Sign in with
Sign up | Sign in
Your question

Questions for Raystonn

Last response: in CPUs
Share
May 3, 2001 1:08:48 AM

Now, you say you work for Intel Corp., and you describe yourself as a software engineer, which makes me wonder... With the SSE2 instruction set out in the P4 core, what exactly will that mean once SSE2 filters down into the mainstream market? Say you have Program A, and it is a gaming program. Now, Program A has the SSE2 optimizations, how exactly will that affect the quality and speed of that game? How exactly would to performance change in relation to that same program without the SSE2 optimization? I use gaming as an example, merely because it's the only thing I can think of, and the main thing I'm interested in. Since you say you are a software engineer for Intel, it made me assume that you would have an intimate knowledge with the SSE2 and the impact it has on programing and what not. Also, does programing the SSE2 set into code require a generous amount of effort, or is it relativly simple to do? My other question has to do with something that was on my mind a while back, but I neglected to post it. What exactly does the SSE2 instruction set do for the processor? I mean, I understand it is a set of instructions that does something positive for the processor, but what exactly is it? I have concluded that it is something that forces a program to upload certain things into the RDRAM that normally aren't, thereby allowing it to be accessed via the 400MHz bus instead of coming off the hard drive. I may be completely wrong, as I am only a college student with a passing interest in computers, but it is really the only thing that I could think of that would give a large enough performance increase that would compel both developers and AMD to take an interest in it and gain the rights to use it.

WIth your connections to Intel, is there anything that you could tell people that would affect their purchasing choice from AMD to Intel, that might not be well-known? I mean, I will be buying a computer this summer, and have settled on the AMD, mainly for the low price/high performance. Plus I like a good challenge. Is there anything that you can say that wouldn't violate a NDA that might persuade me otherwise? I mean, right now Intel has some serious issues with the P4 not performing up to the tech communities standards and the seemingly unnatural union between Intel and Rambus. I have nothing against Intel, my P2 has been a great processor, but even if a quarter of the things one reads about Rambus were true, it is hard to think about supporting such a company. Therefore, until Intel removes the reliance of Rambus from the P4, I could never justify the purchase. Therefore, if there will be a DDR solution for the P4 around the time of the real P4 coming out, I would consider putting off my computer purchase until then. I say DDR, because I truly think that DDR will be the next stepping stone in the evolution of RAM. RDRAM was really a step back, with the pair installation requirement and such. So, if you have any info, it'd be greatly appreciated.

-SammyBoy

Without Evil, there can be no Good. Therefore, without an Intel, there can be no AMD.

More about : questions raystonn

May 3, 2001 1:49:33 AM

Quote:
is there anything that you could tell people that would affect their purchasing choice from AMD to Intel, that might not be well-known?


just keep in mind that he <b>does</b> work for intel and therefore has no more insight on amd's products than the rest of us. unless they do cpu dissections over there...

<A HREF="http://www.512productions.com/lobstermagnet/" target="_new">Hyakugojyuuichi!!</A>
May 3, 2001 2:11:13 AM

'unless they do cpu dissections over there...'

Count on it. Only the enemy tells you where you're weak, and he is strong.
That's why DDR P4 has been given so much consideration.......

BTW, all this stems back to Cash (notice the capital letter).
Related resources
May 3, 2001 2:31:02 AM

"has the SSE2 optimizations, how exactly will that affect the quality and speed of that game"

"What exactly does the SSE2 instruction set do for the processor?"

There are two separate things that should be looked at when discussing performance of applications on a P4. The first is the application being built with a compiler that is knowledgable of the P4's pipeline and the way it does branch predictions. This does not use any P4 specific instructions (it keeps the application compatible with all processors) yet can have a significant impact on performance. Depending on the age of the software, a simple recompile with a newer compiler, such as Intel's C/C++ Compiler 5.0, can increase performance by anywhere from about 20% to over 300%. Most of the benefit comes from eliminating branch mispredictions and will even speed up applications when run on non-Intel processors. (This is why AMD chose to use the Intel C/C++ Compiler when they built and ran the Spec benchmarks. They definately saw an improvement.)

The second factor in application performance is taking advantage of SSE and SSE2. SSE stands for "Streaming SIMD Extensions." SIMD stands for "Single Instruction, Multiple Data." These instructions are optimized in such a way as to allow a single instruction to perform operations on a whole stream of data at once, rather than forcing applications to insert one instruction for each single piece of data. As any programmer will know, operating on a whole chunk of data at once is much more efficient than processing one byte at a time.


"is there anything that you could tell people that would affect their purchasing choice from AMD to Intel, that might not be well-known"

Of course what is well-known depends on who you are. This is why I answer questions on this forum. I'd rather everyone be well informed when they make their purchasing decisions than buy on a rumor or misconception. I understand everyone wants the most processing power for their buck. I understand some people are afraid of wasting their money because of some company's marketing department. I understand everyone is going to assume I'm biased because of where I work, even if I do not work in marketing. This is why I only present facts, usually with links to sources, and try to let others form their own opinions. But I digress. You asked what I could tell everyone that might affect their purchasing choice.

I would hope that all the data I have posted on this forum might have helped influence everyone in the direction of truth and performance over hype and rumor. I'm not going to point out features and flaws of companies, but of products. SSE2 is a "Good Thing"(tm). NetBurst is a "Good Thing". High bandwidth and low latency is a "Good Thing". Most of us know these things. What I've been trying to show in most of my posts is what most people don't know or haven't thought about.

One of these tidbits of information is that RDRAM actually has less latency than SDRAM in medium and high memory load situations. This is because SDRAM has to wait through a few dead cycles whenever it changes from reading to writing or vice versa. RDRAM has no such delay. In a busy system you are going to have reading and writing going on all the time, forcing many dead wait cycles on SDRAM. Most synthetic benchmarks today measure reading and writing in completely separate tests. These won't show accurate measurements of memory usage in real applications because you never switch from reading to writing or vice versa in any of the tests.

Now I'm not going to say I support or oppose the business practices of any particular companies, but I will say that RDRAM is technically better than SDRAM. Everyone is free to decide not to purchase products from a company for political reasons. This does not upset me, even though the lack of the purchases don't actually affect those you are targetting. (Licenses for making RAM are going to be paid by manufacturers regardless of how much of it sells.) But to say that a product is inferior because the company that designed it might be doing something with which you disagree is just wrong. If you are going to boycott a product, do it with full knowledge of exactly what you are going to be doing without. Don't try to pretend the technology is no good to try to make yourself feel better.

If anyone has any more questions on SSE2, RDRAM, or anything else, feel free to ask. If you disagree with something I say, feel free to debate politely. If you have surpressed anger and feel like venting, go look for tbirdinside and amdmeltdown. They are always up for a good fight. ;) 

-Raystonn

= The views stated herein are my personal views, and not necessarily the views of my employer. =
May 3, 2001 2:21:06 PM

Right on Ray (<i>you da man</i>)! :cool:
Anonymous
a b à CPUs
May 3, 2001 2:55:47 PM

Another question.

You've been talking about Intel stuff so much like SSE2 giving C/C++ programs 30%-300% increase in speed. Would you mind telling me about how it compares with the stuff that AMD lets out. Your talk of stuff is impressive, but they all concern Intel stuff. Would you mind telling me some stuff about AMD and its processor and how it compares to Intel? I wouldn't mind if you critize AMD as long as the reasoning is true.

We put text in here and that's our signature?
Anonymous
a b à CPUs
May 3, 2001 3:07:01 PM

Alot more questions just popped up my mind.

DDR-RAM is just a piece of SD-RAM but the bandwidth has been doubled by something right? And if this is true, what is that something that doubles the DDR-RAM speed?

Plus, in a earlier post u described RDRAM has lower latencies compared to SDRAM when doing medium to high end jobs. So would DDR-RAM have higher latency too because it has several dead cycles like SD-RAM? And therefore, in low-end jobs, does DDR-RAM have lower latencies than RD-RAM? What would you classify as a low end job and medium end job and high end job?

We put text in here and that's our signature?
May 3, 2001 3:48:48 PM

Quote:
DDR-RAM is just a piece of SD-RAM but the bandwidth has been doubled by something right? And if this is true, what is that something that doubles the DDR-RAM speed?

I think you're in the realm of QDR, or Quad Data Rate SDRAM.
Where data is sent on the rising and falling edges as well as the high and lows of the clock signal. One of the things I like about RDRAM is the fact that you can have delayed writes, where write requests are stored in a buffer and read requests are processed first.. thus supplying the processor with much needed information.

As per Raystonn's comments.. I myself have no problem with the RDRAM technology, my issues lie with the people behind that technology, which is why I choose not to support it.

Intel Components, AMD Components... all made in Taiwan!
May 3, 2001 7:05:12 PM

well, reads and writes could occur only on edges since digital circuits change their states only on edges. level clocks can be used for signalling only.

DDR could be designed with buffers as the RDRAM, and the data rate could be quadruples by using multiple channels - just as P4 uses. the chipset and the cpu must be capable of doing so. thus the cpu could have two (or more) independent memory channels - thats double number of address as well as data lines, or they could have a wider bus, say a 128 bit one against 64 as it is now.

or there could be analog memories, where one cell could store more than one bit, in terms of voltage levels, which could be decoded directly into the cpu, which will help us make physically smaller but logically wider buses.

girish


<font color=blue>die-hard fans don't have heat-sinks!</font color=blue>
Anonymous
a b à CPUs
May 3, 2001 8:04:40 PM

>Also, does programing the SSE2 set into code require a
>generous amount of effort, or is it relativly simple to >do?

The Intel compilers will do some stuff for you, but to actually realize the big benefits of SSE2 you need to recode your application. And it doesn't look like a trivial exercise for the most part.

Intel has an SSE2 <A HREF="http://developer.intel.com/software/products/itc/sse2/s..." target="_new">tutorial</A> on their site. They mention some abstraction mechanisms (new vector types etc.) but concentrate primarily on assembly code.

For some examples of how to code things in SSE, see this <A HREF="http://developer.intel.com/software/products/itc/strmsi..." target="_new">link</A>
My favorite is the 4x4 matrix inverse. The performance of the SSE version is impressive, but take a look at the code. Yikes.

Also, all may not be peaches & cream with SSE2. See this <A HREF="http://linuxtoday.com/news_story.php3?ltsn=2001-05-03-0..." target="_new">link</A>
It's some e-mail traffic from Linux Torvalds, the head honcho of Linux explaining why some things are slower on the new kernel (2.4 series) then on the old 2.2 series on the PIII cpu. Has something to do with SSE, but I'm not enough of a gearhead to follow it all.

I'm a little confused though, they say PIII and SSE2. I thought PIII only had SSE and SSE2 was only available on the PIV?


In theory, there is no difference between theory and practice.
In practice, there is.
May 3, 2001 9:08:57 PM

"The performance of the SSE version is impressive, but take a look at the code. Yikes."

I thought the code using the SIMD instructions was much simpler than the code without (just the C code using Cramers Rule).
May 3, 2001 9:35:20 PM

"SSE2 giving C/C++ programs 30%-300% increase in speed"

Actually, simply recompiling with a newer compiler such as Intel'c C/C++ Compiler 5.0 will bring these performance benefits. SSE2 is more icing on the cake and can significantly improve performance even greater.

"how it compares with the stuff that AMD lets out"

AMD's 3DNow was equivolent to Intel's MMX. Currently AMD does not have anything comparable to SSE or SSE2. This is why they chose to adopt SSE2 in their future processors.

"Would you mind telling me some stuff about AMD and its processor and how it compares to Intel?"

I make it a point not to discuss companies, just products. The Athlon was AMD's answer to Intel's Pentium III. Both have similarly designed cores that will scale to approximately the same clockspeed. The Pentium 4 is designed to reach clock speeds upwards of 10GHz by the end of its lifetime. It is similar to a 2nd gear processor whereas the PIII and Athlon would be 1st gear processors. You won't see a great deal of difference in speed between the high end of 1st gear and the low end of 2nd gear, but 1st gear will seem to have more power. However, there comes a point where you reach the limits of 1st gear, and to continue pushing faster with the same gear (design) will redline your RPMs and blow out your engine. Before this happens it's necessary to switch gears, moving to a processor core that's capable of the speeds that are to be expected well into the future. This is the purpose of the Pentium 4. I cannot comment on AMD's future offerings, not that of any others, but I do not see anything that has actually been released by them that promises to offer this next generation of speed.

Many will now jump up and yell about the future not being relevant; it's all about the present. This is true to some extent. However, software changes every day. Software is constantly being updated and recompiled, some even adding support for SSE2. As software is rebuilt, even without SSE2, it will show greater performance on Pentium 4 processors than on any other processor in the marketplace. The current batch of old software still in use suffers from a performance penalty of between 30% and 90%, depending on its age, simply due to the branch mispredictions incurred by poorly optimized code. A simple recompile, without adding any SSE code, will lead to a performance increase of 30% to 90% without any real effort on the part of programmers, as these branch mispredictions will be reduced dramatically, if not completely eliminated. What this means is current batches of Pentium 4 processors will start performaing better and better even without taking into account newer/faster versions of the CPU. A couple good examples of this are the Photoshop addon from Adobe (www.adobe.com) that increases performance by 70% on the Pentium 4, and the new version 12 drivers from nVidia that show a performance increase of between 30% and 90%, depending on the game/application. Because of this, if you want to purchase a CPU to last for 3 or more years today, this is the one I recommend.

Remember that when looking at a performance/price ratio, one must take into account the following: the performance averaged over the whole time period of ownership, and the price as a function of time. For example, if a processor performs better and better as time goes on, this must be averaged in throughout the time period you plan on keeping it. Also, if you plan on buying a new CPU every 6 months or 1 year, it will cost you more than holding onto one for a longer period of time. For this reason, prices should be thought of in terms of a dollar amount per month of ownership. If you buy a $180 CPU every 6 months, your CPU is costing you about $30 / month. If you buy a $400 CPU every 2 years, your CPU is costing you about $16.66 / month. You can often get a much better performance/price ratio by buying a CPU that will do better in the future and hanging onto it a bit longer.

-Raystonn

= The views stated herein are my personal views, and not necessarily the views of my employer. =
May 3, 2001 9:42:40 PM

"DDR-RAM is just a piece of SD-RAM but the bandwidth has been doubled by something right? And if this is true, what is that something that doubles the DDR-RAM speed?"

Memory is accessed on the rising and falling edge of the memory clock rather than just on one of them.

"So would DDR-RAM have higher latency too because it has several dead cycles like SD-RAM?"

Yes it would.

"in low-end jobs, does DDR-RAM have lower latencies than RD-RAM"

It can, depending on how much memory you install and what the CAS rating is on your DDR. A low end job would be one where there isn't a high demand for memory accesses, and hence there isn't a whole lot of switching between reading and writing. I should also point out that memory latency has a lesser impact on overall system performance when memory isn't being used much.

"What would you classify as a low end job and medium end job and high end job"

Low end I described above. A high end job would be a system near 100% CPU usage with many threads reading and writing to memory simultaneously. A medium job would be something in between, such as a standard application where you may not have a great deal of multiple threads, but there is a healthy amount of reading and writing to memory and in no particular order. A memory latency benchmark that does its reading and writing in separate batches would be an example of a low end application. The nice neat separate reading and writing is definately not an example of a real world application and carefully avoids showing the latency issues when switching between reading and writing.

-Raystonn

= The views stated herein are my personal views, and not necessarily the views of my employer. =
May 3, 2001 9:45:05 PM

"they say PIII and SSE2. I thought PIII only had SSE and SSE2 was only available on the PIV?"

SSE2 is not available on the Pentium III, only on the Pentium 4. I haven't read the article so I'm not sure what they were trying to show. However, Intel does have a beta Linux version of their Intel C/C++ Compiler coming out this month.

-Raystonn

= The views stated herein are my personal views, and not necessarily the views of my employer. =
May 3, 2001 10:12:25 PM

"Remember that when looking at a performance/price ratio, one must take into account the following: the performance averaged over the whole time period of ownership, and the price as a function of time. "



Excellent point!
Anonymous
a b à CPUs
May 3, 2001 10:13:02 PM

>However, Intel does have a beta Linux version of their
>Intel C/C++ Compiler coming out this month.

This is kernel stuff. Nobody in their right mind is going to try building the linux kernel with the intel compiler. Even the reccomended Gnu compiler for the kernel is several releases back from the current release.

The linux kernel is so complicated, and interacts with hardware at such a low level that it tends to be very picky about compilers.

That's one thing you have to keep in mind when you say "Just recompile it". Sounds like a simple solution, but it's often not. For mission-critical projects it, at the very least, requires a complete validation cycle. At worst it can mean tracking down some very obscure bugs.



In theory, there is no difference between theory and practice.
In practice, there is.
Anonymous
a b à CPUs
May 3, 2001 10:22:12 PM

>I thought the code using the SIMD instructions was much
>simpler than the code without (just the C code using
>Cramers Rule).

Yea, whatever.
Maybe you're an SSE god.
But as somebody who does a lot of scientific & mathematical programming that cramer+sse code is a nightmare.

Aside from all the new functions, you're telling me that all those casts, pointer offsets, and hex addresses make for clean, simple, maintainable code?

I'd certainly hate to go back to it in a few months and figure out what it was I thought I was doing.



In theory, there is no difference between theory and practice.
In practice, there is.
May 3, 2001 10:27:11 PM

I am not claiming to be an SSE anything, much less an "SSE God". I was just simply stating that the SIMD code seemed more straight forward.
I would like to see your implementation of inverting a 4x4 matrix?
May 4, 2001 1:16:00 AM

Rayston can you share any info on the rumor of hammer being "SMP on a chip" 32x2=64bit?

You know AMD dropped EV6 or NUMA1 bus architecture, why would AMD settle for inferior BUS when current Netburst beats the piss out of it? EV6 at dead end?

Hope your not effected by NDA's
May 4, 2001 7:37:04 AM

"Hope your not effected by NDA's"

I am.

"can you share any info on the rumor of hammer being "SMP on a chip" 32x2=64bit?"

According to what I've read, hammer is simply a 64-bit extension of the same old architecture. It's no more "SMP on a chip" than the 32-bit 80386 was as compared to the 16-bit 8086/8088 CPUs. It's far too early to start theorizing what will be in that chip.

According to AMD's public roadmap, the hammer is not due out until the 2nd half of 2002. Judging by the lateness of their current offerings, for example the Palomino, I would say it will probably get pushed back to sometime in 2003, but that's just pure speculation.

-Raystonn

= The views stated herein are my personal views, and not necessarily the views of my employer. =
May 4, 2001 10:28:48 AM

<<AMD's 3DNow was equivolent to Intel's MMX. Currently AMD does not have anything comparable to SSE or SSE2.>>

This is a very dangerous statement because it is completely FALSE. AMD CPUs implement MMX instructions. MMX is used for integer SIMD on integers. AMD came out with 3Dnow! (which is mainly concerned with SIMD on single precision data) and Intel followed with SSE and gained a large following due to their size. AMD added some new 3Dnow! instructions with the Athlon but Intel created SSE-2 mainly to perform SIMD on double precision data. The Hammer series of processors will support MMX, 3Dnow!, extended 3Dnow!, SSE and SSE-2.

<<I cannot comment on AMD's future offerings, not that of any others, but I do not see anything that has actually been released by them that promises to offer this next generation of speed>>

This is not outright false but is misleading. Sure AMD have not released anything to get to clock speeds of much more than 2GHz but they are due to do so with the Hammer series. In any case it is well know that a faster P4 is required to match a lower clocked Athlon. Also remember that the Palomino series of CPUs is due which will have architectural changes to improve the performance per clock even more. The GHz rating is not the most important thing.

Most of the rest of your points relate to how well a P4 platform will in the future which is a very important issue. The 30-300% gains you mentioned from recompilation will work across ALL processors so that isn't a big deal for the P4 alone, in fact it is not true that you will get at least 30% in all cases but we'll let that one ride. If you buy a P4 now you will not be able to significantly upgrade the speed of the CPU because the pin-out is changing. SSE-2 optimisations will also improve performance and will be available on Hammer CPUs.

So let's recap. P4 can go to pretty high MHz rating but doesn't actually give great performance on existing apps. In future we can expect more performance from recompilations but these will improve performance for all CPUs. If we want to use SSE-2 we need to have code which uses lots of double precision numbers and then we have to put a lot of effort into it to make it work.

Basically the P4 is a good buy now if you

1) don’t mind the cost
2) want great performance in a small number of SSE-2 optimised programs (which you don't mind buying)
3) only want to keep the advantage for less than a year when AMD will also implement SSE-2
4) are not worried about not being able to upgrade your processor

either that or you want the CPU with the most MHz and are not worried that it doesn't actually provide any performance boost on the majority of existing applications.

As a final note. The performance gains from the Nvidia drivers are impressive but only have any real impact at resolutions which nobody uses when playing for real.

I am disappointed that you use your technical knowledge to provide half truths and some outright falsehoods while pretending to be unbiased.

L
Anonymous
a b à CPUs
May 4, 2001 2:17:01 PM

"The Athlon was AMD's answer to Intel's Pentium III. Both have similarly designed cores that will scale to approximately the same clockspeed."

You know as well as we do, that the Athlon has a far more advanced core than the P3 (which is basically a PPro on steroids). As for its scaling, we already know it will scale very soon to 1.5/1.7, and everything points in the direction of > 2 Ghz. Overclocked, these speeds have already been reached with the current process, where P3 usually can not be clocked much beyond 1 Ghz (also, with current .18 technology). You could just as well twist your statement around, and say the P4 was intels answer to the Athlon, since its P3 couldnt reach high enough speeds (cf P3 1.13). To keep follow your metaphore, intel shifted from first to third gear (P3 to P4), while AMD is currently in second, perparing the forth gear.

"Judging by the lateness of their current offerings, for example the Palomino, "

Has it ever occured to you, AMD has not really felt the need to launch palomino just yet ? I highly doubt the lateness of palomino has much to do with AMD's inability to produce them. Much more with marketing strategies. They focus on the mobile market right now, because they see a great opportunity to compete there with the mobile Palomino. Intel currently has very little to compete with a mobile palomino (P3 being slower, mobile P4 not being ready yet). To support this theory, this is the first time I ever saw a "new" cpu core apear in the mobile market BEFORE it was available in the desktop market. For years, this has been the other way around (P1, P2, P3, P4,..)

"I would say it [hammer] will probably get pushed back to sometime in 2003, but that's just pure speculation."

It is (pure speculation). And for the sake of your credibility, you might perhaps want to refrain from this sort of biased speculation. Over the last few years, AMD has done a terrific job delivering their products on schedule (often even earlier). Palomino may be an exception, but I think for good reason.

"If you buy a $180 CPU every 6 months, your CPU is costing you about $30 / month. If you buy a $400 CPU every 2 years, your CPU is costing you about $16.66 / month. "

There is one mistake in your calculation.. the remaining value. I tend to upgrade about once a year, but I always sell my "old" cpu for a decent price. If you buy a $400 cpu , it will loose its value much faster than a $180 one. We all know top megahertz cost top bucks, and prices fall much faster in the mid- to high-end than in the low-end range. A P4 1.7 costs $400 today, but may only be worth $120 next year, and $50 the year after. Thats $14 a month.
I upgrade about every 6-9 months, and rarely spend more than $150 for the cpu, and usually get about $50-100 for my old one. thats $4-8 a month

Of course this is always true comparing high end to low end (regardless of type and make), but it skews your calculation anyway. Buying a high end cpu to keep it 3 years is always more expensive than upgrading regulary your mid- to low end cpu. I know.. I once bought a P3-500 when it was top of the line.. it only lasted for two years, and I didnt get much back after these two years.
May 4, 2001 6:51:50 PM

"This is not outright false but is misleading."

It is not misleading at all. I clearly state that I do not see anything that has bene released by AMD that promises to offer this next generation of speed. Vaporware does not count.

"faster P4 is required to match a lower clocked Athlon"

I disagree. A higher clock speed does not indicate a faster processor. the clockspeed in one factor in the overall technical specifications of the processor. Performance is the product of clockspeed and average instructions per clock. If you want to find a P4 and an Athlon that should have equivolent performance according to technical specifications, go find the average instructions per clock for both processors and do the math. You are looking for equal performance numbers (average instructions per second.) I will say that yes, a higher clocked Pentium 4 is equivolent to a slightly lower clocked Athlon. Just remember that clock speed is only one factor in the performance of a CPU. When you decrease the IPC by design, you also increase the clock speed by design to compensate.

"If you buy a P4 now you will not be able to significantly upgrade the speed of the CPU because the pin-out is changing."

I disagree. It will require the purchase of another motherboard, but the upgrade can be made. Most people keep their CPU for about 3 years anyway, after which time motherboard upgrades are nearly always a necessity.

"I am disappointed that you use your technical knowledge to provide half truths and some outright falsehoods while pretending to be unbiased."

I am not the one providing the half truths and falsehoods. Am I biased? Perhaps on the side of truth.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =
May 4, 2001 7:06:15 PM

"the Athlon has a far more advanced core than the P3"

Not really, check the technical specifications. They are very comparable.

"it will scale very soon to 1.5/1.7, and everything points in the direction of > 2 Ghz"

Highly doubtful. We will have to wait and see.

"I highly doubt the lateness of palomino has much to do with AMD's inability to produce them"

Marketing at any company is usually great at coming up with ideas on why they 'decided' to introduce something later than expected. More often than not it's simply a cover for a technical glitch being worked out. Have you seen any mobile Palominos yet?

"you might perhaps want to refrain from this sort of biased speculation"

There's nothing wrong with speculation as long as I let everyone know it is indeed speculation.

"I always sell my "old" cpu for a decent price"

Doing the old trade-in is a pretty good idea, though requires a bit of effort. I usually hand out my old CPUs to family and friends as gifts. The processors are usually worth far more than the market gives. My family and friends value them more than the market.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =
May 4, 2001 11:25:13 PM

"'it will scale very soon to 1.5/1.7, and everything points in the direction of > 2 Ghz'

Highly doubtful. We will have to wait and see."

Well, it is true that the Palomino core is just a revamp of the old Athlon core (K7), like the P2 to the P3, and what not, but the P4 is a new core, completely different from the old P3, so in saying that the athlon core will scale past 2GHz is a valid statement, just not the absolute truth. But it is the case that the Palomino is just the Athlon core with more bells and whistles, and the Thouroghbred is the Palomino with a .13 micron die, and that the Barton (provided it makes it out the door in late 2002) will be the Thouroughbred with SOI technology. So, AMD is stretching out the K7 core, and according to Moore's Law, they will reach about 3.5GHz or so, since it won't be a complete 18 months, and I'm guessing that AMD will push a 2GHz Palomino out around September, provided that the 1.7GHz Palomino will be out in July (or so). Now, I'm sure there will be other revamps, and the Athlon ghost will be faint in the Thouroghbred and Barton, but it will still be a 7th gen CPU (K7), and the Hammer line will be their 8th gen (K8?).

And oh yeah, if you were to buy a 1.2GHz Athlon right now at $151, which is not the bleeding edge, but new enough to last 24-30 months, and then upgrade again in 18-24 months with the same not-quite-bleeding edge, at around the same price (which is reasonable to assume, with Wall Steet worrying about price wars b/w Intel and AMD, which will probably last for at least two years), I will have only spend $300 or so on CPUs, which, while not the best, are good enough that even in that 1.5-2 year time period, they will still be more than enough, unlike getting a bleeding edge P4 for near $375, and trying to squeeze out whatever you can towards the end of the 3 years... like I'm doing now with a P2 300MHz. So in conclusion, it is actually cheaper to buy a chip that is just a month or two back from bleeding edge and upgrade again in 18-24 months as opposed to getting a bleeding edge CPU and trying to push it towards the end. (Some numbers: 150+150=300, over a 3-4 year period compared to 375 over a 3 years period, and add another 375 (or less) for that 4th year... this works when comparing CPUs of the same type as well (1.33 TBird compared to 1.2 TBird, the 1.33 being near 50 more, and then you have to spend that 200 again to get that 4th year...))

Extra Addition: After doing some research, I concluded that the system recommended for most gaming apps <i> that would be built today, </i> (as opposed to a system that you bought a while back), is a 900MHz CPU (brand makes no difference). Now I know, that isn't that future proof, but remember that the 900MHz P3 and Athlon were relased somewhere around 18 months ago. So you can conclude that a system built today with a 900MHz CPU would last around 12-18 more months, and therefore, further proves that getting a CPU at it's intro price is wasteful, and it ends up having the same 3 year life span that every other CPU out there has...


-SammyBoy

Without Evil, there can be no Good. Therefore, without an Intel, there can be no AMD.<P ID="edit"><FONT SIZE=-1><EM>Edited by SammyBoy on 05/04/01 07:52 PM.</EM></FONT></P>
May 5, 2001 5:45:26 AM

Sam -

Suppose I'm a consultant who bills about 150 hours a month. I conclude that if I upgrade to a state-of-the-art system, I can raise my rates $5 an hour.

How much time should I spend considering the differences in monthly costs of today's cpus?

Would I be better off spending that time finding a reliable vendor who ships in 10 days instead of 6 weeks?


beans
May 5, 2001 7:26:02 PM

Not much arguement with what you say, except...

Quote:
I'm guessing that AMD will push a 2GHz Palomino out around September, provided that the 1.7GHz Palomino will be out in July (or so).

While AMD has been secretive about the Palomino's vcore, I believe they stated (in their recent stockholders mtg) that the Desktop Palomino will released (in Q3) at 1.533Ghz.

While we each have our own CPU preferences, my feelings are that over the remainder of 2001 Intel's Northwood will be the only 'enticing' CPU on the scene to re-ignite corporate PC sales.
May 6, 2001 1:54:57 AM

I have to say that the two replies to my post were rather enigmatic. Beans said something about consulting... which lost me completely, as never before had anything beyond the realm of home use been mentioned in this thread. So beans, if you care to shed some light on that, please do.

Now IntelConvert also mentions corporate use:

my feelings are that over the remainder of 2001 Intel's Northwood will be the only 'enticing' CPU on the scene to re-ignite corporate PC sales.

Now, I'm not sure why this has entered into the thread, so if someone wishes to enlighten me, please do. Now as to what I posted earlier that those two commented on, all I was saying was that Raystonn doubting that K7 core would reach 2GHz soon wasn't prudent. Now, I understand that it would be a Palomino instead of a Thunderbird, but both are still the K7 core, unlike the P4, which is Intel's first 7th gen CPU, or so I'm lead to believe.

Yes, when I say that a 1.7GHz Palomino may be out in July, I'm not jumping the gun. Remember that Q3, even though it sounds later, begins in July and ends in September, so the Palomino coming out then isn't all that far-fetched. Also, I remember the previous AMD roadmap stating that the Palomino desktop would start at 1.7GHz, which makes sense, given that Intel recently raised the stakes by putting out a 1.7GHz of their own. Also, since the current Thunderbird "C" can overclock to 1.5GHz+ with no problem, it would rather pointless to release a CPU that is no better than the current one out there (subjective statement, but since it is still the same basic K7 core, only so much tweaking can be done and you have to raise the clockspeed to get performance increases.). So, I still stand by the statement that the Palomino 1.7GHz will come out in mid to late July, maybe early August, and that a 2GHz will probably be out around the end of Q3. Now, on May 15th, AMD is supposedly making an announcement. TheRegister believes that it will be announcing the debut of the Palomino for moblies and servers. It will then be around mid to late June that the CPU's will be shipped. Also, since there will be the release of the server version, there is speculation that the 760MP will be announced as well, since most think that putting out a CPU aimed at the server market without a dual-CPU solution is suicide.
Go to <A HREF="http:// http://www.theregister.co.uk/content/3/18750.html " target="_new">http:// http://www.theregister.co.uk/content/3/18750.html </A> for their take on the situation.

So, all in all, I'd just like to say that Raystonn, unless AMD is lying through it's teeth and the Palomino is a fake, there will be a living 1.7GHz, maybe even 2GHz K7 core out to compete with the P4. So yes, the K7 core still lives, just in the guise of the Palomino instead of the Thunderbird. And IntelConvert, I'm not sure what else you were trying to say, but I believe that both Intel and AMD will have enticing offerings come this fall. And since Intel is having problems with a third-party machine, the Northwood has probably been delayed at least a month. So, Intel, due to a situation out of it's control, is not able to take the advantage of the P4 core for a while yet, and thereby giving AMD yet another chance to snap at their heels.

-SammyBoy

Without Evil, there can be no Good. Therefore, without an Intel, there can be no AMD.
May 6, 2001 5:35:21 AM

Sam -

I wasn't trying to be enigmatic or obscure here -- just trying to look at cpu prices from a different perspective. You, Raystonn, and bbaeyens all analyzed the monthly cost of a cpu; I was trying to show by example that the cost of the cpu is only one piece of a large puzzle, and at today's prices it can be trivial in some circumstances.

Since you said in your first post, "I use gaming as an example, merely because it's the only thing I can think of..." I didn't realize you considered the context as <i>limited</i> to home use. I'm sorry for the misperception.


beans
May 6, 2001 8:07:03 AM

No problem. Sorry if I misunderstood you. It was just wierd that both you and IntelConvert took the same angle almost at the same time. It made me wonder if I had missed something somewhere. The different POV is appreciated.

-SammyBoy

Without Evil, there can be no Good. Therefore, without an Intel, there can be no AMD.
May 6, 2001 11:21:25 AM

Actually you are wrong about 3dnow and MMX, AMD adopted MMX with no problem, and they brought out 3dnow to help the k6 with its weak fpu unit. It wasn't an answer to MMX, because all later k6/k6-2/k6-3/Athlons have the MMX line coding in it. The K6-2s/K6-3s have the MMX coding and 3dnow.

Intel brought out SSE as an answer to 3dnow. Athlons now have enhanced 3dnow which have more instruction lines than SSE but they don't have as many as SSE2.

From studying all the coding, the instructions set were getting more and more similar, so one adopting the other was going to happen sooner or later. Honestly with Intel being the bigger company I think it was a smart move on AMD's to adopt it for their hammer line.

I do have to agree that the instruction sets could have been so much more; awhile ago. Its still smarter to code in the new DX8 vertex language than using these instruction sets.

Also with Athlon's powerful fpu units, it doesn't see as much boost as the k6-2/k6-3 and the P4 does. I still think its a design flaw in the P4 to rely that much on opimization, but time will only tell.....

MeldarthX
!