How to Make Multicore Chips Faster, More Efficient | Hacker News Reader

How to Make Multicore Chips Faster, More Efficient

spectrum.ieee.org

34 pointsjonbaer17 comments

This just reminds me a bit of how programmers could fiddle with memory in the Cell Processor for the PS3. IIRC memory was managed by the programmer and it only made it harder to program for.

On the other hand, once they learned to program it, the benefits were apparent.

The PS3 aged more gracefully than the X360, with programmers being able to squeeze more performace over time as they came to terms with the architecture (the same had happened with most consoles historically).

I feel that this won't happen with the current generation, since both PS4 nad XBONE are basically PCs in a box, and improvement in graphics will probably be much more limited over the years (due to programmers having a good grasp of the architecture from day one, and not because the architecture is inherently bad).

I wager that the PS4 and XBONE will benefit PCs more than previous consoles benefited PCs, specifically because they are slightly slower 8 core CPUs. Most games these days are optimized for quad cores at best, so to get more performance, additional threading is required. Due to everything being x86 and the prevalence of widely used game engines, everyone's performance goes up. Personally, I would like better AI.

I feel like we are unfortunately sitting on the tail end of the integrated circuit S-curve and are now waiting for the emergence of the next processing medium.

I think you're right, it seems like graphene offers a promising next step though.

Actually I think both the processing capability and communication capability are almost reaching their limits regarding the sequential performance, I bet that's why multicore and bandwidth are what producers advocate, and also people are working hard to utilize the parallelism. Missing the good old days when 'faster' truly meant the sequential performance for CPU and latency for networking, and you could always get 'faster' by simply increasing the frequency...

Unfortunately? I find it fortunate that we are looking at the point in which companies have to go for truly revolutionary changes to processor technology.

I'm still waiting for a proper 8 core consumer x86 CPU. AMD's chips don't count because they perform worse than the ones from intel.

What is the consumer application for an 8 core CPU?

I can't think of one, as of 2015, that i'd not call a commercial application.

There are definitely prosumer applications; I'd quite like faster builds at home but am not quite willing to pay for a Xeon, say. Bit of a niche market, though.

You're only saying that because most programs are incapable of taking advantage of multiple cores, but if 8 core CPUs were the norm, I guarantee programs (where performance matters) would change.

If 4 cores are the norm now, and most programs cannot take advantage of them, what it the rationale for which things would change in the case of 8 cores ?

4 cores aren't quite the norm. The computer I'm currently at is a Dell XPS 13, the CPU is an i7-4510U, which is dual core. Regardless, in most circumstances, 4 cores are only just above break even point for parallel code, so the pay off isn't big enough.

* Transcoding large video files, especially in real time.

* Various Hobbyist projects.

* Extreme multi-tasking (running a game, 100 firefox tabs, home server, encoding, etc)

* Seti@home/folding@home -> computational philanthropy.

Could you be more specific about why a consumer 8-core CPU isn't the consumer 8-core CPU you wanted?

I don't know why he's complaining.

http://www.anandtech.com/bench/product/1340?vs=1316

AMD's newest processor compares quite well to the nearest Intel octo-core that costs 3x as much.

I've thought for a while that the hardware should manage threads and scheduling, not the operating system.

Reply on news.ycombinator.com