I believe the inquirer article is hogwash.
I don't know what planet the author comes from. But VLIW, locally recompiled code, individually compiled code for every cpu and generally great dependency on compilers, are all very old ideas. Old ideas that have been around for decades and have been invested very heavily into both in terms of billions of $ and in man years of work. And they are sofar failures.
It would be fantastic for AMD if Intel were to squander their vast advantage in engineering resources and funds on such adventures.
It's perhaps a good moment to remember that Intel really have never, not ever in their entire history, done anything good by themselves. 8086, 80286 and Itanium, for instance, deserves being recognized amongst the worst CPU architectures of all time.
8086 and '286 were commercially successful only because of the credibility that the IBM badge earned for the 'PC' amongst business people.
And the limited success that Itanium enjoys is only due to various business ties/dealings, - Intel-HP-DEC which have killed off HP-PA and Alpha.
And Netburst is arguably only successful because of Intel's heavy handed and 'criminal' flexing of their monopolistic muscles.
The clever '386 32-bit ISA, that saved the 'PC', and which we have been using since the advent of Win95 and Linux, was basically a design requirement from Microsoft.
The P6 core ( Pentium Pro through to PIIIe, and arguably, PentiumM ) that Intel's fortunes have relied so heavily upon in recent years was "borrowed" together from DEC and Nexgen.
And the AMD86-64 ISA that the future depends upon was, of course, designed by AMD.
During all this, Intel's contributions have been limited to leaning on benchmark providers to rig their tests to favor Netburst. "AMD is best for gaming" - no, frankly, AMD is best for pretty much everything, but the gaming benchmarks are the ones that correlate to reality.
So both because of this and because of Intel's recent technical experiences with the failures of Netburst and EPIC, I don't think Intel will climb far out on a limb with Conroe.
And it would be foolish to expect continued design ineptitude from Intel indefinitely. Companies change, and Intel have over the time acquired a lot of experience, knowledge and tools.
...But then, more in the Intel tradition - if it can't be invented by Intel -, there is the purchase of Elbrus... (though I somehow think this is more related towards getting the 'epic' Itanic to take off)
So there are very good reasons to be curious and even excited about the Conroe.
My belief about what the Conroe will be haven't changed much since I joined this forum. But I have absolutely no knowledge.
My view of the Conroe is that it will be a CPU with CISC decoders. AMD86-64 decoders more precisely.
I also think the basic groundwork core layout is going to be something roughly along the lines of the K8, but wider and smarter.
I think it will have four decoder pipelines. I also think it will have four execution pipelines.
I think the core will be designed to be used in multicore CPUs. Multicore, not just dual core.
I don't think Conroe will feature hyper threading. It's basically worthless and a lot of trouble.
I'm guessing that the advanced novelties are going to be found partly in caches, branch prediction, speculative execution, stuff that supports the OoO sheduling, - and partly in the getting the work in and out of the CPU, the memory interface.
But it's just a guess, and not an imaginative one. The crux, as I see it, is to get as much load as possible on the four execution pipelines. This seems hard, and would thus be where Intel may have a technical breakthrough.
**************
Maybe we should step back a moment and consider general movements and requirements of computing.
The Conroe's design will surely reflect what Intel are thinking about that and the future.
The problem areas I see are:
1: Containing power requirements with a very large, and increasing, amount of transistors on the chip. The ever increased number of transistors is required by the demand for higher performance at conditional computing. The inquirer article author obviously think the solution here is to reduce the number of transistors, and thus he has an entirely different angle on this.
I believe the solution is to hold back clock rates and increase IPC. (Power requirements, I think, increase about seven times faster than clockrate.)
...- besides process improvements, of course.
Again this belief is not particularly imaginative, and only follows the direction that the success of PentiumM and Athlon64 (and Elbrus) points to. There are reasons to be curious about the Conroe.
2: Getting the work into and out of the CPU. This is where AMD's integrated memory controller and hyper transport have triumphed a bit lately. But I suspect there is a lot more to be done here.
Again, the inquirer article author has a different angle because he seems to think that the problem is execution speed inside the CPU. I don't know why he thinks that. Maybe it has to do with his idea of limiting the amount of transistors. Or maybe he's drunk too deep from RISC religion and anti-CISC propaganda.
3: For some reason (I can't understand), the consumer market is moving towards laptops. This is surely also one of Apple's reasons for interest. Low power requirements. See 1: above.
4: Blade servers and heat. See 1: above.
***************
AMD - Intel relationship.
I believe we are going to see some reversal of properties, in the immediate future. Conroe vs.K10. AMD may have higher clock rates.
Also, my guess is that AMD will go for more cores per chip and less performance per core than Intel. That is really sticking my neck out, though.
AMD does not have the market support enough to be able to compete with Intel engineering, once Intel gets something right. This is extremely serious. I doubt very much many of you understand how really serious this is for computing in general, and for Wests longterm leadership in computing technology.
Finally, to reflect on something in the inquirer article, and give it some due:
Intel would like to use their market dominance to somehow exclude and kill off AMD.
For that reason, introducing a new, different, proprietary ISA must be very tempting.
In fact it seems Intel have already tried that and been held off by Microsoft. But they might try it again. And maybe, maybe, maybe Microsoft actually folded, and we're just waiting on Longhorn? Maybe that's the real reason for MS recent lack of interest in the Itanium.
They must be backwards compatible with x86 and yet not fall under the X86 cross licensing agreement. It implies some software stage or morph.
Combined with this, might also be an attempt to reintroduce EPIC on the desktop. Though I think in this case it is more likely to be something completely new, but related.
But anything like this could be an opportunity for AMD.
Facebook
Twitter