Intel to demonstrate 64bit Xeon/P4 in February

[Duvie]

Originally posted by: Pariah
"Poll the public and ask them what they think of AMD getting out 64-bit first for the home and you'll get one of two responses: (1) who is AMD? or (2) what is 64-bit? Basically the public doesn't give a damn."

It's not that the public doesn't give a damn, it's that they just don't know. Once again proving that AMD is its own worst enemy and has failed itself again. They've had the buzzword advantage since the A64 was released, yet have still managed to completely fall on their face because they haven't marketed their product at all in the mass media. How many months did it take a major OEM to release an A64 desktop system after the chip was released? Intel is going to manage to jump into the market before the public learns about the technological "advantage" that AMD had. Intel will have their usual marketing blitz when they release their 64bit CPU, and any advantage AMD had will be gone in a blink. AMD will never really succeed if it doesn't learn how to market better. Apple practically kept the company alive for years on shrewd marketing alone.

This pretty much says it all.....

AMD has had an opportunity here to gain much recognitionand hence marketshare but they choose not to advertise and they already lack oem support from many of the big ones....

So in the end INtel coming into the game late will be able to come in introduce the public to 64bit (cause once again the average public does not know about AMD, trust me I build them for ppl) steal the limelight and regain any little amount of marketsahre they have even lost to AMD.....

It is quite funny...AMD seems to be a rather poorly ran company as far as I see it....I have been harping this since the days of 1ghz athlons and I was amazed they were not preaching the power of their chips...

 

[IPLaw]

I bet Intel needs a neck brace for changing directions so fast and so often.

 

[ViRGE]

Originally posted by: Pariah
"Poll the public and ask them what they think of AMD getting out 64-bit first for the home and you'll get one of two responses: (1) who is AMD? or (2) what is 64-bit? Basically the public doesn't give a damn."

It's not that the public doesn't give a damn, it's that they just don't know. Once again proving that AMD is its own worst enemy and has failed itself again. They've had the buzzword advantage since the A64 was released, yet have still managed to completely fall on their face because they haven't marketed their product at all in the mass media. How many months did it take a major OEM to release an A64 desktop system after the chip was released? Intel is going to manage to jump into the market before the public learns about the technological "advantage" that AMD had. Intel will have their usual marketing blitz when they release their 64bit CPU, and any advantage AMD had will be gone in a blink. AMD will never really succeed if it doesn't learn how to market better. Apple practically kept the company alive for years on shrewd marketing alone.

Mind you however, AMD is managing to sell everything it has on hand the last couple of quarters; Flash memory and CPUs alike. In the short term at least, marketing is a cost that they can't re-coup.

 

[Eug]

Itanic is toast.

 

[Adul]

Originally posted by: Eug
Itanic is toast.

itanic can make toast

runs hot enough at least.

 

[jm0ris0n]

Haha

 

[Chadder007]

Itanium = Owned

 

[Accord99]

Originally posted by: Chadder007
Itanium = Owned

Of course, given the significant support that Itanium has from OEMs, excellent performance in all configurations, superb products like SGI Altix, and being the server replacement chip for HP's PA-RISC, Compaq's Alpha and SGI's MIPS, its likely that Itanium can stand on its own now.

 

[jhu]

itanium with wide oem support? the only main ones are hp and possibly sgi. dell and ibm sell itaniums as well but they seem more like afterthoughts. by releasing their own x86-64 intel is sending the message that they fvcked up with itanium's dissemination. there's an article on aceshardware detailing a good way of killing off x86 (but that's with 10 years of hindsight, which is always 20/20). i think the immediate losers here are hp and intel. sgi can just go back to using mips, dell will be business as usual, and ibm will just chug along with their power4 and soon to be power5 (and opteron as an afterthought).

 

[digitalsm]

Originally posted by: jhu
so people don't need 64-bits right now, huh? alright intel, then why the fvck are you introducing a 64-bit x86 processor?!? hypocrites...

anyway, i think itanium will go the way of the alpha: superior technology bested by better advertising and commoditization

Can you read? They arent releasing anything right now, they are just demonstrating their technology. Allegedly Prescott contains it, but 64bit wont be turned on/released until sometime in 2005, about the year Intel was planning on migrating to 64bit in the consumer market.

 

[RaynorWolfcastle]

I for one wouldn't be surprised if the 64-bit technology that Intel is introducing is a hardware emulation of IA64, while IA64 isn't ready for prime-time just yet it would provide a migration path to IA64 through the LV Itaniums. I also wouldn't be surprised if it turns out to be nothing more than an x86 extension a-la x86-64. I just have my doubts that Intel will destroy their multi-billion dollar investment in IA64 just like that.

 

[Accord99]

Originally posted by: jhu
itanium with wide oem support? the only main ones are hp and possibly sgi.

HP is the largest server vendor in the market and they have stopped development of PA-RISC and Alpha in favor of Itanium. There are also NEC and Unisys, both large vendors.

by releasing their own x86-64 intel is sending the message that they fvcked up with itanium's dissemination.

Or maybe Intel is confident of the growing momentum of Itanium that a 64-bit extension of some kind to x86 would not impact Itanium, especially in the >4P market where Itanium based products are strong.

losers here are hp and intel. sgi can just go back to using mips, dell will be business as usual, and ibm will just chug along with their power4 and soon to be power5 (and opteron as an afterthought).

SGI can't go back to MIPS, since it is hopelessly obsolete in performance and SGI has more or less bet the company on Itanium, though Altix is a killer product that is beginning to pay off big for them.

 

[MadRat]

I don't understand why everyone doesn't belive Intel will demonstrate IA64 emulation on the Xeon. No way they don't support IA64.

 

[ViRGE]

Originally posted by: MadRat
I don't understand why everyone doesn't belive Intel will demonstrate IA64 emulation on the Xeon. No way they don't support IA64.

Because IA64 is entirely different from x86; they're like oil and water. Take a look at the Itanium for example - its x86 emulation is at best, 1:1; for all the theoretical power it has, it can only pull off 1mhz of effective x86 P4 power for every 1mhz the chip runs at. Trying to emulate IA64 on an x86 chip would be even more disasterous, as it's trying to run massively parallel 64bit code expecting a lot of cache on a chip that's anything but parallel, and doesn't have nearly as much cache. Emulation would work, but emulation needs to be fast to be worthwhile.

 

[Cerb]

Originally posted by: MadRat
I don't understand why everyone doesn't belive Intel will demonstrate IA64 emulation on the Xeon. No way they don't support IA64.

They might support IA64, but why on a Xeon? Defeats the purpose, doesn't it? Why not just make Itanium-based Celerons? It seems like it would be better for them, especially now that, while it's hardly a smash hit, the Itanium has a real foothold. Address space extensions or x86-64 (with the extra registers and all), Which M$ might be able to force on them, seem a better way to go, as that way the COTS parts don't compete in any way with the supercallafrajalistic expensive parts. Extreme parallelism seems like a great idea for databases or running lots of single or double-threaded calculation thingies (yeah, there went my vocabulary), but it just isn't practical for your average user or developer--even your average 2-4 CPU server user. IMO, while you can find holes in that logic, it's a very good reason to keep them separate and charge out the a$$ for Itaniums (and in 8 and 16-way boxes, they do kick a$$).

Also, with HP behind it so firmly...Itaniums won't just die. Even if they fail, the sinking of the Itanic will be a sllw one. Intel has great power...AMD has good aim (in this case, excellent performance/cost for 2-4 CPU machines).

On a side note, I'd really like to know why it is many RISC machines have such good scaling after the 2nd CPU. It seems almost all x86 servers have good (50+%) scaling for the 2nd CPU, and then it just drops off...none of them seem to have any real bandwidth advantages. Is it just design of the chips and compilers, or is there some major inefficiency in the x86 instruction set(s) (since they also don't seem to have any better performance per CPU)?

 

[dexvx]

Originally posted by: jhu
sgi can just go back to using mips...

Thats the funniest statement I've heard in a long time (akin to "64 bits doubles performance over 32 bits").

MIPS is hopelessly outdated. The most current MIPS incarnation is several generations behind Itanium/Power4 and even the now defunct Alpha/HP PA. A single Itanium will probably be better than the latest 8 way MIPS server. If SGI goes back to MIPS, they need to pull something magical, probably similar to nVidia going from TNT to GeForceFX 5950 Ultra or ATI going from Rage 128 to Radeon 9800 XT in 1 product cycle.

SGI has pegged their entire future on Itanium. So far the bet is paying off with Altix.

 

[MadRat]

Originally posted by: Cerb
On a side note, I'd really like to know why it is many RISC machines have such good scaling after the 2nd CPU. It seems almost all x86 servers have good (50+%) scaling for the 2nd CPU, and then it just drops off...none of them seem to have any real bandwidth advantages. Is it just design of the chips and compilers, or is there some major inefficiency in the x86 instruction set(s) (since they also don't seem to have any better performance per CPU)?

Perhaps because no x86 OS has parrallel thread support like RISC machines have in their OS options. RISC by default is built for scaling the processors and using a simple CPU layout, with more power using more CPUs in parrallel and scale the number to match the workload. The x86-32 machines are built on the x86 ISA that works fine for two processors, but was never intended to scale beyond that. Software emulation has to be used to run x86 code in a parrallel style, and no true centralized hardware control of the entire system. The software has to emulate the control that RISC hardware, in say like in an IBM P-series server, would have assumed by default.

Add in the phsical limitations of Intel's IHA and North/Southbridge architectures, with their GTL\GTL+ bus, then you run into design limits that don't even scale well with two processors. Profusion and Opteron architectures are limited by the OS more than their theoretical hardware barriers, and theoretically SHOULD show scaling with each and every processor up until the eighth. Theory and practice, however, we know are two different things altogether.

When .NET and its XML scripting of intraprogram communications becomes the norm then we won't need to worry if its being run on AMD64 or IA64, because there will be no platform specific API to account for in the programming. MS Longhorn project is going to rewrite the need for platform specific programming on a level that makes Java look like it WAS a good idea at one time but is obsolete. At least MS believes it to be true...

 

[Cerb]

Originally posted by: MadRat

Originally posted by: Cerb
On a side note, I'd really like to know why it is many RISC machines have such good scaling after the 2nd CPU. It seems almost all x86 servers have good (50+%) scaling for the 2nd CPU, and then it just drops off...none of them seem to have any real bandwidth advantages. Is it just design of the chips and compilers, or is there some major inefficiency in the x86 instruction set(s) (since they also don't seem to have any better performance per CPU)?

Perhaps because no x86 OS has parrallel thread support like RISC machines have in their OS options. RISC by default is built for scaling the processors and using a simple CPU layout, with more power using more CPUs in parrallel and scale the number to match the workload. The x86-32 machines are built on the x86 ISA that works fine for two processors, but was never intended to scale beyond that. Software emulation has to be used to run x86 code in a parrallel style, and no true centralized hardware control of the entire system. The software has to emulate the control that RISC hardware, in say like in an IBM P-series server, would have assumed by default.

Add in the phsical limitations of Intel's IHA and North/Southbridge architectures, with their GTL\GTL+ bus, then you run into design limits that don't even scale well with two processors. Profusion and Opteron architectures are limited by the OS more than their theoretical hardware barriers, and theoretically SHOULD show scaling with each and every processor up until the eighth. Theory and practice, however, we know are two different things altogether.

When .NET and its XML scripting of intraprogram communications becomes the norm then we won't need to worry if its being run on AMD64 or IA64, because there will be no platform specific API to account for in the programming. MS Longhorn project is going to rewrite the need for platform specific programming on a level that makes Java look like it WAS a good idea at one time but is obsolete. At least MS believes it to be true...

Not counting the last paragraph, that makes sense. So basically there would need to be robust extensions to allow for more hardware control of scheduling. Also explains why the current Opterons in 4-way appear to scale linearly and then just kinda...stop (though I imagine Intel will follow with a similarly-performing Xeon, they are just taking their time, where AMD doesn't have the time to waste--or use), where they seem to have a nice curve in 1- and 2-way. The CPUs end up sitting there, and the main OS scheduler isn't able to send everything out every which way faste enough and keep up all the other threads (kinda like how PATA and SATA currently--and PATA RAID sometimes mroeso--perform similarly to SCSIs...and then you add those one or two extra users to it and they fall apart, because it was never made for it?).

MS probably will do an excellent job. Except for the OS and browser having too many holes, they generally do a very good job.

 

[Pocatello]

Originally posted by: NFS4
AMD: 1
Intel: 0

But is it good for AMD in the long run? If intel produce a 64 bit x86 processor, Dell will stick with intel and other companies don't have to buy AMD Opterons anymore, now they have a choice. AMD had quite a nice run when the first Athlons came out, but it didn't last very long did it? I mean, I didn't mind paying $50 for an 1700+ Athlon XP and OC the hell out of it, and I'll be happy for a $100 for a 3000+ Athlon FX, but it won't help AMD financially. Someone has already stated that the consumers are the real winners, perhaps MS as well.

 

[JYDog]

The consumer public may not care much for 64bit, they still desire the latest and greatest they can get their hands on. If M$ releases Longhorn at least near schedule(4th Qtr?), 64bit simply becomes THE primary specs for the PC, as it is in the G5 for Macs.

 

[Cerb]

Originally posted by: Pocatello

Originally posted by: NFS4
AMD: 1
Intel: 0

But is it good for AMD in the long run? If intel produce a 64 bit x86 processor, Dell will stick with intel and other companies don't have to buy AMD Opterons anymore, now they have a choice. AMD had quite a nice run when the first Athlons came out, but it didn't last very long did it? I mean, I didn't mind paying $50 for an 1700+ Athlon XP and OC the hell out of it, and I'll be happy for a $100 for a 3000+ Athlon FX, but it won't help AMD financially. Someone has already stated that the consumers are the real winners, perhaps MS as well.

...but if they sell, they sell. Overclockers are in the minority. The success of the Athlon XP, Athlon64 and Opterons is in the hands of OEMs like HP, eMachine, and white-box dealers. So far it has been good. How many 1700+ and 1800+ PCs do you see sold under the HP, Compaq and eMachine names? Not many. Quite a few 2200+ and 2600+ ones, though (last I was at Sam's, anyway).
Also, however, I have a feeling AMD had many more ideas than they used for the spec, and if they were smart, used mostly the ones that favor their more efficient CPUs...they'd lose a bit of Intel made an x86-64 CPU, but since the Slot-A, they've managed to stay competitive, even if just in a few key areas (in this case, performance/cost and gaming, both of which surely help out a lot). It's quite clear a lot of what makes the Opteron a success is as much their business decisions as good engineering and fabs...while getting to Joe Consumer is a bit tough for them (why?), they don't seem to have as much trouble w/ businesses, and all the open source folks were on the edges of their seats...and then were not dissappointed.

 

[Accord99]

Originally posted by: JYDog
The consumer public may not care much for 64bit, they still desire the latest and greatest they can get their hands on. If M$ releases Longhorn at least near schedule(4th Qtr?), 64bit simply becomes THE primary specs for the PC, as it is in the G5 for Macs.

Except that Longhorn is supposedly delayed till 2005, or 2006 and is primarily a 32-bit OS. And OSX is not 64-bit either.

 

[SSibalNom]

Originally posted by: thorin

Originally posted by: Lonyo
MS: 1
AMD: 1/2
Intel: 0

SCO -1

IMHO this is no big news. Everyone knew Intel was going to be forced in to this position. Hell Intel obviously knew Intel was going to be forced in to this position or else their x86-64 debut wouldn't be for another year (if not more). I still see no benefit to Average Joe owning a 64bit processor (maybe 2 yrs from now it'll make sense).

Thorin

but if it's 64-bit or not, the current amd 64-bit setup is cheaper and outperfoms mosts p4's at a much lower cost as far as gaming goes, and most average joes are just playin games on the pc

 

[RaynorWolfcastle]

Originally posted by: Accord99

Originally posted by: JYDog
The consumer public may not care much for 64bit, they still desire the latest and greatest they can get their hands on. If M$ releases Longhorn at least near schedule(4th Qtr?), 64bit simply becomes THE primary specs for the PC, as it is in the G5 for Macs.

Except that Longhorn is supposedly delayed till 2005, or 2006 and is primarily a 32-bit OS. And OSX is not 64-bit either.

Longhorn isn't delayed for 2005-2006, it's scheduled for 2005-2006. Longhorn is a MAJOR overhaul of the operating system, much more so than the transition from Windows 95 to Windows NT/2000. Longhorn has an enirely different design philosophy in which the objective is to abstract the hardware away as much as possible. Like MadRat said, the idea for most programs will be to use a JIT compiler on MSIL code. As such, writing code a native binary (as is commonly done today) would only be necessary for the programs that require absolutely the best performance.

The advantage of MS' aproach is that as the JIT compiler improves, so does ALL code. Also, if Intel one day decides to release SSE49 it doesn't matter if your program supports it, a simple JIT compiler udate will update the way your code compiles and you'll immediately get the benefits.

I also stand by my prediction hat it is quite possible that CT is in fact related to IA64, we'll see soon enough.

 

[MadRat]

Cerb,

I should have simply referred to 64-bit CISC or RISC, rather than AMD64 or IA64, in that last statement. When it comes down to it, there will be two camps here in the next few years, IA64 and the rest. The rest will be so fractured and require so much more support than IA64 that they won't matter as far as the mainstrean goes for 64-bit programming. I was referring to Longhorn being less platform specific than its predecessor, therefore it won't matter if its ran on an x86 compatible platform or one native to IA64, it will run as good as the compiler allows it to run.

On a side note, has anyone contemplated what could happen if Prescott had AMD64 support and IA64 support in separate modes? What would it take to bypass the CISC decompiler so that their code could run in the RISC-like architecture as native IA64?