We’ve known for many years that Intel keeps making its chips more complex for increased performance and a higher price, but the Haswell architecture is a definite step towards putting the OS and its applications in silicon.
- Power management which would normally be a role for the OS and its schedulers is now going into silicon pausing unused portions of the chip to save power.
- High-level graphics functions are being placed in silicon right down to Codecs and image filters.
While these two strategies will increase performance they do so at a price. Increasing the complexity of chips this way does nothing to reduce the size of chips and the number of transistors but does the opposite. Silicon is much more costly than software so the price/performance ratio keeps getting worse for the end user/consumer. Locking in certain algorithms in silicon may lock in users to Wintel as well, something that does not benefit end users/consumers. It also guarantees that sooner or later our hardware will be subject to viruses/vulnerabilities, not just our software and data. If you want Patch Tuesday to mean changing the microcode on your CPUs this is the way to go.
No. There are benefits to integration of this kind but the costs outweigh those. There is a reason layers of abstraction work in IT. Mucking with the layers is a disaster waiting to happen. Microsoft and Intel are thrashing around trying to remain relevant but they only make things worse. Simpler is the way to go forward and ARM and */Linux have that sewed up.
Deep, deep dive inside Intels next-generation processor • The Register.
17407
12730
206
3
2
23841 
11830 
11682 
4612 
4236 
1641 
198 
14 
2 
0 
0 
0 
http://mrpogson.com some of this kills the pay for patent codec issue. Its in the cpu.
Yes lack of codec isolation does cause problems.
Silicon is much more costly than software so the price/performance ratio keeps getting worse for the end user/consumer.
There are a lot of consumers who are ready to pay, don’t forget Apple. And they have nothing against being locked.
And what about energy saving? You know electricity has some price, and batteries are not infinite. I think there are a lot of people who are ready to pay some extra $20 for an extra half-hour of battery life.
And why not to create a processor which will execute Microsoft bite code natively? And will support such features as memory management (allocation/garbage collection/boundery checking), UTF encoding and much more?
Imagine the boost of performance and energy savings. Even now Windows is more energy efficient than Linux so called desktop distributions.
There is a reason layers of abstraction work in IT
Less these layers better is performance and consumers are happy.
While these two strategies will increase performance they do so at a price. Increasing the complexity of chips this way does nothing to reduce the size of chips and the number of transistors but does the opposite.’
I wonder if you would say the same thing if it was an ARM chip rather than an x86 microprocessor.
Anyway, here’s an example:
http://www.maximintegrated.com/products/video/h.264-codecs/
By the way, this so-called “system-on-a-chip” technology is pretty much the common theme for ARM-based systems. So, what is it now? “Small cheap thingie” or “complexity”? Make up your mind, please.
That Exploit Guy wrote, “what is it now? “Small cheap thingie” or “complexity”? Make up your mind, please.”
Comparing the complexity of an ARMed system to Intel is very easy. Just count transistors, die size, power consumption. If it weren’t for new power management Intel wins on complexity and users lose on price/performance. The most complex ARMed chip is tiny and simple compared to any Atom chip. ARM doesn’t need to code that huge instruction set nor worry about heat-flow nearly as much.
iLia wrote, “Even now Windows is more energy efficient than Linux so called desktop distributions.”
Let’s see… I need 4gB RAM, a 500gB hard drive and a quad core CPU to work slower than GNU/Linux on an 8 year old PC with 40gB hard drive. How is that more energy-efficient? The old CPU may use 30W and the new beast 95W. The old RAM a watt or two. The new stuff 10W. The hard drive might be a wash but that other OS needs a faster hard drive to give it any advantage at all over GNU/Linux. I know. I compared a NIB(new in box) machine with specs like these with our old P4 machines and all the kids could say was “It’s so slow!” with all the body language a teenager can muster when they saw that other OS on a brand new machine.
‘Comparing the complexity of an ARMed system to Intel is very easy. Just count transistors, die size, power consumption.’
So, what’s your justification for these things being proper metrics for “complexity”?
Also, what about fabrication techniques? Transistor density? Number of components? Instructions per cycle? Do they count as part of your definition of “complexity” as well?
‘I compared a NIB(new in box) machine with specs like these with our old P4 machines and all the kids could say was “It’s so slow!”’
P4 was notorious for its lack of efficiency due to the use of deep pipelines (see “Netburst”) to sustain operations at high clock frequencies. This is in contrast to the AMD Athlon processors of the time that offers the same (if not better) performance though generally clocked at less cycles per second.
I was a proud owner of an Athlon, by the way.
Also, Intel has long since abandoned Netburst in favor of architectures derived from Banias (Pentium M).
That Exploit Guy wrote, “what’s your justification for these things being proper metrics for “complexity”?”
I would have thought it was obvious. More details = more complexity. Compare a short poem and an epic saga. Which do you think is more complex without any definition of complex? ARM is minimalist. There’s hardly anything in there that’s not needed to keep the motherboard extremely simple. Just hook connectors to it. At the same time very few costly heat-generating components are involved. The Intel stuff has hundreds of millions of transistors in a CPU to do the same stuff a million transistors can do in ARMland. Lots of Intel CPUs now have 4 or more cores each with thousands of millions of transistors in them. The first Atom CPUs had just a few tens of millions of transistors and ARMed CPUs have about a quarter of that. The die size of Corei7 is ~65mm2 per core while an ARM core is about 5 mm2.
‘More details = more complexity. Compare a short poem and an epic saga. Which do you think is more complex without any definition of complex? ARM is minimalist.’
This is interesting because, base on the same reasoning, I have arrived at a completely opposite conclusion – that an ARM-based SoC is anything but minimalist. After all, it’s a self-contained system on a single chip! How minimalist can you possibly expect it to be?
I am not even counting SoCs with moving parts such as digital micromirror devices. That’s a story for another time.
I think Pog believes that because a system on a chip is smaller that means it’s “minimalist”. Minuscule and minimal are not the same thing.