4 ms·

The cost of copying your executable is near-zero. The cost of producing a CPU die is non-zero. Suppose that ECC support takes a certain die area, then either ev

The cost of copying your executable is near-zero. The cost of producing a CPU die is non-zero. Suppose that ECC support takes a certain die area, then either everyone is already paying for it while buying the CPU or everyone who licenses the feature is paying for the silicon for everyone who doesn't enable it. So, if you are licensing a CPU feature, you may be paying twice for it. Of course, whether you actually do so, depends on the relative cost of producing that area of the die vs. the development cost of that functionality. Another issue is that this probably would not happen in a healthy, competitive market. In fact, the opposite would happen, CPU vendors would try to provide more features at a price point in order to compete (which is probably why AMD does support ECC on many Ryzen SKUs, as an underdog AMD still has to compete). However, if you need a x86_64 machine, you only have two options. Or if you are unlucky and required to purchase from a large vendor like Dell, you may only have one option.

magicalhippo 5y ago
> The cost of producing a CPU die is non-zero. While strictly true, the actual die is a relatively tiny cost when producing a CPU. The majority of the cost is the upfront cost of designing the masks. Once you have those, using this or that mask doesn't make a huge difference. > Suppose that ECC support takes a certain die area, then either everyone is already paying for it while buying the CPU or everyone who licenses the feature is paying for the silicon for everyone who doesn't enable it. As mentioned the majority of the cost of the ECC support is in the R&D, not the physical manufacturing. As such, moving this to a software level can be a win-win for both parties. By selling ECC support to more customers, they can afford to price the bare CPU cheaper. That's exactly what happens with our software. There's no way we could sell to single-person companies at the price we do without having features behind "license walls". > Another issue is that this probably would not happen in a healthy, competitive market. In fact, the opposite would happen [...] One of the reasons people love cloud computing these days is that the cost scales with activity. Get more customers generating more activity? No problem, just add some extra capacity from your cloud provider. Need some additional functionality to service a new market? No need to pay for years for something you don't use until you need it.

magicalhippo 5y ago

> The cost of producing a CPU die is non-zero. While strictly true, the actual die is a relatively tiny cost when producing a CPU. The majority of the cost is the upfront cost of designing the masks. Once you have those, using this or that mask doesn't make a huge difference. > Suppose that ECC support takes a certain die area, then either everyone is already paying for it while buying the CPU or everyone who licenses the feature is paying for the silicon for everyone who doesn't enable it. As mentioned the majority of the cost of the ECC support is in the R&D, not the physical manufacturing. As such, moving this to a software level can be a win-win for both parties. By selling ECC support to more customers, they can afford to price the bare CPU cheaper. That's exactly what happens with our software. There's no way we could sell to single-person companies at the price we do without having features behind "license walls". > Another issue is that this probably would not happen in a healthy, competitive market. In fact, the opposite would happen [...] One of the reasons people love cloud computing these days is that the cost scales with activity. Get more customers generating more activity? No problem, just add some extra capacity from your cloud provider. Need some additional functionality to service a new market? No need to pay for years for something you don't use until you need it.