Pentium 4 Discussion - Round 2

Quote:

---

Originally posted by Arcadian:
Well, let me address you first question first. The Itanium will not use SSE or SSE-2, and the reason is that it won't have to. The Itanium actually has a more advanced SIMD operation native in the hardware.

SSE gives the Pentium III and Pentium 4 128-bit SSE registers. By filling these registers with four 32-bit single precision numbers, SSE instructions can execute the contents simultaneously. SSE-2 allows 2 double precision 64-bit numbers to be used instead, which can accelerate apps that need the extra precision (scientific apps, for example).

I believe the Itanium has 128 80-bit floating point registers. The extra bits allow for extended precision calculations, something that even SSE-2 doesn't allow. The Itanium is able to simultaneously execute the contents of 4 of these extended precision floating point registers, for a total of 4 extended or double precision floating point calculations, or 8 single precision floating point calculations EVERY CLOCK! So even though the Itanium doesn't clock higher than 800MHz, it packs a whallop in every one of those hertz. Consider it the antithesis of the Pentium 4 https://www.sharkyforums.com/images/.../2005/06/5.gif.

Regarding your question on ramping up megahertz, I definately think it's possible. However, the last thing Intel wants is another 1.13GHz Pentium III fiasco. In other words, Intel will not release a new Pentium 4 unless it has plenty of headroom. This should be good news to overclockers, as they will probably be able to squeeze a couple hundred megahertz out of every Pentium 4 they buy. This means that if Intel decides to release 2.0GHz in Q3 2001, they should be able to actually produce them up to 2.2GHz. And this is all on a .18u process!

I believe scalability to be much higher on .13u. To address a concern made by Phoenix, I believe that Intel has already thought ahead, and is already prepared to deal with the problems associated with > 2.0GHz clock speeds. The EMI grounding clamp and new 12 volt power supplies are just a few of the obvious precautions already being implemented. You didn't think that 1.5GHz actually needed these things, did you? https://www.sharkyforums.com/images/.../2005/06/5.gif

There are actually some other goodies that I know about, but am not at liberty to divulge. There are also some VERY interesting features that will come on a soon to be released update to the Pentium 4 architecture that I have recently been made aware of, but yet again, I am not at liberty to divulge. I will say one thing, though. Do not count Intel out of the server market because they don't have a dual processor solution just yet https://www.sharkyforums.com/images/.../2005/06/5.gif https://www.sharkyforums.com/images/.../2005/06/6.gif https://www.sharkyforums.com/images/.../2005/06/5.gif.

---

Hi Arcadian! - So the Itaniam will not use sse/sse-2? - then why will the 3d-studio programers (and lightwave etc...) rewrite from x87 to sse-2 if only the p-4 will use it?. Also x87 uses 80-bit and sse-2 uses 64-bit. You state a logical outline in post one, but you assume sse-2 will be adopted (post "c") and become mainstream!. MMX is not mainsteam and its had 5-years!. I still think slow x87 is suicide - and the de-facto standard (whether you like-it of not). WRT to memory bandwidth - the k-7 will have 400 mhz soon and the p-4 will really start to look bad. intel needs to add performance to the x87 fpu (if their smart they will for the northwood), - I know your an Intel guy, and no disrespect intended, but Intel does'nt have the power they used to (now 80-percent market), and I don't think they have the influence to "force" sse-2 adoption (in the majority of software - Quake-4 asside).

Quote:

---

Originally posted by gaffo:
Hi Arcadian! - So the Itaniam will not use sse/sse-2? - then why will the 3d-studio programers (and lightwave etc...) rewrite from x87 to sse-2 if only the p-4 will use it?. Also x87 uses 80-bit and sse-2 uses 64-bit. You state a logical outline in post one, but you assume sse-2 will be adopted (post "c") and become mainstream!.

---

The Itanium is meant for a different market than the Pentium 4. It will have its own optimizations meant for a completely different architecture. It is unlikely that very much about the Pentium 4 and Itanium will be similar.

Quote:

---

Originally posted by gaffo:
MMX is not mainsteam and its had 5-years!. I still think slow x87 is suicide - and the de-facto standard (whether you like-it of not).

---

I have to disagree with that. I believe MMX has become quite ubiquitous in software. In fact, you probably use it in most of your programs and don't even know it. Even AMD processors include MMX extensions, so I have a hard time believing that programs don't use a feature that is so easy to develop for.

Quote:

---

Originally posted by gaffo:
WRT to memory bandwidth - the k-7 will have 400 mhz soon and the p-4 will really start to look bad. intel needs to add performance to the x87 fpu (if their smart they will for the northwood)

---

Will they? That seems unlikely to me, given that AMD is not currently able to do a number of things they have already promised by this time. If you don't know what I'm talking about, then AMD marketing seems to have worked on you. Where are DP Athlon servers? They were supposed to be here by now, yet have been delayed to Q2 2001. Where is Palomino? That should have been here last month, yet that has been delayed until Q1 2001. AMD makes a lot of promises, but some of them are harder to deliver than others. In light of AMD's recent delays, and the fact that 400MHz front side busses are nowhere at all on their roadmap, and the fact that 266MHz Athlons are still not available, I would have to conclude that you won't see 400MHz front side buses for a long time, and probably not through next year.

Quote:

---

Originally posted by gaffo:
I know your an Intel guy, and no disrespect intended, but Intel does'nt have the power they used to (now 80-percent market), and I don't think they have the influence to "force" sse-2 adoption (in the majority of software - Quake-4 asside).

---

Gaffo, I see your general point, and it is well taken. The best Intel can hope for is to lead the horse to water, but even they can't make him drink. In other words, Intel is able to show that optimizations are possible. They can also show that optimizations yield large performance increases, and they can even make optimized libraries available. However, there are even programmers out there afraid to hit the optimize switch on their compilers, so how are they supposed to be able to make SSE-2 popular?

If this is the case, then I think this stems from a bigger problem in the industry, and that is that software innovation is not able to keep up with hardware. The old x87 FPU has reached the point of limited returns. Even Athlon designers would have sworn up and down about 2x or 3x FP improvements in their core before they released it, and it still takes a little work to reach 20%-30% gains.

Hardware can't hope to get huge performance any more in the x86 world. Most improvements in hardware at this point require a certain amount of optimizations in order to achieve better performance. The Athlon is the result of some of the best engineers brute forcing the best theories and ideas, and even after it all, it bearly keeps up with a Pentium III. The Pentium 4 is a crippled processor without enhansements, and when the designers were thinking about how to implement certain ideas 5 years ago, they must have certainly thought that software would be a lot different than it is today.

I'm sure they were expecting great improvements, and yet here we are still using the same word processors and spreadsheets that we used 5 years ago (with little added value). Your favorite program, Gaffo, is 3D Studio, and if you think that program is not in need of a dire upgrade, than I don't know what is. If anything, I would expect that Kinetix would be overjoyed to have Intel optimizations in their rendering package. If anything, it would increase how compelling the package would be if it could gain a 50% improvement in speed. If you do not think this is the case, then I see a dark future for the industry in general.

With fewer software optimizations, I would say that bloatware is pretty much the software of the future. Maybe Microsoft has the right idea to create sliding windows, and user dialogues to take up all that CPU time, rather than actually putting it to good use. If that is the future, though, then I am ready to retire.

However, I would hope that isn't the case, and that developers will embrase SSE with time, especially when they see the improvements that are possible. MMX was a different story because it was hard to code, and only allowed for small performance gains. If Intel can show SSE to be easier, while showing the performance it has in SPEC scores, etc, then I believe they can be pretty successful with getting developers excited. This is just my opinion, though.

I have to add something I don't believe has been touched here yet about a feature brought to us by the P4 that I really like...as did Tom. That feature is the new PC spec. the P4 requires. Some poeple think this sucks and say that it shouldn't be neccesary but I think they are flat out wrong. We have seen many problems with AMD chips compatibility because they did not require better power supplies and such. I also really like the idea of a grounded heatsink for blocking EMI and the fact the heatsink doesn't require the motherboard PCB to support it (specially since it is so damn heavy!). I think AMD would be smart to try and adapt the same new standards Intel is trying to push.

------------------
My box said "WindowsME/2000 or better" so I installed Linux.

I honestly don't know weather it'll be a success or a failure but I'm leaning more towards success, since I own a old classic
P-150 and here I'm running on a blazingly fast P3-933 and wondering how can it be the same core?
and since the P4-1.4ghz is basically the next step from P3-1ghz, we already knows that it performs all the way on top.(maybe it doesn't beat it but it's plenty fast enough) so I think Intel will enhance the willamette core and after 5years we'll be at the same situation where there's a new core.
I get this mainly because I belive history repeats itself https://www.sharkyforums.com/images/.../2005/06/6.gif
I heard some ppl talking about the limits of the X86 (or X87 don't know much about it, is it a machine code?) I think the p4 will be one of the last ones for it 'cause most ppl here seem to be saying that unless the software has the optimizations it will not perform any better. and I think it is true.
too old I guess (what was is 1979 when 8086 was released? that means atleast 21yrs)

------------------
boo! bah!

Hey Arcadian - I agree about AMD's lack of mobo DDR speeds. This is becuase of AMD's small size and reliance upon 3-d party mobo makers (ALI and VIA). Not a lot they can do to speed mobo improvements - there chip has been held back by poor memory/mobos for over a year now - so ya it'll prob. be a least one more year before 400 mhz is here - but the EV? bus supports that speed - so it'll show up sometime.
Hey about the x87 and the k-7's speed being only a little more than the p-3! I think mostly that is due to lack of Athlon optimized code! - since the k-7's x87 fpu is 33-percent faster, then optimized x87 code would run 33-percent faster on the k-7 compared to the p-3. Most apps don't (but 3d-studio does, and I'm curious if this app is optimised for the k-7 since it really clobbers the p-3). The other thing holding back the x87 is the same as the rest of the ahtlon chip - poor memory bandwidth, rel. poor branch predictor. The k-7 is a x86 alpha WRT classic fpu.
About the 3d-studio needing a rewrite. This program really is one of the best. Release 3 is a new rewrite with a new rendering engine - with specular and rediosity and NURBS and metaballs. Its a lot more powerful than Lightwave (IMO) and has a great render quality - Maybe the 3d best x86 rendering app (just behind Softimage and Maya).
I also like Linux and I doubt that many opensource/free ware apps will rewrite for sse. Sse seems to be a "Wintel" thing - so if its not windows or another big player, the apps may not have support. About "bloat", I think most bloat is commercial justification to upgrade more than limits to the x86 code base. A lot of the free linux apps are often twice the speed and half recource dependant than windows apps (even the ones with extenion code). Gimp and Blender-2 are very fast apps, I although I don't know they don't have MMX, or SSE, I'd be surprised if they did.

Quote:

---

Originally posted by gaffo:
Hey Arcadian - I agree about AMD's lack of mobo DDR speeds. This is becuase of AMD's small size and reliance upon 3-d party mobo makers (ALI and VIA). Not a lot they can do to speed mobo improvements - there chip has been held back by poor memory/mobos for over a year now - so ya it'll prob. be a least one more year before 400 mhz is here - but the EV? bus supports that speed - so it'll show up sometime.
Hey about the x87 and the k-7's speed being only a little more than the p-3! I think mostly that is due to lack of Athlon optimized code! - since the k-7's x87 fpu is 33-percent faster, then optimized x87 code would run 33-percent faster on the k-7 compared to the p-3. Most apps don't (but 3d-studio does, and I'm curious if this app is optimised for the k-7 since it really clobbers the p-3). The other thing holding back the x87 is the same as the rest of the ahtlon chip - poor memory bandwidth, rel. poor branch predictor. The k-7 is a x86 alpha WRT classic fpu.
About the 3d-studio needing a rewrite. This program really is one of the best. Release 3 is a new rewrite with a new rendering engine - with specular and rediosity and NURBS and metaballs. Its a lot more powerful than Lightwave (IMO) and has a great render quality - Maybe the 3d best x86 rendering app (just behind Softimage and Maya).
I also like Linux and I doubt that many opensource/free ware apps will rewrite for sse. Sse seems to be a "Wintel" thing - so if its not windows or another big player, the apps may not have support. About "bloat", I think most bloat is commercial justification to upgrade more than limits to the x86 code base. A lot of the free linux apps are often twice the speed and half recource dependant than windows apps (even the ones with extenion code). Gimp and Blender-2 are very fast apps, I although I don't know they don't have MMX, or SSE, I'd be surprised if they did.

---

I was not aware of some of the features available in Release 3 of 3D Studio MAX. It does seem that they have outdone themselves in the area of features.

The explanation of why the Athlon does well in 3D Studio is because of the Athlon's dual FPU pipelines for ADD and MULT. Athlon designers knew that 3D would be a popular technology, so they designed the Athlon around that. 3D transformations usually use a lot of ADD and MULT instructions, so having a separate pipeline for each allows the Athlon to excel in 3D transform code.

However, this kind of code lends itself even better to SSE, and I still believe Kinetix to be foolish to miss the opportunity to include SSE optimizations in their code.

As for Linux, I see the complete opposite happening than you suggest. Open source communities are more likely to implement new instruction sets, as they have less of an agenda in bringing products out. You will notice that the Linux community has always made optimizations when possible, which is one of the reasons why Linux and Linux programs run so fast on PCs. I would think that SSE optimizations to be quite popular in the Linux world.

You even agree that Linux applications are the antithesis of bloatware. It may come to pass that the PC world embrases the Linux community because of this fact. Microsoft can't keep bloating their software forever, as it will surely turn a lot of people towards other alternatives. I think it may be possible for Linux to save the PC if it ever catches on.

We'll have to see if this has any chance in happening.

Saying the OpenSource community won't adapt things like SSE is kinda rediculous. Take a close look into things and you will see operating systems and programs spawning from the opensource community have support and optimizations for more platforms then most people have ever even heard of. Us members of the OpenSource community believe in bringing software to as many people as possible in the best way possible. I guarentee linux and other opensource os's as well as apps will use SSE optimizations at some point.

Why is it people automaticaly assume linux isn't going to support something when if you actually look into it you will see linux (and it isn't alone..BSD and many other unix-likes are in the same situation) have more support for different things then Windows by a long shot? It is often just different kinds of support and support for devices gamers don't often come accross.

Sorry for the rant... but maybe somebody will listen this time... not sure why I bother.

------------------
My box said "WindowsME/2000 or better" so I installed Linux.

[This message has been edited by jtshaw (edited November 22, 2000).]

Quote:

---

Originally posted by jtshaw:
Saying the OpenSource community won't adapt things like SSE is kinda rediculous. Take a close look into things and you will see operating systems and programs spawning from the opensource community have support and optimizations for more platforms then most people have ever even heard of. Us members of the OpenSource community believe in bringing software to as many people as possible in the best way possible. I guarentee linux and other opensource os's as well as apps will use SSE optimizations at some point.

Why is it people automaticaly assume linux isn't going to support something when if you actually look into it you will see linux (and it isn't alone..BSD and many other unix-likes are in the same situation) have more support for different things then Windows by a long shot? It is often just different kinds of support and support for devices gamers don't often come accross.

Sorry for the rant... but maybe somebody will listen this time... not sure why I bother.

---

Jtshaw, I read a lot of your posts, and find them very informative. I just wanted you to know that I for one appreciate your feedback. Rant on... rant on. https://www.sharkyforums.com/images/.../2005/06/5.gif

PS... I also agree that the Linux community to be among the first to embrace new technologies.

Quote:

---

Originally posted by Arcadian:
I wanted to do a little analysis to see if your estimation prooves useful. I'm going to average actual scores from Sharky's, Anand's, Ace's, and Tom's Hardware web pages in productivity, games, video/audio, and other. The order will be Pentium III 1GHz (on i815 if possible), Pentium 4 1.5GHz, Athlon 1.2GHz, Athlon 1.2GHz+DDR. The Pentium III has a base of 100%

Productivity:

Tom's Hardware Sysmark:
100% 95.2% 104.3% 115.5%

Anand's Content Creation Winstone:
100% 110.9% 111.1% 117.6%

Sharky's Content Creation Winstone:
100% 116.8% N/A 121.6%

Sharky's Sysmark:
100% 101.5% N/A 113.6%

Average:
100% 106.1% 107.7% 117.1%

3D and Games

Tom's Quake III:
100% 132.7% 104.5% 116.7%

Tom's MDK:
100% 121.9% N/A 123.6%

Anand's Unreal Tournament:
100% 102.4% 104.1% 112.2%

Sharky's 3D Mark (Win98):
100% 127.8% N/A 117.5%

Average:
100% 121.2% 104.3% 117.5%

Video/Audio

Tom's Mpeg4 compression:
100% 160% 114.3% 127.7%

Ace's Hardware Soundforge (couldn't find similar bench on Anand's):
100% 105.6% N/A 131.9%

Ace's Hardware GoGo MP3:
100% 113.6% N/A 129.3%

Sharky's Windows Media Encoder:
100% 194% N/A 125.5%

Average:
100% 143.3% 114.3% 128.6%

Other
Here I will take a wide range of tests where the Pentium 4 both underperforms and overperforms.

Tom's 3D Studio MAX:
100% 93.3% 175% 175%

Tom's Linux Compile:
100% 102.2% 122.7% 123.8%

Anand's Spec Viewperf (average):
100% 121.6% 115.5% 127.8%

Sharky's Dragon Naturally Speaking:
100% 114.5% N/A 112.5%

Average: (Athlon 1.2GHz SDR weighs strangly due to excellent 3D Studio scores and missing Dragon score)
100% 107.6% 137.7% 134.8%

Now if we take Productivity, 3D games, Audio/Video, and other, and take a final average, this is what I get for overall processor advantage.

Pentium III 1GHz on i815: 100%
Pentium 4 1.5GHz on i850: 119.6%
Athlon 1.2GHz on KT133 : 116%
Athlon 1.2GHz on 760 DDR: 124.5%

This is using a wide range of programs that show the strengths and weaknesses of the Pentium 4. I believe this to be a fair representation of the programs out there, and although the Athlon at 1.2GHz with DDR outpaces the Pentium 4 at 1.5GHz, the amount is not as much as you have made out. I think this is a much better comparison. Any comments from anyone?

---

Oki, thanks for adding up the scores. I just did some very rough estimations and weighted in thing I find more important (such as FPU stuff) slightly more heavily than other stuff. But if you take the whole application spectrum into account your summations are probably a very good genereal performance indicatino.

Quote:

---

Originally posted by Arcadian:
I also wanted to add to my analysis the fact the SSE/SSE-2 optimizations should really improve upon the Pentium 4 is a lot of those scores, so I can see the Pentium 4 really leading by a larger margin by the end of 2001. I also wanted to mention that the Palomino Athlon may further AMD's lead as well, so between Intel's optimizations and AMD's new processor, things may get interesting, and either one can take the lead. Right now, though, the Pentium 4 is not a bad performer. It just costs too much. https://www.sharkyforums.com/images/.../2005/06/5.gif

---

I expect SSE to continue to get more support while SSE-2 will be less important since it's only useful to scientific applications which needs the added precision.

Quote:

---

Originally posted by Arcadian:
I can't find where I read it, but someone on Ace's Forums had a theory that the compression codec found in mpeg and jpeg files is a very optimized codec with very predictable program flow. This is really what the Pentium 4 accels at, and also what it was designed from the beginning to do well in.

---

Yes, the JPEG/MPEG coding consists of many highly parallel operations. Only the BitStream packing and ZigZag reordering parts can't take any significant advantage of SIMD optimizations, but most other stuff such as DTC/iDCT, colorspace conversion and shifting gains a lot from SIMD.

Quote:

---

Originally posted by Arcadian:
I have to disagree with that. I believe MMX has become quite ubiquitous in software. In fact, you probably use it in most of your programs and don't even know it. Even AMD processors include MMX extensions, so I have a hard time believing that programs don't use a feature that is so easy to develop for.

...

MMX was a different story because it was hard to code, and only allowed for small performance gains

---

Eh, is MMX hard or not? https://www.sharkyforums.com/images/.../2005/06/5.gif

Actually, MMX isn't used in many applications since most compiler don't generate MMX code and MMX is only applicable to certain types of applications, the same goes for SSE/SSE-2/3dnow.

Quote:

---

Originally posted by Elxman:
I heard some ppl talking about the limits of the X86 (or X87 don't know much about it, is it a machine code?)

---

x87 is the FPU.

hey guys - check out tom's updated p-4 review - looks like the high quality setting for DVD/mpeg conversion under the flask(something) program uses x87 fpu, and the k-7 wins two to one.

Quote:

---

Originally posted by gaffo:
hey guys - check out tom's updated p-4 review - looks like the high quality setting for DVD/mpeg conversion under the flask(something) program uses x87 fpu, and the k-7 wins two to one.

---

The results aren't too surprising, are they? Man, that guy would do anything to stick it to Intel.