HYDRA SIEGE Project

[4metals]

Reading this thread made me realize that the issue with producing reliable data is a function of certain variables which we have little control over such as variations within the material (ie. consistency of plating thickness between lots, manufacturers, mil spec or non mil spec. and dates of manufacture), or processing method chosen to recover the values, or even how completely the material was prepared before the refining.

An example of preparation is the often heard term "closely trimmed fingers" A lot of fingers that were trimmed very close can vary by 10 or 15% from identical fingers trimmed "pretty close". That makes the yield per pound vary considerably. We could list our own first hand variables all day on different types of pre treated scrap, so it is easy to understand what I'm saying.

And the last variable concerns the sample. Both Chris and I always stress the importance of the assay and knowing how much gold is in your material and that involves assaying. And the truth of the matter is that a precisely run assay isn't worth anything if the material was improperly sampled.

If there is enough interest, I will start a thread about sampling, or more appropriately representative sampling. There are many many different techniques to sampling and different materials require different techniques. But we have a good number of members here who can offer valuable input on this topic if there is enough interest. Once you learn how to sample properly, it is possible to sample and send representative samples to a lab for fire assay or instrumental analysis. Or if you're really motivated we can let the discussion evolve into fire assaying and even running an AA.

I could deal with a fresh new topic. Opinions?

 

[aga]

A Generic thread on Sampling sounds like a good idea.

Defining any results of sampling specific items might be more trouble than it's worth, as each 'sampler' could argue the results.

 

[Topher_osAUrus]

I like the sound of that too.

A sampling and assaying thread would be awesome.

Maybe another thread going over all the fancy (expensive) testing machines as well. Their pro's and con's, etc..

 

[goldsilverpro]

I'm all for a sampling/assaying thread.

Assaying is easy. Sampling is hard. All an assay does is tell you what's in the sample. A bad sample produces a worthless assay.

 

[anachronism]

Chris whilst I agree that sampling is hard could you explain how assaying is easy for the average home refiner please? To put some meat onto the bones of my question:

Whilst taking one processor (as an example) and recovering all the gold/silver/PGM into solution is pretty easy if done correctly, the quantitative part seems a lot more tricky to get right. I appreciate that technically one could dissolve ten or twenty identical processors in one solution to even out the bumps and obtain a measurable amount to calculate the yield per processor, however having a large enough sample size isn't always going to be possible.

For "one offs" surely the best and most accurate way would be to dissolve, destroy the nitric, make a known volume liquid and use an AA, and that's where it deviates a little from the average home user. The AA will be able to tell you all the relevant metal ppms and then you can calculate the yield per piece.

If I am missing something here that will give the kind of solid accuracy this database is looking for then I'm all ears but I am wracking my brains to think of it. I currently can't see beyond the AA for providing this level of detail for such small potential samples so I'm open to suggestions :?

Regards

Jon

 

[kurtak]

A thread about sampling & assay would be great

If I am missing something here that will give the kind of solid accuracy this database is looking for

Jon

I don't think GSP &/or 4metals are talking about discussing sampling & assay for the purpose of coming up with an accurate database of yield "per type" component such as Alexander has suggested here

Rather - I believe they are talking about sampling & assay as a way of determining value of a batch/lot done per batch/lot

If that is the case (GSP &/or 4metals please correct me if I am wrong) then sampling & assay should be discussed in its own thread --- on the other hand if the sample/assay is for building a reliable/accurate database - then it should be discussed here

Personally - having been at this for something like 7 or 8 years - as a means of making a living (meaning for profit) I just don't see coming up with a database that is so to speak "reliable/accurate" based on a sample/assay of a particular component/item (ie. fingers, pins, chips etc.)

The best you could do ("after" running many samples) is provide a list that says - I have had yields as high as X & as low as Y on this item

In my 7 - 8 years of doing this - this is what I can say --- I have done batches that made my eyes bug out because the yield was "way" higher then I expected - BUT - I have also done batches that made me scratch my head with a frown on my face because the yield was "way" lower then I expected

And for the record - when I run a batch - I "do not" throw ANYTHING away until I am done with the batch --- if the batch comes out in the ball park of what I expected or higher I call it good to go --- if it comes in "under" my expectation - I go back through "everything" to double check for loss - rarely - if ever has doing so brought my disappointing yield up to what was expected (meaning the "expected" gold just was not there in the first place)

That said - granted - for the most part batches of a particular item (say fingers) will run "a bit" to the plus or minus side (but none the less - still plus or minus) of expectation --- BUT you will also have batches that come in so low its very disappointing - & as well batches that come in so high you jump with joy - & that is especially true if/when you start running "larger" batches on a regular bases

Kurt

 

[4metals]

I think this deserves a separate thread and when I get home (where the good books live!) I will begin a thread in the Sampling / Assaying / Testing section.

The value of proper sampling is more skewed to larger lots, and I agree with Chris that the good sampling is the hard part, not the analytics, once you have learned the process. But we can all express our opinions about this in the new thread.

Kurt is correct, the proper sampling will help the small and intermediate refiner get better results when shipping out material or keep their in house processing in line with expected yields. I do not hold out a lot of hope for a database such as the HYDRA SIEGE project producing valuable data, although properly sampled and assayed materials will improve the results. Much like the old computer programming saying "garbage in, garbage out", the same is true with assay results. Garbage in = unrepresentative samples submitted for assay, garbage out is the assay result of those bad samples.

I will start writing on my laptop and post a thread shortly (day or two) in the Sampling / Assaying / Testing section and in time I will come back here and link the thread.

 

[g_axelsson]

As it happens, I'm running a number of batches on CPU:s I have to get a better picture of how much gold I have in my scrap. The view that components can hold different content of gold depending on a number of factors are correct.
After treating a kilo of 486 CPU:s in aqua regia I discovered a couple that still had gold braze holding the die in the ceramic cavity but when I cleaned them out I discovered that some in fact looked to be glued in. There also were a large difference in die size so the amount of gold used to braze a CPU could differ with the exact model number.

So this is the two 486 CPU:s I could identify as the dies was still loosely attached to the ceramics.

Both chips would be identified and sold as 486 CPU:s, for example by boardsort. My guess is that it can differ quite a lot in yield between the two CPU:s.
These two dies were still attached on the bodies when I picked them up out of the acid. It turned out to be made by two different manufacturers (Intel and IBM) and differed about four years in development. One was a DX model while the other one was a DX2 model.


Seeing this with my own eyes I don't trust my own yield numbers any longer. My sample wasn't random as the IBM CPU:s were over represented in this test sample. I started with the most mangled ones and these had heat sinks glued to the top and got a bit smashed up when I removed them.

Göran

 

[anachronism]

Goran

Firstly that's two different 486 procs. You've got a range of them from sx-16 procs all the way through to DX4-100. Some are gold lidded whilst others are not, even within the same type.Then you've got the different years of manufacture and to add to this you have the different factories they were made in. I only mention all this to back up what you're saying.

In essence we have an expression in the UK that says it's like "comparing apples with oranges." Nobody can say that a 486 has a particular yield. In order to present a database that is accurate, all the yields for every above variation would have to be included on a separate line. Anything less would be a generalisation, and inherently misleading.

 

[g_axelsson]

I was surprised to see the glue, but not to see the different sizes of dies. Still, boardsort and others have a unit price for 486 CPU:s and I have never seen a sale where 486 CPU:s were sorted into different batches.

So not only do you need to have different entries for every subtype of CPU, you need to convince the seller to sort the CPU:s according to that list too.

It's more like comparing different apple sorts when you go down into details like this.

Göran

 

[goldsilverpro]

I've seen quite a bit of epoxy (so I've been told) used to secure the chip, especially on ceramic DIPs. I've also seen silver braze used on DIPs with silver plated legs.

Another thing to consider is the size of the chip. As everyone should know by now, the gold braze used to attach the lids and the chips usually makes up more than half of the gold found on these parts, even when everything else is gold plated. The gold braze used is in the form they call braze preforms. These are small sheets of eutectic braze, about a mil (.001"), or so, thick. The larger the chip, the bigger the preform. Therefore, those ICs with large chips tend to contain more gold than those with small chips.

 

[kurtak]

Goran

Your pics are a "perfect" example of what we have been talking about - thanks for posting them :!:

Goran

Firstly that's two different 486 procs. You've got a range of them from sx-16 procs all the way through to DX4-100. Some are gold lidded whilst others are not, even within the same type.Then you've got the different years of manufacture and to add to this you have the different factories they were made in. I only mention all this to back up what you're saying.

In essence we have an expression in the UK that says it's like "comparing apples with oranges." Nobody can say that a 486 has a particular yield. In order to present a database that is accurate, all the yields for every above variation would have to be included on a separate line. Anything less would be a generalisation, and inherently misleading.

Per the underlined - correct - as in ABSOLUTELY CORRECT

Now translate that into IC chips & the variations of yields on the same "type" are even greater

Considering the fact that I run - "on average" - about 500 pounds of chips "per year" & have been doing so for 4 years (maybe even 5 years now) & I run them in batches of "at least" 30 pounds & preferably batches of 50 - 60 pounds - I can tell you "from experience" you are going to have batches (of the "very same" type chip) that are VERY disappointing - granted you will have some that run better then expected also --- (if I had to put a number on it I would say you will see 2 disappointing batches for every 1 better then expected)

The bottom line is - & I don't care how many small batch test you runs &/or how many fire assays you run on a particular "type" chip - you are going to have "highs" & "lows" & "many" numbers in between (presuming you use a scale accurate to the 4th decimal & NOT a scale only accurate to 1 decimal)

The best you can do - is "after" running MANY test/assay - on the same type chip - is come up with a yield number that is based on "the law of average"

When you buy on the law of average - be prepared for disappointment - sometimes VERY disappointing

You can buy on the law of average & make out "OK" - in the LONG run --- BUT - you also have to buy in large enough volume for the law of average to work in your favor - other wise buying on the law of average is a flat out gamble

Edit to add - bottom is there is a difference between a list that represents an "average of yield" and one that claims to be yield data

An "average yield" list is just that --- it represents the average - of a given number of samples &/or batches - of equal size

Data - is a set number - of expectation

Two "different" things :!:

Kurt

 

[4metals]

When faced with a customer that is selling you the material and demanding payment vs. toll refining, as a refiner you have no choice but to use the historical low as a payout figure. This is where record keeping comes in handy.

But if you were toll refining, and the yield was high, the customer makes out better. This is exactly why it pays to develop a relationship with a refiner you can trust, people today are greedy and will trade loyalty of an honest refiner for a few percent. Then a lower grade lot comes in and they swear it is the refiner. Truth is the variances can be extreme for good reason, as Goran posted a few posts up from here.

I learned the same kind of lesson years ago when dealing with contractors. Some insist on a fixed quote for a job. Often there are hidden what if scenario's that a contractor must charge for if he is to be held to a quote. I always chose contractors who I trusted to give me a fair day's work, and just had them work for time and materials.

 

[g_axelsson]

When faced with a customer that is selling you the material and demanding payment vs. toll refining, as a refiner you have no choice but to use the historical low as a payout figure. This is where record keeping comes in handy.

I totally agree with you. I didn't know how much gold a 486 CPU contained before but I did a calculation based on boardsorts prices and some approximations. It looks like I'm within percents on what I expected for a general 486 even on this lot that now seems like a low yield test. Looks like boardsort have adjusted their prices to the low yielding CPU:s of a category so they still earn money if someone was to only sell them chips with dies glued in.

Time to start another test...

Göran

 

[goldsilverpro]

I've always said that, "In any buying/selling PM scrap negotiation, the person that has the greatest amount of knowledge concerning the actual PM value of the material in question, always wins." This knowledge of values has to come from YOUR OWN evaluation. Anything else will lose money in the long run. By your own evaluation, I am including sending the samples that you took out for assay. I do think that every one should pull their own samples. Almost anyone selling PM scrap will allow you to pull samples.

 

[4metals]

I think a good example of this can be polishing sweeps. I've run literally thousands of fusions and cupellations on jewelers sweeps but if anyone ever asked me how much a pound of raw jewelers sweeps is worth I could give you an answer. But I couldn't give you an answer that you could use, or that you would feel good about if you paid for that answer. There are just too many variables.

It depends on the type of polishing rouge, red or green, how heavy handed the polisher is, what karat the jeweler was polishing, and if there were brushes and buffs mixed in with the sweeps.

So what makes this any different from asking what 100 pounds of mother boards is worth? Or 100 pounds of plated pins?