A critical question would be whether the benchmark is really created to be useful for design engineers, or whether it is manipulated to give consortium members marekting ammo.
Both. If you monitor any of the standards making efforts, you see lots of jockeying for position, and "We firmly support standards! Do it our way!" posturing.
I'd expect less of the latter in this instance, simply because there would likely be fewer participants with an investment in an existing approach that they wish to preserve.
I think of the need as analogous to what a software engineer does when developing code. They'll do profiling to determine where their code is spending its time to speed things up and reach performance targets. (And the places that need optimizing often aren't where the developer might expect.)
I wonder what developers of stuff that could use this sort of benchmark use now? I can see some engineer or the engineer's employer saying "This is how we do it." and propose it as a base for how others mighjt do it.
Thanks for your input as always DMcCunney. Obviously a key part of the challenge here is defining what is needed in a benchmark for use within the industry. But that still leaves unanswered the question of who will lead this effort, if anyone. A consortium would seem the most logical choice, but we know how the politics of consortiums can be. A critical question would be whether the benchmark is really created to be useful for design engineers, or whether it is manipulated to give consortium members marekting ammo.
I would expect some sort of industry consortium, but that would require a perception by the industry that standard benchmarks are needed, with a standards effort to define what the benchmarks measure and how they do it as the first step.
So the question becomes what purposes the benchmarks will serve and who the projected audience will be. I don't see such benchmarks as being intended for consumers (though they will almost certainly be used that way when they exist.) If you are in the industry, what benchmarks will you need, and how will you use them?
Since the main addition I can see to such benchmarks is measuring platform efficiency of a particular design, if I'm an engineer, I want to see whether my design is hitting the platform efficiency targets I have, with data to indicate why I'm not and what changes I have to make to hit my targets if I'm not.
They're the sort of measurements I'd expect to apply in the prototype stage to help me reach what is wanted in what my employer wants to sell.
It will take broad industry participation to develop effective benchmarking standards that represent both current and future usage models and system configurations. The question is who will lead this effort?
So what's the solution? In the comments on the earlier articles, we saw very clearly that many people disregard benchmarks altogether. Is there any hope of overcoming these challenges and making benchmarking a meaningful and worthwhile activity? Or should we just disregard them?
Blog Doing Math in FPGAs Tom Burke 2 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...