In the wake of ARM's first quarter results Warren East, who heads up ARM, has been talking about the possibility of ARM-based servers hitting the markets in 2011. There has been some questioning as to why people might use ARM and whether it represents a threat to Intel etc. The simple reason is that it doesn't, nor does it have to. ARM and those who licence its designs are often driven by the need to provide functionality within systems that require thermal outputs to be kept within well-prescribed limits. So for, say an embedded server, whether for print or disk servers or for other similarly-constrained systems such a design might be ideal.
On the other hand - if you wanted to build a low thermal output, but more mainstream server, then it might be that the Total Cost of Ownership/chip count equation could work in your favour with something like an ARM-based design. In the end it will probably be comms bandwidth that is going to discriminate for or against a particular solution in such circumstances.
Multicore is about so much more than simply high performance computing.
Friday, 30 April 2010
Tuesday, 27 April 2010
How much better is it for you?
Just a side thought. Would you like to try answering these questions (respond in the usual way by leaving a comment):
Are you using MCPs (where multicore means four or more processor cores on a chip)? If the answer is "Yes", then can you estimate how much the throughput of whatever activity - or activities - you are involved in (business, particular application or whatever) has improved (let's say expressed as an approximate percentage *) as result of deploying them?
Finally - and this is the big one: Are you in a position to able to estimate the financial benefit to your organisation from the improvement? Most people won't be able to answer that one I expect, but...
* So, if your process has doubled its throughput through the application of MCPs then the improvement is 100%; if it's half as good again as it was, then 50% etc. Don't forget that we are talking about the whole process so it is quite possible for a quad-core system to enable a ten-fold increase in application throughput, for example
Are you using MCPs (where multicore means four or more processor cores on a chip)? If the answer is "Yes", then can you estimate how much the throughput of whatever activity - or activities - you are involved in (business, particular application or whatever) has improved (let's say expressed as an approximate percentage *) as result of deploying them?
Finally - and this is the big one: Are you in a position to able to estimate the financial benefit to your organisation from the improvement? Most people won't be able to answer that one I expect, but...
* So, if your process has doubled its throughput through the application of MCPs then the improvement is 100%; if it's half as good again as it was, then 50% etc. Don't forget that we are talking about the whole process so it is quite possible for a quad-core system to enable a ten-fold increase in application throughput, for example
Friday, 23 April 2010
In the beginning...
In case anyone is wondering this blog is about multicore processors (MCPs) and parallel computing and related issues.
Why write a blog about Multicore processors? Well, hardware and software manufacturers are telling the truth: multicore is going to change the way that we "do" computing. MCPs are a technology that has been waiting in the wings for a number of years. The classical variant of Moore's Law (really his "observation...") started to run out (too much power needed, too much heat to get rid of, starting to get towards limits of feature sizes (probably), memory bandwidth issues and so on) so that engineers reverted to a tried and trusted technology, Parallel Computing; and MCPs are just its implementation on a chip. Or at least some are... (more of that another time probably).
There's a lot going on in the multicore world and this blog will hopefully provide a venue to discuss them and I welcome serious, considered responses to any piece that I write.
It's true that I already contribute to Concertant's intermittent blog (here) and will continue to do so. Concertant focusses on very specific topics and its blog is in effect peer-reviewed, as all Concertant's docs are.
I discovered after a while that I wanted greater flexibility to write pieces which were more wide-ranging (see for example my piece "Google, China, India, France and Digital Britain" on the Concertant blog), which really didn't have their place on there. So here we are...
Why write a blog about Multicore processors? Well, hardware and software manufacturers are telling the truth: multicore is going to change the way that we "do" computing. MCPs are a technology that has been waiting in the wings for a number of years. The classical variant of Moore's Law (really his "observation...") started to run out (too much power needed, too much heat to get rid of, starting to get towards limits of feature sizes (probably), memory bandwidth issues and so on) so that engineers reverted to a tried and trusted technology, Parallel Computing; and MCPs are just its implementation on a chip. Or at least some are... (more of that another time probably).
There's a lot going on in the multicore world and this blog will hopefully provide a venue to discuss them and I welcome serious, considered responses to any piece that I write.
It's true that I already contribute to Concertant's intermittent blog (here) and will continue to do so. Concertant focusses on very specific topics and its blog is in effect peer-reviewed, as all Concertant's docs are.
I discovered after a while that I wanted greater flexibility to write pieces which were more wide-ranging (see for example my piece "Google, China, India, France and Digital Britain" on the Concertant blog), which really didn't have their place on there. So here we are...
Subscribe to:
Posts (Atom)
Posts
