Athlon 64 - Intel is jealous?

[OddTSi]

Originally posted by: pspada
I meant in compaison to AMD chips at the same clockspeed.

Well, if you were to inject my legs with novicaine and out of nowhere put me in a wheelchair race with no time to practice, I guarantee you that I would lose. However, put me in a situation, with no novicaine in my legs of course, where speed really counts (like getting your a$$ out of a burning building) and I guarantee I'll be sippin my Sprite by the time the first wheel chair guy comes out there.

What's my point you say? It's this: Who cares which one's faster when you have to cripple one of the competitors, like comparing the two chips clock-for-clock? Ever hear the expression "run what ya' brung"?

I'm really sick of AMD people saying "AMD chips are better because they have higher IPC". That's like an Intel fan saying "Intel chips are better because they run higher frequencies." Neither statement makes sense. It's a combination of IPC and clock rate that makes a chip fast. So comparing an Intel chip at the same clock-rate as an AMD chip is like comparing an AMD chip at the same IPC as an Intel chip.

 

[Accord99]

Originally posted by: chsh1ca
Actually, that's untrue, there are many many niche markets that can make use of > 4GB memory addressability and certainly any app that relies heavily on 64-bit integer and floating point operations. Why do you think that the Operton's preliminary performance numbers put it on par with a similarly configured Xeon system clocked upwards of 1GHz higher?

All of those performance gains comes from the on-die memory controller and/or the increased bandwidth and/or larger L2 cache that the Opteron has over the Xeon. None of these performance enhancements requires 64 bitness. And double precision floating point operations don't gain anything from 64 bitness, as these registers are already natively >64 bit on existing x86 CPUs.

 

[Wingznut]

Originally posted by: chsh1ca

Originally posted by: Wingznut
That being said, what part of "PC market is not ready for 64 bit computing" do you disagree with? How do you think could you benefit (on the desktop) from a 64bit cpu?

Anyone requiring more than 4GB of ram on their system for one. Anything that can use it from Games to 3D Rendering to Video Editing to whatever uses you can find for it. The thing is, there is already a market out there for it.

It's good of Intel to speak on behalf of the market, but obviously Microsoft disagrees, otherwise they would not have invested the money in developing a version of XP that will run on it. Microsoft expects some kind of ROI, and it's very telling what Microsoft sees the market as wanting or ready for and what Intel insists the market wants. Intel's comments on 64 bit on the desktop have been strikingly similar to AMD's comments on Hyperthreading. It's typical marketing, and nothing more. 🙂

You are correct... Any application that benefits from addressing more than 4gb of RAM will benefit from being designed for 64-bit... But I can't think of a desktop application that fits that description.

Games? Not hardly.
3D Rendering? That would be a HUGE chunk of data, to require more than 4gb. Anyone needing that kind of power would be on a multiprocessor workstation, not a desktop.
Video Editing? See above answer.

But you are right... There could very well be a couple of niche applications that might benefit from 64-bit. I don't really know which ones those might be, though.

As for MS... It would be incredibly stupid of them not to design an OS for Hammer. They need to be prepared for whatever direction the market might take them.

Originally posted by: mechBgon

That being said, what part of "PC market is not ready for 64 bit computing" do you disagree with? How do you think could you benefit (on the desktop) from a 64bit cpu?

Are you saying that there will be zero performance improvement from 64-bit desktop OS and code, Wingznut? The interview with Epic's Tim Sweeney quotes him as saying that he expects to see some improvements in the effective processing power available. He also seemed to be foaming at the mouth to get a 64-bit x86-compatible platform for his development team, but that's due to the ability to address more RAM.

I'm not so much agreeing or disagreeing with the above statement. I was just asking what would be the benefit on the desktop, at this time (or the near future.)

But to answer your question... I'm not a programmer, so I'm not sure what part of the OS would benefit from the added precision of 64-bit, that cannot be obtained with 32-bit. Nor am I sure what kind of (desktop) application could be improved with that kind of precision.

From what (very) little I do know about programming, I'm not aware of any desktop applications that would benefit.

Maybe there's a programmer reading this that can point me in the right direction?

Btw... Does anyone even make PC3200 in sticks greater than 1gb? The only 2gb sticks that I saw cost ~$1200 (and that wasn't from a reliable source.) But assuming that's accurate, who is going to pay $2500+ for memory on their desktop PC?

 

[mechBgon]

Just to remind you, Epic is the creator of UT2003 and they do expect real-world improvements in game performance with a 64-bit OS and 64-bit-optomized code. Something on the order of 15%, he estimated, versus the same hardware running 32-bit.

3D modelling and animation certainly can burn through the RAM. My dual-P3 system (well, one of the two) managed to eat 896MB and then grab another ~900MB of swapfile on just an 1800-frame animation of a panther walking through fog. Granted, that was in Caligari trueSpace 4.3, the Ford Pinto of the 3D modelling/animation universe... 😛 but it's what I can afford, and I've used the trueSpace family since I got my first PC, a refurb Cyrix PR166 with 32MB of EDO (ewwww! 😛).

 

[Wingznut]

Originally posted by: mechBgon
Just to remind you, Epic is the creator of UT2003 and they do expect real-world improvements in game performance with a 64-bit OS and 64-bit-optomized code. Something on the order of 15%, he estimated, versus the same hardware running 32-bit.

Yeah, I'm quite familiar with who Tim Sweeney is. 🙂 And yes, I recall the interview you are referring to. He said:

I think you'll probably see a clock-for-clock improvement over Athlon XP of around 30% in applications like Unreal that do a significant amount of varied computational work over a large dataset.

I personally think that statement is not realistic at all.

He also said:

We've been up and running on 64-bit Linux for months, and will release it publically at the consumer Opteron launch. Stay tuned for news regarding a 64-bit Windows version.

Did that ever happen? I don't recall seeing a x64 port of UT2k3 anywhere.

 

[Mingon]

You are correct... Any application that benefits from addressing more than 4gb of RAM will benefit from being designed for 64-bit... But I can't think of a desktop application that fits that description.

Photoshop - with my 62yr old dad at the controls ' I tried to scan at 1800dpi and it seemed to go slow' - 'no dad youve set it to 1800 Dots per centimetre'

 

[KF]

>That's what an Intel engineer was doing - stating that the AMD Athlon 64 is no good for the
> market but what is he comparing it to???? And why would thje market accept the Itanium and not AMD's Athlon 64?

Maybe it would be simpler to put it this way: Why ask a GM executive about the new Ford cars. What is he going to say?

Intel never shows all its cards, at least not without an NDA. (Non Disclosure Agreement) But here is what Intels plan appears to have been. I say appears. I have no inside information. The Itanium is radically different than any x86 CPU has been. It is not an x86 CPU. It is "far out." But Intel planned on superceding P3s with some Itanium-like processor after it got the super-duper real Itanium going well. That took longer than planned, and AMD's XPs seemed to be gaining on P3s, so instead of advanced P3s, Intel filled in with the P4, revised the first Itanium into Itanium2, and lengthed the timetable. With all the changes in the P4, it has become good enough to keep Intel mostly leading AMD. Intel is aware that the address space needs to be over 4 Gigabytes within the next year. To go over 4G, you need more bits in the address than 32. At this stage in technology, doubling it to to 64 bits is not hard. In all of its previous history, Intel has been ready way in advance of the time that memory was cheap enough to exceed its CPU's memory address space, so Intel is now way behind where you would have expected them to be. But Intel doesn't want to undercut the Itanium-like home processor it will becoming out with in the next couple of years,which will have as one of its features that it is 64 bit. That has put AMD in a difficult position in designing its next generation Athlons. They needed it to be 64 bit to be on the expected, normal time line. Without Intel defining what a next generation, 64 bit processor should be, AMD had to hang their own ideas out there and see if people go for it; and people have never gone along with AMD. They always go Intel's way. There are a lot of practical reasons why.

AMDs 64 bit extensions are about as simple and direct as they could possibly be. To use 64 bit addresses you need to have 64 bit wide registers. With the amount of registers that even present CPUs have behind-the-scenes, doubling the available general purpose registers from 8 to 16 is trivial. Programmers alway like more registers. It makes programs go faster, and you don't have to juggle things on an off the CPU so much. The SSE2 instructions that Intel orginated have proven to speed things up on some pretty popular CPU uses, so AMD has doubled the number of 128 bit registers that these instructions can use. Notice something here: while Intel's marketing spin-meisters are pooh-poohing 64 bits, Intel's SSE2 registers are 128 bit, the MMX registers are 64 bit, and floating point registers are 80 bit. People also should know that memory sticks have been 64 bits wide for years and years. It is about time the general purpose registers went to 64 bits.

The downplaying of 64 bits is going too far IMO. To some extent, the negativism was required if people were thinking doubling the number of bits is going to double the speed. No. It could come close to that only if the data type you are working on is 64 bits wide, or comes in 64 bit chunks. Only in special parts of certain types of programs is that true. Only in those parts could the speed be doubled or quadrupled. But for most of that type of data, present CPUs are already 128 bits wide, as I mentioned. That leaves some of the in-between data processing that won't have to be narrowed to 32 bit pieces to get done; data types for which there are no special SSE2 style instructions; and data types that the future will define. Believe it or not, there was a time when there was no MPG, no mp3, and no jpg. There will be new data types in the future.

The major use for 64 bits is the larger amount of memory. (and more memory in itself usually speeds up things.) Maybe 4G seems like a lot. But right now people don't get worked up about 2 512K sticks = 1G. When the new generation of memory chips comes along, they quaduple the size. So in a roughly a years time, or maybe 2, 4G will cost about what 1 G does now. It won't stop there. It never has. If anything, the growth is accelerating.

For most people 512K is so cheap they get that amount even though they could do with less. They do enjoy some benefit from the 512K. Some people do a little more ambitious things with their computers, and therefore do not consider the price of 1 G very much. Where computers make money for their owners, the price of 4G or 16G is nothing to get concerned about. In a few years that will be minimal in a home computer. In order for that to happen, development has to begin now. But development cannot begin without a defined CPU in place. That can only happen once a CPU is released. That's why 64 bit computers should be out now. Intel is not saying what the 64 bit CPU is going to be, so no one can develop for it. They are holding back computer technology in the mass market, when they have in the past alway been ahead of it; way, way ahead of it. Maybe they are behind because they tried to make a leap with the Itanium and can't make a go of it.

AMDs streamlined method of handling memory in 64 bit mode is going to yield greater and greater speed increases as the memory size grows far beyond 4G. I'm sure Intel has some good ideas about what the future 64 bit computer is going to be, but they are not commiting to anything. Instead the spin-meisters are pretending 64 bits is not going to happen. Anybody that has followed computer technology for the last 20 years know that it is.

My second computer had 32K, which cost $500, and $500 was a lot more money back then. That was the Intel 8080 era. The 8080 was considered an 8 bit computer, but it had a 16 bit address width. The full address space was 64K. That was an immense amount and would have cost a fortune when Intel introduced the 8080. I got in pretty late in the cycle and the price wasn't so bad. People wondered what you could do to use all 64K. But in a year or two everybody had 64K. Intel was introducing the 8086, and it had a whopping 1M address space, which boggled the imagination. Who could afford 1M of memory? It had 16 bit registers, which was exciting in a futuristic sort of way, but who really needed 16 bit computing power? The point: Intel was always leading the way, and paving the way. Yet, what seemed unreachable, was surpassed in a few years, and today you can only chuckle. 1G would have cost many millions of dollars!

If Intel were on its schedule, and leading the way, 4G would not be affordable.

 

[Mingon]

For most people 512K is so cheap they get that amount even though they could do with less. They do enjoy some benefit from the 512K

😀 bill gates

 

[MrEgo]

People do not need 64-bit processors right now because there are no 64-bit applications. There are no 64-bit applications because there is no 64-bit processors.

The hardware market was afraid to release a 64-bit processor because there were no 64-bit apps. The software market was afraid to release 64-bit apps because there is no 64-bit hardware to use it with.

AMD finally took the plunge and, as we speak, "the snowball is rolling down the hill", meaning that now software companies can now comfortably, or at least a lot more comfortably than before Opteron/A64 was released, develop 64-bit software.

It was just a matter of who made the first move...congrats AMD.

In a way, you're seeing the same situation in the Video Industry. Since many of the new graphics cards were optimised for DirectX9, software companies are now starting to create games that utilizes DirectX9, i.e Half Life 2. Since ATI planned much better than NVIDIA did (at least it looks like it right now) when developing DirectX9 compatible video cards, ATI has captured the performance crown and probably a lot more market share relative what they had in the past.

 

[beyoku]

i dont think the a64 does emulation. I thought it is a 32bit chip with 64 extensions
yeah its run 64 native -

 

[beyoku]

VERY well put!
Yes the anti 64 has gone too far.
We should be happy with ANY new technology that we can get.
I have said this before but we dont NEED a lot of stuff.
we dont need 8xagp ?
You dont need dual channel ram
or 2ghz + processors
we dont need to overclock?
but we do, we take whats given and run with it.
I think intel is a little jelly when it comes to A64.
And who really cares it it will run 32 faster than its little brother AXP - why would we complain that it does extra.
Thats like complaining that your car has a hartop convertable? Or thinking its not useful that a car can go on land and water?

 

[zephyrprime]

But to answer your question... I'm not a programmer, so I'm not sure what part of the OS would benefit from the added precision of 64-bit, that cannot be obtained with 32-bit. Nor am I sure what kind of (desktop) application could be improved with that kind of precision.

Well, I am a programmer and even though I'm not an os programmer, I think I can say a few things about the topic. An OS would essentially see no speed benefit from going to 64bit. If anything, things could get slower because the instruction stream would get bigger with the presence of 64bit address arguments and the need to use a more sophisticated virtual memory page table scheme. With 32bit address space and the x86, you can just use a simple flat table for virtual memory translation which results in a 4MB big table. For a 64bit address spaces, such a table would be 4 petabytes big! Yikes!

The one point of benefit, which is peculiar to the A64, is the speed gain from having twice as many registers.

Also, I know that it's common to say that "64bits offer more precision" but I think that the phrase is misleading. 64bit integers don't offer more precision. They offer bigger numbers. Only floating point numbers offer more presicion with increasing bittiness. But anyway, basically no/very few parts of a 32bit os benefits from having "higher precision" except for the file positions pointers which don't need to be high speed anyways since the physical disk is so slow. However, if the OS needs to manage more than 4GB of memory, then of course 64bits would be extremely useful since page swapping sucks.

This is a little beside the point but let me give an odd example of when moving from a 16bit os to a 32bit os is useful. Windows 3.1 used a 16bit signed int to represent screen resolution. But printers are considered a kind of screen in windows. There now exist some really wide carriage printers that have more than 32768 "pixels" of resolution so windows 3.1 wouldn't be able to print to such a printer!

64bittiness benefits applications not operating systems. The current application that will benenfit most from 64bittiness is database servers which can also use more memory.

 

[Wingznut]

Thanks, zephyr... I appreciate the input.

 

[redpriest_]

AMD's 64-bit implementation isn't just about 64-bit pointers or addressing space (even though it's in reality 40 bits physical and 48 bits virtual, I believe, in current implementation, but that's off the top of my head). It's also about extending the register set, which prevents register spilling to main memory for a lot of common algorithms; this alone can improve performance in *recompiled* applications on order of 10-15%, sometimes even more. Also, anything requiring "big number" computing will benefit from the 64 bit wide registers (cryptography for instance, or large digit integer number type applications).

 

[Jeff7181]

I'm not going to bother quoting everything I want to respond to...

Intel says there's no need of 64-bit processing in the home... IBM said there was no need for computers in the home.

So it's not a yes or no question... it's a question of WHEN. Now? No, probably not. Tomorrow? Probably not. Next month? Again, probably not. Next year? Possibly. Two years from now? Even more possible. Five years from now? More than likely.

Programs don't get any more simple, if history repeats itself, programs will get more complex and require more processing power and more RAM to run properly.

Where could you see benefits from 64-bit processing now? Server applications definately... video editing definately. And my own personal belief is that portable computing would benefit from it as well. If you can run a processor at a slower speed, on less voltage, and get the same performance, that means longer battery life, and every notebook user would enjoy longer battery life.

Maybe Wingnutz can shed some light on that (or maybe not)... like what type of increase in transistor count to offset the 27% decrease in voltage, as well as the reduction in current draw from the SOI process.

Also... 32-bit computing WILL benefit from the A-64 when we get a 64-bit version of Windows... I'm just going to copy and paste to avoid typing and/or misquoting...

Compatibility mode gives you none of the advantages of a 64-bit architecture on the application level, as it is designed for running 32-bit apps on a 64-bit OS (hence the name compatibility); The extra registers and 64-bit register extensions are ignored in this mode. Compatibility mode is important because of the 2GB process size limitation under Windows OSes. Although 32-bit Windows offers support for a maximum of 4GB of memory, each process can only use a maximum of 2GB of memory - the remaining 2GB is reserved for the OS. By running a 64-bit version of Windows (when released) and a 32-bit application, compatibility mode allows for each 32-bit process to have up to a full 4GB of memory, with the OS using anything above that marker.

- according to Anand Lal Shimpi

Also something else to consider is that AMD has been using ceramic substrates for the Opteron. Switching to an organic substrate should allow higher clock speeds.

Rumor has it IBM and AMD have already begun working on the .065 micron SOI process as well... there has been talk of dual Opterons in the same packaging for months now... and while that's speculation, I surely wouldn't dismiss it as an impossibility. In fact, I'd say it's very likely... you'd get the power of dual processors at a reduced cost due to the reduced packaging cost.

 

[chsh1ca]

Originally posted by: Accord99
All of those performance gains comes from the on-die memory controller and/or the increased bandwidth and/or larger L2 cache that the Opteron has over the Xeon. None of these performance enhancements requires 64 bitness. And double precision floating point operations don't gain anything from 64 bitness, as these registers are already natively >64 bit on existing x86 CPUs.

That's an excellent response, except that factually it falls very very short. The On-die memory controller only lowers latency. The FSB of the chip is still the FSB, and DDR333 will still only operate at 333MHz, which is a far cry from RDRAM's base of 800MHz effective. The L2 cache on the Opteron is identical in size to that of the Xeon (both are 1MB, at least, both I've seen tested together). Higher precision floating point operations would by necessity gain speed from a 64-bit wide register. The addition of registers will certainly help out, and as has been pointed out, that's chip-specific.

Originally posted by: Wingznut
You are correct... Any application that benefits from addressing more than 4gb of RAM will benefit from being designed for 64-bit... But I can't think of a desktop application that fits that description.
Games? Not hardly.
3D Rendering? That would be a HUGE chunk of data, to require more than 4gb. Anyone needing that kind of power would be on a multiprocessor workstation, not a desktop.
Video Editing? See above answer.

Well, I will take a venture here, and suggest you don't do much of any of those three categories.
Games most certainly do, there are quite a number of them that are RAM hungry, and routinely force a system to use more than 4GB taking swap space into consideration. Mainly I would consider these to be MMOGs (Planetside is one I play and the thing eats memory like a swarm of locusts eat plants), and highly graphically-intensive FPS or Simulation type games.
3D Rendering absolutely will use any and all resources it has. Even some rather simple scenes in 3D Studio Max use upwards of 1GB of RAM, and the higher the polygon count the more RAM it will eat.
Video editing is one area I'm surprised you didn't admit to, since it is to me the most readily obvious. Having used Adobe Premiere quite extensively, I've found that its optimal performance working on a DV-cam video capture was about the 2GB of ram mark. That's changed recently as well, and it eats more RAM in the newer versions.

But you are right... There could very well be a couple of niche applications that might benefit from 64-bit. I don't really know which ones those might be, though.

The more common examples given are scientific research (number crunching), and renderfarms.

The thing is, you have to think of the 'average home PC' as a changing concept. The entire concept of the PC is going through a metamorphosis. In the homes of people I know, the PC is replacing a lot of entertainment equipment. Not necessarily TVs, but I know more and more people whom I wouldn't classify as 'advanced' users who are using their PCs for things like DV capture, editing, and burning to VCD or DVD, a stereo replacement, etc..

PCs are moving towards becoming the entertainment hub, and while that may be a niche market now, remember that PCs themselves were once a niche market. The more and more data a PC has to work with, the more memory it will need. We may not see an app that lists 4GB of RAM as a requirement tomorrow, or a year from now, but it will come eventually.

 

[Accord99]

Originally posted by: chsh1ca
That's an excellent response, except that factually it falls very very short. The On-die memory controller only lowers latency. The FSB of the chip is still the FSB, and DDR333 will still only operate at 333MHz, which is a far cry from RDRAM's base of 800MHz effective. The L2 cache on the Opteron is identical in size to that of the Xeon (both are 1MB, at least, both I've seen tested together). Higher precision floating point operations would by necessity gain speed from a 64-bit wide register. The addition of registers will certainly help out, and as has been pointed out, that's chip-specific.

The on-die memory controller is by far the most important enhancement to the K8 architecture in terms of performance, as it effectively halves the memory latency, which is least as important as memory bandwidth for performance. DDR333 gets you 2.7GB/s of memory bandwidth, 16 bit 800MHz RDRAM gets you only 1.6GB/s and RDRAM has been deprecated from the Intel lineup. The L2 cache of the Opteron is larger than the majority of Xeons which only have 512KB, there are some with larger L2s but are rare, and some new ones have a 1 MB L3 cache. Floating point registers are already 80-bit in x86 CPUs, only the integer registers increase from 32 bit to 64 bit. The Opteron gets 8 additional registers but they're only in 64 bit mode.

 

[Wingznut]

chsh1ca... I would put money on the fact that you will not notice any speed improvement in any game when going from 4gb of RAM to any amount higher. Sorry... but no.

I agree that scientific research and renderfarms would benefit from 64bit... But I certainly wouldn't define that as desktop usage.

And yes, there are instances where 3D rendering and video encoding/editing will benefit from more than 4gb of RAM... But those people doing that are using multiprocessor workstations, not desktop PC's.

DV capture, burning cd's/dvd's, audio playback... None of those things would benefit from more than 4gb of RAM for the desktop user.


Look, I'm not saying that there is no need for 64bit, or that there never will be. I just personally can't think of an example where 64bit would benefit the desktop user.

 

[Jeff7181]

I just personally can't think of an example where 64bit would benefit the desktop user.

Distributed Computing 😀

 

[MDE]

Originally posted by: Jeff7181

I just personally can't think of an example where 64bit would benefit the desktop user.

Distributed Computing 😀

And who's one of the major sponsors of F@H?

 

[SlipperySam]

I would just like to interject somethig here.
All the preliminary testing seems to point to the conclusion that the Athlon64 will run my games faster than the fastest P4.
Nothing else matters.

 

[MDE]

Originally posted by: SlipperySam
I would just like to interject something here.
All the preliminary testing seems to point to the conclusion that the Athlon64 will run my games faster than the fastest P4.
Nothing else matters.

Couldn't have said it better myself. Care to comment Wingznut?

 

[sandorski]

These comments are really High Praise though, it used to be that AMD didn't even exist, if you listened to comments(or lack thereof) from Intel. 😉 The official line used to go something like this: "We do not comment on others' product lines"

 

[Cerb]

Originally posted by: metroplex
I read a CBS market watch report that said something along the lines of:
there was a engineer talk to engineer conference in the past 3 days in San Fancisco, CA, Intel says AMD will launch Athlon 64 next week, they expect negative reaction from market because PC market is not ready for 64 bit computing and lack of application support/etc...

But why would the market readily accept Itanium instead of Athlon 64??

The market hasn't accepted the Itanium. If it weren't for Intel and HP developing it, it wouldn't even have the shaky hold it has now.

Am I missing something here?

Yes. 64-bit computing has NOTHING to do with it right now on the desktop.
Key phrase: "right now"

 

[Wingznut]

Originally posted by: MonkeyDriveExpress

Originally posted by: SlipperySam
I would just like to interject something here.
All the preliminary testing seems to point to the conclusion that the Athlon64 will run my games faster than the fastest P4.
Nothing else matters.

Couldn't have said it better myself. Care to comment Wingznut?

Ask and ye shall receive. 😉

SlipperySam has the right attitude. To the desktop user, the strength of the A64 is not that it can run x64 code... It's strength is that it looks like the A64 will likely be a strong 32bit performer.

Originally posted by: sandorski
These comments are really High Praise though, it used to be that AMD didn't even exist, if you listened to comments(or lack thereof) from Intel. 😉 The official line used to go something like this: "We do not comment on others' product lines"

I believe that is still Intel's policy. We don't really know the source or context of the heresy in metroplex's initial post, but I doubt it was any kind of official statement.