Vespill :
Sorry, I do not understand. Do you mean 20MB, 25MB, and 13.75MB?
No, I mean the total amount divided by the number of cores. The more cores a CPU has, the more cache it would need to keep those cores fed properly.
k1114 :
I think you all missed the much larger amount of faster L2 cache. No need for more L3 when you have more L2. I don't know about L1 but chances are, I bet that's more too.
Even if you didn't see the L2 amount, it wouldn't be cost because they're going to use the same dies as the xeons and it's unlikely cache will be a cause for binning. But seeing L2, that's more expensive than L3 so that kills that theory. Also since amd is pushing back, they would want to up them however they can and widen the gap. If they are just getting by, people will see it as intel faltering and that's not an image they want. Intel knows exactly what they are doing.
I'd put architecture as the significant performance factor. Cache isn't something you even look at. But in the end, performance matters and you don't really need to pay such close attention to most specs.
Good point. Though Skylake-X still works out at less combined L2+L3 per core than Broadwell-E, it's by a much smaller margin. I wouldn't be surprised if the L2 in Skylake-X is slower than in Broadwell-E, but... it's surely better than L3. So a little less overall cache but considerably faster -> probably better for a lot of workloads.
I'll still be interested to see how much cache the Xeons come with though.