It's a failure from a technological point. Not from a sales point.
I think even the biggest Intel fanboy would like to see NetBurst disappear.
If intel decided to work off the northwood core instead of doing a brand new prescott, people would be saying different things right now. Netburst still would have died off, but it might have been a graceful exit. Netburst isn't really that bad, but the prescott project totally smeared netburst.
- Sep 9, 2001
- 1,256
- 0
- 0
Netburst was a complete fiasco since its conecption, based on the idea that "more is better" targeted to average Joe's all over the world. The fact that a new way of processing, so to speak, based on the clock speed instead of the opc that AMD adopted, showed it was completly out of any technological/performance base.
It was shameful to see Pentium 3 @ 1Ghz beat Pentium 4 at 1.4Ghz, IF paired with Rambus. Remember Rambus, dontcha? But very few ppl actually got that Rambus memory: as soon as i845 with SDRAM support stepped in, performance stepped out, but it made Pentium 4 affordable for the masses. That made P4 @ 1.6Ghz perform... well... I'm not sure if its ok to call it "performance" at all, like a P3 @ 933Mhz. Not good.
As many correctly stated: Nortrhwood was what Pentium 4 should've been from the begining... but that's just about the only thing that Intel has done right in the past 5 years regarding desktop processors. How isn't that a fiasco, from a technological point of view?
Since the dawn of computing (well pretty close), there have been two methods to do a problem: the Brainiac approach (high IPC, low clock) versus the Speed Demon approach (low IPC, high clock). People are confusing Netburst with the ideology of the latter's approach.
To say that a Barton is nothing more than a 64 bit Athlon with an integrated memory controller is a high exaggeration. The A64 is a totally new generation from the Barton, much like the P4 is a totally new generation to the P3. And to say a Pentium-M is nothing more than a high clocked P3 is just as inane. It has been proven time and time again the Pentium-M is far closer to a Pentium4 than a Pentium3. Simple proof is that converter (which is nothing more complicated than a Slot1 -> S370 or SlotA -> socketA converter) exists.
Back then when the software was unoptimized, sure a 1Ghz P3 = 1.4Ghz P4. I doubt you'll find many benchmarks comparing 2 older processors, but I can assure you that a 1Ghz P3 wouldnt stand a chance against a 1.4Ghz P4 (with DDR or RDRam) using any up-to-date software (encoding, games, etc).
To say that a Barton is nothing more than a 64 bit Athlon with an integrated memory controller is a high exaggeration. The A64 is a totally new generation from the Barton, much like the P4 is a totally new generation to the P3. And to say a Pentium-M is nothing more than a high clocked P3 is just as inane. It has been proven time and time again the Pentium-M is far closer to a Pentium4 than a Pentium3. Simple proof is that converter (which is nothing more complicated than a Slot1 -> S370 or SlotA -> socketA converter) exists.
Back then when the software was unoptimized, sure a 1Ghz P3 = 1.4Ghz P4. I doubt you'll find many benchmarks comparing 2 older processors, but I can assure you that a 1Ghz P3 wouldnt stand a chance against a 1.4Ghz P4 (with DDR or RDRam) using any up-to-date software (encoding, games, etc).
Thankyou, both to you and dmens for making this point.
The topic of this thread lacks point, IMO.
It's as silly as saying that a Lexus is a Ford T. The proof is not in that it has four wheels.
Well, software in general cannot be 'optimized' for the P4. Only special tasks, like benchmarks and encoding, can. In modern media centered PC use, there is a lot of encoding going round, so it's no trifle. But a 1GHz P3 probably still runs with a 1.4GHz P4 on many things. Encoding, - no. But games, the P3 would IME do quite well.
By optimized, I was specifically referring to SSE2. When P4's first came out, no software took use of SSE2. Nowadays, almost every piece of software that can utilize SSE2 does.
http://www.tomshardware.com/cpu/20041221/cpu_charts-13.html#opengl
THG is the *only* site I know of that has a massive CPU chart.
A 1.3Ghz Willamette (256K L2) using i850 RDRam dominates the 1.0Ghz Coppermine (133FSB 256K L2) using i815 and PC133 SDR. What's also amusing is that it nearly matches a Thunderbird (133FSB 256K L2) using KT133 and PC133 SDR clock for clock. Back then, a T-bird usually owned a Willamette clock for clock.
Yes I'm sure THG has a lot to gain from showing that a 4 year old Pentium-4 system that is no longer being sold today is better than a 4-5 year old Pentium-3 system that is also not being sold today.
- Sep 9, 2001
- 1,256
- 0
- 0
Well, my conclusion was based in the readings I (and most of us) had when Opteron came out. I quote from this page:
"When we first looked at the K8 around two years ago we noted that, architecturally, not much had changed from the K7. Many attributed this to the old axiom, "if it ain't broke, don't fix it," and to a certain extent they were correct. The K7 microarchitecture proved to be an excellent performer, and even to this day has managed to at least remain competitive with Intel's NetBurst microarchitecture found in the Pentium 4. "
I found that, as much as I personally dislike THG, I must accept that from a tecnological explanation point of view, they're pretty good. My point is taken here.
I quote:
"On closer inspection the attentive observer might notice that the AMD Hammer's physical foundations look identical to the old Athlon core in many of its details. "
And... I saw similar remarks on Hexus.net and OCWorkbench.com, but its 3am in the morning and I am to lazy to look them up.
Its more than obvious that the branch prediction, on die memory controller, stage pipelines and other enhancements clearly make it different from the K7, but as the statements I posted above suggest, its still, in escence, an updated K7.
Similar things can be found on the Pentium-M, but I think I've proven the reason why I posted this thread, and hope you don't find it without a point anymore.
SSE2 is just FP instructions. Pretty much anything that uses FP can use optimizations to a certain degree, obviously some more than others. Just look at the link I gave to THG. Most of the times (games, synthetic, encoding) a 1.3Willamette/i850/RDram pulls ahead of the 1.0Coppermine/i815/SDR. The only difference between now and when it first launched is the software (mostly the same applications, but newer versions). The hardware has not changed (P4's and P3's dont become slower as they age), so the deduction is that with software optimizations, the playing field has changed.
Die size is pretty much meaningless. Nowadays the L2 cache comprises around half the total die size.
However, I don't agree with Anand or THG's take on the premise that the K8 is merely an extension of the K8. By that account, we can assume:
286 [+ 16 bit instructions + Math coprocessor] =
386DX [+ upgraded clock speed and clock multiplier] =
486DX [+32 bit instructions] =
Pentium [+MMX] =
Pentium-MMX [+Pentium Pro] =
Pentium II [+SSE and higher clock] =
Pentium III Katmai [+on die cache] =
Pentium III Coppermine [+higher clock speed and SSE2 + Netburst Bus] =
Pentium-M Banias [+L2 cache and clock speed] =
Pentium-M Dothan [+dual core] =
Pentium-M Yonah
And it goes on.
Edit: On an interesting topic, you could more or less say the Itanium2-Madison is just a Pentium-M + more L2 cache + L3 cache + IA64.
It's very risky to do a major shift, in the CPU industry as it takes awhile for the software to catch up, and you could have a poor performing processor if the current code isn't optimal for your architecture, hence it is much less risky to enhance what you have already got.
286 [+ 32 bit instructions + Math coprocessor] =
386DX [+ upgraded clock speed and clock multiplier] =
486DX [+superscalar, higher clockspeed] =
Pentium [+MMX] =
Pentium-MMX [+Pentium Pro, 3 wide issue architecture] =
Pentium II Klamath [+higher clockspeed, better process] =
Pentium II Deschutes[+SSE, higher clockspeed] =
Pentium III Katmai [+on die cache] =
Pentium III Coppermine [+ more L2 cache, higher clock speed, better process] =
Pentium III Tualatin [+more LV2 cache, higher clock speed, SSE2, Netburst FSB, micro-ops fusion, cache power down, longer pipeline] =
Pentium-M Banias [+more L2 cache, clock speed, better process] =
Pentium-M Dothan [+shared cache, SSE3, DMB, Dual Core, enhanced process] =
Pentium-M Yonah [+macro-ops fusion, longer pipeline, EM64T, 4 issue wide architecture]
Merom/Conroe
286 [+ 32 bit instructions + Math coprocessor] =
386DX [+ upgraded clock speed and clock multiplier] =
486DX [+superscalar, higher clockspeed] =
Pentium [+MMX] =
Pentium-MMX [+Pentium Pro, 3 wide issue architecture] =
Pentium II Klamath [+higher clockspeed, better process] =
Pentium II Deschutes[+SSE, higher clockspeed] =
Pentium III Katmai [+on die cache] =
Pentium III Coppermine [+ more L2 cache, higher clock speed, better process] =
Pentium III Tualatin [+more LV2 cache, higher clock speed, SSE2, Netburst FSB, micro-ops fusion, cache power down, longer pipeline] =
Pentium-M Banias [+more L2 cache, clock speed, better process] =
Pentium-M Dothan [+shared cache, SSE3, DMB, Dual Core, enhanced process] =
Pentium-M Yonah [+macro-ops fusion, longer pipeline, EM64T, 4 issue wide architecture]
Merom/Conroe
I believe if you run cpu-z, it identifies a pentium M as something like P6-III or something like that, so it's still based on P6. In fact, I'd say that the backend is probably almost exactly identicle, rather it's the front end that has significantly changed.
BTW, is the G5 really new technology? I thought it was based off of one of IBM's ancient server cpus? Well, anyhow, one look at its transistor count could probably tell you that it couldn't compete, the chip is soooo much less complex than a modern intel or amd processor.
Both are heavily based off the previous though, isn't the P4 basically an even more super pipelined P3 with an insanely fast cache?
And the A64 is a Barton with some tweaks, 64 bit support, a vastly improved cache, and an integrated memory controller?
The P-M uses the same memory bus as the P4, that's why its compatible. It communicates the same way as the P4, but I'd say the core is still closer to the P3.
P-M is pretty much P3 with some tweaks, and the addition of everything that was added to the P4, minus the increased pipelining. So basically it's a P4 without a focus on clock speed, which is what 99% of the focus of the P4 was. Either way works I guess, P-M is a P3 + new memory bus, super fast cache, some tweaks, or a Netburst with less pipelining.
Intel stopped making 1.4ghz p4s by the time P4s reached clock parity with the P3s.(though in some things, the P4s still haven't reached clock parity in because there are some things where a P3 will match a P-M in)
The real shame was that the 800mhz(or the closest mhz to that) Athlon C outperformed both the 1ghz P3 and the 1.4ghz P4.(depending on the task btw, for Intel or P4 optimized apps, like Quake 3 or encoding, the results would not look like this, just there were quite a few tests where an athlon 600mhz below a p4 would beat it)
Well, that makes up for the P4's floating point deficiency, but SSE2 is much harder to use than....well not using it.(besides, P3 had SSE, how much better is SSE2 over SSE?)
thecoolnessrune
Diamond Member
- Jun 8, 2005
- 9,673
- 583
- 126
I'd like more modulation as opposed to integration. Id like to be able to select my own CPU, Northbridge, Southbridge, GPU, GPU memory, system memory, and all that stuff individually. Not bundled together with junk I dont want.
- Sep 9, 2001
- 1,256
- 0
- 0
You are mostly speaking of synthetic apps, and we all know those where HEAVILY optimized for Pentium 4 processors. Remember the fisaco that came from BAPCO because their software was written to take advantage of Netburst and make it look like the best thing since sliced bread?
I disagree. Its like Sisoft Sandra showing Skt939 processors having almost 5700MB/s memory bandwith and Skt754 having 3000Mb/s, but still coming, in most ocations, 1-3% behind its "dual channel memory" counterpart. Flashy numbers for products that needed to look flashy.
Even if it means more performance? I'd rather every cpu have an integrated memory controller, performs far better and you don't get stuck with a crap one.(say, something made by VIA?)
Now then, the rest of the stuff doesn't need to be integrated since it wouldn't add to performance.(and since there's no price difference between DDR and DDR2, the memory doesn't matter much, I'd just like them to stick with a standard)
An old quote I always like to tell people:
There are two kinds of fools: one says, "This is old, therefore it is good"; the other says, "This is new, therefore it is better." - William Ralph Inge
There are two kinds of fools: one says, "This is old, therefore it is good"; the other says, "This is new, therefore it is better." - William Ralph Inge
Rock Hydra
Diamond Member
- Dec 13, 2004
- 6,466
- 1
- 0
Yeah, that's so difficult when the foundation for math was basically entirely laid out. Not like you make up rules to math.
I always have been wondering about this. What is the point of the hardware companies not producing modular hardware for video cards, etc? Is there some sort of financial reason that I don't see?
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter