What's the ipc like compared go zen, 25%?
ARM is dead
Christopher Peterson
Gabriel Butler
How is it bad for workstations to have 190W CPUs.
Threadripper is at that range
Angel Phillips
Because they have nowhere near the same IPC let alone AVX 256 performance. I was expecting around 95W but I guess I expected too much.
Dylan Reyes
> ARM is dead
It's just JS won.
Isaac Moore
>Muh cornz!!
>Muh cinecornz!!!
Connor Nelson
Yeah man, most consumer cpu workloads consist of high i/o database tasks.
Benjamin Kelly
>Nice, how many decades before it can compete with the zen 2 x86 ISA processors coming out in a few months?
You don't understand the point of high performance RISC architectures. Here the point is not just IPC, but scalability in *multi-socket* systems.
IBM's year old POWER 9 has 24cores 96threads, and you can have 2 of them on the same board, for 48cores 192threads. And insane RAM amount and bandwidth.
You can get these systems now. What does x86 have to compete with that?
Clusters are not the answer because some workloads require single systems, otherwise latency becomes unusable. If you need those workloads, you *need* a scalable risc processor, even with worse IPC, because it ends up much much faster by keeping it single system vs cluster.
And mind you, POWER is not the most scalable risc architecture.
Sparc M8 for instance is 32cores 256threads. And you can put 8 in a board (the M8-8 server), for a total 256cores 2048threads.
Let that sink for a moment. Nothing x86 can come remotely close.
ARM, another risc architecture, even though not designed with scalability as first priority, is still infinitely more scalable. Proven by the fact that Fujitsu is building their next supercomputer, Post-K, with ARM processors.
In the case of ARM, it has even an IPC advantage over current Intel and AMD x86. Apple's A12, designed for low power consumption, is with 15% in single core *performance* from current x86 chips, at a 2Ghz *disadvantage*. Which indicates crazy higher IPC, in the order of +50-60% compared to x86.
The x86 ship has sailed user. It just lingered on life support due to legacy desktop software.
Owen Davis
>In the case of ARM, it has even an IPC advantage over current Intel and AMD x86. Apple's A12, designed for low power consumption, is with 15% in single core *performance* from current x86 chips, at a 2Ghz *disadvantage*. Which indicates crazy higher IPC, in the order of +50-60% compared to x86.
>The x86 ship has sailed user. It just lingered on life support due to legacy desktop software.
Let me guess, you base this off gookbench and not FP64 linpal sustained benchmarks, right?
Brandon Sanchez
>JS just happen to benefit a lot from it since it uses float all the time?
This.
Jaxson Harris
So tl;dr, power9 is dogshit outside of database or connecting GPUs together with almost no performance loss as it scales up for supercomputers?