Hammer Prototype Test

[Jerry Jones]

If a 64-bit processor is available NOW...

...that doesn't necessarily mean one will be able to tap its potential NOW.

But the fact a given chip can run on future 64-bit operating systems (i.e. 64-bit Windows) certainly gives the 64-bit chip an advantage, in my mind, over any legacy 32-bit chip.

Would it be "crazy" for a purchaser to buy a 64-bit Hammer chip, using a a 64-bit Hammer motherboard, and run an existing version of Windows and existing 32-bit software in anticipation of upgrading within 18 months to a 64-bit version of Windows?

I think not.

With such a chip, one could expect a significant performance increases from the OS upgrade alone.

Later, one could expect another significant performance increase when a 64-bit application (MediaStudio Pro-64) were to be released with enhanced code to take full advantage of 64-bit processing.

I think Atkinson-Barr's comments (posted earlier) reveal several areas where 64-bit processing would offer video editors, specifically, some pretty amazing performance gains.

Jerry Jones

http://www.jonesgroup.net

[Hulk]

I'm only saying that in my experience it's not worth it to buy a piece of computer related gear expecting future features to be implimented. I buy a system based on what it can do for me immediately. I mean, I waited a few weeks to get a 533fsb P4 over a 400fsb P4, but I am extremely leary to buy based on future benefits. I've been burned too many times. Waiting for SSE apps, waiting for SSEII apps, waiting for 3D apps, etc...

Yes, those apps did finally come along. I would have done better waiting for the SOFTWARE, then buying the hardware to support it rather than the other way around.

In addition, by waiting, not only do you not lose performance since the software isn't ready, but you save money by not buying the latest/greatest, AND the second version of the CPU usually is enhanced and has better motherboard support.

Witness the original P60/66. Expensive, and used large process designs. The P90/100 released a few months laters were smaller process, cheaper upon release and much faster. These chips fulfilled the promise of the Pentium advertising campaign.

The initial PII 266's were also using a larger process than the soon released PII 350/400's which also used the 100fsb which was MUCH more suited to the PII design.

Same thing with the first incarnation of the PIII's. First ones used .25u process and did NOT have on die secondary cache. I bought PIII with the introduction of Coppermine and the 0.18u process and on die/full speed L2 cache.

Again, this happened with the P4. Initial ones used 0.18u process with 256kb cache. Second generation had 0.13u process and 512kb L2, along with other improvement.

AMD follows this same trend if you look back.

I personally stay away from the first generation of any new chip. History shows the first release to be mainly built on aggressive advertising. The quoted performance gains usually come to fruition in the first revision.

Mark

[Jerry Jones]

These are all valid points, although I'm not worried about the first generation AMD Hammer chips, personally.

But is it possible that even given EXISTING Windows and EXISTING software... Hammer will be able to do more NOW than Intel's fastest 32-bit offerings?

We'll see soon enough.

It'll be fun to read the first real world benchmarks.

Jerry Jones

http://www.jonesgroup.net

[Hulk]

Agreed.

Mark

[Brian Ellis]

Also, when Bill has finished adding all his stupidities to 64 bit Windows, it will probably be slower than 32 bit or even 16 bit.

Remember DOS? When you added Windows 3.11 to DOS 6.2, it was slowed down. Then came DOS 7 with WIN95, which was still essentially 16 bit and was slower again. NT4 was the first 32 bit and was roughly on a par with 95, but 98, which was mostly 32 bit, was slower than 95 (as many here testified). In the days when I wrote software for complex instruments with much abstruse maths, I did it initially under DOS 5. When W95 really became the de facto norm, this software was translated for it. Under DOS, it worked on a 386 with 387 at 25 MHz with 1 Mb RAM (minimum). Under W95, the minimum to allow it even to work was a PII at, I think, 200 MHz with 16 Mb RAM and, even then, the number crunching was slower.

By future extrapolation, I would guess that the advantages gained will be minimal, if at all.

I still have a few specialised 16 bit DOS apps that I run under 6.2 on a PIII 450. You have NO idea, unless you do the same, how fast things work. Under Linux, I have a complex CAD system that I beta test and it works almost twice as fast as the same programme under Win 98 or 2000.

Someone above says that a 64 bit Linux is already available. Now, I imagine that this may sprint like in the Olympics 100 m, but Windows? Well, let's wait and see whether Bill's checking of all the security leaks (and memory leaks) will really allow us to gain an advantage. As a sceptical Scotsman, I hae ma doots :-)

[Ray Austin]

First off HT or Hyper threading is basically the same as dual processor systems. If the OS has dual proc. capability it has HT capability. There is no need for a special written program to take advantage of HT. PROVIDED....that it was a program that has multi thread capabilty then it will take advantage of the HT of the P4. Provided that you are using an Operating System that support Dual processors. [I.E. Windows NT, Windows 2000, Windows XP Professional]

Basically that's what Hyper threading is... 2 processors in one. The OS sees it that way and therefore it is.

So don't make this more than what it is 'gize.


It is true that the all this talk about 64bit needs an OS that will support it. It's been in the wind that M$FT will be there. They want to be the only ones selling software so you can be sure Mr. Bill will be there.

So to put this in perspective, don't buy a HT processor if you are running Win9x/me, XP Home... 'cause it won't matter if your program supports HT or not... it won't even know it's there.

[Jerry Jones]

Ray,

I understand Hyperthreading requires Windows XP Pro.

Windows 2000, I'm told, doesn't fully support it, as this article explains:

http://www.theinquirer.net/?article=2571

Jerry Jones

http://www.jonesgroup.net

[Hulk]

Have any of you heard of a Multitrack audio editing program called SAW? Software Audio Workshop by IQS (Innovative Quality Software), written by Bob Lentini.

This was a multitrack audio program that was also quite adept at mastering from the early '90's. When the competition (Sound Forge and others I can't remember right now) were fumbling around with routines that they still basically use today, but on computer MUCH, MUCH slower (Pentiums, no II, no III, no 4, just plain old Pentiums), Bob Lentini had SAW which was fully written in assembly and was as fast on my old P5 90 as Sound Forge 6 is today on a machine AT LEAST 10 times as fast.

SAW and IQS fell out of vogue as computing horsepower overwhelmed the need for smart, efficient code. That program was truly amazing. It came on ONE floppy disc, ran from the executable, and installed NOTHING on your computer!

It would be nice if some of the more time consuming routines in MS Pro could be hand coded in assembly language. Things that are used often like crossfades, some of the video filters, etc. I bet we would see a HUGE increase in performance.

I know this would take A LOT of resources and wouldn't provide much advertising on the side of the box, but it would be a great benefit to the consumer.

I remember doing some assembly programming on my Atari 800 20 years ago and it was a bear!

Mark

[Brian Ellis]

Yes, HT requires XP Pro.

However, XP Pro and HT wll NOT give you any improvement on software not SPECIFICALLY written for it. Quote from an article in PCPro, February 2003:
"For example, when we first tested Hyper-Threading in issue 90, we were using a version of LightWave with good SMP support, but without Hyper-Threading support. It rendered nearly 30% slower when running four threads across four logical processors. However, when we ran the same render test using LightWave 7 on a dual 2.4 GHz Xeon, we gained a 7 per cent performance increase from 4 threads."

Most software cannot take advantage of HT, or dual processors, because it is written as a single thread. This means that, if HT is enabled, it is slower because each instruction may be executed in both threads and the chip takes time to choose which it really means to use. If it doesn't do this, half the registers will not be working, anyway.

To make it clear: HT is NOT two processors on one chip. Only a small number of items have been added to a Pentium to allow it to work. These allow the single processor to use idle cycles and more clearly defined to allow two threads to work in virtual parallel. The motherboard sees an HT chip as if it were two processors. If you like, it's like a dual-lane carriageway running alongside a motorway, rather than two motorways. The OS sees the logical "second" processor as if it were a physical one. Theoretically, W2kPro may be able to run a single HT, as it can support two processors, but certainly not a dual HT systrem. Whether it can practically, I don't know. Intel recommends XP Pro The problem is that HT has its own specific instruction sets.

[Ray Austin]

Ok I have officially been corrected. I thought I had read an article at THG that said what I was saying above. However, I dug deep and couldn't find it.

I did however find what you gize are saying. Intel does claim that only small mods need to be done to the code to make it compatible with HT. In a nutshell I think Intel is just trying to eke out more performance from it's chips to stay ahead of amd who run's their's at a lower rating, but more efficiently.

I also saw a few Lightwave tests with and without HT. I think the code was not optimized either as the HT enable ones didn't do so well.

So far I think Intel will just use this as an excuse to charge even more for their chips. With that said I hope AMD64 punishes INTEL severely. Maybe by 2005 the world will become 100% 64Bit... maybe...

Although a few more years like 2002 for AMD will likely cause extinction.

[Brian Ellis]

Of course, you can buy a 64 bit workstation today, from Hewlett-Packard or Sun Microsystems. The HP ZX6000 has dual 900 MHz or 1 GHz Itanium 2 processors. The UK price is a mere GBP 9,410 + VAT, say, about USD 14,000. The native OS is the HP UX11i v 1.6 64 bit OS, with 32 bit emulation of Windows. It will also run Red Hat Linux. I'm told that with ordinary Windows apps, it feels like a Pentium II/250 I don't know whether the emulation works with every software. Interestingly, the graphics card is a simple 32 MB ATi Radeon in an AGP 4x slot!

My forecast is certainly not a 2005 world running 64 bits. There is simply no need for it. 32 to 64 is not the same situation as 16 to 32 bits In 1994, there were any number of 32 bit RISC processors available, not to mention the Pentium (launched in 1993). This means that the hardware was the driving force, because cheap computers were available even 8 years before a fully 32 bit Windows (XP) became available. Today, it is unlikely that Hammer will sweep the world and Itanium 2 even less so. Why? Because very few users NEED the extra decahexal precision real number calculation possibilities. It's not going to let you type a Word document any faster, is it? And I'll wager that probably the only Excel workbooks that would profit from it (assuming a 64 bit Excel existed, of course) would represent 0.0001% of the total.

I'll tell you where 64 bit working will be useful: for corporate servers and this is where the market is. That is why Intel are going for the Itanium series with its massive cache of 6 Mb (on I3, codenamed Madison).

Hammer, which is closer to the legacy x86 series, is aiming more for the advanced PC market. Because of this, AMD have retained a 32 bit CISC core with a 64 bit RISC shell. This will allow 32 bit software to run well as the shell can convert the data to the core and vice versa, very rapidly, but it will be no better than a 32 bit CPU for these apps. In fact, probably slightly slower, because there have to be several RISC ops for each CISC op. Where the Hammer falls down in its present config is the insufficient number of registers and small cache. The HyperTransport system will partially make up for this lack for ordinary use but it will not be a clever system for server use. Here, Itanium will win hands down.

The Intel Deerfield project may seek to compete with Hammer, but I don't think we'll know for another year, after Montecito is launched later this year (this will be fast multi-GHz version of Itanium).

[Jerry Jones]

Brian,

You haven't convinced me.

I believe you are W-R-O-N-G.

Sorry.

:-)

Jerry Jones

http://www.jonesgroup.net

[funky-d-munky]

I agree. The A64 will perform better clock for clock than any 32-bit x86 chip ever made. And its 64bit mode will be instantly usable under Linux helping that platform secure even more market share of high-end apps. As well this chip will be seen as somewhat future proof for when win64 arrives.
I for one think Intel and Windows will be stung hard. I mean is there anyone here who would rather have a p4 3.5 than a A64 3500+?

[Ray Austin]

Brian to say 64bit isn't needed is to also say a 4Ghz cpu isn't needed either or even a 3ghz chip for that matter.

Why did we jump from 16 to 32 bit? Progress. Plain and simple.

It doesn't matter if it's needed or not. It's to compete. There is a theoretical limit as to how fast a processor can go in 32bit. I'll chance it and say that we are very close to that threshold. [I am talking air-cooled just so we all don't get slap-happy.]

I think 64bit will be mass-market by 2005 for very obvious reasons. "Even tho Intel made 64bit for the server market.... what better way for AMD to disrupt their plan."

• AMD is losing money and ground to intel
• Big numbers always sell... 64bit is a magical number.
• 64bit is larger than 32, therefore it must be better.
• Chipset Manufacturers are lining up

I realize that code needs to be re-written to take advantage of 64bit, but AMD has that covered too. "backward compatible with 32 bit...." It's a win - win for AMD.

So it's not about what is better.... it's about what will sell and create enthusiasm for a presently gloom market. In the long run it will be better. As far as not needing it? Hi-Def video is right around the corner, I think we can think up plenty of ways to suck the power out of it.

[Elie]

Welcome to MURC Ray

I agree with you, AMD reported a loss in revenue to Intel fourth Quarter mainly due to lower sales and restructering etc.

As I said before, If AMD releases the 64 bit CPU to the market by mid 2003 at higher clock rates, it will hurt Intel, but not as much.
I know Intel has stuff up their sleeve ready to announce any time soon to answer to AMD's 64 bit technology.

Cheers,
Elie