Favorite Processor

[Braiii]

I must admit being an AMD XP 1800 owner I almost applaud the competition, I don't take sides...I only go after the best value and plus this is a battle of the fittest and to me (just an opinion) Intel wins hand down. Yes, you'll pay a whole lot more for Intel but hey, new Ferrari’s don't cost the same price as a new BMW...speed and performance winds hands down for Intel but you get a better value with AMD.

Yes, I'm aware of new chip's coming out in the next couple of years on both sides and I think that it we'll become a heated battle for all criteria...performance, value and speed, and if I was you I would invest in both companies...

I mean don't get me wrong here but what AMD owner wouldn't want an Intel OC'd to 3.5GHz? I know my XP 1800 won't get there the best I can hope for OC'd is 1732MHz and that's only like a XP 2100...and for the AMD owners who wouldn't want that 3.5 please explain why I'm open for all reasons...remember I'm not taking sides.

- Rob
: Vote AMD

[$1500-P4 gamer]

Yes at 1.2gig thats so true. But we both know as the P4 scales higher to a point, it increases the fpu significantly. I say to a point because after a certain speed it saturates the bus, the new 133fsb helps with that though and 3gig was a barrier as of no more. Now to say the P4 architecture in it self is that bad at fpu you should look at the evolved chip not the first screw-up they pumped out the door. Look at the Northwood B with 133fsb where it belongs. At this speed with the larger L2 cache the fpu is doing alot beter than ever before. The very first P3's werent anything but sped up P2's with sse as you said. You have to keep that in mind. So the 1.2 was going to be a flop and we all know it but I think you must judge the line of chips, the entire P4 deseign off the best ones. You know what I mean or am I just babling? Anyhow I dont think it is a Weak FPU at all. I think it has to do with how the FPU unit is incorperated in the cpu. I mean we are looking at a 20 stage pipeline. If it mispredicts, well you know, lost cpu cycles. That is all going to kill FPU/ALU and overall speed of the chip. You have to take that into acount. So in that way I think it is the architecture of the chip (aimed at high mhz) that is the FPU limitation not the actual FPU unit itself.

[Rugor]

P4 Gamer, I beg to differ.

The only significant architectural difference between P4 Willamette and P4 Northwood is the larger cache on Northwood. Any other differences are no greater than those between steppings of a given core.

I didn't want to compare a faster P4 largely because this was the only speed grade I had numbers for. However, at 1.2GHz the P4's FPU should be able to operate in as efficient a mode as possible. Remember, at this speed the P4 has more bandwidth per clock tick than at any higher speed grade. Not even a Nortwood B can come anywhere close to saturating the bus as much as this one. My goal was to compare FPU efficiencies and you can't do that unless you use the same clock speed.

The Athlon uses a triple unit fully pipelined FPU, able to perform 3 FP operations per clock. The Pentium III core has a dual unit semi-pipelined core that can perform 1 FP operation per clock, alternating the two units for improved performance. Theres a good comparison of the two FPU's in this article at Ars Technica. The P4, like the PIII is limited to a single FP operation per clock, but its less robust hardware ends up producing weaker results.

Personally I think that the P4 Northwood B is producing much better results than the original P4 not because of a stronger FPU, but because of the larger cache and higher clock speeds and FSB. The processor itself has a relatively weak FPU but it is running fast enough that it can afford to spend the multiple cycles that the single FPU requires without too much of a real-world problem. Also, both SSE and SSE2 do help with performance, especially as SSE2 enabled applications enter the marketplace. However, the P4 remains at a disadvantage when forced to use its FPU in raw mode, as the benchmarks show.

Personally I think any design should be able to show its strengths, and weaknesses at any speed. I do agree that the chip's architecture is the source of its weak FPU but would explain it differently. The FPU is weak because in order to keep the transistor count and die size down they implemented a simpler FPU which also allowed for better clock speed scaling with reduced performance per clock.

[$1500-P4 gamer]

I understand what you are saying there. Thing is though if Raw FPU is so important still to gameing why is it that the Game benches I've seen dont reflect this at all. Serious sam is the only game where a dif. can be noticed between the XP and the P4. Thats the only game out of the many upon many benches I've seen based on this very discusion. And now the P4 even wins in the Serious Sam bench just proveing that the newer (not even the B) P4's are getting a faster fpu out of them regardless of how it is happening. I see sse2 as the same thing that happened with MMX. Everyone said it will never happen yet it did. My nvidia card is sse2 optimized. I have sse optimized games here too. Even the 3dfx voodoo 3 was optimized for the P3's sse. I see game developers useing it if Intel sticks with it and the P4 is proof it is going to stick around awhile-thats all game developers needed to know. Keep the coments comeing Rugor- I'm enjoying this discusion. I just for the life of me cant see the great benifit of the super fpu on current apps. No benches show it unless you are riping off dvd's and compileing them into divx or similar. There it is only the case because they are hackers that cant right software optimisation for the life of them. Whats sse2?

[$1500-P4 gamer]

"I beg to differ.

The only significant architectural difference between P4 Willamette and P4 Northwood is the larger cache on Northwood. Any other differences are no greater than those between steppings of a given core. "

No thats is where I disagree. There is a HUGE dif. between the williamette and the Northwood. First of all there is the L2 as you said. This is a big help on the performance of whole cpu. The more you hold in the L2 the more data ready to be fed to the L1 instead of fetching from the system. Second there is the .13 fab vs. .18, now admitt that counts or why are you all waiting for amd to get there? Third there is the copper interconnects on the Northwood vs. crappy alluminum on the williamette. This too was something amd fans held on us but now that we have it some dont think it matters?? The higher fsb of the AMD was also a feature everyone thought amd had intel over the barrel on, now the P4 is on 533fsb and there is no dif between it and the 400 or even the 266fsb??? How can one even say that! All this yeilds a faster cpu and higher O'clocks with less heat and lower cost. They are not just cpu revisions, there a dif. cpu all in all. The stages of the pipeline are the same and etc. but there are noteable dif. between the two cpu's.

What I was saying about useing the flagship to judge the entire lineup is that the P4 is on 533fsb now. Thats more bandwidth which on my tests prove that the P4 yeilds a much higher fpu. Anyone wants to try that O'clocks there P4 just run it at stock, bench it then run at O'clocked like I did. The fpu will be much higher than that of a = clocked P4 at stcok. Why because the higher the mem bandwidth the faster the P4 all around, Intel knows it thats why they are moveing to 133fsb/533. I think they had this move in mind all along as my P4t-e has had the 133fsb option all along. Like my 1.5 wimpy williamette beats any 1.8 when I clock at that speed, the only dif is the mem due to fsb.

So I still dont think it is just to compare the first released P4, to a aged tried and true tested P3 and AMD. Its not about mhz rugor and you know that. There really is no = mhz comparison betweeen the P4 and P3 or AMD's it just isnt there. So by running a 1.2 P4 against a P3 1.2 you already know whats going to happen just because of the longer pipeline and the single fpu unit. No point in even showing that comparison its irrelevant unless you want to make the P4 look bad. But anyone who runs one gameing and such isnt going to see one thing that yo mentioned in performance, why? That is the answer I'm seeking.

[kazuza]

This little picture right here on the left is my answer!!!

[Rugor]

I guess I wasn't clear enough in my original post.

What I was trying to do was clearly show the relative strengths of the three CPUs FPU units. Regardless of how anyone looks at it the FPU on the P4 is weaker than that of either of the other two CPUs, and the numbers I posted show by how much. I then explained why the P4's FPU is weaker, because it can do less per clock cycle than either the PIII or Athlon. I would not consider a 1.2GHz P4 as a competitive processor to either the Athlon or PIII at the same clockspeed, but that's not what I'm doing. What I'm doing is comparing the designs to show relative strengths and weaknesses. I can't compare the architectures without using the same clockspeed. A P4 1.5GHz would be a better performance comparison, but it wouldn't illustrate my point, because then you wouldn't see how the FPU compares at the same clockspeed, which would obscure any architectural differences.

Without demonstrating how the FPU's differ and why it becomes difficult to really explain why the P4 needs so much more clockspeed to be competitive. It's true the P4 has taken the speed crown (in benchmarks not just clock cycles) but it's also true the Athlon hasn't reached a 2500+ Performance Rating. I expect a 2.53GHz P4 to beat an Athlon XP 2100+.

If I wanted to compare with regards to which processor you should buy I would use dollars rather than clock speed as my measuring point. I would take a P4 Northwood RDRAM system, and an Athlon XP DDR system and see how much I had to spend to get a given level of performance, but that wouldn't give me any insight into differences in processor architecture and how they reflect in performance. The P4 is faster, and it needs to be faster, my table just illustrated one of the reasons why.

FPU is a big component of performance, but not the only component. System bandwidth also counts, and the P4 has gobs of bandwidth, especially with the faster bus. Games require FPU and bandwidth, and a weakness in one can be to some degree made up for by strength in another. The P4 is now winning at Serious Sam because it is now fast enough that its strengths outweigh its weaknesses.

I still hold by my contention that apart from the larger L2 Cache the P4 Northwood has very few if any architectural differences from the P4 Willamette. You made the points that the Northwood is produced on a smaller process and uses copper interconnects. Both of these are true, but they are not what I would consider architectural differences. They are manufacturing differences. Moving to copper interconnects and a smaller manufacturing process do not change how the processor handles data by one iota. What they do do, is allow processors with the same design to scale to higher clock speeds. They don't grant efficiency or performance improvements other than those resulting from the higher clock speed. As to the faster bus speed, that is not in any way an architectural change as it's a completely external change. Yes it grants better performance, but that's because it's using the internal architecture more efficiently, not because it's a different architecture. The only performance enhancing change to Northwood is the larger cache. The other improvements allow the processor to run faster and thus gain performance, but don't improve performance in and of themselves.

I am looking forward to Thoroughbred's shift to .13 microns, not because I believe that the .13 micron process will add anything more to the processor than the ability to scale to higher clockspeeds. Yes, AMD does need a faster bus, probably more than Intel did, but moving to a faster bus doesn't change the nature of the processor.

[$1500-P4 gamer]

Alright I get you rugor.
"The P4 is now winning at Serious Sam because it is now fast enough that its strengths outweigh its weaknesses. "

That makes sence. The real question is how far will the P4 fsb go? With pc1200 (166fsb?) it hints of another fsb down the road as the bus gets saturated. I mean look at the P4 3gig on 100fsb, its junk compared to the P4 2.4 on 133fsb. I think 4 or 4.5 is when the move will be needed.

[Rugor]

Ummmmmmmmm Gamer, the P4 may need to go to a 166 (666) FSB faster than you you think.

The P4 2.4B has an 18x Multiplier, whereas 3GHz at the 400MHz FSB would be a 30x Multiplier. If you put an 18x Multiplier with a 166 FSB, you end up at 3GHz, so a faster FSB may be needed sooner rather than later.

[$1500-P4 gamer]

Actually mine and most other P4 boards atleast on the !850 goes to a 24x multi and then there is the ability to increase that multi even higher through a bios update. But even at 24x a 133fsb is 3.2gig and at 166fsb thats 4gig. But yeah the 166fsb is gonna be needed sooner than I was thinking. but hey with a 24x (3.2gig) I can run at 155fsb so that is a good O'clock at 3.7gig. I'm happy with that being a final speed on this mobo. That is if they dont update it through the bios. There is also the ability to change the model steping of the chip so really multiplyers dont matter at all. Remember the P2 233 it runs at a multi of 1.5 x66mhz! What I ment by needing a higher fsb at 4.0 or 4.5 is the bus saturation not the hardware limitation. Suposedly (as far as intel says) the 478 can hit up to 8gig without that limitation. But Im sure mem bandwidth will be a prob way before that. Thats what I was refering to.

[Rugor]

I was referring entirely to bus saturation in my previous post. The P4 3C has the same degree of bus saturation on a 166FSB, as the P4 2.4B has on the 133FSB. I was assuming the multiplier had no limits other than those that come from running so high as to cause data starvation.

P4's need data, the faster they are the more they need. Scary isn't it.

[BlackIris]

Ok start flaming

Having a celeron 633 @ 1064 with stock HSF. Using core v. 1.75. the temp has never gotten above 38C. Why would I want to spend the $ to get the higher priced chips?

[Rugor]

Why would you?

I wouldn't upgrade til you feel you need it.

I wouldn't buy a Celeron 633 if I were in the market for a new system, but if I were running your system I wouldn't bother with an upgrade yet.

[$1500-P4 gamer]

There is a easy solution to that though rugor. You know it too! Dual DDRII. Even the same thing as nvidia did would work. Or even Quad RDRAM. Just quad pump the fsb right through the north onto the rdram. I dont know if quad ddr is possible but I think with the seriel nature of rdram it would be possible. Can you imagine that bandwidth!pc1066 wold be 8.4gb/sec and pc800 would be 6.4gb/sec. There are so many solutions to that. But you right it is a mountain to overcome.

[BipolarBill]

Dual Pentium Pros, Baby!

J/K...although it was a nice platform and is still in use in legacy servers.

I like the Duron and Athlon, but I also like to fry my eggs in a pan.

Current favorite is the Tualatin in any incarnation - P3 or Celery. It's forgiving, and I need forgiveness!