Hacker News Reader

Comment by gjsman-1000 | Hacker News Reader

For me, think about that memory bandwidth. No other CPU comes even close. A Ryzen 5950X can only transfer about 43GB/s. This thing promises 400GB/s on the highest-end model.

View on HN

As always, though, the integrated graphics thing is a mixed blessing. 0-copy and shared memory and all of that, but now the GPU cores are fighting for the same memory. If you are really using the many displays that they featured, just servicing and reading the framebuffers must be...notable.

A high end graphics card from nvidia these days has 1000GB/s all to itself, not in competition with the CPUs. If these GPUs are really as high of performance as claimed, there may be situations where one subsystem or the other is starved.

No consumer CPU comes close. Just saw an article about the next-gen Xeon's with HBM though that blows even this away (1.8TB/s theoretically), but what else would one expect from enterprise systems. Getting pretty damn excited about all the CPU manufacturers finally getting their asses into gear innovation-wise after what feels like a ridiculously long period of piss-warm "innovation".

Thanks to Apple in this case for taking a holistic approach to making a better computer.

The AMD chips in the PS5 and new XBox reach 448GB/s and 326GB/s bandwith respectively with their unified memory.

Not entirely true. Xbox Series X has two different memory bandwidths. The first 10GB has 560GB/s and the last 6GB has only 336GB/s.

Yeah it's an interesting setup. I believe the 560 is prioritized for graphics processing and 336 for more general tasks.

And only 10 cores so a 5950x completely wreck an M1.

A 5950x uses ~4 times as much power.

What application makes full use of 5950x all cores?

Video encoding, ray tracing. Prime95 if you're trying to stress test your CPU and memory.

make -j32

Compiling stuff.

To add to the other comments: Spark, Stockfish, some games

Reply on news.ycombinator.com