Submitted by With the Xeon 7600 line, Intel is finally using the ‘R’ word: RISC. It’s targeting the mission-critical market dominated by Sun SPARC and IBM Power with the new chips, a first. Can the Xeon E7 processor deliver Intel’s final blow to the RISC market, which includes its own Itanium?
RISC market, which includes its own Itanium?
Are you implying that Itanium is RISC?
Yeah, it’s a VLIW architecture or something along those lines. Not RISC or CISC.
No, they’re saying that Itanium competes in the market once dominated exclusively by RISC architectures…
Huh?
x86/x86-64 only dominates the PC market due to the backwards compatibility pressure of the market. In fields where such an issue isn’t present, x86/x86-64 is just another processor, and not as good as many other competitors. See phones/tablets/netbooks as an example – x86/x86-64 doesn’t hold a candle to ARM. Look at consoles – all three major consoles are PowerPC based. Look at handheld consoles – they are MIPS or ARM.
It’s the price. Commodity hardware servers are the norm today and x86 is a commodity architecture.
Prices for System p or SPARC are just plain ridiculous…
No, it is many things, backwards compatibility being one of them.
The principal reason why x86 dominates in the middle and high performance spaces is that it simply has the best price/performance ration in that segment.
A POWER or SPARC processor may cost an order of magnitude more than the comparable x86 part. So unless the POWER or SPARC system brings a significant value added, most business tend to go for the cheaper price. Which is exactly what has happened.
A POWER or SPARC may cost more, but not the PowerPC, ARM, SuperH, or MIPS. All are either very competitive, or actually much less expensive. And that goes back to my point that if an x86 is NEEDED for some reason, like BC, it’s not used; instead, the PowerPC, ARM, SuperH, or MIPS is used.
Which is why I said clearly that I was referring to the middle and high performance market segment.You know the one the article was referring to.
Can you find a SPARC or POWER processor with similar performance/price ratio as a modern x86_64 part?
Not the high-end, but the middle could have been dominated by PowerPC except for BC issues.
And before that MIPS could have dominated, and before that SPARC, and before that 68k, and…
They coulda, shoulda, woulda, but they didn’t.
X86 has always had the advantage that it has rode both curves: process and microarchitecture. While most of its competitors were riding only one. PowerPC may have had an initial microarchitectural edge, but it was on a stagnant process.
Because businesses demanded BC. If IBM had picked anything other than the 8086/8088 for their first PCs, that would have been the main CPU today.
And if my grandma had balls we would have called her grandpa.
What is your point exactly?
The same one I made in the original post with one caveat – without the need for BC, x86 would not dominate any market with the exception, perhaps, of the high end. Even there, x86 only dominates because the middle and low end pushed the x86 along and provided the money needed to do the R&D for high end chips. ANY PLACE BC ISN’T NEEDED EXCEPT THE HIGH END is now dominated by RISC, with the examples being set top boxes, consoles, handhelds, PDAs, netbooks, smart phones, etc. CLEARLY x86 only dominates in areas where BC is the major concern of BUSINESSES.
It doesn’t matter what made x86 chips better for the markets they dominate in*, the simple fact is that they are better.
*which is mostly: effects from economies of scale in one market making the chips better for few others; those effects (and overall the sort of evolutionary pressures acting also with technological “organisms”) BC being often desirable are a simple facts of life.
Edited 2011-09-26 23:55 UTC
Not even always both, there were some not very impressive stretches. But it had enough.
Edited 2011-09-26 23:50 UTC
Interesting facts:
IBM POWER6 servers was several times faster than x86 cpus, and costed 5-10x more.
Latest IBM POWER7 servers is 10% faster than Intel Westmere-EX and costs only 3x more.
Future POWER8 will be slower than x86? Then IBM has to lower the price even more, so POWER8 costs 2x as much as x86 servers. Why would you then buy POWER8 servers, if they are slower and more expensive? IBM does only high margin business, and kills off low margin business. POWER is going to be low margin.
IBM Unix dialect AIX runs only on POWER servers. If IBM kills POWER servers, then AIX will die too.
Coincidentally, IBM has officially said that AIX will be killed off and replaced with Linux:
http://www.zdnet.co.uk/news/application-development/2003/01/29/ibm-…
On the other hand, Solaris runs on x86. And it is open sourced in OpenIndiana distro, so it will not die.
Curious how Apple switched to x86 a few years back, against their BC concerns.
Or how Linux enthusiasts, who can mostly compile to anything, are almost virtually on desktop / laptop x86 machines.
Intel used to be part of the RISC, business, indeed. They only cut themselves out with the great idea of selling their XScale division to Marvell.
Don’t forget the original Processor Architecture for Windows NT the i860.
http://en.wikipedia.org/wiki/Intel_i860
Beware, potential wiki-loop there.
When is that not a risk? (and let me use this opportunity to introduce you to, say, tvtropes or everything2
)
none of the writing of this story makes sense, just delete it all and link the hp blog if you want
Well, (and this is a response to multiple comments) there are two RISC markets, essentially – the embedded market (which includes phones and consoles), and the server market.
The server RISC market is the “serious iron” (other than mainframes, anyway) that Intel hasn’t quite been able to touch with their x86 stuff, for various reasons. And, despite Itanium not being a traditional RISC CPU, POWER isn’t exactly RISC nowadays, either – it has memory-memory instructions and micro-ops, both the hallmarks of a CISC CPU.
That’s what Intel’s claiming they are able to compete with now.
I think it was Linux Torvolds who said that at the end of the day you are performing the same amount of work whether your processor is CISC or RISC. In full disclosure I work for Big Blue on the software side. My opinion is my own and doesn’t represent anyone else living, dead, or yet to be born.
Will X86 appear in the high end? Of course they will. However as the article states RISC is not going away. At the top you are already seeing hybrid mainframes made up of x86_64 and power/RISC processors. The hybrid approach is to address specific workloads. The interesting thing is the intel machines that make it on the top500 are mainly driven by GPU and not CPU power.
The top is about raw efficiency and governance. As I quoted before, the difference between RISC and CISC workloads is negligible but it is what surrounds the processor that makes the difference. RISC based mainframes have nearly 100% workload efficiency which is very important at the top. It may not be important for a desktop but for data centers it is. Since data centers drive the Cloud services that provided content for many of our connected ARM devices, this market is growing. If anything I believe X86 will be reducing in number. It will be slowly replaced from the bottom by ARM.
Now of course someone is going to state that you can buy enough cheep x86 machines to top the 500 list but you end up paying the difference and then some in electricity. Unless you are from a country that has cheap electricity or have your own renewable energy generator (wind turbine, geo thermal, etc…).
In short I agree that Intel will increase their numbers in the high end market but I don’t think the RISC processor numbers will decrease, I think the high end market will get bigger mainly driven by cloud services and virtualization.
I am sure you know what you are talking about, but there is a small mistake in there IMO. IBM’s mainframe processors are not actually RISC but CISC. I am sure they share a lot of their internals with IBM’s power processors, but the ISA is CISC. Ofcourse what you probably meant was x86 Vs the rest of the market where the rest of the market was labelled as RISC. And the rest of the market is often RISC, the only exceptions are the Itanium processors and IBM’s Z series.
IBM’s* z line CPUs(the mainframe stuff) are not Power. They have different tasks at hand, they were built for a different environment. There is literally no other CPU that was specifically built like z was and not a single one can compare.
The z CPU was never intended for high load processing, it was created for total throughput. Thus Watson is not on a z CPU, but on Power. And IBM’s Supercomputers are on Power instead of z. In fact, Power CPU was a result of the early IBM’s supercomputers.
* – my former employer
The POWER CPU is the result of an early IBM text processor.
There are very few true RISC processors left. The longer processors are on the market the greater the instruction set gets. You are correct. I used the term RISC to mean the market which includes IBM, HP, SUN/ORACLE, etc…
There are also very few true CISC processors left. The complex instructions are programmed in microcode running on a very simple instruction set implemented in hardware. You could almost say that RISC and CISC sort of converge towards each other to a more optimal golden middle, but not quite. There are still differences between the two, such as special purpose vs general purpose registers, address modes, etc.
CISC and RISC, and even VLIW, have been meaningless acronyms ever since microarchitecture and ISA became decoupled concepts a while ago.
The average x86_64 CPU tends to a very modern microarchitecture who happens to be able to execute old x86 instructions. Same can be said about SPARC, and POWER for example. Not one of those architectures happen to execute most (if any at all) of their ISA’s natively. All of them break them down into micro ops which are not visible to the programmer.
SPARC and POWER, for example, had to do some serious architectural gymnastics in order to allow aggressive superscalar, vector, and out-of-order execution of their original ISAs. Just as intel and AMD has to do with x86.
The main advantage most RISC CPUs had over old CISC designs, in the desktop/high performance areas, was that they offered a cleaner path to 64-bits and marginally faster pipelines… when transistor budgets for logic were still an issue. But that was almost 20 years ago.
Edited 2011-09-20 23:16 UTC
Well they still are in most (units shipped) processors, where low power usage matters; like in ARM, still quite RISCy, and at least some time ago still not very microcoded.
Edited 2011-09-27 00:15 UTC
I think you are confusing servers and supercomputers here. They are different things with different purpose.
In this regard it is not. The top is all about RAS.