the cpu the memory controller dumps.
So if the cpu is unloaded the memory controller will be able to do its job better.. and with less errors. Thats all it is.
Does that make sense?
I also have a week 37 retail winny and it fails prime95. I tested after clearing my CMOS and booting into a fresh install of windows and runnin p95 and it failed. My system also freezes if i try to run any liveCD for linux (installers, knoppix, etc) and even some dos based tools like data lifeguard for WD HD's.
I talked to amd tech about the freezing like a month ago and they said it was most likely a motherboard or ram problem, but i tryed a friends NewCastle and it worked fine. :stick:
aonic
RMA the chip. I bet you'll get some 0447 or even later that will O/C to at least 2.7 GHz :D
Please excuse for my lazyness, but i'm don't have the nerves to read 16+ pages.
Can someone give me a short summery of what's going on? :)
Thank you very much :D
another, related question....
I have an OEM 3500 week 37 here....NOT tested since i am still waiting for a nforce 4 board. the 60 day warranty run out a few days ago. I KNOW.
So what do i do once i see it would fail prime or any apps at stock ?
People say they cant even run Linux installs/kernel compiles etc....so..in IOTHER WORDS: The CPU is f****g broken if it doesnt do basic things under stock speeds !
I cannot RMA an OEM CPU where the warranty ran out (AFAIK !) - so the only option would be ebay to some poor soul ?
What rights do i have when i paid $315 for a CPU which would turn out defective ????? Btw. i bought from monarch. I also asked them after i got the CPU if it is possible to purchase longer warranty afterwards - but they said i cant.
Do i have ANY way to pressure either the retailer or AMD (well..its defective CPU, duh !) in case this CPU turns out a dud ? Need some help here..i would even go so far and write to AMD etc....AFAIK the programmer of P95 already claimed there is nothing wrong with P95 !
>>So what do i do once i see it would fail prime or any apps at stock ?
Swear and buy a retail cpu with 3yr warranty next time for $10 more
Ya, I am going to RMA when the busy shipment season is over :)
Some Winchester 939 90nm CPU's fail Prime95 Blend Test at stock 200fsb speeds, using relaxed chipset timing and RAM timings. These are mostly week 37 models.Quote:
Originally Posted by Swing
If you need to know more read inly the important looking posts. :)
Been doing some Testing on my systems and getting some wierd results...
All was tested on my MSI K8N Neo2 Plat with 1.3 Bios. system is water cooled.
all testing done with Prime95 Ver 23.8
Winchester Core (week 0441)
Passes at 260x10 with my Gskill Tccd 2x256Mb
Fails Prime at stock settings with My Ocz PC3700 Gold Rev3 2x512Mb
Newcastle Core (week 0427)
Fails Prime with and FSB over 220 FSB with My Gskill Tccd 2x256mb
Presently Passing at 237x11 with my Ocz Pc3700 Gold Rev3 2x512Mb
hag6br
Ah.. 2 memory banks workie, 4 memory banks NO workie. Newcastle - why are ya testing it? Just test stock..
Pass WHEN?
Fail WHEN?
Look people, there is no 8 hour magic mark on Prime95.
If your system (mobo, memory, cpu, psu, something else) is unstable, it is more than likely unstable at non-predictable times and not all the time.
This means that there is NO single test turn (esp. only 8 hours long) that can prove your system is 100% stable.
If your system does not fail in two hours, is it stable? Perhaps, but not guaranteed.
If it does not fail in 8 hours, is it still not a 100% guarantee of stability.
If you really want to be sure, run it for at least 3-5 times 12 hours each at different times of day with different loads on the computer and your electricity grid.
The more you run and the less you crash, the higher the likelihood of real system stability.
It's not an ON/OFF feature where your system either is fully unstable or fully stable.
regards,
halcyon
Only reason I tested the Newcastle is that it is a new chip I just picked up and I was seeing how well it overclocked..Quote:
Originally Posted by IvanAndreevich
:cool:
There you go. More evidence of the issue.Quote:
Originally Posted by aonic
Are you all having trouble with only week 37 cpus? Or all CPUs before a certain time? Or is is just random?
I have a 0412 NC and an 0431 Winnie both work just fine so I am just curious.
It's more of a random thing. Not all week 37 CPUs fail and not ane we are not sure what other weeks might be affected.
Anyone tried using the floppy version of Prime95? Just to see.
Download available here.
anyone have any ideas of what i can say to make the amd tech's believe that the cpu is the problem so they let me go through with the RMA?
aonic
You can hmm.. lie? It's your karma :p
DoGMaN
People have tried it to no effect AFAIR
give me some suggestions on what to say :pQuote:
Originally Posted by IvanAndreevich
Ive never RMA'd before lol
you have a boxed retail CPU ? If YES, consider yourself a lucky focker :)Quote:
Originally Posted by aonic
I think unable to run Prime at stock and Linux and whatever should be enough to RMA a retail CPU. (3 years warranty, thjats what is FOR).
In comparison i am REALLY screwed since i got an OEM CPU (with warranty already gone.) If it turns out this CPUI is a piece of :banana::banana::banana::banana:e i cant do ANYTHING than ebaying it and hope the new seller doesnt know about Prime,
There is no way in hell AMD exchanges my OEM CPU (even if 'proven' defective, eg. unable to run some apps at stock)...and there is no way i can get an exchange from monarch. (I even asked them a few weeks ago if i can purchase a longer warranty, they denied my request. I still am unable to even test this CPU lack of parts/motherboard etc.).
In case it turns out my CPU is defective i could have even taken the $315 i paid for it, put it in the toilet and flush it down.
I know i was a fool not to purchase a 6 month or so warranty....but the problem was that i did NOT expect to get the CPU and then being unable to test it for the 60 day warranty period i had. I thought 60 days should be fine and i should get my board and everything. I was wrong. Besides the CPU was expensive enough so i didnt want to pay more for extended warranty.
Still...this is an odd situation.
Man thats low. Knowingly selling a part on ebay you think is defective. Then hoping the buyer doesnt find out. Great.
we could debate whether it is REALLY "defective" - dont forget that even AMD *claims* there are no issues with these CPUs.Quote:
Originally Posted by BeerCan
Its "defective" for me since i always used Prime95 to test stability and therefore am dependant on P95 for overclocking/setting up my system. If it doesnt pass Prime i really have no reference to check my system for overclockability.
The CPU might very well work with every other application.
I am hoping to find a buyer who doesnt care about Prime95 and i know many do.
If you call me "low" because i MIGHT (in case IT IS defective) sell this CPU....what do you think about AMD then giving out defective CPUs, denying that ANYTHING is wrong with it (see INQ article)...?
THEORETIC SCENARIO:
I would VERY, VERY much like to see an AMD representative here, running a system at stock, default settings. NO OVERCLOCKING.
Loading Prime95 and see the CPU give errors, and this VERY same person telling me (and others) in their face "THERE IS NOTHING WRONG WITH THIS CPU"
(I do of course assume that my other components are all working fine, tested, and the CPU failures can be reproduced by putting in another CPU).
Also..if i have another way to echange this CPU i would be the first person to jump on this opportunity.
*** and this all only applies in case it would turn out this week 37 cpu bheaves like that. Right now i am just paranoid and planning beforehand...it could very well be that everything is ok with my cpu ***
also:
If we had a more "scientific" list and statistical data that REALLY a significant number of those CPUs acts like that..this might help too. At least in TRYING to convince eg AMD. that there is something wrong and the fault is actually on THEIR side.
I know enough forums/sites where i/we could make SOME waves and get some reaction from AMD. If they sell CPUs which dont work under certain circumstances they should take these back. The customer should not be the fool at the end.
But:
We dont have such a list. We only have randomness here and not really anything substantial.
Therefore (right now) we would have some problems "proving" to anyone that there is something wrong with these CPUs...thats the problem.
Flexy,
Enough. If your setup Fails P95 when you get it running, then Bi*ch.
Not to be rude, but why complain about something when it hasnt happend yet.
Hrm, can someone explain what this means to me, because I sure have no clue.
My Winchester 3500+ at 250x10 primed without errors for 24 hours (in-place large FFT) at 1.6V. The moment I try 250x10.5, prime trips out and keeps saying "Illegal sumout error" or something like that, but doesn't stop priming! I have to manually stop the test and close prime for it to stop priming.
Anyone have any ideas?
you miss the point. I can see early batches of ANY CPU being worse overclockers than later batches. Thats normal.Quote:
Originally Posted by bachus_anonym
But we are not talking about which CPU/week is the best overclocker.,.BUT about serious (alleged) flaws of chips who cant run what they're supposed to run at F****G STOCK SPEEDS.
This is prime95 at STOCK with normal 200 HTT and specced CPU frequency...or Knoppix linux/Kernel compilations whatseover.
Please dont even mention RMA and overclocking and HTT of 290 in one sentence...otherwise you would lose credibility..not in this thread. Or we would need another thread about "what CPU week is the best overclocker"....but in my opinion the important subject here is (POSSIBLE) hardware flaws of certain AMD chips (WHich STILL has to be determined IMHO).
achaye
That has nothing with the "fail at default" bug being discussed here ;) Do a search on how to use prime OR make a new thread if you don't find anything.
My 3000+ week 42 ran prime blend at 2625 with memory at 269 overnight which was a little over 8 hours.
Very nice. The later weeks seem ok. :toast:Quote:
Originally Posted by MUCHO
for me to pass memtest I have to drop multi to 7.
but test 5 stable at 245 2-2-2-10 1t
Devils Rejection has a week 44 chip that he says has problems, so I guess that hypothesis is bunk too...Quote:
Originally Posted by andyOCZ
yea thanks for backing me up wicked, im going to a newcastle now lol.
That was an attempt at being diplomatic. heheQuote:
Originally Posted by WiCKeD
I have been using an old install of XP and just removing / re-adding drivers when switching boards (from 754 to now)
Standard Prime blend was stable on my winnie to 2.65 Ghz. (100% cpu util at default settings)
Finally, did a clean install a few times over the last few days of the 'winxp virus' trying to get prime to work the same again, but now it can't seem to get 'started' (never stays @ 100% cpu utilization) even @ stock (must reduce mem used from 927 to < 910 mb to go). Same stuff installed as b4. :upset: :bs: :dammit: :brick: :cussing:
This is w/ both xp w/ sp1 and sp2.
Will be experimenting more.... :mad:
Doh, I have just read the WHOLE thread!
This way you will never accomplish anything. You have to be more systematic.
I will suggest that you create another sticky thread called "Report Prime95 failing at stock speed with Winchester core A64 here". That way you would get the idea of how many people are affected.
In my opinnion there are three options here:
1. Since the memory controller is in the CPU I presume it is directly (well, almost) wired to the RAM chips. Put simple, that should mean that the voltage from the RAM chips is getting into the CPU. Maybe this is just some electrical incompatibility between 90nm memory controller and some memory sticks or even chips?
2. If this is not caused by RAM or controller flipping a bit and making calculation wrong, then there must be some assembler instruction or even instruction mix which Prime95 uses that is responsible for the error in calculation. It should be possible to find and isolate that sequence and even run it outside of Prime95 and produce consistent result.
3. Mathematical algorithms are tough stuff especially if they are based on estimation. If you are estimating some parameters and expecting them to convolve there is a small chance that because of error in algorithm caused by rounding, bad compiler optimization, etc the estimation does not convolve every time (i.e. the result is not approaching the right result). That could lead to the program that continues calculating but will never finish because derived value is not even close to the expected or to the program itself aborting calculation if it has built in detection for this case.
Hopefully this will help directing your thoughts toward the solution of this problem.
Btw, I am not affected by this and can't help by testing because I am an Intel fan (Pentium 520 -- LGA775 D0 stepping).
Week 44 Winnie 3200+ Socket939
I can run 3DMark05 at 2.72GHz, BOINC (Seti) at 2.66GHz but for Prime to run stable I need to tune down to 2.55GHz.
I have removed the IHS from the Winchester and the core is tiny (9x10 mm) compared to my older Clawhammer. Maybe its an issue of localized overheating on parts of the die that the IHS is unable to remove on the earlier Winchesters?
Killeroy.
ok this is sooo weird ... noticed that i was prime stable for more than 8hours @ 2430MHz (i canceled the test myself) and then when i switched back to TCCD chips and went @2400 i couldn't prime for more than 1-10 minutes
Quote:
Originally Posted by esdee
Funny you say that, I just had the exact same thing happen. Now I put some BH-5 back in and it passed more than 10 hours (overnight) at 250x10. :stick:
in my first attempt i was using some twinmos UTT chips, aka Winbond ...Quote:
Originally Posted by DoGMaN
Now that's weird!
how many of you guys who are failing Prime are trying to run super low latencies like 2-2-2-5?
Hi, I tested with tight timings, loose timing and different kits of RAM and still failed prime95 on my week 37 Winchester. Testing was done at 200fsb with aggressive timings turned off.Quote:
Originally Posted by scarface
Same here. Tried everything. With one stick of RAM no problems. I have tested the sticks I have up to 235MHz 2.5-3-3-8 in single-channel mode and they passed PRIME95 for 14 hours each. At stock speed, dual-channel mode, it will fail in less than 4 hours regardless of the timings ot memory voltage I set. Sometimes it fails in 5 minutes, sometimes in 3 hours, but it always fails before 4 hours. BTW : it's a crappy week 37 as well :(
Very nice, but what week CPU do you have?Quote:
Originally Posted by misteroadster
0448 Spaw with 55 stock HSF ;)
My 0445 FX55 do the same , incredible. :slobber:
FYI I got the *exact* same rounding error. :oQuote:
Originally Posted by andyOCZ
3500+ Winnie
Prommy Mach 1
OCZ Plat rev 2 2-2-2-5
250fsb
HTT x4
My winnie is being replaced by a newcastle due to cold shutdown error. :(
Hardly "exact" when it only mentions the rounding error returned 0.5 instead of something closer to 0.4 :stick:Quote:
Originally Posted by r00tcause
If you both get something exactly like rounding error 1.25455784558 then it's either coincidence (highly unlikely considering the precision levels), or there is something wrong with Prime95's coding, which is also highly unlikely since the programmers themselves on the Prime95 forum have clearly checked this and explained the problem, ie not in the coding.
or there is something wrong with Prime95's coding, which is also highly unlikely since the programmers themselves on the Prime95 forum have clearly checked this and explained the problem, ie not in the coding.
This thread has gotten very long and its hard to find any of the informative bits heh. You mention Prime95 forums (which I cant seem to find).
Do the Prime95 coders actually acknowledge a problem with the A64 and P95?
Although it's certainly possible for the A64 archeteture to be a little different and maybe not perform some of the same ops exactly the same (especially if its dropping from 64b to 32b precision), i can imagine how this would affect Prime95 only when CPUs are running at certain speeds.
Pi and Prime95 routinely fail while im trying new overclocks. Pi generally fails when the CPU is too high, and Prime95 is more sensetive to memory issues (and more advanced OP codes maybe). However, I can correct these faults by increasing voltages. That makes me think its not a logic error but a standard overclocking error. Prime95 will just pick up on different things that dont make other apps crash. Its handy to have several different apps to test stability. I've written my own burning app which checks math against preformed data (its extremely similar to Pi tho so I just use that).
I'm not really forming an opinion either way, but to me it seems that its not a Prime95 flaw on my system (av8, 3500+). Prime95 is stable even at very high FSBs on my system as long as the system itself is stable. I havent gotten any repeatable errors.
Prime95 warms up the CPU a lot higher than Pi tho, ive noticed that.
That still doesn't explain why PRIME95 fails at stock speeds. And why in single channel mode it works perfectly and fails in dual-channel mode (at least my chip does this).
It's never failed at stock for me in dual-channel. If it was a coding problem, the problem should be fairly consistant and reproducable at the same spot?
My week 37 also stinks pretty bad.Quote:
Originally Posted by andyOCZ
The max OC i get is ok.. (2.6 GHz at 1.7V)... but its not phenomenal... especially b/c I have it watercooled... it still fails prime95 at stock sometimes... not always, though.
I'm thinking of RMAing it. The mem controller on it really weak and has problems pushing my EB RAM.
Hopefully this will be fixed with the newer CPU week... like 50
I just got a new system and it will freeze while running prime 95 (week 44 winnie and neo2 plat.). It's got 1 stick 512 of ddr400 in it. What gives? Should I rma my chip or my mobo or both?
my ADA3500DIK4BI CBBFD 0436XPBW on a MSI Neo2 Plat. is prime stable at stock (2.2Ghz @ 1.4v) and overclocked to 2.3. It's been my experience that Sumout or Rounding errors are due to undervoltage on the cpu. usually a slight bump will solve that.
http://www.xtremesystems.org/forums/...id=22747&stc=1 :cool:
Quote:
Originally Posted by cardnut99668
you should do neither. Its nothing to worry about aslong as your comp doesnt give you anymore problems
1st of all, Wrong Thread. could be a weak power supply.Quote:
Originally Posted by cardnut99668
I've got a 480w thermaltake. Is that too weak? I wasn't sure where to put this. So should I try disconnecting stuff and see if it passes?
Not slamming Thermaltake, but most of thier PSU's are rated too high IMHO. Try running just the basics. A hard drive, CPU fan and nothing else.Quote:
Originally Posted by cardnut99668
I'll give it a shot.
I disconnected everything except the raptor w/ the os, 1 optical drive, my 6800gt gs, and the mobo and it's priming fine. Does this definately mean it's a psu issue? Should I go with the OCZ powerstream 520?
The 520w Powerstream should fix you up fine. 600w is more future proof, but more money. Bummer about your PSU. As you found, with A64 the PSU is critical!
Guys if you have a winni that fails prime could you list the full part and model number for your CPU...Especially the 3rd line of test on the cpu package.
Im helping narrow down if there was an issue with specific weeks cosed etc.
Since I am testing 'full' load (ZoneAlarm security + AV, acoustic edge sound pannel, etc). Other than that normal blend has been running for ...(looks at timer on main PC) 14 hrs @ his settings in sig. :banana: Gonna let it run until the morn (crosses fingers) when I have to 'cleaning lady' safe the PC for the day.
That sounds like a purely-processor-core related problem.Quote:
Originally Posted by saaya
I would check the processor temperature. It's likely overheating.Quote:
Originally Posted by cardnut99668
I tried to read most of the thread and absorb the most important info, I'm pretty sure I got the gist of it but probably missed some details so feel free to correct me. Having said that, I guess my results might kinda throw a monkey wrench in the whole Winchester/week 37 theory, sure glad I got lucky if some are really bad though. This chip was bought in November, take a look:
http://img120.exs.cx/img120/9890/maxstableoc7oy.th.jpg
http://img31.exs.cx/img31/144/maxsta...lftt2qf.th.jpg
I'd crop but I figure some might wanna take a look at the temps or whatnot and I'm too lazy to edit 'em and re-arrange it all. :p: In any case, first one was of my first test with a clean install after settling on 2.3GHz for now (more on that later). As you can see, Prime95 (blend) + looped MP3 + rhtdribl. See sig for system info, timings, vcore, etc.
Second one was a small FTT run as suggested here, it did put the CPU at 100% util which Blend alone didn't. However it didn't hiccup once, had HL2 DL'ing in the back even. The code on my A64's IHS reads as follows:
ADA3000DIK4BI
CBBFD 0437WPHW
1093569I40169
assembled in Malaysia
I mentioned I settled on 2.3GHz for now... Not very high, I know, 'specially by the standards you guys 'round here set. Alas, at 2.4GHz not only was it not Prime95 stable but it just wasn't stable while idling in Windows... It'd just lock up without any errors, rather quickly too. It's odd and I haven't been able to find out why. I've tried just about every setting/timing configuration.
The only things I haven't checked out at 2.4GHz are A) running at 1.4-1.45v vcore (I tried up to 1.65v and it lasted hours as opposed to minutes but was heating up too much) B) 4x HT multi as opposed to my 3x (most tell ya to stay under 1,000 but some have reported more stability actually going over 1k so worth a try).
My temps aren't the best (tropical island, concrete walls, no A/C, oi) but they're not terrible and they weren't any higher at 2.4GH than 2.3GHz so I doubt that's the prob in getting to 2.4GHz. My 12v line is usually pretty low (11.5-ish) and there's not much draw on it. I dunno if my PSU's just weak or the 2-pin Panaflo 92 L1A is drawing too much, I've gotta try having it on a molex just to make sure. Gotta check under the mobo NB HS too...
Anyway, I mention all this just in case it has any relevance, may help some, or in case anyone has any ideas that might get me 2.4GHz... But as you can see there's some week 37s out there that have absolutely zero issues with Prime. I used the latest version straight from the project site.
Thanks to ImageShack for Free Image Hosting
It was never proposed that ALL week 37's were bad. Other weeks up to 41 seem to have similar issues, but fewer in number. The reality is that the cores get made, then assembled and dated, so the cores surely get distrubuted over a more than one week.
If you have a good one, we are all happy for you. :) Thanks for the input.
We have a similar thread going over on the ocforums:
http://www.ocforums.com/showthread.php?t=348124
This problem is most bizarre, and needs clearing up - the Prime95 problems could well be indicative of bad processors that need to be returned. We have established that this is likely a hardware issue, and for what it's worth, I have sent an email to AMD about this.
Having corresponded with AMD several times before, I did my best to word things in order to best guarantee a helpful response. In particular, I want to know if owners of these "bad" chips should be RMAing.
I did not mention the main date stepping (0437) in my email, as this has proven to be inflammatory in my past correspondences: for whatever reason, AMD does not give answers to people who write like overclockers, or imply overclocking, and general stepping curiosity.
I will be sure to post any response to this email, both on the ocforums, and here.Quote:
My Email to AMD
Recently, a large number of Athlon64 processors have proven to have stability issues right out of the box, at 'stock' clock speeds.
Users are finding that their Athlon64 processors (particularly "Winchester" core Athlon64s) are consistantly unable to pass Prime95 CPU stability testing at stock speeds and voltages, even with AMD approved processor heatsinks, and as a group, are wondering why this stability issue exists.
Should these processor owners be returning ("RMA") their apparantly faulty processors? Does AMD know about the source of this problem?
Is there a specific telephone hotline that these users can call for assistance?
Thank you for your time,
You folks are doing some great work over here - good stuff! :)
I will point AMD to this thread, I will get them to look at OCforums also
Quote:
I will point AMD to this thread, please also post a link to the thread over at ocforums so they can go look there.
Here it is again :):
http://www.ocforums.com/showthread.php?t=348124
I have received the first automated response to my email. Basically, it details the processor owner's warranty rights, and gives a telephone number that all of you can call for help :rolleyes:. Typical, I guess my email wording was a little bit too rinky-dink.
I resent it as "unresolved", hopefully we'll get a run-down on the actual source of the problem soon.
I'm wondering, you Neo2 users... does turning "DRAM Drive Strength" to weak do anything to your Prime95 stability?