7 Comments

Some rumors claim AMD had to scrap increased cache on the iGPU due to Microsoft requirements for the NPU.

Expand full comment

If so, AMD made the choice of still having more CPU cores than, for example, Intel did with Lunar Lake. That SoC also has a big NPU.

That all being said, your point (or the question behind it) is well taken. The usefulness of spending the area on a large NPU rather than for a larger GPU and/or larger Caches (L2 and SLC) is still undetermined. In Smartphone SoCs, having a somewhat beefy NPU has shown some utility, but there they have concrete uses, such as image enhancements. In laptops, those tasks can be accomplished by the CPU or the GPU.

Expand full comment

For the consumer I think AMD has found a more convincing balance for those still seeking x86 laptops in 2024, whereas Intel are clearly trying to stop the dawn of the ARM laptop era by squeezing performance to achieve almost ARM-like power efficiency.

The NPU count is still a bit of a mystery, there doesn't seem to be grand plan for these. AMD laptops like Zenbook S 16 aren't even listed as "co-pilot+" PCs despite advertising 50 TOPS for dedicated AI processing, more than the new Snapdragon X chips. It's clearly just the prevailing "mood" of the chip market, propelled by Apple, that devices without dedicated AI processors are futureless.

Expand full comment

Impressive, and Arrow Lake is apparently the same arch, but half the size? If so I will have to go with Strix Point still in a small form factor.

Expand full comment

Why is there no test on arm's mali gpu?

Expand full comment

Intel's Linux drivers for Lunar Lake are rather subpar. Does it affect microbenchmarking?

Expand full comment

GPU-side benchmarks were run on Windows

Expand full comment
Error