“Both Microsoft and AMD have taken the unusual step of re-affirming their collaboration on 64-bit computing.
The companies issued a joint press release in Taiwan today claiming that future Microsoft Windows operating systems and AMD’s K8 family, formerly codenamed Hammer, would be developed closely together.” Read the report at the Inquirer.
not that i mind amd getting some good money(which ofcourse they will in the long run from this deal) what happened to intel’s ittanium? and why isn’t ms hookin up with intel like they have been for so long?
I’m sure that Microsoft is working with Intel too. The real news here is that MS favoring the Hammer 64-bit ISA over Intel’s own is another nail in the coffin of IA-64.
Remember that by now everybody was supposed to be using Itanium CPUs, especially vendors who had been using so-called RISC chips. But the RISC vendors, including IA-64 partner HP have decided to keep on making their own CPUs, citing Intel’s problems in producing Itanium silicon. Then AMD comes along and extends the tried and true x86 ISA so that software vendors don’t have to recode for a totally different architecture.
For a company like Microsoft, what AMD brings is a godsend. It’s de rigeur for MS developers to tack this and that onto the spaghetti code that is behind Windows, but overhauling it all for a new CPU is a big job! No doubt MS remembers the confusion and mistrust that came out of Apple’s decision to switch hardware in mid-stream. They don’t want to repeat that, and AMD gives them a way to stall for time.
But as time drags on, and Intel has to go begging for vendors to support IA-64, the market becomes more and more wary of IA-64. IA-32 works — the only people who complain about it are those who market non-IA-32 systems. So AMD’s plan to simply extend what’s a known quantity, and is known to work looks better and better. Hardware vendors have a hard time selling the expensive Itanium boxes with no software for them, so that market is stagnating. So what does Intel do? What Intel is doing is they are copying AMD! That’s right, Intel is now designing chips that use AMD’s 64-bit x86 architecture!
So Napoleon had his Waterloo, IBM had their MicroChannel, and Intel will have Itanium.
For a company like Microsoft, what AMD brings is a godsend.
<p>
Can you blame Microsoft for embracing x86-64? From what I understand, IA64 will not be able to run 32-bit apps, while Hammer will retain backwards compatibility.
Now consider Microsoft having to develop and market a new Windows for these new chips. One one hand, the version for IA64 won’t be able to run older 32-bit apps, while the x86-64 version will. Backwards compatibility–to a certain extent–is an important feature in an OS.
I see either one of three likely outcomes:
1) The IA64 port of Windows falls by the wayside and goes the way of NT for the Alpha
2) Intel delays IA64 based on market feedback to build bacwards compatibility
3) Intel plows ahead and releases IA64 without backwards compatibility and realized it made a mistake and releases a backwards compatible version later on.
Like it or not, folks, because of its relative market share versus AMD’s Intel is the tail wagging the Microsoft dog here. I seriously doubt Microsoft would say “F*ck Intel–AMD has a chip out that does what we want for our new OS *and* is backwards compatible to older apps” even though that would be the coolest thing.
This is still great news though…with this new endorsement from Microsoft, open-source developer interest in porting Linux and BSD to Hammer can only increase:
http://www.x86-64.org
I don’t understand why people always wait to see Intel do something about AMD products in the last 3/4 years.
In this period Intel has been looking like a Reactive Company rather than an Active Company.
I think that AMD has proven that they no longer have to give credit to Intel plans. We just have to remember the bad choices Intel has made after the rfelease pof the P4 line – RDRAM, bad performance for an high final price, resistance to work out the mistakes and follow consumers demand for DDRAM P4 chipsets, bad licensing policies for VIA P4 chipsets , wich are superior to Intel’s chipsets).
However, I have a Celeron Tualatin 1100A machine for now ๐
Never thought about change to Duron mainly because of heat/cooling and because of processor reliability-confidence in AMD.
I would feel like I would be screwed by Intel marketing if I bought a P4 for such a high price.
If this x86-64 Hammer/Thoroughbread proves to be reliable and a cool running processor I am one of the (very) old Intel customers to switch to AMD-64bit with all my software for Windows.
In this case I will not mind the “Activation” Process of the actual winXP since I will only have one 64 machine to install it, currently I run win2K.
If Intel don’t want (or couldn’t) give Microsoft chipsets and processors that could run 32 bit software on a 64 bit platform why would Microsoft turn AMD down ?
Can you blame Microsoft for embracing x86-64?
No, I don’t blame Microsoft for doing what is prudent. OTOH I’m not thrilled that they might not have to port the code, presumably leaving many bugs intact.
From what I understand, IA64 will not be able to run 32-bit apps, while Hammer will retain backwards compatibility.
From what I understand, the Itanium includes an IA-32 core to run IA-32 code on. In fact, one review that I read mentioned that the Itanium boxes boot to real mode, and run MS-DOS just fine! You’ll have to revise 2 and three because of that, and the fact that they are shipping…
From what I understand, the Itanium includes an IA-32 core to run IA-32 code on. In fact, one review that I read mentioned that the Itanium boxes boot to real mode, and run MS-DOS just fine! You’ll have to revise 2 and three because of that, and the fact that they are shipping…
Yes the above is very true, and i’d like to reinforce this. But the truth about the IA-32 core in the Itanium is that the fastest Itanium can only run a non IA-64 program as fast as a P100. As you can see running IA-32 software like DOS would be fine, since DOS ran well on 386’s but when you try to run a older version of Windows you really start to run into walls.
I don’t particularly care about backwards compatibility. IA-64 is clearly a superior architecture over x86. I don’t want to see yet another x86 extension become the mainstay. If you’ve ever programmed in assembly on a x86, you’d know how much of a pain in the ass it is. Limited registers, and the fact that you can’t address a full 32-bit word at one time. Now, with the 64-bit extension you will probably have to shift registers twice just to get at the whole 64-bit word. In most cases, applications would only need a simple recompile to run natively on the Itanium. Only in cases of very low level code, where bit ranges are an issue, would need modification.
I like AMD, and I have several Athlon boxes, but I don’t like this idea of extending x86 yet again. x86 needed to die a long time ago. Yet, here’s the chance, and I see everyone going with the inferior product simply because it’s cheaper and because it’ll run old binaries faster.
Amen to that, the x86 is so dirty when you have to do assembly for it. Adding another layer of dirtyness doesnt solve the problem. I used to be a Hammer fan, but i converted to Itaniumism for the simple fact that it was _different_. The x86-64 actually does very little as far as improving the x86, making registers bigger isnt THAT big a deal.
Question for you: When you spoke of “assembly on a x86 … limited registers … can’t address a full 32-bit word at one time” were you were you talking about real mode (DOS) assemblers, or were you talking about 32-bit protected mode assembly that goes through the host OS?
And a comment: It seems to me that you’re dismissing the AMD product as “inferior” based solely on it’s drawbacks, without considering its benefits. Intel’s success has come in a large extent from their backwards compatibility. It must not be that hard to work with, or else everybody would have jumped ship when RISC showed up.
The day Intel releases an Itanium that doesn’t cost a kings ranson is the day I will consider it as a viable alternative to the Hammer Processor and RISC processors being sold by SUN, HP and MIPS.
$US1300 for the low end, Itanium processor, and what is worse, there is NO, I REPEAT NO! applications for it, either in the server or workstation arena. No Photoshop, no Maya, no Bryce etc etc, all applications that could benefit from the extra power of this NG CPU from Intel.
As for compiling old programs for Intanium, there has yet to be a decent compiler to produce reasonable quality executable code for the Itanium processor. So the assumptions that applications can be easily ported to the Intanium is a load of rubbish.
About the only thing I would change in terms of the Hammer processor and motherboard is the removal of the BIOS and replace it with the an OpenFirmware like what the *NIX RISC stations use. This will finally get rid of alot of the limitations imposed because of the bios and the need for eternal backwards compatibility.
Superior architectures don’t always win unfortunately, remember 68k, Alpha etc, list is endless.
I too used to loath the x86 as it was 10-15yrs ago, but you need to look at x86 today. It still isn’t that great but…
The registers both data & address have been 32bit for a very very long time so it can often look and feel as much like a risc as any other true risc. The issue with x86 is still it’s tiny user reg set (8 or so), & the very complex instruction encodings. Given 2 similar ops that differ in 1 simple way, the x86 codes could be vastly different in sz & timings & this would impact asm programmer if you doing alignment optimised loops etc.
The x86 cpu internally keeps many many more regs than 8 & uses renaming to effectivley increase the set, still not as good as having 32-128 or so direct user regs but effectivly better than 8. In addition the Level 1 cache effectively is your extended register set since Level one cache hits are 1 cycle penalty.
Also in the Hammer the reg set is now fully 64bit for data & address IIRC & for the heck of it, AMD doubled the no of regs as well to 16 or so. I am sure the opcodes are really tortured but most C/asm programmers don’t see that low down & the Silicon hides the decoding penalty very well.
One of the most interesting things about the Hammer is the HyperTransport (HT) bus, different Hammers can include 1 or more of these, each time a HT bus is added, the Hammer can be built into more complex connected Multi Processors. Reminds me of Transputers of long time ago (AMD marketing pics look same). The HT bus may also be known as links. Only the pervasive HW supported threading is missing here. Of course HT as well as the current Athlon bus originated with the Alpha.
Also FPGAs that can be used for say Encryption can also include a HT bus & could go right next to cpu (instead of much sloooower PCI bus). Such an FPGA could easily define in soft HW any critical operations the cpu is missing to counter anything the IA64 may have on chip. Developers could also design soft HW for such an addon once the framework has been worked out! This is what I am most waiting for. Tool flows exist now for soft HW design using PCI hosted FPGAs but are expensive.
If I were in AMDs shoes having rested control from Intel, I might try to figure out if the internal micro architecture can be exposed as a new risc ISA that makes all the goodies available outside, ie 16*64 bit reg available in a classic 32bit opcode format that can coexist with the crappier x86. After a few yrs of use, this cleaner ISA would become the dominant ISA & the x86 baggage would be the backward support. Exposing this ISA could allow the architecture to move forward in the right direction & eventually jetison x86, one can dream on.
The 1st Hammers are supposed to be 75W but will go down, so cool/small is out for a while. The 1st Hammer ref boards are being cloned in Taiwan as we speak.
And finally Intel is well aware of the AMD scheme & is supposed to have done the same 64 bit addons in the Yamhill project, so upcoming P4s can switch this on if needed.
Interesting times ahead
(Note: I’m no MS troll here)
Isn’t it kinda strange? I mean, remember the email about Linux in Microsoft? They wanted to harm all those who support Linux, which happens to be Intel… And then AMD went to testify in favour of Microsoft in the antitrust case. I was Microsoft anyway, I would still support AMD, but have greater support for Intel. Intel is an household name, and they didn’t get it for free. Instead of advertising its brandname, AMD decided to “educate” the public. AMD is still not an household name. If AMD spend all their advertising budget on brandname, I’m sure AMD could beat Intel Inside…
Besides, Intel doesn’t really need MS support for Itanium. Microsoft is loosing market share in the server market, and the market they are targeting, RISC servers, don’t use Windows Server anyway. So, they could get HP to port their Tru64/HP-UX/OpenVMS whatever to Itanium, and they already got Linux, just get BSD and Solaris along, and viola, they might even get in (maybe not be #1). AMD is developing x86-64 to get the 64-bit workstation market, and to expand their market by bringing legacy x86 apps. Sure, x86 is an decades old montrosity, but it has the most amount of apps.
What Intel is doing is they are copying AMD! That’s right, Intel is now designing chips that use AMD’s 64-bit x86 architecture!
Yamhill is vapourware, just like G5. There have been rumours about it left and right, but nothing from Intel (okay, virtually nothing). So until the day Intel doesn’t reluctantly admit its existance, and starts promoting it, its vapourware.
I like AMD, and I have several Athlon boxes, but I don’t like this idea of extending x86 yet again. x86 needed to die a long time ago. Yet, here’s the chance, and I see everyone going with the inferior product simply because it’s cheaper and because it’ll run old binaries faster.
Unfortunately, not many people are willing to throw out x86 legacy apps. Maybe if WIndows come with PPC support, and encourage developers to compile their apps for both x86 and PPC (maybe VS.net could compile for PPC on the x86 version and vice versa)… but it seems that is not gonna happen anytime soon. The person with the most power to move the entire industry to another platform, Microsoft, isn’t interested in doing so.
Does anyone know where to find the press release in Taiwan?
I’m from Taiwan, but I haven’t seen this news or any Asian release in any Chinese web site…
The day Intel releases an Itanium that doesn’t cost a kings ranson is the day I will consider it as a viable alternative to the Hammer Processor and RISC processors being sold by SUN, HP and MIPS.
The day you see a Hammer machine is the day when that starts to become true. Intel’s latest round of price cuts were on the high end, where AMD is eating their lunch.
$US1300 for the low end, Itanium processor, and what is worse, there is NO, I REPEAT NO! applications for it, either in the server or workstation arena. No Photoshop, no Maya, no Bryce etc etc, all applications that could benefit from the extra power of this NG CPU from Intel.
In fairness to Intel, they made it quite clear that Itanium isn’t for desktop applications (this time for real, LOL), and every example that you cited was a desktop application. Are you certain that there are no server apps? I have no idea, just asking.
As for compiling old programs for Intanium, there has yet to be a decent compiler to produce reasonable quality executable code for the Itanium processor. So the assumptions that applications can be easily ported to the Intanium is a load of rubbish.
What?!? No GCC? ๐
About the only thing I would change in terms of the Hammer processor and motherboard is the removal of the BIOS and replace it with the an OpenFirmware like what the *NIX RISC stations use. This will finally get rid of alot of the limitations imposed because of the bios and the need for eternal backwards compatibility.
I’ve been saying this for a long time, even before IA-64! I don’t know about Open Firmware though. I like Sun too, but think that something along the lines of the ARC console that’s on Alpha boxes would be a better fit. The menu system resembles the current PC CMOS setup programs, so there would be less culture shock. Plus it already loads Windows!
My vision is to extend the preboot environment, using some current technology. Without DOS we will need a more complete program execution facility. Programs to configure hardware that used to run from a DOS floppy would be installed right into NVRAM or some auxillary media like CF or SD memory cards. Other programs that run from DOS floppies or NT at boot time (like Partition Magic) could be available regardless of OS. In fact, they could be used to partition disks before any OS was installed!
Use your imagination… Anything is possible. How about a minimal Linux installation that could be used to download install files, patches or virus definitions before the main OS loads? Or a supervisor program that downloads new e-mail every once in a while (without booting the main OS) when you’re on vacation? Think about it…
Yamhill is vapourware, just like G5.
While the G5 can still be called vaporware, IBM’s Power5 chip (5th gen ppc) is anything but vaporware. I can’t wait to see it kick Intel and AMD’s ass.
>>Yamhill is vapourware, just like G5. There have been >>rumours about it left and right, but nothing from Intel
EETimes has repeatedly carried the story front page, thats something most Intel engineers read or comes across their desk weekly. Being in the biz, I believe it for now, its not like its that difficult to do. If AMD runs ahead with 64 bit desktop & workstation, Intel has no choice but to support same features or come out with their own x86-64 extensions.
I have read that so far only a few thousand Itaniums have been bought, this is sounding a bit more like the IAPX32 story of 20yrs ago all over again. Always waiting for the next silicon to catch up.
I saw the Hammer ATX board a few weeks ago at a Hyper Transport show, so I guess AMD is on track.
Which means I’ll be able to upgrade my K7 when the price is right!
Yamhill is vapourware, just like G5.
No it’s not, have a look here ๐
http://dynabook.com/pc/catalog/dynabook/020424g5/index_j.htm
While the G5 can still be called vaporware, IBM’s Power5 chip (5th gen ppc) is anything but vaporware.
They’ve announced they are working on it and even talked about it’s new features, and also Power6 for that fact, a very interesting article here:
http://zdnet.com.com/2100-1103-892836.html
A couple of things they missed:
While everyone else was around 200MHz IBM demoed a prototype PowerPC which had multi threading and ran at 1.1GHz, it was never released as it was meant to be a test bed only but this was *years* before intel had multi-threading.
Also they mention TCP/IP acceleration in hardware – Some Motorola chips alread do this.
I can’t wait to see it kick Intel and AMD’s ass.
Why? Power4 does this quite nicely already ๐
The day you see a Hammer machine is the day when that starts to become true. Intel’s latest round of price cuts were on the high end, where AMD is eating their lunch.
AMD just announce price cuts, most notably is mobile athlon xp (thoroughbred).
In fairness to Intel, they made it quite clear that Itanium isn’t for desktop applications (this time for real, LOL), and every example that you cited was a desktop application. Are you certain that there are no server apps? I have no idea, just asking.
Intel never claim that Itanium was for desktops or even workstations.
Anyway, there is a bunch of apps for Itanium for servers (like Oracle, Linux and so on, it’s a pretty long list).
Anything written in portable code (eg: C or C++), porting is painless.
Use your imagination… Anything is possible. How about a minimal Linux installation that could be used to download install files, patches or virus definitions before the main OS loads? Or a supervisor program that downloads new e-mail every once in a while (without booting the main OS) when you’re on vacation? Think about it…
Just asking, wouldn’t this slow down the whole booting session, especially if you don’t have a fast connection? Besides, Linux too every once in awhile get security problems and need security patches…
While the G5 can still be called vaporware, IBM’s Power5 chip (5th gen ppc) is anything but vaporware. I can’t wait to see it kick Intel and AMD’s ass.
Did I call Power 5 vapourware? (Besides, I never heard of it…)
EETimes has repeatedly carried the story front page, thats something most Intel engineers read or comes across their desk weekly. Being in the biz, I believe it for now, its not like its that difficult to do. If AMD runs ahead with 64 bit desktop & workstation, Intel has no choice but to support same features or come out with their own x86-64 extensions.
Still is vapourware. I have seen conflicting rumours of Yamhill. Some say its a processor that runs both IA32 and IA64 apps natively, some say it uses x86-64, some say it is Intel’s own standard…
Besides, for 64-bit, for normally desktops, there isn’t really any real advantage (the limitations of 32-bit hasn’t been a problem for normal desktops). So, all AMD gets is advertising rights.
Anyway, Intel schedule IA32’s dismiss (IIRC) on 2006… i really doubt another extension to it.
Which means I’ll be able to upgrade my K7 when the price is right!
K7= Athlon, Athlon XP, Athlon 4, Duron.
K8= Hammer (Athlons and Opterons).
As for price, you could get a K7 processor (Duron) at $50 (malaysian street price). Really, is it that expensive? Please don’t tell me you still are using K6…
No it’s not, have a look here ๐
http://dynabook.com/pc/catalog/dynabook/020424g5/index_j.htm
So, you are Japanese. But I’m not. English?
But from what I see at the site that I can recognize is that Intel Pentium 4-m logo. How does that prove Yamhill or G5 isn’t vaporware?
While everyone else was around 200MHz IBM demoed a prototype PowerPC which had multi threading and ran at 1.1GHz, it was never released as it was meant to be a test bed only but this was *years* before intel had multi-threading.
Why didn’t they release it? Oh wait, IBM sucks in marketing, how did I forget?
I can’t wait to see it kick Intel and AMD’s ass.
It may kick Itanium’s ass (if it have an ass anyway) cause it is in the same market. It’s like comparing Alpha and Duron. Power 5 and 6 IIRC is plan to arrive first on mainframes…. Would be much better if you wrote
I can’t wait to see it kick Sun’s ass
The day Intel releases an Itanium that doesn’t cost a kings ranson is the day I will consider it as a viable alternative to the Hammer Processor and RISC processors being sold by SUN, HP and MIPS.
The day you see a Hammer machine is the day when that starts to become true. Intel’s latest round of price cuts were on the high end, where AMD is eating their lunch.
– What high end price cuts? the fact that the P4 still has the dreaded 21 stage pipe line, and is slower than my grandmother with a walking helper. The Xeon is more of a disappointment. After looking at the abismal benchmarks I wounder why people are still sucked into this “intel inside” rubbish, and why a person would pay an extra $AUS1800 for a Xeon workstation with equal performance of a P4 desktop.
$US1300 for the low end, Itanium processor, and what is worse, there is NO, I REPEAT NO! applications for it, either in the server or workstation arena. No Photoshop, no Maya, no Bryce etc etc, all applications that could benefit from the extra power of this NG CPU from Intel.
In fairness to Intel, they made it quite clear that Itanium isn’t for desktop applications (this time for real, LOL), and every example that you cited was a desktop application. Are you certain that there are no server apps? I have no idea, just asking.
– Did you actually read what I wrote? “WORKSTATION”, you know, big friggin, number crunching computers used to render multi-gigabyte films like Lord of the Rings. That is where the IA64 could grab a market hold, but when I can get a high end SGI O2 workstation with all the software and trimmings from SGI, why should I wait for Itanium to deliver? As for the server space, again, it is crap. JVM performance is worse than the 32bit counterpart, there has been very little enthusiasm by the server software companies like Peoplesoft, iPlanet, Oracle, Sybase and IBM RDBMS.
As for compiling old programs for Intanium, there has yet to be a decent compiler to produce reasonable quality executable code for the Itanium processor. So the assumptions that applications can be easily ported to the Intanium is a load of rubbish.
What?!? No GCC? ๐
– Yes there we go again, read the bloody reply, there are NO GOOD compilers that can produce quality VLIW code, simple as that. Neither Intel, Microsoft or the GNU community can produce a quality compiler due to the complex nature of the EPIC instruction set.
About the only thing I would change in terms of the Hammer processor and motherboard is the removal of the BIOS and replace it with the an OpenFirmware like what the *NIX RISC stations use. This will finally get rid of alot of the limitations imposed because of the bios and the need for eternal backwards compatibility.
I’ve been saying this for a long time, even before IA-64! I don’t know about Open Firmware though. I like Sun too, but think that something along the lines of the ARC console that’s on Alpha boxes would be a better fit. The menu system resembles the current PC CMOS setup programs, so there would be less culture shock. Plus it already loads Windows!
– Anything is better than BIOS, yeck, replace the bloody thing with Kickstart to say the least. Reason I suggested OpenFirware is that it is openstandards based, and the programming language, forth, is pretty good.
My vision is to extend the preboot environment, using some current technology. Without DOS we will need a more complete program execution facility. Programs to configure hardware that used to run from a DOS floppy would be installed right into NVRAM or some auxillary media like CF or SD memory cards. Other programs that run from DOS floppies or NT at boot time (like Partition Magic) could be available regardless of OS. In fact, they could be used to partition disks before any OS was installed!
Use your imagination… Anything is possible. How about a minimal Linux installation that could be used to download install files, patches or virus definitions before the main OS loads? Or a supervisor program that downloads new e-mail every once in a while (without booting the main OS) when you’re on vacation? Think about it…
– Why not chuck the whole OS onto a removable EEPROM chip and store the configuration on the hard disk, kind of like the RISC OS. By doing that the installation of OS’s will be done by simply turning off the computer, replacing the EEPROM chip, and then turn it back on for reboot.
>So, you are Japanese. But I’m not. English?
>But from what I see at the site that I can recognize is
>that Intel Pentium 4-m logo. How does that prove Yamhill
>or G5 isn’t vaporware?
It’s called humour!!!
It’s a Toshiba Laptop called the dynabook G5 and yes it is fitted with a Pentium 4…
and no I’m not Japanese either.
>Why didn’t they release it? Oh wait, IBM sucks in >marketing, how did I forget?
Probably because (to quote myself)
“it was meant to be a test bed only”
They did a paper on it at the ISSCC (IC conference) in 1998.
What im wondering is what makes a processor multi-threaded. I mean, i’ve worked on an OS for the x86 out of pure assembly, and atleast the x86 cant execute 2 threads at once. The threading is done by the OS, but flipping all the registers at a given point so that and telling the processor “no no, now you gotta go decode the code thats THERE…”. I have yet to see processor specs where there are different registers for different threads.
>[i] “What im wondering is what makes a processor multi-threaded. I mean, i’ve worked on an OS for the x86 out of pure assembly, and atleast the x86 cant execute 2 threads at once.The threading is done by the OS,”/[i]
As I see it, Multithreading means to assign several threads to diffrent cpu simmetrically, so without multiple cpu’s (more then one) there is no multithreading ๐ When a thread goes to cpu0 the next is already being processed on cpu1 simultaneously, if the application can generate multithreads.
“A x86 cpu can’t process 2 threads at once” – obvious, … the threading is donne by the OS if it has real-time capabilities and chooses the priority of the threads while telling other threads to wait untill the higher priority thread is finished on the single cpu (see -> http://www.qnx.com ).
and no I’m not Japanese either.
You could have fooled me there. I mean Japanese guys having weird Western names (hehe, humour)… is common.
Rajan (try to read what you’re about to comment) and Russel Jackson (you can’t access full 32bit words on IA32?! You must really suck as an assembly programmer) made so many wrongs up there I won’t even try to begin to correct them. Instead I’ll shoot for Mr. Gardiner who wrote:
“Did you actually read what I wrote? “WORKSTATION”, you know, big friggin, number crunching computers used to render multi-gigabyte films like Lord of the Rings.”
You don’t render (final render) stuff on workstations, you model and animate on workstations. The rendering is (mostly) done by dedicated render-farms with big friggin, number crunching *SERVERS*
While the animators and modellers like responsive boxen, I’m not about to blindly accept your contention that it would be the perfect fit for IA-64.
IBM POWER CPUs — Am I the only one here who knows that IBM POWER is not PPC? If not, I’m certainly the only one who cares! Silly Mac-heads…always playing games of semantics, building paper tigers. Watch out boys! The Amiga-heads are providing competition in the “tall tales” front.
Multithreading — Multithreading is a programmatic thing, not a CPU thing. Threads are “lightweight processes” that share the same environment space. This is a big deal in Windows, where it takes a lot of work (and time) to set up and tear down full-fledged processes. Threading allows a single program to more efficiently queue a bunch of tasks. For example, before multithreading, web browsers would just sit there, unresponsive while loading a web page. The users didn’t know whether it was waiting or if it had frozen up. Threading splits up the networking tasks and the rendering tasks, so the user sees a little of the page before it’s all there. Much nicer!
Multiprocessing is the act of dividing processes and threads among multiple CPUs. Multithreading programs can take advatage of multiprocessing, but it’s not a requirement. For example, if you’re running a distributed number-crunching client on a SMP box, you need to start a separate instance of that program for each CPU. That’s because most distributed clients execute in only one thread. If they were multithreaded, you could start one, and it would spawn enough threads to saturate all CPUs automatically.
Hyperthreading is a proprietary name for Intel’s technology that makes a single CPU act as if it was two. From what I hear, there are some real benefits to this. Time will tell.
BIOS Replacements —
“Just asking, wouldn’t this slow down the whole booting session, especially if you don’t have a fast connection? Besides, Linux too every once in awhile get security problems and need security patches…”
Obviously “data center” features like this will be of less use to the home user. But home users could also benefit. You could have the morning’s news downloaded to your laptop as you sleep, all ready for you to read as you ride a train to work, or fly to a meeting. If all you have is a dial-up modem line, the ROMed OS could manage downloading of large software updates while you sleep, and when phone rates are low. Then when you boot your main OS, the updates would install quickly.
A small ROMed OS has far fewer security risks than a full-blown one does. But the ROM OS could be updated when the main OS is running. One hand washes the other…
“Why not chuck the whole OS onto a removable EEPROM chip and store the configuration on the hard disk, kind of like the RISC OS. By doing that the installation of OS’s will be done by simply turning off the computer, replacing the EEPROM chip, and then turn it back on for reboot.”
Yes! Very good! EEPROM is kinda old and slow for the job, but flash cards would do the job nicely. It would make it a cinch to multiboot!
>IBM POWER CPUs — Am I the only one here who knows that
>IBM POWER is not PPC?
The answer to that is both Yes and No.
For those who don’t know POWER CPUs are mighty beasts which run in IBM servers and workstations, PPC is reserverd for embedded, Desktop (i.e. Macs) and possibly also low end workstations.
They did have different instruction sets however they merged some time ago, Power3 etc. uses the PPC64 instruction set.
It’s not difficult to see why they did this, PPC evolved from the Power line in the first place. The first PPC, the 601 was a modified Power chip with a bus system taken from the Motorola 88,000 RISC.
I guess it made sense to have a single instruction set across both CPU lines, at very least saves having two different compilers.
>Hyperthreading is a proprietary name for Intel’s
>technology that makes a single CPU act as if it was
>two. From what I hear, there are some real benefits
>to this. Time will tell.
About 20% performance increase from what I’ve read.
However Intel really seemd to have done a very bad job of this. Both IBM and previously the Alpha people had talked about 100% speed improvements.
I agree with your description of Multithreading V’s Multiprocessing however the Alpha 464 would have been able to make use of both Process (i.e. different programs) and Thread level parallelisim, I imagine there would be a need to modify the OS to take advantage of it though.
IBM POWER CPUs — Am I the only one here who knows that IBM POWER is not PPC? If not, I’m certainly the only one who cares! Silly Mac-heads…always playing games of semantics, building paper tigers. Watch out boys! The Amiga-heads are providing competition in the “tall tales” front.
I don’t know… but did anyone claim it to be a PowerPC processor?
Obviously “data center” features like this will be of less use to the home user. But home users could also benefit. You could have the morning’s news downloaded to your laptop as you sleep, all ready for you to read as you ride a train to work, or fly to a meeting. If all you have is a dial-up modem line, the ROMed OS could manage downloading of large software updates while you sleep, and when phone rates are low. Then when you boot your main OS, the updates would install quickly.
So, in other words, the computer would be on 24/7. Which could wear the computer off by an decade…
>Hyperthreading is a proprietary name for Intel’s
>technology that makes a single CPU act as if it was
>two. From what I hear, there are some real benefits
>to this. Time will tell.
>About 20% performance increase from what I’ve read.
>However Intel really seemd to have done a very bad job of >this. Both IBM and previously the Alpha people had talked >about 100% speed improvements.
Hyperthreading has been described in the literature at least 25yrs ago in ACM Sigarch etc, but the cost was too high back then. In it’s modern form it requires the cpu to increase the size of all state & register sets by N (say 2 or 4 or more). This would actually slow things down a little since deeper registers, memories are always slower, (follows log of size) & more expensive. But the payback is 2 fold, 1st if all opcodes are selected cycling through the N threads, the pipeline gets alot simpler since interlock HW may be removed completely so it can clock faster. 2nd if a cache access fails, the thread can be descheduled for atleast the mem access time of a few 100 cycles, letting another thread move in (or not), the cpu can keep busy on almost every cycle. In this model the cpu resources are evenly divided over N threads (including caches) giving the illusion of N slower but more efficient cpus. The OS is still in charge, seeing N virtual cpus. That would be the Intel, Alpha, IBM? model IIRC, & I don’t yet see much enthusiasm from MS for it.
In another approach, the threads switch only when needed say on cache fail or after a time slice limit. The interlock HW is still needed & the registers are still much deeper. This version is not pretending to be N cpus but instead is a fully multi threaded single cpu with OS style process switching, scheduling & message passing in HW, Transputer style. Don’t see MS liking this one either.
Of course they can both fall apart if all threads start doing random mem reads etc, then thread swapping becomes a limit, but apart from that scenario there should be a big speed bump either way. Better to say that the cpu performs closer to the idealised limit v sometimes falling way behind on cache fails.
Myself, I wouldn’t dream of designing a cpu without threading, but what can you expect from an old Transputer guy.