(11-29-2017, 07:53 AM)JonnyH Wrote: That would be true if all those statements were correct - but the top 3 processes (Intel, TSMC, GF & Samsung) all have strengths and weaknesses - even if it's just yields and costs, or targeting different points on the area/power/performance curves.
But it's not really "transistors" individually that you care about, the "process" includes a library of logic cells that are the building blocks actually used, and within them multiple options that may be focused on different things (optimised for power, speed, area etc.), plus placing and routing all those things takes significant effort (and skill), and can against be designed to maximise particular goals. So even on the "same" process, there's the opportunity to have vastly different characteristics of the end product. Plus it's a massive undertaking to harden a design to an actual silicon implementation on a specific process, especially to do it well.
The simple fact that Intel likely have the "best" (on average) process for (relatively) high-power, high-frequency desktop CPUs, but that's likely as much due to the tight integration with the people working on the silicon implementation and effort put into the design and layout as much as the physical characteristics of the end silicon gates. And they seem to be struggling to get the low power efficiency required for mobile devices (you could argue that's as much design issues, as you can't quite compare apples->apples as an atom isn't the same as an arm cortex - but these days x86 decode is relatively small so I somewhat doubt it's a fundamental issue with the ISA rearing its head).
And they're pretty much nowhere with GPUs - if they did have the 'massive' process advantage some people seem to claim, they'd just be able to power through any design deficit. And in CPUs the ryzen stuff shows the others can at least be competitive in many situations.
I think that was a rather long-winded way of saying "It's not that simple, it's really hard to directly compare processes".
I would agree with your overall point and most of your points, but a few things I would add.
Intel's IGPs are very competitive if you take compute applications into account, which is what they're mainly optimized for. They use much more complex "cores" (ALUs mainly) which of course sacrifices raw throughput for greater flexibility. Thus the reason that on a per watt basis AMD/Nvidia beat them in 3D rendering but lose in compute. AMD is doing the same thing now with GCN but to a lesser degree. And Nvidia is pretty much heading in the exact opposite direction with their latest vliw offerings.
As for ultra mobile devices (smartphones/tablets) back when Intel was actually competing in this space their chips ran circles around the competing flagship arm chips performance wise. They also consumed a lot more power though. Performance per watt though was slightly better in most cases. But that's not what consumers needed. They needed chips that were "good enough" that could offer good battery life which arm did a better job of providing. Intel simply didn't "dumb down" their pipeline enough for that market. But that seems more like a design problem than process. They also really struggled with the I/O and wireless aspects of the SoC. The cell radios on arm SoCs were so much more efficient it was ridiculous, now THAT was likely a process issue. There is no way the x86 decode hindered them to any significant degree. Modern arm cpus do cisc style decoding anyways and the complex instructions in x86 reduce bandwidth and cache usage as evidenced by the struggles of Itanium and RISC cpus back in the early 2000s despite their massive buses and caches.
I was always under the impression that "process" was short for "manufacturing process" which would include the hardware used for manufacturing but I would think the logic library would fall under "design" and not "process".
Edit: Been googling for awhile now. Everything I can find also seems to list it as part of the design, not the process. I think that's where the confusion between us lies.
"Normally if given a choice between doing something and nothing, I’d choose to do nothing. But I would do something if it helps someone else do nothing. I’d work all night if it meant nothing got done."
-Ron Swanson
"I shall be a good politician, even if it kills me. Or if it kills anyone else for that matter. "
-Mark Antony
-Ron Swanson
"I shall be a good politician, even if it kills me. Or if it kills anyone else for that matter. "
-Mark Antony