For the tech folks: I've added a stats dashboard at /__stats to monitor cache performance, plus a cache purge API to instantly refresh content when needed. Security headers are properly configured while maintaining affiliate compatibility. (4 of 4) #CloudFlare #Worker #Developer #HitRate #Security #Website #Performance

The fact that Infinity Cache #Hitrate isn't getting a large boost with #RDNA3 makes me worried about #Navi32 and #Navi33

The less IC you have, the lower the bandwidth multiplier it is, and the narrower the memory bus, the less bandwidth you have in the first place

This means bandwidth will likely be a much more severe bottleneck for Navi32 and Navi33 (especially 32, which is targeting >1440p resolutions)

Knowing that #Navi31 has a 960GB/s Mem BW, a 5.3TB/s Infinity Cache BW, and a 3.5TB/s "Effective" Bandwidth, we can calculate the #Hitrate of N31's Infinity Cache with the formula AMD used for Effective Bandwidth in the past:
MemBW + Inf$BW * HR = EffBW

Solving for Hitrate, we get 48%

Oddly enough, that's LOWER than Navi22's hitrate at 4K (53% w/ same 96MB) according to this AMD slide

(annotations by @Locuza)