sigh good bye cheap processors

[Madpacket]

Since we're all speculating here my take is ARM technology will unseat X86 and Intel over the next 5 years as the leader of CPU design. All major growth markets now are mobile and ARM is leading the way, Intel is not catching up, they're further behind ARM than AMD is to Intel.

 

[BenchPress]

yeah, yeah i know all this stuff....but, with skyrim and physX using x87 code for soo long, i still have my doubts

One of the reasons it used x87 is because it takes real effort to write good SSE code. It typically involves assembly code or compiler-specific intrinsics. This effort has to be justified in expected increase in revenue. Also keep in mind that before the Core 2 Duo, SSE was executed on 64-bit execution units only so it didn't offer a great speedup for the effort and this negative image made the adoption very slow.

AVX2 is different in that it allows auto-vectorization of scalar code in any programming language. So it can easily be used with existing code. It's also executed on 256-bit execution units so the gain is huge for very minimal effort. It really won't be ignored like its predecessors.

 

[LOL_Wut_Axel]

Since we're all speculating here my take is ARM technology will unseat X86 and Intel over the next 5 years as the leader of CPU design. All major growth markets now are mobile and ARM is leading the way, Intel is not catching up, they're further behind ARM than AMD is to Intel.

Looks like you don't really know much about this market.

Intel already have a good SoC to compete with ARM; ARM doesn't have any good high performance processors.

 

[BenchPress]

Since we're all speculating here my take is ARM technology will unseat X86 and Intel over the next 5 years as the leader of CPU design. All major growth markets now are mobile and ARM is leading the way, Intel is not catching up, they're further behind ARM than AMD is to Intel.

Just because it's a growth market doesn't make it the biggest market. People still spend way more on desktop and laptop CPUs. Also, did you miss the news that Intel's Medfield x86 chip is competitive against ARM designs? And they've also announced that mobile chips will be moved to newer processes faster. So any ARM manufacturer will have a really hard time keeping up with Intel. ARM's 64-bit chips will also make them less power efficient, and out-of-order execution isn't helping them either in relative terms.

 

[Madpacket]

It's no secret that performance per watt is what matters when batteries are involved and this is where Intel is struggling in the smartphone market. OOO CISC x86 is will have to compete on superior fabrication advances as it's simply less efficient than ARM at the same die size. Long term this is a bad strategy for Intel because there is an end game to all this die shrinking magic. If Intel is really serious they will license ARM IP and would use their fabrication advances to get ahead of the competition. I suspect something like this may happen after they fail to attract any serious ODM designs. What they have out now is third world worthy.

 

[LOL_Wut_Axel]

It's no secret that performance per watt is what matters when batteries are involved and this is where Intel is struggling in the smartphone market. OOO CISC x86 is will have to compete on superior fabrication advances as it's simply less efficient than ARM at the same die size. Long term this is a bad strategy for Intel because there is an end game to all this die shrinking magic. If Intel is really serious they will license ARM IP and would use their fabrication advances to get ahead of the competition. I suspect something like this may happen after they fail to attract any serious ODM designs. What they have out now is third world worthy.

LOL, whatever you say.

And that's Intel's first try at the smartphone market on a relatively unrefined architecture. Needless to say, reality doesn't agree with you.

 

[Idontcare]

It's no secret that performance per watt is what matters when batteries are involved and this is where Intel is struggling in the smartphone market. OOO CISC x86 is will have to compete on superior fabrication advances as it's simply less efficient than ARM at the same die size. Long term this is a bad strategy for Intel because there is an end game to all this die shrinking magic. If Intel is really serious they will license ARM IP and would use their fabrication advances to get ahead of the competition. I suspect something like this may happen after they fail to attract any serious ODM designs. What they have out now is third world worthy.

Given that (1) Intel did have an ARM license and product line (XScale) and so they do know from first-hand experience the strengths and weaknesses of ARM SOCs, (2) Intel knows exactly what it takes based on first-hand experience to make technologically superior technology nodes on a timeline that bests everyone else in the industry, (3) Intel knows from first-hand experience the strengths and weaknesses of its design teams and microarchitecture pipeline, and (4) Intel knows from first-hand experience exactly what it takes to be highly profitable while technologically dominating a marketspace...I would say they probably know what they are doing in ways that no other company ever has or will.

I know of no other company that has all four of those feathers in their cap. The ARM guys know what they are doing for the markets they are in and the competition that they have encountered thus far. But they haven't encountered competition from a company that has access to superior process technology the likes of Intel's, nor the depth of IC design talent the likes of Intel.

AMD comes close, but again if you look at AMD and you add up all they knew about what they were doing then you see they also came to the conclusion that ARM was not the answer.

If x86 was destined for the dustbin then Intel would be freely licensing it to anyone and everyone in a desperate attempt to keep it relevant to the industry. Instead you see them doing the exact opposite, not letting Nvidia have a license and so on. They know they have something golden there and they are intending to cash in on it.

 

[mikeymikec]

The problem is that these are two sides of the same coin. Modern processors aren't designed to maximize performance, but to do so for a given thermal envelope. The reason AMD can't compete with Intel isn't that they can't crank out as much performance, but that they can't crank it out as efficiently as Intel can.

I assume that AMD was trying to achieve all things at once with Bulldozer - more performance, more efficiency, new architecture to allow further expansion for faster processors. What I was suggesting was to do away with 'more performance' and 'further expansion' as high priorities, and just concentrate on greater efficiency for the same performance.

 

[amenx]

And hello over priced crap like the last batch of P4's

http://www.tomshardware.com/news/amd-intel-cpu-apu-processors,15741.html

The P4s were not over-priced, the Athlon 64s were. AMD were gouging customers in 2005.

 

[VirtualLarry]

The P4s were not over-priced, the Athlon 64s were. AMD were gouging customers in 2005.

Oh no? $550 to $570, for a 2.8Ghz P4 CPU at launch, is not overpriced?

http://www.sharkyextreme.com/hardwa...61_1452751_9/Intel-Pentium-428-GHz-Review.htm

Surely, compared to todays' top chips, it was overpriced. (*)

(*) Well, my theory is that today's CPUs prices should be indexed to volume, and that is why today's CPUs are cheaper, comparatively, to yesteryear's CPUs.

 

[AtenRa]

Oh no? $550 to $570, for a 2.8Ghz P4 CPU at launch, is not overpriced?

http://www.sharkyextreme.com/hardwa...61_1452751_9/Intel-Pentium-428-GHz-Review.htm

Surely, compared to todays' top chips, it was overpriced. (*)

(*) Well, my theory is that today's CPUs prices should be indexed to volume, and that is why today's CPUs are cheaper, comparatively, to yesteryear's CPUs.

The Pentium 4 2.8GHz equivalent CPU of today is the Intel Core i7 3930K at the same cost of $500-600 as 10 years ago.

CPUs are not cheaper, in 2002 90% of Intel CPUs sold were Celerons. Today 90% of Intel CPUs are Pentiums(Celeron equivalent). It may seam to you that you have more performance spending $200 with a Core i5 2500K/3570K today but it was the same with the Core 2 Quad 4 years ago. You had the same performance then as i5 2500K has today, for the applications of that time against CPUs 4 years older than Core 2 Quads.

 

[cantholdanymore]

Since we're all speculating here....

You're speculating, some people in this thread know for a fact

 

[amenx]

Oh no? $550 to $570, for a 2.8Ghz P4 CPU at launch, is not overpriced?

http://www.sharkyextreme.com/hardwa...61_1452751_9/Intel-Pentium-428-GHz-Review.htm

Surely, compared to todays' top chips, it was overpriced. (*)

I did say 2005. My memory is a bit more sketchy for 2002.

The 2.4ghz Athlon x64 4000+ on release (2004) = $729

http://www.anandtech.com/show/1517

I bought a 2.8ghz Northwood C for around $260 around that time.

 

[beginner99]

but that's the problem...internet doesn't go faster from i3 to i7

The problem here is bandwidth. Your processor will always be a lot faster than what can be transferred over a network. I have 6 TB of storage but there are programs out there to minimize your JavaScript from 100kb to 20 kb. Go figure.

Talking about JavaScript. AFAIK it's not multi-threaded yet. That could help a lot.


Note that I'm talking about consumer stuff that runs on their laptop and not server stuff. On server side it's IMHO easier like web server: 1 thread per request or virtualization. On the client/laptop it's different. Besides the most common apps (Browser, office) most software is single-threaded with GUIs blocking...

 

[sm625]

It's not the storage, in fact the through put on the 7,200 RPM drive in the notebook is more than my old Raptor 74.

It doesn't matter that the apps are single threaded when you are running a lot of them. This thing is a turd. Maybe if it were 3.0+GHz but it's a 2.5.

Everything I do on the desktop is faster, smoother, etc. I can tell a difference - so can most people here at work (SAN engineers).

Dude, it IS the storage. The throughput on a notebook drive is not an issue. Dont look at MB/second. The raptor runs circles around every notebook drive in real daily use because the access times. Random small I/O latency is what kills every notebook drive and makes them feel so slow.

I run a E6600 with a raptor 74 GB, and this thing runs circles around every single PC in my building, save for the couple that have SSDs. It is so much faster it is like a bad joke. The 74GB raptor is going to feel faster than notebook hard drives made 5 years from now (except maybe when you are copying huge files, but really that's not what makes you want to toss a computer out the window)

 

[BenchPress]

Talking about JavaScript. AFAIK it's not multi-threaded yet. That could help a lot.

HTML5 Web Workers.

 

[SickBeast]

Here's the latest info about Piledriver:

A 26% improvement is not enough IMO.

 

[ShintaiDK]

Here's the latest info about Piledriver:

A 26% improvement is not enough IMO.

Over Husky? Aint Husky Llano?

Ouch ouch ouch. I feel another AMD disaster hitting.

 

[SickBeast]

Well, they're refining a terrible product into a bad product. There's not much more that you can expect.

 

[podspi]

That slide is talking about Trinity, not Vishera.

 

[Vic Vega]

...