Why are engineering sample CPUs illegal to sell?

Due to the pre-production nature of the engineering sample processors, they are generally only loaned to OEMs, ODMs, and ISVs for pre-production test and evaluation work under specific contractual terms and conditions to assure the protection of assets and confidential information.

Engineering sample processors are not made available to the general public by Intel."

In effect, you're in possession of stolen goods.

The whole idea is, no matter what happens, those goods are not to be circulating in the hands of the public. You could give them back to the local rep, and he could have them shredded. (Some factories shred their e-waste to prevent recovery by waste removal people.)

Those samples could have defects, maybe they don't have a 100,000 hour operating life (early mortality). They might not even compute properly at full speed. Like an ES 3GHz processor, there might be an errata sheet in the box, stating you're supposed to run them at 2GHz.

Intel could also mark them with sufficient information, to trace them back to who received them. To determine who is leaking them and violating a contract term.

With other manufacturers, those parts are the equivalent of the "qual barrel". And the stuff in the qual barrel, is definitely not production quality.

Not allowing them to be sold, is to protect *you* from receiving inferior goods.

Paul

No need to have the same model number.

That is a lie, he did not say they were tests.

No, if I buy an engineering sample advertised as such, I don't expect it to work perfectly.

Newsgroups reinstated.

I've only seen this one post, but I know pre production models of things can slip out for whatever reason. I had a computer once where the pcb had been manufactured with a fault, tend you can clearly see back then where the tracks had been manually cut and wires bridged the contacts to the right places, leaving the tracks as orphans. Likewise a number of Sinclair ZX Spectrums in the early days were made with known faulty ULA chips and a logic chip glued to the surface with its legs splayed and wired to make the circuit work. The first batch of Phillips CD100s the very first CD player on the market had quite a lot of wires cut tracks and components wired in odd ways inside it. I have to say that none of these what we might call bodges ever caused any trouble during the lives of the products. The CD100 is in fact still working, although its tendency to jump if a gnat walks across the floor shows it does not have the memory buffers in modern players. Brian

"Fly-wire" or "spider wire" mods were pretty much SOP in the early years - the first several revs of new AMI motherboards ALWAYS had a few - and "boogered" chips were pretty standard too - transistors bridged onto logic chips to "fix" a bad gate were also quite common. (along with "clip-on" heat sinks stuck onto chips that were running at their thermal limits.

Remember all those "sx" processors? DX Processores where the floating point processors built on the fie were defective, and rather than discard everything they just blew a few "fuses" on the chip and sold them as SX on boards where an external FPU could be mounted if you needed the FPU function

I've never seen a transistor bridged onto a logic chip, nor can I even conceive how that would ever be possible, even with early, IC logic chips, forget about in the PC motherboard era.

That's another myth. At introduction, the SX's used the same silicone as the DX product, with the floating point unit untested and disabled. If Intel were to have relied on die where the floating point unit was defective, there wouldn't have been enough to meet the enormous demand for SX product. This was high yield production at Intel not Joe's crap line. And within a year Intel had a new SX version with the floating point unit removed, which made the die smaller, less costly to produce and more profitable. That was the plan all along.

The practice is called binning, and is used by all processor manufacturers to this day. It generally falls into three buckets:

It costs significant amounts of money to design and manufacture a processor, and the typical die yield is far less than 100%[*]; without fusing/binning, processors would be far more expensive than they are now - particularly as we transition to 5nm and 3nm process nodes where the yield is currently significantly lower than older process nodes (yields can be less than 50% on a new process node).