Compared 3 different Bulldozer FX-8120's. Want to know the difference? - Page 2

Server chips like Evil Penguin said are supposed to be held to a higher standard.

Reason being is that server chips are generally more expensive for pretty much the same performance. Where it differs really is that they can supposedly take more heat and are supposed to be more stable because people spending money on a server are almost paying for insurance that the chip will run mission critical servers/applications with 99.99% uptime 24/7 and that those chips aren't going to burn out until they go for a component refresh/upgrade.

I've been running this thing at stock settings so that VID is what I have then.

Its perfectly fair,
Server chips are always the top of the ones coming out of the line, They then clock them even lower so that instead of there being a error every 1000hours its more like they only make a error every million hours. Which is important for servers since you dont want to have your banks website blue screening every few days.

Then the next binning is the 8150, Which I suspect most its chips are getting re-binned to server chips so they are hard to get ahold of. Then the ones that fail the 8150 tests go into 8120's which is what most of us have. They are in quite high demand so they proably have a lot less strick tests to pass and they only have to run at 3.1ghz so 4.8ghz is a major increase in speed for a low binned chip.

5ghz out of a fx 4100 will still proform less then 4.5ghz out of a 8120 well except in single threaded stuff, But if you really want to be that picky disable some modules on your so its only a fx 4100, I bet youd get higher speeds out of yours since you can choose the best quality modules.

They only test for stability at the model rated speed, they do not guarantee overclocks. The only thing about overclocks was mentioned to review sites, and I believe 4.6 was stated as the approximate limit on air, 4.8 was on water, and 7-8 on ln2 or lHe-4. I assume 5.0+ would only be attainable with sub ambient cooling such as phase change/cascade(with all modules active).

They do mention 5+Ghz if all but one module is disabled.

I see what you're saying but from all the results I've seen the difference between an 8150 and 8120 is marginal at best. 8120s are more then capable of beating out lesser 8150s, suggesting there isn't really any aggressive binning between the two going on. I don't know why there's a shortage of 8150s because they seem to be 8120s with a stock overclock.

Since your post is long ill break it down a bit.


You would shut them off in the bios, My bios doesnt have the option visable unless I un-hide it with a bios editor. Havnt tried using it since I was proably hidden for a reason.
On my chip I know core 3 and core 8 are the weakest 2, So that would mean Id disable module 2 and 4. The bios just lables them Core 1-2, Core 3-4 and so on.
Dont think you can disable just a single core, But maybe some one will make a schedule editor that will let you do it.

The chips are different but the die's are the same.
When they make the die they have a ruff estimate of which ones will be the best, Then they xray them also to check for quality and thats the first binning.
Best ones will get placed on a server Chip which has different pinning but runs the same die, This is also where they set the settings of the chip before they stick the IHS on it.

While the 8120 and 8150 are pritty much the same chips the internals of them will be of different quality, The xray will reveal this straight away.
This is the easyest test to figure out which ones need cores disabled because some cores will have defective/broke tracks and such in side them.
They dont test it by overclocking them, there main tests are xrays then finally they will bench test them.
Stick them in a special cpu testing machine which makes them crunch a very big number and it has to be correct and the stock speeds for so many hours. They allow for some erros with desktop cpus since what does it really matter if every 1000 hours it gave you 1.923523423421 instead of giving 1.923523423422. A end user would never notice this where in a server it could mean the difference between reaching mars in a space mission or over shooting it. Ages ago there was a cpu that failed this test every time cant remember if it was a Pentium 1 or a Pentium 2.
So it really does come down to luck of the draw, When they give a guide line like it should be good for 5ghz they are going off the architect should be good for that. Not that specifc chip.
The longer the memory pipe lines the higher frequency a chip will run at. The bulldozer has very long ones so thats why it can run at high frequency, But the draw back is it makes per clock processing slower.

[/quote]

You are only guaranteed the processor will work at its stock and turbo core speeds. Any overclocking you get is a bonus and normally means you can expect some CPUs to come out at higher speeds stock in the future. Bulldozer is still very new so they dont have the full speed chips out yet but I expect will see something like a 8170 which is 4ghz, 8180 at 4.2ghz and a 8190 at 4.4ghz (which would be the rarest and highest binned desktop chips they have.) I imagine if mine was re-binned it would come in at a 8180 since it can do 4.2ghz at 1.28v

The more demand there is for 8150's the lower the quality of the 8120's will be. If there was near 0 demain for a 8150 then they would be rebinning them to 8120's. Im guessing thats what they did with the first batch of chips since they seem to be of higher quality then the second batch and because you cant find many batch 1 8150's.

The overclocking margin will be all over the place since 32nm processing is still new to AMD, Over more revisions they will figure out what works best and how to produce the chips so each one in the batch is more alike.

I think can be directly correlated to how they bin their chips, they run them at the stock speed and figure out what voltage they need to run that without errors. In manufacturing the chips there are subtle differences in every chip, this is especially apparent in a new process, be it smaller, or a whole different architecture. So therefore, if a chip has slightly larger gates it would need more voltage to activate these larger than normal gates, but these gates would also be able to handle higher speeds and voltages because they are larger and can take more current.
I have no way to back up this theory beside the evidence given here, and my knowledge of how chips are work, made, and binned.