The AMD Execution Thread [2007 - 2017]

Status
Not open for further replies.
1) Battery life under typical mobile conditions
Battery life is a qualitative concept. Show me comparative energy usage for a different set of tasks and I get interested.

2) Integrated graphics - if you don't plan on having a discreet GPU then AMD win hands down
Like I said: AMD is better at the peripheral but fails at the main actor. There appear to exist some people for whom this matters, but there are obviously not a lot of them.

The reason is simple: the biggest use case of a GPU is to scroll text. Even Intel graphics are good enough for that. (This is the same argument I've been making for tablet SoCs.) Intel has been putting just enough GPU power on its CPUs it can get away with, because it understands very well that putting a faster one won't materially increase their value, unlike CPU performance. It's only now that even blazing fast CPUs are starting to succumb to Moore's curves that they are piecemeal increasing their share of the die. You can only put so making CPU cores on a desktop machine. But I don't doubt for a moment that they'd not have done so if there were bigger increases possible in single threaded execution.

3) Tablets: Atom is shit and IVB is incredibly expensive and still too hot for fanless. Jaguar will the be ideal mobile chip for many people, if they ever get the chance to buy it
4) Anything to put in a console
3) is a largely non-existing market for now. Once it's not, it won't take much to come in and take over.
4) is a niche market with a nice steady income that's probably sufficient to pay Intel's electricity bills but little more than that.
You might as well have mentioned discrete GPUs then.

When talking about anti-competitive practices, it doesn't make much sense looking at small markets where the aggressor chooses not to play.
 
The reason is simple: the biggest use case of a GPU is to scroll text. Even Intel graphics are good enough for that. (This is the same argument I've been making for tablet SoCs.) Intel has been putting just enough GPU power on its CPUs it can get away with, because it understands very well that putting a faster one won't materially increase their value, unlike CPU performance.

It seems counter-productive to me to freeze a market into the status-quo. Putting hardware into a device which can do more than you asked for creates the foundation for evolving a market into other directions. The desperate search of the gaming-industry to get more customers finds very rich soil to bloom on if the not-suppose-to-be-good-enough-for-gaming-hardware is suddenly/surprisingly more than sufficient.
<rant>If some companies could think further than their managers nose, they should even consider subsidizing AMD (paying the whole of the GPU-block, Dell just buys at the cost of a CPU). Instead AMD stretches itself in this uphill battle. It's sad there's not much synergy [possible anymore] in the higher spheres of business, everyone waits for the moment when a company exhausted their investment into R&D, so they don't need to spend on it, and if they don't get bargains they just wait again until they can pick on the corpse or become the corpse. Bah. Anywho, what does AMD or intel matter in the sea of money the world drowns itself in. :cry:</rant>
 
So basically according to silent_guy, it's all crap apart from I guess the markets that intel is winning in?

I think intel is more concerned about the markets it's winning in evaporating in front of them. Good luck winning in tablets and phones where they aren't even a bit player.
 
So basically according to silent_guy, it's all crap apart from I guess the markets that intel is winning in?

I think intel is more concerned about the markets it's winning in evaporating in front of them. Good luck winning in tablets and phones where they aren't even a bit player.
Intel should be, and is, very much concerned about other markets: the one Qualcomm dominates (who has a bigger market cap now than Intel.) It just so happens that this is not a market AMD is competitive in either.
 
Intel should be, and is, very much concerned about other markets: the one Qualcomm dominates (who has a bigger market cap now than Intel.) It just so happens that this is not a market AMD is competitive in either.

Then again at least AMD has experience of operating against the odds and an uneven playing field. I get the feeling that we're going to watch the slow demise of a giant as it fails to understand the market it so desperately wants to dominate.

There are too many big players for ARM for Intel to be able to just bully and bribe their way to success, but it sure will be fun watching them try.
 
Intel should be, and is, very much concerned about other markets: the one Qualcomm dominates (who has a bigger market cap now than Intel.) It just so happens that this is not a market AMD is competitive in either.

Let's see what Temash can do. I don't expect AMD to compete against Qualcomm in the latter's core businesses (phones and radio chips) but they do have a shot in tablets, I think. Of course, they don't exactly have a great partner in Microsoft, but maybe Windows 9 will suck less.

Then again, Intel is working pretty hard on Android for x86, so perhaps AMD will once again be able to ride on their coattails.

As for phones, I can't see it happening with Temash, but on 20nm, paired with a radio chip from Qualcomm, why not?
 
Then again at least AMD has experience of operating against the odds and an uneven playing field.
They have a lot of experience not making a whole lot of money in said playing field.

I get the feeling that we're going to watch the slow demise of a giant as it fails to understand the market it so desperately wants to dominate.
I think Intel understands 1 thing incredibly well: that market that you claim they don't understand commands a dramatic lower price/mm2 than their CPU business. Say $20/SOC instead of $100-$1000 per CPU. They have very good reasons to stay out of that race to the bottom until they really have no other choice.

There are too many big players for ARM for Intel to be able to just bully and bribe their way to success, but it sure will be fun watching them try.
They don't need to bully and bribe: massive process superiority goes a long way.
 
I think Intel understands 1 thing incredibly well: that market that you claim they don't understand commands a dramatic lower price/mm2 than their CPU business. Say $20/SOC instead of $100-$1000 per CPU. They have very good reasons to stay out of that race to the bottom until they really have no other choice.

Left it far, far too late.

They don't need to bully and bribe: massive process superiority goes a long way.

Wake me up when they have a chip that takes advantage of it.
 
I think Intel understands 1 thing incredibly well: that market that you claim they don't understand commands a dramatic lower price/mm2 than their CPU business. Say $20/SOC instead of $100-$1000 per CPU. They have very good reasons to stay out of that race to the bottom until they really have no other choice.

I would argue that they are very much at that point already. And given their efforts regarding power consumption, whether it's on Haswell or Atom, I think Intel agrees.
 
For people that still live in a world were the Atom architecture is not competitive in phones and tablets, even on an older process, I suggest checking these reviews/articles:

http://www.anandtech.com/show/6330/the-iphone-5-review/10
http://www.anandtech.com/show/6529/busting-the-x86-power-myth-indepth-clover-trail-power-analysis
http://www.anandtech.com/show/6536/arm-vs-x86-the-real-showdown/14

And for people who aren't fooled by Anandtech's ability to always find the best case for intel, there is this:

http://www.phoronix.com/scan.php?page=article&item=samsung_exynos5_dual&num=1

Where the A15 totally dominates the Atoms including the D525.
 
Comparing to a 45nm 3 year old Atom makes a lot of sense :rolleyes:
To not mention they don't even try to understand performance per watt.
I'll stick to the least imperfect analysis, thank you ;)
 
Comparing to a 45nm 3 year old Atom makes a lot of sense :rolleyes:
To not mention they don't even try to understand performance per watt.
I'll stick to the least imperfect analysis, thank you ;)

Yes that 45nm D525 chip has very similar performance characteristics as the Z2760, so it does make a lot of sense.

It's very difficult to ascertain perf/watt with so many of them hiding the true numbers as well, plus overall system draw can vary from part to part.

The Atom may draw slightly less power at the chip level but its performance isn't up to the A15's outside of Anandtech's cherry-picked benchmarks.
 
I would argue that they are very much at that point already.
Yes, I agree.

ARM SOCs and Intel CPUs are on an interesting collision course wrt features/performance/power. All things equal, you'd expect Intel to win this in the long term purely based on process and pricing power. I don't really buy into the claim that ARM architectures have an inherent advantage, especially on SOCs where the CPU area isn't too large to begin with.

(Thinks back about Byte Magazine cover, in real paper, in the 90ies claiming Intel in danger because of PowerPC threat.)
 
Last edited by a moderator:
Yes, I agree.

ARM SOCs and Intel CPUs are on an interesting collision course wrt features/performance/power. All things equal, you'd expect Intel to win this in the long term purely based on process and pricing power. I don't really buy into the claim that ARM architectures have an inherent advantage, especially on SOCs where the CPU area isn't too large to begin with.

(Thinks back about Byte Magazine cover, in real paper, in the 90ies claiming Intel in danger because of PowerPC threat.)

I agree about the collision course, but Intel ultimately winning is less obvious to me. They do have a process advantage, but there is some question as to exactly what it will amount to in a couple of process nodes, given that the power benefits brought by each generation seem to decrease. On the flip side, the power advantage inherent to the ARM ISA, or rather to ARM decoders, is real, but small. Technically, this looks like a draw, or at least a close call.

Therefore, I think deciding factors are to be found elsewhere, and the ARM ecosystem does have a few advantages: being more fragmented and less vertically integrated, the risks and costs are spread over more companies, making the whole more resilient to mistakes, failed projects, etc. For instance Samsung's loss could be Qualcomm's gain, and vice-versa. Of course, this also means that some of the synergies possible for Intel are less likely with ARM, and there is some redundancy.

Difficult as it may be, I think it is useful to try to project ourselves into the "distant" (by this industry's standards) future: five years from now, what will SoCs look like? It is possible that some new need for ever faster hardware will emerge, but there is a real chance of commoditization of the SoC market, which I do not think would be to Intel's benefit. In fact, highly integrated companies like Samsung, or even Apple, would be more likely to profit from this, as they could still make money from their finished devices with high added value.

But what OEM would be willing to pay $200 for a SoC/CPU when a $50 (at most) alternative could provide perfectly adequate performance, perhaps 70~90% of this more expensive SKU? This could be a very difficult adjustment for Intel. Ironically, AMD's relatively small size, fablessness and versatility (especially with HSA) could turn out to be a big advantage in such a situation.


Then of course is the whole software side of things, which is harder to predict, but I suspect that Microsoft's x86 legacy will be about as much of an obstacle to ARM as Android's reliance on the latter. That is, with Windows RT on one side and Android/x86 on the other, or rather their respective successors, nothing insurmountable.
 
Yes that 45nm D525 chip has very similar performance characteristics as the Z2760, so it does make a lot of sense.

It's very difficult to ascertain perf/watt with so many of them hiding the true numbers as well, plus overall system draw can vary from part to part.

The Atom may draw slightly less power at the chip level but its performance isn't up to the A15's outside of Anandtech's cherry-picked benchmarks.

Slightly less? It draws a LOT less power than the A15. The Exynos 5 consumes up to 4x the power depending on the workload. Considering that the Atom architecture hasn't changed significantly (it's mostly been with an eye towards reducing power consumption the past 3 years), it shouldn't be surprising that a modern CPU core should be able to outperform it (while also consuming significantly more power).

The fact that a 3 year old CPU architecture on an old Intel manufacturing process (32 nm) is even able to compete with a brand new arch. is quite significant.

Regards,
SB
 
AFAICS, Intel's biggest hurdle is simply others not wanting to do business with them (for various reasons). Well, that and "good enough" performance.
 
I've always thought that was dumb anyway. Besides, Ferrari and Acer's respective brand images sort of cancel each other out, don't they?
 
Status
Not open for further replies.
Back
Top