Personal though about the whole UTT stuff ...

[wtz54321]

It's this year we start to see some DRAM module made by winbond UTT chip ... Apparently there are some problem now ...
Although the module maker claim and made it to become a high voltage + high speed in official spec ... My personal though is that
this is something , err , ridiculous ... UTT memory chip is even untested from winbond at 2.5V to make sure every single chip is a good one
within it's chip spec , forget about 3.5V plus life time warranty ...
I will not be surprise to see if module made by UTT chip can not work after some weeks or months even at default voltage ...
I will not buy anything that is untested by it's "original" manufacturer ...

Think about it ...

If nvidia made a lot of UTT NF4 chipset , and sold them to motherboard manufacturer ... Someone start to sell these board made by
UTT NF4 chipset , they even claim and gurantee that this board is capable of running at 400mhz fsb and 2.0V chipset voltage just because
they have test it before they ship it to you ...

Will you buy any board made by UTT NF4 chipset like this no matter you want to use 400mhz fsb and 2.0v chipset voltage or not ?

I have a classmate who work for IC packaging and testing company ...
According to what he told me , IC testing is quite complicated and is a very professional process which made it to become one
separate business from IC chip fab ... It's not something like put the chip into final product , running real or testing application , and it's
working good , ok , done ...

As far as I know , IC testing goes through different process compare to real application testing like what we do ... And the IC
testing will require the complete and very detail testing data which is specially designed for testing propurse only from the original
manufacturer which might be a business secret to original manufacturer ...

Since the UTT chip is sold because of it's not tested by professional testing company to save the extra testing cost , I don't think
that the one who buy it will send it back to the IC testing company to do it again ... So they should skip the IC testing level and test it
in application level which might contain uncertainty in it ... Application testing level may not 100% solid no matter how the application is programmed ...

Until now , I did not see First tier module company like kingston or corsair to produce any UTT based module --- within or over the chip spec ...

There must be some reason they did not choose UTT memory ...
For me , I will not buy anything made by UTT chip no matter overclocking or not ...

[The Mofo]

I hear what youre saying. However, since memory is so cheap nowadays, esp DDR-1, winbound is dishing these out to please the low latency crowd without the overhead of testing them. The only reason UTT has a bad rep right now is because of how far a person can take those modules... That person, in a nutshell, is inexperienced. Think back around 2 yrs ago when volt modding was a science. Rarely did you hear about people killing their ram modules. They initiated a cleaner volt mod and were cautious on how far they took their memory. Any joe can go buy a DFI and crank the voltage out of the box. The funny thing about this whole scenerio is those who know they fubar'd their ram arent man enough to admit it, therefore UTT and DFI catch blame for this.. To be honest, i saw this coming.

I do believe that UTT isnt the same as old BH-5... 3.4V is probably the limit on these modules.

[wtz54321]

I hear what youre saying. However, since memory is so cheap nowadays, esp DDR-1, winbound is dishing these out to please the low latency crowd without the overhead of testing them. The only reason UTT has a bad rep right now is because of how far a person can take those modules... That person, in a nutshell, is inexperienced. Think back around 2 yrs ago when volt modding was a science. Rarely did you hear about people killing their ram modules. They initiated a cleaner volt mod and were cautious on how far they took their memory. Any joe can go buy a DFI and crank the voltage out of the box. The funny thing about this whole scenerio is those who know they fubar'd their ram arent man enough to admit it, therefore UTT and DFI catch blame for this.. To be honest, i saw this coming.

I do believe that UTT isnt the same as old BH-5... 3.4V is probably the limit on these modules.

The root problem now is that how to make sure every untested chip from winbond is 100% ok even within the chip spec ? It's impossible for winbond to have 100% yield rare ... There must be some chip which is not 100% ok in chip level testing , even it's only one transistor compare to the other million ones ... And it's possible that application testing or module level testing can not find it ...

[EMC2]

Before you say anything regarding testing... read these

Mushkin testing OCZ testing

And one other thought to ponder... the A64 memory I/O interface is rated for an "absolute maximum" of 2.9V, which is lower than the absolute max rating on the UUT, whic is 3.6V

Not saying there may or may not be some issue with UTT... just some things to think about is all

Peace

[wtz54321]

Before you say anything regarding testing... read these

Mushkin testing OCZ testing

And one other thought to ponder... the A64 memory I/O interface is rated for an "absolute maximum" of 2.9V, which is lower than the absolute max rating on the UUT, whic is 3.6V

Not saying there may or may not be some issue with UTT... just some things to think about is all

Peace

Oh well , does that mean they have very expensive IC testing machine ? My friend told me a good IC memory testing machine cost around 3 miliion US dollars ... That's really expensive ...

Can they post more picture of their IC testing machine instead of just type something on the web page and what's the spec they use for testing to convince all of us that it's not UTT anymore ?

[highoctane]

They would definitely have to test modules in one way or another in order to bin them for speed and timings but I wonder what voltage is used for testing and for how long.

Just like the cpu's you have allot of people doing their suicide runs just trying to see what the max speed possible is but odds are if they left things pushed that far to the edge at quite a bit higher than std voltage cpu life would definitely be shortened but for short periods there's less chance of damage occuring.

[drunkenmaster]

has anyone catagorically seen a document saying ch-5 had the same voltage specifications as bh-5? its spec'd for higher timings for one thing so is already different, afaik only bh-5 was ever really cared about but we all "know" this is rated to 3.6v(i personally haven't seen the document, nor have i heard from people that have seen the official ratings under what cirumstances 3.6v is the max rated voltage ). utt is not bh-5, its nothing, anything can be untested, its an abbreviating not a product name, there is ch-5 and bh-5 available in these sticks from what i've read and seen.

every single cpu/gfx, everything that has a die shrink that i can recall has lower specified voltages, so is it safe to assume that ch-5, based on the 0.13 process as opposed to the higher process of bh-5(0.15 or 0.18 i can't remember) , has lower tolerance to voltage? in my opinion yes.

as for making a glaring statement on a forum about not purchasing something that is untested blah blah blah. well, its simply a case of, for instance its a what, 3 or more year older process, its been made in mass quantity during that time, yeilds over that time became very high. its simply a case of they can test it after production and add cost, but everyone is perfectly happy to buy "untested" modules. if you get one of the few that are faulty you send them back, its cheaper for US and for THEM to do it this way round. cos instead of testing every single chip to leave the production line, as in hundreds of thousands, and hundreds of thousands of hours testing them when added together, simply replace the 1/2% that come back faulty.

Also the chips themselves, assuming the 3.6v isn't a rumour, are rated to that, where has anyone read that the pcb's that say twinmos use have thick enough traces and are designed for those voltages? two companies have mad higher voltage rated products, not all of them. who says the resitors and other bits on the pcb's are rated for that high voltages. no one, i for one have bought bh-5, ch-5 and new UTT versions of both, they all work well, i've overvolted them all and after throwing 3.8v into them i have no intention of sending them for rma if they die(which they haven't).

i got cheap memory, if you want to pay £200 instead of £80 for the same chips but extra specially tested, instead of risking having to spend a week rma'ing a faulty set to get a good one, thats up to you. 99% of the customers buying it are very happy to save the money, as am i, we are happy companies pass on price cuts and tell us why and how they make them possible and make us aware of any possible problems.

an nf4 chipset is worth about $30, $50 off for the sli version, cost to the board makers, each single chip off bh-5 or ch-5 is worth a max of like $3. thats the difference here. if the memory was worth 15x as much then testing it , costing the same price to test it , makes up a much smaller percentage of the final cost.

[[cTx]Philosophy]

my speed premiums have bh-5 stamped on the chips itself

[EMC2]

Look at the docs yourself for Winbond CH and for AMD CPUs ... Section 7 of both docs

Peace

[wtz54321]

Look at the docs yourself for Winbond CH and for AMD CPUs ... Section 7 of both docs

Peace

So , it's possible that not the voltage kill the UTT memory ? From the spec point of view , A64 memory controller should die if the voltage exceed what they can handle

I have notice that it's absolutely maximum rating , but it does not say for how long ?

There is two more interesting thing in the CH5 datasheet

1 . 70 degree celcius absolutely maximum rating
2 . It can handle 1W power dissipation per chip only

The power consumption of digital logic circuit will be Power=f * C * VCC*VCC
We have frequency from 200mhz up to 250mhz now , it's 1.25 times more than original spec ...We have the same Capacitance load in the same condition ... The VCC rise to 3.5V from 2.5V now , 1.4 times than original spec ... So the power of each chip will be 1.25 * 1.4 * 1.4 = 2.45 times now ....

Power = I * V , I = Power / V

1W / 2.5V = 400mA per chip can handle in it's absolutely maximum rating ... if the working VCC is 3.5V , apparently the current per chip can handle is 285mA only ...

In the CH5 datasheet page 10 , it says the random read maximum current will be 270mA , that should be at 2.5V + DDR400 setting ...

We are talking about 3.5V , DDR500 operation ... Unless it's only 15mA increase from 2.5V DDR400 enviroment ... It seems that only 285mA per chip @ 3.5V is not possible ...

270mA * 2.5V = 0.675W , if we raise to 250mhz and 3.5V , 0.675W * 2.45 = 1.65375W = 3.5V * I ... The random read maximum current will be 472.5mA at 3.5V ...

[MUCHO]

I'm sorry I don't have any scientific evidence but I've not heard of many people who had their BH5 randomly die on them (except on the DFI board) but I have had many people (including myself) who have had their UTT working one day and then suddenly not.

My own example I ran BH5 at 3.2 volts for months without issues.

I ran UTT for one month at 3.2 volts and one stick would not post.

Both had active cooling.

[wtz54321]

Hello guys ... I just ask my friend working for chip testing company once
again , apparently I am wrong to think that the UTT chip is totally
untested ...

He told me the biggest difference of these so called UTT chip and normal chip
is not the Wafer Level Funtional Testing , every packaged chip comes from
the fab will pass the wafer level probe dc parameter functional test ... He
send me a link for me to explain the difference between UTT chip and normal
chip .... And it's quite a big surprise to me because this article say that who use
these UTT chip is so called " Downgrade module manufacturers "
... And apparently the whole UTT stuff is already exist in the module
industry for a long time ... Check this page , you will be surprise to see what
it says ...

http://www.simmtester.com/page/news/...Testing&num=83

[wtz54321]

Some very important information here which might be the root cause of the UTT death problem ...

In the previous article about the difference of module made by UTT chip and normal chip are

1 . Chip(packaged) Function Test
2 . Chip(packaged) Level Burn-IN
3 . Chip(packaged) Final Function test after burn in

Now , Let's take a look at this powerpoint document , page 7

http://humanresources.web.cern.ch/Hu...5Apr02_2.ppt#7

This article explain clearly that it's a normal situation for semiconductor device to have 1% ~ 5% failure rate in the first 1000 hrs ...

That means even every chip is tested ok in wafer level and module level , a noticable percent of them may still fail within the first 1000 hrs ...

Key point now ...

The three process are not done with UTT chip , UTT chips are sent directly to module assembly ...

I think the highend module maker using UTT chip will still spend some extra money and do process 1 after they got these UTT chip from chip company to ensure these chip will meet of the original spec ...

The problem is " Is the chip level burn-in well done to find out these chip that may fail quickly before they are assembled into module "

If the anwser is " NO " , it's no wonder that so many UTT dead case here as user who buy these module are doing the " BURN-IN machine job " for these module ...

And Over Voltage + Over Clocking + Extra Heat will accelerate the burn in process to find out these fail quickly chip that may still fail after a longer period in normal situation ...

And the highest possible failure rate of these module will be 16 times higher than the chip failure rate becasue once there is one of the total 16 chips on the stick is bad , the whole module will be " MALFUNCTION ONE "

And that's based on the estimation that the chip is designed 100% ok for the 3.5V , 250mhz , 2-2-5-2 working enviroment ...

If the chip is not ok with these " not guranteed by original chip company spec " , the failure rate will be even higher !!! (Take a look at the failure rate cure becasue of bad design chip in the powerpoint file)

That's my 2 cents about the root cause of the UTT death mystery ... And I believe it's very close to the truth if is't not exactly the anwser

Let's wait and see what's the finla answer !!!

[xs64]

Interresting topic , keep it up guys

[EMC2]

--- wtz ---

Be careful when reading datasheets The 70C temp rating is external environmental temperature (say the case temp of your system) and is for when operating the devices at worst case conditions (highest power consumption in this case... no pun intended). The 1W power rating is for that same worst case condition of 70C.

All of it put together (power & temp) is based on keeping the junction temperature of the actual semiconductor device below a certain temperature (usually 125C for ICs). So, if your case is 40C instead of 70C, the parts can handle more than 1W. If your cooling is better than a bare chip in no air flow, it can handle more than 1W. Hence the use of fans on the chips when we OC and well ventilated cases

You also need to be careful on power calcs... all the current isn't dynamic current (based on frequency)...

Regarding the testing side of things... accelerated burn-in when properly done can be accomplished in 24 hours to get past the 90% point of the infant mortality stage. 160 hour burn-in is the point under less stressful conditions. BTW... here's a link that will start to explain infant mortality and burn-in processes to greatly reduce them in products. Notice the failure % without burn-in btw... Figure 4 of the link.

Peace

[wtz54321]

--- wtz ---

Be careful when reading datasheets The 70C temp rating is external environmental temperature (say the case temp of your system) and is for when operating the devices at worst case conditions (highest power consumption in this case... no pun intended). The 1W power rating is for that same worst case condition of 70C.

All of it put together (power & temp) is based on keeping the junction temperature of the actual semiconductor device below a certain temperature (usually 125C for ICs). So, if your case is 40C instead of 70C, the parts can handle more than 1W. If your cooling is better than a bare chip in no air flow, it can handle more than 1W. Hence the use of fans on the chips when we OC and well ventilated cases

You also need to be careful on power calcs... all the current isn't dynamic current (based on frequency)...

Regarding the testing side of things... accelerated burn-in when properly done can be accomplished in 24 hours to get past the 90% point of the infant mortality stage. 160 hour burn-in is the point under less stressful conditions. BTW... here's a link that will start to explain infant mortality and burn-in processes to greatly reduce them in products. Notice the failure % without burn-in btw... Figure 4 of the link.

Peace

Thanks for the correction about the temperature and working current

But I did not get it about what you talk about the burn-in process ...

If the burin-in device failure rate is 0.5% , non-burn-in device failure rate is 1% , but it's chip level ... It's possible that 8% difference will be presented in module level ...

Could you explain more about what you thought of the burn in failure rate ?

[Tulatin]

Hmm. personal thought is seeing the recent increase in death of these modules as a combination of things.

As you've all said, the quick death rate is a combination of PCB, Chip and memory design, coupled with inexeienced users pushing hardware past their abilities. Of course UnTesTed modules are going to burn out at 3.6V when they're being pushed to a full load all the time with zippo airflow over them. While it may be true that on some boards (DFI at the moment seems to be encountering this issue) the spikes may be the reason for the season, i'm starting to think it's more module issues and user issues. After all - just how many have burned out BH5/CH5 or TCCD? I'm pretty sure the people here who actually have that stuff, that clocks damn high protect it like family, children and attached organs. Airflow over said modules and their voltage regulators is a must. Perhaps there should be a little sticker on all UTT memory: ACTIVE COOLING REQUIRED WHEN OVERCLOCKING!

Just my two cents though...

[wtz54321]

Hmm. personal thought is seeing the recent increase in death of these modules as a combination of things.

As you've all said, the quick death rate is a combination of PCB, Chip and memory design, coupled with inexeienced users pushing hardware past their abilities. Of course UnTesTed modules are going to burn out at 3.6V when they're being pushed to a full load all the time with zippo airflow over them. While it may be true that on some boards (DFI at the moment seems to be encountering this issue) the spikes may be the reason for the season, i'm starting to think it's more module issues and user issues. After all - just how many have burned out BH5/CH5 or TCCD? I'm pretty sure the people here who actually have that stuff, that clocks damn high protect it like family, children and attached organs. Airflow over said modules and their voltage regulators is a must. Perhaps there should be a little sticker on all UTT memory: ACTIVE COOLING REQUIRED WHEN OVERCLOCKING!

Just my two cents though...

That's what I thought about the whole disaster ... It's not possible that the
board is the only reason that kill the memory when there is other board and
DDR booster kill the memory as well ... I did not say that all the death is not
related to the motherboard , but from the reported case ... We can see
clearly that these UTT chip made module is much more easy to be burned
compare to the old BH chip ... If the root cause is that those chip that may
fail quickly did not found by chip company , that explain what we see now
perfectly because those BH5 chip that may fail quickly has already been
sorting out before they put onto the module ... So the old BH5 module that
we could buy will mostly base on the low failure rate chip ... That's in the
3.5V will not kill the memory situation ... If the new UTT chip is not able to
handle 3.5V in it's design or fabrication process , we know what will appened
in the next ...

[Crankster]

I have heard of little UTT dieing under a NEO2 for example.
IMO the core of the problem is the ADMITTEDLY faulty design of the DFI 5V jumper.
Read around here and you'll find the fix.

[wtz54321]

I have heard of little UTT dieing under a NEO2 for example.
IMO the core of the problem is the ADMITTEDLY faulty design of the DFI 5V jumper.
Read around here and you'll find the fix.

Can you show us the thread about where it's confirmed , I did not see it here ~~

[wtz54321]

UTT death with 3.3V jumper location does exist ...
http://www.xtremesystems.org/forums/...541#post912541

[wtz54321]

I found the industrial information about the burn in condition of memory chip ... Apparently it's not possible for module maker to do some testing like this ...

http://www.wbc-europe.com/services/p...1043242062.pdf
http://www.jianyeradio.com/English/chanpin/laohua7.htm
http://www.avt.et.tu-dresden.de/FILE...r-Infineon.pdf
http://www.parc.com/research/publica...files/5318.pdf
http://www.parc.com/research/publica...files/5318.pdf

125 degree celcius is not possible for module level to do such kind of testing ...

[bachus_anonym]

Honestly, I've been on the same side of the barricade as you since we started seeing OCZ VX dying. I've said too much about already
I totally do not agree on blaming boards for it's death. To me it's simple - Winbond UTT long-term reliablity and tolerance to higher voltage is questionable. They clock very nicely, bu just too many of them die pre-maturely. Or at least, they are useless and need to be RMA-ed.

[STEvil]

out of all the UTT i've had only one stick has gone south on me, and it was based on CH-UTT.

I've got two BH-UTT's that have been running at 3.3v+ for approximately 8,000+ hours (2 months near 24/7 operation).

I would tend to agree that the ones dieing are just the 1-5% (not all wafers are equal) that would usually die.. although CH-UTT seem to be far more prone to this than BH-UTT. Most likely due to the smaller production process.

[masterofpuppets]

I've got tons of BH-5 (including old-school Winbond BH-5 dies, and new BH-UTT). They all run at insane voltages, no deaths or boot problems yet, and they memtest for hours with no errors. Meh.