CISC/RISC argument is long dead, particularly for x86. The reason is simple. Only a relatively fixed amount of logic is required to decode x86 'CISC' like instructions into a more RISC-like core. Except for some minor tweaks, this hasn't changed in many years. However, because feature size on chips has gotten a lot smaller, the portion of chip required for this decoding logic has also gotten smaller. To the point now where the amount of space it takes up and the power it consumes is essentially zilch. A second reason is that it is NOT possible to improve performance these days by simplifying existing cpu core logic. All of those simplifications have already been done. The only way to improve performance is by adding fairly complex bits of logic such as bigger caches, larger TLBs, better branch prediction algorithms, on-cpu call/return address stacks and optimizations, register rename tables, and so forth. (There are concepts that can improve aggregate performance with simpler designs by moving to many (32+) cores, but software has a decade of catching up to do to take advantage of those sorts of cpus).
To the author: I liked the article, though I should point out that in-order vs out-of-order execution is only a small part of a super-scaler design. Intel has the advantage of being able to start from much higher-end architectures (xeon, sandybridge, ivy, etc) and simply reduce cache sizes and remove features until the desired power envelope is met. The original Atom started from an older Pentium design and worked upward.
What Intel is not really telling people is that the real golden egg for power efficiency is moving downward from the higher-end cores, not moving upward from the original Atom. At some point they will meet. I have no doubt Intel will pull more tricks to make them meet more quickly. Ultimately the power envelope for higher end cores will be measured by nanoamps per instruction executed rather than measured by core frequency or scaling optimizations. This means that in the next year or two Intel will be completely outgunning ARM cores on the low-end.
ARM has a much more difficult task ahead of it. Intel has already solved issues related to high-performance execution and only needs to fit those features into a power envelope. ARM, on the other hand, has to research and invent high-performance execution solutions, which is a MUCH harder task.
from MattZN2 on Apha
Some very positive commentary here. I do think Intel will run into another unforeseen dilemma when its atom chips are second to none. Are you all sure that Intel won't be the recipient of a concerted backlash, with cries of trying to monopolize the whole chip industry, if they are successful in the handheld mobile device segment?
Sentiment: Strong Buy
What you have brought up is actually something I'm sure Intel has considered carefully. Intel has to be concern about turning into a monopoly since AMD is now exiting the x86 market. I believe Intel will be content to develop their technologies in such a way that the Atom SoC, their future low-end SoC, will remain competitive with arm's best while positioning their future Core SoC for the high-end. So in the future (14nm and down) you will see EOMs come out with smart phones and tablets with Atom SoC inside, which represent the low end products for Intel but these will be competitive with the very best of arm's. On the high end, since at 14nm and down Core (Broadwell) SoC will fit into a phone, you will see EOMs creating much more sophisticated smartphones that will run full-fledged Windows/MacOS/Chrome OS. At 14nm (just a little over a year from now), arm's best cannot compete with Intel's core and Intel will basically own the high-end market. This is why when in my view, Intel will be in the strongest postion it has ever been in it's history. Buy and hold here. Do not believe the mass media #$%$ and hype. Intel doesn't fear arm, it fears that arm won't be competitive enough, and Intel would then turn into a monopoly, which it doesn't want. Arm fanboys think arm has intel by the ropes, when in fact Intel is toying with them.
I think you arm fans are missing the point of what Matt is saying. Once the process gets to 14nm and below, arm will not be competing with Atom, it will be competing with Intel's Core processors because the power envelop will be equal at that point. I believe Intel's strategy has always been to move the core down into the Atom's power envelop...once we get to 14nm, Atom will be the low end SoC while the high-end SoCs will be running Broadwell. arm's job is alot tougher, they have to improve performance while still maintain the power envelop and they are 4 generations behind. In terms of the "good enough" argument, well 1) prices go up you build higher performing processors and processes 2) Given the choice, I would pick a faster processor over "good enough" anyday
"I believe Intel's strategy has always been to move the core down into the Atom's power envelop."
ARM's success is partly due to that was exactly NOT INtel's strategy...
"In terms of the "good enough" argument, well 1) prices go up you build higher performing processors and processes 2) Given the choice, I would pick a faster processor over "good enough" anyday"
Do you use the i7 Extreme ?? Or is your CPU good enough...
Just to clarify, Intel does not have to modify anything with the Core processors to get it to the same power envelop as the Atom's...Once it gets to 14nm, the PROCESS itself will take care of it. They are modifying Atom now as a Low-end architecture...onces the two meet, at 14nm...they will be basically the same....but Core will be high-end while Atom low-end. This is why are you hearing rumors that Broadwell might not have sockets and will be BGP only.
"Intel has the advantage of being able to start from much higher-end architectures (xeon, sandybridge, ivy, etc) and simply reduce cache sizes and remove features until the desired power envelope is met."
Those great speedups to the architecture are what eats up the power...
INtel's dirty little secret is memory sizes matter..
If they bumped up the ATOM cache size, it would cut into other CPU SKUs...
"This means that in the next year or two Intel will be completely outgunning ARM"
Textbook INtel cheerleader...
"INtel will come out with something next year, and the competition is thru..."
"ARM has a much more difficult task ahead of it. "
YOu got me laughing... Thank You !!!!
Intel has been trying to get down to the power envelope of mobile for about 6 years now. Medfield is the first version of their chip that borderline fits. Their vantage point, therefore, continues to be the high power end. From there, its easy to speculate that getting your power consumption into a budget is simply a matter of optimizing from amongst a range of performance options that you've already developed. In other words a very tenable task. But if you look at their results from performing this very optimization over the last 6 years, they are depressing. It leads me to believe they don't know what they're talking about because they've never been there. ARM on the other hand has gone way lower than the mobile phone battery operating point. e.g. Their M0+ which is used in embedded.
Secondly, if you look at design trends amongst computing gadget makers such as Apple, they are moving away from performance driven improvement to a more use-centric approach. This greatly favors the low end of the power spectrum because with advances in process technology, the low end is going to become more and more performant until it is good enough for most applications. No-one can tell the difference between a game running at 30 fps and 130 fps because after a certain speed, you can't tell the difference. The high end, Intel's vantage point, therefore gets relegated to niche uses such as super-computers.
You should go write a puff piece about Arm on SA I'm sure they would let you, give all the reasons why 75x P/E is just peachy keen and investors should BUY BUY BUY... In th meantime I will short every bump like there's no tomorrow, CES is going to put a damper on your pipe dreams just like last year, ohhhhh this will be fun!
Sentiment: Strong Buy
"Medfield Atoms already blew away Tegra 3 and is competitive with Samsung's ARM. Next iteration would be even better."
Tegra 3 is 40nm versus Medfield at 32nm? I wonder why Google chose Tegra 3 for their tablets instead of Medfield?
Lets see how Medfield does against Tegra 4 on 28nm. My guess is that it won't even be close. Tegra 4 is a screamer.
anand came to the same conclusion in his review of Haswell....at 14nm, arm will not just be going up against Atom, it will be going up against Broadwell, which will be in the same power envelop!
This is the same conclusion from Anand Lal Shimpi at Anandtech. When process gets to 14nm and below, arm will be going up against the Core CPUs not Atom. Google "Intel's Haswell Architecture Analyzed"
This is the same conclusion from Anand Lal Shimpi at Anandtech. When process gets to 14nm and below, arm will be going up against the Core CPUs not Atom:
"With Core architectures dipping below 10W, I do wonder if that slide was a bit of misdirection. I wonder if, instead, the real goal is to drive Core well into Atom territory"