Hacker Read top | best | new | newcomments | leaders | about | bookmarklet login

Wait till you see the memory bandwidth that 1 thousandth of a cpu gets you.


sort by: page size:

But its not available to the CPU. A couple posts up the chain has a link showing the CPUs are incapable of coming anywhere close to the full bandwidth of the memory.

It’s not the CPu it’s the memory bandwidth, isn’t it?

I think the author has discovered memory bandwidth. When you have a simple function and just scale the number of cores it's easy to hit.

Note that this is considerably faster than the amount of memory bandwidth available per core.

With any sane implementation, this just benchmarks the memory bandwidth.

The memory bandwidth is for the benefit of the GPU, not the CPU.

Memory bandwidth as well.

No, this is about memory bandwidth.

memory bandwidth gets to be a limiter quickly with many cores for certain workloads.

> Usually when we say bandwidth, we mean bytes/second, to and from the caches and main system memory.

I thought that the registered memory of high end server processors that could support 100+ GB of main memory were significantly slower in bandwidth than the DDR3/4 main memory of consumer processors.


The memory bandwidth result is impressive.

The theoretical processor bandwidth is one thing; the actual throughput on motherboards with existing slow memory chips (DDR2-667 according to the specs page) is quite a different matter.

This becomes a pretty big deal when you have that many cores all hammering on it at once. Divide by 32...


How much more memory bandwidth does that require?

Memory is orders of magnitude slower than the cpu at the moment, so you'd need orders of magnitude more bandwidth to come close. So sure, it's going to be faster, but the cpu is still going to be waiting for memory more than the other way round.

You're comparing strictly CPU memory bandwidth to CPU + GPU. If you add CPU + GPU bandwidth for a PC you'll get similar numbers.

If only CPUs could get data from memory at these speeds.

Genuine question: how often am I memory bandwidth constrained?

Or to put it another way, if I had infinite memory bandwidth, how much faster would browsing/gaming/compiling become?


> they have 400mb/sec of memory bandwidth

Per chiplet.


This is kind of pointless without commensurate memory bandwidth. Those CPUs are just going to be waiting around doing nothing.
next

Legal | privacy