On Sun, Nov 27, 2005 at 09:11:11PM +0000, Russell Wallace wrote: > I've seen a figure of 20 gigabytes/second for Cell main memory bandwidth. I
Bandwidth is nice enough (XBox ERAM has 250 GByte/s, but it's small), but it's the latency where it's at. They claim a few cycle latency to set up a DMA connection as compared to several 100 cycles when there's a cache miss. > don't know whether this figure is correct or not, if it is, then while > (obviously, necessarily) slower than on-chip memory, it's faster than the > main memory bandwidth of any other processor I know of. Yes, it's a nice fat pipe. Not nearly as nice as on-die RAM, but then, also not nearly as limited and expensive. > I'm guessing main memory _latency_ isn't going to be amazingly hot, given > that Cell is designed primarily for vector-style workloads. (Of course, main > memory latency isn't very good on any modern processor, though it's less bad > on some (e.g. Athlon-64/Opteron) than others.) It's different. There is no cache. IIRC, the pipeline is really shallow, too http://www-306.ibm.com/chips/techlib/techlib.nsf/techdocs/E815CC047A60914687256FC000734156/$file/ISSCC-07.4-Cell_SPU.PDF > Assuming you're clever about overlapping fetch of one chunk of data with > processing of another, I think performance will be in large part determined > by whether you can arrange your data in long sequential chunks with > contiguous addresses, i.e. you slurp many kilobytes in one go before moving > somewhere else, rather than fetching a couple of words, then following a > random pointer to somewhere else, fetching another couple of words etc. Absolutely. You have to map your data structure RAM geometry and access patterns. This has already been true for all modern CPUs for some time already. Nothing changes, things only become more so. > Or (roughly speaking, for illustration purposes only, you need not remind me > why this is not entirely accurate): Fortran-style arrays = fast, Lisp-style > linked lists = slow. Actually, modern CPUs really like arrays. Randomly linked lists (like in the Lisp machine) are a no-no. Not very novel, either. > Of course, this is to a considerable extent going to be true of all > processors for the foreseeable future, so work done to optimize your data > structures for this won't be wasted even if you end up not using Cell. It has been true at least for the last 3-5 years. -- Eugen* Leitl <a href="http://leitl.org";>leitl</a> ______________________________________________________________ ICBM: 48.07100, 11.36820 http://www.leitl.org 8B29F6BE: 099D 78BA 2FD3 B014 B08A 7779 75B0 2443 8B29 F6BE ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
signature.asc
Description: Digital signature
