Processor Architecture

PLOF

Junior Member
Feb 17, 2003
4
0
0
I'm sure all of you have seen the Athlon XP 3000+ review, and how its basically equal to the 3.06 Intel, but I found a review I found where it compared a Intel 2.4GHz with an AMD 2600+ at 2.4GHz. The AMD won hands down, no match. At the same clock rate, AMD wins. But if AMD has the less efficient architecture, should Intel really be worried about Hammer? I say yes, becuase Hammer will own all of Intel for a couple quarters. It will be amazing what will come from AMD, but I would like to see a comparison between a 3GHZ HT and a 3GHz Hammer with a 64bit version of Windows(Longhorn), I hope some people will know what i'm talking about, Anand-you certainly do, heh-you da man btw.

peace

Quadract.com
 

ant80

Senior member
Dec 4, 2001
411
0
0
I honestly dont think the clawhammer is going to be a success as soon as it comes out. Can't comment about the sledgehammer though. Would b interesting to see. I think amd has to eventually incorporate ht into its processors. But that would b in 2004-5.

AMD has not made any major advancements to its architecture since the athlon 500mhz. (I am sure people exist with other opinion, so dont flame me for saying that). Athlon has scaled to nearly 300% its original clock speed. Tomshardware explains that it would take the p4 to reach 5.4ghz to reach that level of scaling.

At any level of technology, there exists a particular number of pipelines that gives you optimal clock/performance. Of course, as technology develops, there is an increase in this treshold level. P4 has (if i remember right,) 24 pipelines, which is way more than the optimal at this level. So, it is going to be highly scalable. But wait till athlon catches up.

Just u wait, la intella just you wait.
 

pm

Elite Member Mobile Devices
Jan 25, 2000
7,419
22
81
At any level of technology, there exists a particular number of pipelines that gives you optimal clock/performance. Of course, as technology develops, there is an increase in this treshold level. P4 has (if i remember right,) 24 pipelines, which is way more than the optimal at this level.
How do you figure it's too much to be optimal? On what basis?

For a given instruction set and instruction set mix for common apps, the ideal number of stages in a given processor's pipeline is a complex calculation that involves quite a number of variables with BTB accuracy and cache hit rates pretty much playing the highest significance in the calculation. I would be very curious if you have any factual/measurable basis for guessing the number of pipeline stages that would be optimal in a processor right now... or if you are just talking.
 

AndyHui

Administrator Emeritus<br>Elite Member<br>AT FAQ M
Oct 9, 1999
13,141
16
81
A 64-bit Windows will not run on a Pentium 4, so it won't be an apples-to-apples comparison.

Don't make the mistake of assuming that Intel is in a state of rest with their processors. They are working on something new all the time, just like AMD.
 

Mday

Lifer
Oct 14, 1999
18,647
1
81
Originally posted by: AndyHui
A 64-bit Windows will not run on a Pentium 4, so it won't be an apples-to-apples comparison.

Don't make the mistake of assuming that Intel is in a state of rest with their processors. They are working on something new all the time, just like AMD.

32bit windows should work on hammer though.
 

AbsolutDealage

Platinum Member
Dec 20, 2002
2,675
0
0
32bit windows should work on hammer though.

Yes, but you will see absolutely Zero performance gain out of your extra 32 bits, so essentially you would have payed entirely too much money for something you would not be using.

I honestly dont think the clawhammer is going to be a success as soon as it comes out. Can't comment about the sledgehammer though. Would b interesting to see. I think amd has to eventually incorporate ht into its processors. But that would b in 2004-5.

I agree. The claw hammer will draw the high performance market at its debut, but it is doubtful that software will be able to catch up to the hardware in time for its release. Following the months after hammer's release, you will start to see optimized software, and that is when hammer will show its power.

As far as the HT goes, we may see AMD steer clear of it for a while. As you can see, the hammer architecture is drastically different then Intel's offerings. AMD has decided to go with integrated north bridge/memory control along with the 64-bit architecture. Essentially, they have tried to clear some of the biggest bottlenecks in all of the pathways. Intel, on the other hand, has gone a completely different path, deciding to implement HT. If AMD decides to invest in a multi-threading solution, you may see it a few years down the road, but it will probably come in the form of full-blown vector registers as opposed to the weaker HT.

Anyway, the next couple of years will probably spell the end of the "Clock Wars" that we have seen for so long. At a certain point, I think that the consumer will be forced to look at other performance metrics other that that "magic number". At a certain point (5GHz?, 10GHz?, who knows) clock speed will be nearly unimportant, software will no longer be able to keep up with the hardware.

Just my 2 cents.
 

PLOF

Junior Member
Feb 17, 2003
4
0
0
But the 64bit version of the processor AMD is releasing will basically be like having HT, if the OS is setup correctly. I wonder if the processor will be bigger, like Barton is over Thoroughbred. I wouldn't mind if AMD sent me a Hammer, and Intel sent an equally clocked processor with a motherboard(since I don't have one). just some thoughts.

Quadract.com
 

ant80

Senior member
Dec 4, 2001
411
0
0
I would be very curious if you have any factual/measurable basis for guessing the number of pipeline stages that would be optimal in a processor right now... or if you are just talking.

Ok, I read in some article that this was so. I dont remember where, or when. Perhaps at the introduction of p4 sometime ago. Then again, I cannot prove it. Sorry.

The article said, the pipelines were increasing with the passage of time, and so was the technology. By correlation, technology was driving the increase in pipeline stages. However, I cannot say anything about the cause and effect here.

P4 is not, by a long shot, far superior technology than athlon xp. The only reason why intel increased the pipelines was to win the clock war. That way, it is going to be a lot worse for amd to compete with the intel marketing machine. The only difference between P4 and P3 was SSE2. Correct me if I am wrong, but this is too insignificant an addition to the instruction set, to warrant such a massive increase in the pipelining. This was what I wanted to point out.
 

AndyHui

Administrator Emeritus<br>Elite Member<br>AT FAQ M
Oct 9, 1999
13,141
16
81
The only difference between P4 and P3 was SSE2
Hyperthreading, Execution Trace Cache, better branch prediction unit, huge bandwidth...

Don't be so quick to bash a long pipeline. If you have too short a pipeline, your processor just isn't going to scale. An architecture can only handle so much for a given pipeline length. The P6 architecture really had it at 1.4GHz, after starting off 150MHz with the original Pentium Pro. If you don't do something, where are you going to go next when you are limited by pipeline? The Pentium 4 has plenty of headroom left, and arguably the Athlon doesn't nearly have enough headroom left.
 

borealiss

Senior member
Jun 23, 2000
913
0
0
"But if AMD has the less efficient architecture, should Intel really be worried about Hammer? I say yes, becuase Hammer will own all of Intel for a couple quarters."

we don't know what the performance numbers of prescott are yet. this is a bit presumptious.

"It will be amazing what will come from AMD, but I would like to see a comparison between a 3GHZ HT and a 3GHz Hammer with a 64bit version of Windows(Longhorn), "

this would be a pretty bad comparison imo. a 3 ghz hammer would win in most cases. this wouldn't be comparing amd's and intel's latest and greatest. by the time amd hits 3 ghz, a 4 ghz p4 might be out, with more cache.

"P4 is not, by a long shot, far superior technology than athlon xp. The only reason why intel increased the pipelines was to win the clock war. That way, it is going to be a lot worse for amd to compete with the intel marketing machine. "

this could not be further from the truth from an engineering standpoint. you break up your execution stages into smaller ones so that you minimize your critical path, ie the operations that happen within 1 pipeline stage take about the same time to execute. this is much better from an efficiency standpoint. in theory, these execution gaps that are introduced with higher IPC can be averted by breaking them up into smaller jobs. hopefully the clock speed increase that this approach entails will beset the branch-misprediction penalties.

"But the 64bit version of the processor AMD is releasing will basically be like having HT, if the OS is setup correctly"

nooooo. nothing the hammer has will be equivalent to HT on the p4.

 

chsh1ca

Golden Member
Feb 17, 2003
1,179
0
0
I think the fundamental difference between Intel and AMD's thinking is that Intel is saying "let's make things more expensive, but give 1 processor and 1 pseudo-processor on each die", while AMD is saying "let's introduce standards and expanded memory addressing support that will allow the processor to scale well, and produce it cheap enough that everyone has affordable MP systems".

They're taking the same approach: Working on executing multiple threads at the same time. They're just going about it in different ways.
IMHO, AMD has addressed three major issues in the server market with Opteron:
1. 64 bit memory addressing on the cheaper to manufacture x86 architecture.
2. System bandwidth restrictions (Hypertransport).
3. Scalability.

These are three things that the Athlon MP and the P4 (by design) don't deal with well. Keep also in mind that while it may take a while for DESKTOP applications to go 64 bit, a lot of SERVER applications are much closer to realising that goal, thanks to Intel's Itanium. This is, as I see it, principally why AMD has opted to delay the desktop Hammer (Athlon 64) so long, but is still pushing to reach their release target for servers.

Keep also in mind that this will be the inexpensive server chip that people are looking for when they want to turn their box into a single processor server. AMD has a history of outperforming much higher clocked P4 Xeons, principally because server operations are by design Floating Point operations, IOW: mostly math. The Athlon's superior FPU is what gives it this rather impressive advantage (see the benchmarks Anand came out with between even dual Athlon 1.4GHz vs P4 2.0GHz), and sticking 64 bit addressing on top of that means that we could see the Opteron competing with the Itanium, rather than with the Xeon.

If we do see that, with some kind of OEM support, would be a massive blow to Intel.

Then again, we'll just all have to salivate over the idea until we see some benchmarks.

 

Wingznut

Elite Member
Dec 28, 1999
16,968
2
0
Wow... there are a lot of things said in this thread that are pretty far off base. I can't address them all, but I do request that you guys keep an open mind about Highly Technical architecture. I've been working at Intel for over two years now... Prior to my employment there, I believed all sorts of things that I learned on message boards that turned out to be very untrue.
Originally posted by: ant80
P4 is not, by a long shot, far superior technology than athlon xp.
I'm not going to sit here and debate which architecture is "superior". But there's no way that one architecture is "by a long shot" better than the other. Considering how many problems AMD has had in delivering their products this year, it can easily be argued that Intel is doing something very right.
Originally posted by: ant80
The only reason why intel increased the pipelines was to win the clock war. That way, it is going to be a lot worse for amd to compete with the intel marketing machine.
This is exactly what I was referring to in my first paragraph. And this is also very untrue. Intel went this direction with the goal to manufacture the highest performing desktop cpu in the world. AND be able to DELIVER it to the market. Looking over the past year, it's pretty hard to argue the success of that strategy.

Also keep in mind that it takes several years to design and bring a cpu to the market. The P4 design started long before there was any "clock war".
Originally posted by: ant80
The only difference between P4 and P3 was SSE2. Correct me if I am wrong, but this is too insignificant an addition to the instruction set, to warrant such a massive increase in the pipelining. This was what I wanted to point out.
As stated above, this is also inaccurate. There are A LOT of differences between the P4 and P3.
Originally posted by: chsh1ca
I think the fundamental difference between Intel and AMD's thinking is that Intel is saying "let's make things more expensive, but give 1 processor and 1 pseudo-processor on each die", while AMD is saying "let's introduce standards and expanded memory addressing support that will allow the processor to scale well, and produce it cheap enough that everyone has affordable MP systems".
What makes you think that Intel "makes things more expensive"? I'm not so sure that the cost per die of a Hammer is any cheaper than a P4. Especially considering that the dies are about the same size and Intel is using 300mm wafer in a couple of fabs now.

Not that die cost really matters much to the market price.... It's basic economics. The corporation (no matter what product.... cpu's, toothpaste, tires, etc.) will charge the most that they can and still be able to sell the target quantity. Do you think AMD enjoys charging below market value for their cpu's, and taking a loss for the past 7 quarters or so? Trust me, if they could, they'd rather be selling their cpu's for as much as Intel charges. And if you look at their latest releases, they are targeting the same pricepoint as Intel.
Originally posted by: chsh1ca
IMHO, AMD has addressed three major issues in the server market with Opteron:
1. 64 bit memory addressing on the cheaper to manufacture x86 architecture.
2. System bandwidth restrictions (Hypertransport).
3. Scalability.
1. Unless you are talking about die size (which x86 doesn't actually affect), I'm still unsure as to why you would think that the x86 architecture would be cheaper to manufacture.
2. HyperTransport is a really nice technology. However, there are other bandwidth restrictions (memory speed, etc.) that will limit the effectiveness of HyperTransport.
3. I guess the jury is still out on that one. Hammer won't initially be released at "at least 2ghz", as was promised. The first Opterons will be 1.4ghz, 1.6ghz, and 1.8ghz.

 

imgod2u

Senior member
Sep 16, 2000
993
0
0
I think he means scalability as far as multiprocessing scalability goes, and for that, Hypertransport and dedicated memory pockets arguably releave almost any memory/IO bottlenecks that exist in current multiprocessing systems.
 

Wingznut

Elite Member
Dec 28, 1999
16,968
2
0
Originally posted by: imgod2u
I think he means scalability as far as multiprocessing scalability goes, and for that, Hypertransport and dedicated memory pockets arguably releave almost any memory/IO bottlenecks that exist in current multiprocessing systems.
Doh! I misunderstood.

In that case, then I definitely agree. HyperTransport is a very nice solution for multiprocessing capabilities.
 

chsh1ca

Golden Member
Feb 17, 2003
1,179
0
0
Originally posted by: Wingznut
Wow... there are a lot of things said in this thread that are pretty far off base. I can't address them all, but I do request that you guys keep an open mind about Highly Technical architecture. I've been working at Intel for over two years now... Prior to my employment there, I believed all sorts of things that I learned on message boards that turned out to be very untrue.
[...]
What makes you think that Intel "makes things more expensive"? I'm not so sure that the cost per die of a Hammer is any cheaper than a P4. Especially considering that the dies are about the same size and Intel is using 300mm wafer in a couple of fabs now.
No offense, but you're rather biased. It appears that you're arguing that Intel is taking a larger profit margin than AMD, and that Intel processors are actually cheaper to manufacture. This would not surprise me, because to be frank, any time I've looked at buying any PC hardware (processors, fax boards, network cards, etc), I've found that the cost to print "Intel" on the boards is rather exorbinant.

Not that die cost really matters much to the market price.... It's basic economics. The corporation (no matter what product.... cpu's, toothpaste, tires, etc.) will charge the most that they can and still be able to sell the target quantity. Do you think AMD enjoys charging below market value for their cpu's, and taking a loss for the past 7 quarters or so? Trust me, if they could, they'd rather be selling their cpu's for as much as Intel charges. And if you look at their latest releases, they are targeting the same pricepoint as Intel.
True, however from what I've read, the price point they're targeting is the same as the P4 Processor here, for something competing with the Xeon. You tell me if double the cost will be worth it. Wait, you work for Intel, don't. I can't imagine I'd get anything other than a sales pitch from someone who is employed by either AMD or Intel when asked that question.

Originally posted by: chsh1ca
IMHO, AMD has addressed three major issues in the server market with Opteron:
1. 64 bit memory addressing on the cheaper to manufacture x86 architecture.
2. System bandwidth restrictions (Hypertransport).
3. Scalability.
1. Unless you are talking about die size (which x86 doesn't actually affect), I'm still unsure as to why you would think that the x86 architecture would be cheaper to manufacture.
Perhaps manufacture was a superfluous word. I was, in my mind, comparing it to the relatively pricey RISC hardware that's out there.

2. HyperTransport is a really nice technology. However, there are other bandwidth restrictions (memory speed, etc.) that will limit the effectiveness of HyperTransport.
Possibly, but then, wasn't that true of Rambus as well? Does that lessen the effectiveness of the technology? No, if anything Rambus showed that speedier busses on processors were a good thing, as long as its inexpensive.

3. I guess the jury is still out on that one. Hammer won't initially be released at "at least 2ghz", as was promised. The first Opterons will be 1.4ghz, 1.6ghz, and 1.8ghz.
Scalability was listed in the sense of more than simply clock speed.
 

Wingznut

Elite Member
Dec 28, 1999
16,968
2
0
chsh1ca, I realize that you don't know me, or have seen many of my posts here. But, while I am employed by Intel, it is in the capacity of making cpu's, NOT selling them. Feel free to do a search for my name, and you'll see that never once have I given "a sales pitch" or even suggested someone purchase an Intel product over a competitor's.

I'm not going to sit here and tell you that I have absolutely no bias. But I am very confident in saying that there are MANY others here that are much more biased than I.

So, please don't discard what I have to say just because of my employment. Your first quote of me in the above post is a perfect example, when I was talking about cost to manufacture vs retail price. I said nothing remotely pro-Intel, yet you felt the need to point out that I am biased.

At any rate, welcome to the forums. And please take a look at what I have to say before you judge what I have said.



By the way, I didn't say (or mean to suggest) that Intel cpu's are cheaper to manufacture. I certainly don't know AMD's projected cost per wafer for Hammer, since cost per wafer is probably the single most guarded number in this industry. I'm just saying that while the cost to manufacture a cpu is a direct reflection of net profit, it has little effect to the price on the shelf.
 

AndyHui

Administrator Emeritus<br>Elite Member<br>AT FAQ M
Oct 9, 1999
13,141
16
81
any time I've looked at buying any PC hardware (processors, fax boards, network cards, etc), I've found that the cost to print "Intel" on the boards is rather exorbinant.
Every time I see Intel printed on the board, I know I am purchasing a top quality component that Intel has spent a lot of money on in research, development, manufacture and testing.

There's a lot more to cost than just manufacture.
 

ant80

Senior member
Dec 4, 2001
411
0
0
Intel went this direction with the goal to manufacture the highest performing desktop cpu in the world. AND be able to DELIVER it to the market. Looking over the past year, it's pretty hard to argue the success of that strategy.
Somehow, I doubt that. If you dig out the benchmarks when the P4 1.4 was released, there were several applications where the P3 actually beat the P4. Running at lower clock speeds. How is it possible the the famed intel engineers didn't notice such a performance hit!!!


The only difference between P4 and P3 was SSE2.
That was in response to the following.
For a given instruction set and instruction set mix for common apps, the ideal number of stages in a given processor's pipeline is a complex calculation that involves quite a number of variables with BTB accuracy and cache hit rates pretty much playing the highest significance in the calculation. I would be very curious if you have any factual/measurable basis for guessing the number of pipeline stages that would be optimal in a processor right now... or if you are just talking.


Every processor architecture has its highlights.
Don't be so quick to bash a long pipeline. If you have too short a pipeline, your processor just isn't going to scale. An architecture can only handle so much for a given pipeline length. The P6 architecture really had it at 1.4GHz, after starting off 150MHz with the original Pentium Pro. If you don't do something, where are you going to go next when you are limited by pipeline? The Pentium 4 has plenty of headroom left, and arguably the Athlon doesn't nearly have enough headroom left.
The original p4 didn't have ht, branch prediction unit arises because of the pipeline. Bandwidth was because of rambus. Trace execution cache is a new feature. I maintain that the major difference between p3 and p4 was the SSE2. Especially with respect to the ISA. Yes. Athlon doesn't have enough headroom left. True. However, athlon is a previous generation architecture. P4 is intel's latest and greatest. As I said, technology drives the increase in the pipeline stages. However, such a drastic increase in the pipeline is unwarranted. Probably, the only reason why intel did it was to win the clock race. After they lost face when athlon beat them to the all-important 1000MHz barrier. I guess we will never know.
 

Wingznut

Elite Member
Dec 28, 1999
16,968
2
0
Originally posted by: ant80
Intel went this direction with the goal to manufacture the highest performing desktop cpu in the world. AND be able to DELIVER it to the market. Looking over the past year, it's pretty hard to argue the success of that strategy.
Somehow, I doubt that. If you dig out the benchmarks when the P4 1.4 was released, there were several applications where the P3 actually beat the P4. Running at lower clock speeds. How is it possible the the famed intel engineers didn't notice such a performance hit!!!
Well sure... But "the famed intel engineers" took a look at the big picture... At the future. Not just one or two speed grades. Like I said, look at 2002. Intel consistantly had the best performing desktop CPU, and delivered every release on time with enough quantity.
Originally posted by: ant80
However, such a drastic increase in the pipeline is unwarranted. Probably, the only reason why intel did it was to win the clock race. After they lost face when athlon beat them to the all-important 1000MHz barrier. I guess we will never know.
So, you are saying that Intel designed and manufactured the P4 in quantity "AFTER they lost face when athlon beat them to the all-important 1000mhz barrier."??? Sorry, but not even "the famed intel engineers" can design and perfect the process in the six months between the release of the 1ghz Athlon and the Willamette P4. Like I said, the P4 design started years before that.

 

Sohcan

Platinum Member
Oct 10, 1999
2,127
0
0
Probably, the only reason why intel did it was to win the clock race. After they lost face when athlon beat them to the all-important 1000MHz barrier. I guess we will never know.
That's simply not true...I don't think you appreciate the amount of time it takes to bring a microprocessor from architecture specification to circuit design to the market. I know of a few grad students here that were involved in the P4 microarchitecture spec back in early '96, and the design certainly started in '95...at the time (in '96, certainly when the design goals of the microarchitecture were solidified), IIRC the Penitum MMX was at 200 MHz (maybe 166 MHz?) and AMD had the K5 at 90Mhz. In addition, the latest and greatest Pentium clearly had a large performance lead over every other x86 microprocessor. The K6 and the Athlon simply had no influence on the design of the Pentium 4 microarchitecture. An implication of this is just speculation.

The reasoning behind pushing clock rate in the Pentium 4 was based on a focus on wire delays. In a simplistic view, when a circuit is shrunk with an improvement in lithography technology, transistor delays decrease with respect to the size of the circuit, but wire delays do not. Because of this aspect, wire delays increasingly dominate the delay in a circuit. To combat this tide (which the designers felt would hurt performance), a number of aspects designed to reduce wire delays were incorporated. A long pipeline allows for a better circuit layout and distribution of wire delays; the P4 pipeline allows for two stages that are devoted to propogating signals. The pipelined ALUs, smaller, low-latency L1D cache, and higher-latency complex operations relax their influence they have on the critical path in the microprocessor. There may be arguments against the design decisions that influenced the P4 core, but these are the technical reasons that influenced the design. I'm sure that the marketing department was not upset about the end result of the P4 design, but it simply did not have an overriding influence on the engineers.

Many people seem to think that Intel first explored this high clock rate/lower ILP path with the P4. This is also simply not true. There have been other points in microprocessor design where new paradigms lead to high-clock rate designs. In the early 90's, microprocessors were beginning to explore dynamic (out-of-order) instruction scheduling and high-issue rates. But dynamic scheduling with high-issue rate has a nearly N^2 complexity relationship with the number of instructions N that are "looked at" each cycle. Early dynamic scheduled designs (MIPS R10000, IBM RS6000) had high IPC, but with the relative inexperience with the design paradigm, clock rate scalability was quite bad. The Alpha EV5 team, which stressed performance above all other factors (power, cost), decided on a simple, high-clock rate design. Where other microprocessors had out-of-order scheduling with 3- or 4-way instruction issue, the EV5 had static (in-order) scheduling with 2-way (integer) instruction issue. Through their architecture decisions as well as innovative circuit design, the EV5 was released at around 500 MHz; the second fastest microprocessor at its release was (IIRC) the 200 MHz HP PA 8000. Despite being clocked 2.5X faster, the EV5 was around 10-20% faster in SPECint, though no one argued otherwise that the EV5 was clearly (and continued to be) the highest-performing microprocessor.

Take another change, IBM's switch to a more automated design (rather than the higher-performing, yet more expensive full custom design) with the POWER4. IBM's reliance on automated design means that when all else is equal, their design will likely not be clocked as high as a full-custom microprocessor design. To combat some of this deficit, the POWER4 has a very long pipeline for a RISC/server class MPU; 14 stages vs. 7 stages on the Alpha EV6/EV7 and 8 stages on the Itanium 2. In addition, despite its higher instruction retire rate compared to many RISC server-class microprocessors, the POWER4's grouping restrictions leads to lower respective performance. In the end, the POWER4 reaches 1.3 GHz at 180nm, compared to 1.15 GHz (EV7) and 1 GHz (Itanium 2) on the same feature size. The POWER4 has a lower ILP, but one cannot deny that IBM's design decisions have lead to one of the highest-performance server microprocessors, as well as excellent compactness of its core that has led to its dual-core designs.

Somehow, I doubt that. If you dig out the benchmarks when the P4 1.4 was released, there were several applications where the P3 actually beat the P4. Running at lower clock speeds. How is it possible the the famed intel engineers didn't notice such a performance hit!!!
Make it a fair comparison. The P3 hit a wall at 1GHz at 180nm, whereas the P4 stopped at 2GHz on the same 180nm process technology. Searching through some SPEC results, the Pentium 3 1 GHz yields 462 in SPECint and 340 in SPECfp (base). The 2GHz P4 (Willamette) achieves 684 in SPECint and 745 in SPECfp, an increase in performance of 48% (SPECint) and 219% (SPECfp) on the same process technology. I don't see that as a mistake by the P4 engineers.

* not speaking for Intel Corp. *
 

chsh1ca

Golden Member
Feb 17, 2003
1,179
0
0
Originally posted by: Wingznut
chsh1ca, I realize that you don't know me, or have seen many of my posts here. But, while I am employed by Intel, it is in the capacity of making cpu's, NOT selling them. Feel free to do a search for my name, and you'll see that never once have I given "a sales pitch" or even suggested someone purchase an Intel product over a competitor's.

Perhaps that was a rather presumptuous statement, but given the plethora of people who work for hardware manufacturers I've spoken to, they all take pride in their work (especially the engineers) and they are, in my experience, infinitely more likely to tow the company line when discussing things their company is directly involved with.

I'm not going to sit here and tell you that I have absolutely no bias. But I am very confident in saying that there are MANY others here that are much more biased than I.

Certainly, I didn't say that thinking you were the most biased person I have run into, merely that it was something to keep in mind.

So, please don't discard what I have to say just because of my employment. Your first quote of me in the above post is a perfect example, when I was talking about cost to manufacture vs retail price. I said nothing remotely pro-Intel, yet you felt the need to point out that I am biased.

Oh, by no means did I discard what you said, and I don't want you to think that my "let's make it more expensive" was necessarily a bad thing, I was just saying that a single processor that does 1.5-1.8 processor's worth of work seems to be the way Intel is headed, while AMD looks to be taking the "make SMP cheaper on the whole" approach. Don't get me wrong, I personally would love to see AMD adopt HyperThreading, because the Multithreaded app performance is killer on the P4s with HT enabled.

The only bump in the road I see for both companies is that a significant percentage of apps are written for single-processors, meaning the multithreading capabilities both companies are pushing are merely adding cost to the processor in 90% of cases. Now, if I have to pay more now for my processors to be able to get apps that are written to take advantage of that down the road, then I'm all for it. I guess the real question then becomes how long it will be until a significant portion of software companies start writing multithreaded apps.


At any rate, welcome to the forums. And please take a look at what I have to say before you judge what I have said.

Thank you. I didn't want to impress upon anyone I was discarding what you said, merely that I was taking it with one huge grain of salt, just as I would if reading an AMD employee's response.

Originally posted by: AndyHui
Every time I see Intel printed on the board, I know I am purchasing a top quality component that Intel has spent a lot of money on in research, development, manufacture and testing.

There's a lot more to cost than just manufacture.

Sure, however the quality component can be obtained cheaper via other companies. Take 3Com, rather notorious for marking up their Network cards. However, a 10/100Mb WOL - capable 3Com NIC, while being $10-15 more than a cheaper NetGear/D-Link, is in turn $10-15 cheaper than a comparable Intel card. I have never had any issues with neither 3Com nor Intel's support or card quality, so where is the $10-15 price increase justified?
 

borealiss

Senior member
Jun 23, 2000
913
0
0
"Somehow, I doubt that. If you dig out the benchmarks when the P4 1.4 was released, there were several applications where the P3 actually beat the P4. Running at lower clock speeds. How is it possible the the famed intel engineers didn't notice such a performance hit!!! "

"However, such a drastic increase in the pipeline is unwarranted. Probably, the only reason why intel did it was to win the clock race. After they lost face when athlon beat them to the all-important 1000MHz barrier. I guess we will never know. "

ant80.. please try to be a bit more objective. if you look way back when, 486's running at 133 initially beat the newly introduced pentiums. this is not to say that the pentium was an inferior architecture.

increases in pipeline lengths allow more instructions in flight. while the effectiveness of this may be limited by whether the instructions can be executed out-of-order, or how much out-of-order they can be executed and shoved down the pipe, increasing a pipeline generally requires MORE engineering ingenuity than does a shorter pipeline. i can only imagine what type of scheduler is present in something that has 20+ stages, forwarding mechanisms, etc... couple this with the fact that the p4 can juggle 2 threads at once with something like 5% increase in die space and i think you get the picture.
 

ant80

Senior member
Dec 4, 2001
411
0
0
There may be arguments against the design decisions that influenced the P4 core, but these are the technical reasons that influenced the design. I'm sure that the marketing department was not upset about the end result of the P4 design, but it simply did not have an overriding influence on the engineers.

I guess I was confusing that with other matters. Probably was a bit out of line there. Didn't mean it.

But though I might have been wrong about the reason, I still think there is something suspicious about how long they made it. Going from the low 10's to the mid twenties, in effect, completely doubling it, is not something that I know how to explain.
 
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |