It’s interesting; In Lunar Lake, there will be TSMC + Samsung/Micron LPDDR5X cores and Intel is also “making something”, I assume the Xe2 iGPU uses this internal RAM?

Of course the question is whether the NPU*/GPU also has “cache coherent” access to that memory via CXL2.

I think it’s a very cool cut and paste operation from Intel! I’m curious what packaging they use for this, and who will do the packaging. For example, for NVidia video cards, the TSMC CoWoS packaging was the bottleneck.

I think Intel bought a fair amount of ABF capa a few years ago (read: paid a billion euros to a supplier for a brand new fab), I’m curious to see if the shortage will happen. This is a huge logistics operation for Intel, and I believe it is the largest ever.

AMD has already gotten used to this, and there are Ryzens floating around that say they are made in the US (GloFo IO core), Taiwan (Zen core) and assembled in Malaysia. I wish Intel good luck with this, and I hope they can do it as well as AMD.

*NPU has been a marketing term for the past few years, and may have been a rebranded DSP and a license purchased elsewhere, just as Apple has always done.

[Reactie gewijzigd door kidde op 9 januari 2024 02:53]