SavantStrike wrote:

Case in point, the N64 didn't need RDRAM.

It must have been chosen for its high bandwidth on a narrow (cheap) bus. FPM would have needed to be 64-bits wide to match up on that aspect.

Also, EDO and FPM were quite expensive until 1996. There was a RAM market collapse that year. This was instrumental in allowing Voodoo1 to be priced relatively cheaply. But the N64 chipset had been in development for years at that point and was designed for RDRAM.

Tetrium wrote:

I think swaaye is correct in assuming the cache was halved to save on manufacturing costs. Remember that the CPU cache is a quite large die area.
I don't know if a 1.8Ghz Willy is faster then any Coppermine/Tualatin, but it sure eats more power. Personally I think I'd prefer Athlon XP over Netburst, but perhaps that's also because I also have more experience with Socket A...and they kinda resemble Socket 370.

Willamette and Northwood may be able to hold their own, but Prescott...any point to it?

A 1.8Ghz Willie is faster than a 1.4Ghz Tually in gaming benchmarks...look up old Tom's benchmarks. The motherboards it sits on more importantly perhaps are miles ahead in terms of quality, robustness, features and layout of whatever socket 370/ slot 1.
A'd also like an AThlon XP- but they're considerably rarer here in Temeria 😢.

About Prescott having a point- look up my sig rig. I have a choice of nearly any P4 chip yet that's what I chose to plonk in there. It's way faster than my AThlon 64 2.5ghz rig BTW.

I'll add that I was a Netburst hater back in the day (P3 and s939 user from back when), but having owned and used my various P4 rigs recently has led me to dramatically rethink my previous stands.

sgt76 wrote:

A 1.8Ghz Willie is faster than a 1.4Ghz Tually in gaming benchmarks...look up old Tom's benchmarks. The motherboards it sits on […]
Show full quote

Tetrium wrote:

I think swaaye is correct in assuming the cache was halved to save on manufacturing costs. Remember that the CPU cache is a quite large die area.
I don't know if a 1.8Ghz Willy is faster then any Coppermine/Tualatin, but it sure eats more power. Personally I think I'd prefer Athlon XP over Netburst, but perhaps that's also because I also have more experience with Socket A...and they kinda resemble Socket 370.

Willamette and Northwood may be able to hold their own, but Prescott...any point to it?

A 1.8Ghz Willie is faster than a 1.4Ghz Tually in gaming benchmarks...look up old Tom's benchmarks. The motherboards it sits on more importantly perhaps are miles ahead in terms of quality, robustness, features and layout of whatever socket 370/ slot 1.
A'd also like an AThlon XP- but they're considerably rarer here in Temeria 😢.

About Prescott having a point- look up my sig rig. I have a choice of nearly any P4 chip yet that's what I chose to plonk in there. It's way faster than my AThlon 64 2.5ghz rig BTW.

I'll add that I was a Netburst hater back in the day (P3 and s939 user from back when), but having owned and used my various P4 rigs recently has led me to dramatically rethink my previous stands.

At 4ghz clock speed and ~1066mhz FSB, that would be true, it's going to be faster than the A64 at 2.5ghz.

I had a Prescott too, albeit a Pentium D. I didn't even bother overclocking it as I didn't need any extra performance, but at stock speeds it was a bit slower than a 2.4ghz A64 X2 a friend had. It also cost less too 😀.

I agree with you though, the P4 is under rated. Especially the earlier models which commanded a premium price. Back then though, one was still in the grey area where that extra 10 or 20 percent performance actually meant something (now I'd argue that it means something but not as much).

I just wish Intel had gone with DDR from the start. Their dual channel DDR offerings were top notch and just as fast as the Rambus stuff.

sgt76 wrote:

A 1.8Ghz Willie is faster than a 1.4Ghz Tually in gaming benchmarks...look up old Tom's benchmarks. The motherboards it sits on […]
Show full quote

Tetrium wrote:

I think swaaye is correct in assuming the cache was halved to save on manufacturing costs. Remember that the CPU cache is a quite large die area.
I don't know if a 1.8Ghz Willy is faster then any Coppermine/Tualatin, but it sure eats more power. Personally I think I'd prefer Athlon XP over Netburst, but perhaps that's also because I also have more experience with Socket A...and they kinda resemble Socket 370.

Willamette and Northwood may be able to hold their own, but Prescott...any point to it?

A 1.8Ghz Willie is faster than a 1.4Ghz Tually in gaming benchmarks...look up old Tom's benchmarks. The motherboards it sits on more importantly perhaps are miles ahead in terms of quality, robustness, features and layout of whatever socket 370/ slot 1.
A'd also like an AThlon XP- but they're considerably rarer here in Temeria 😢.

About Prescott having a point- look up my sig rig. I have a choice of nearly any P4 chip yet that's what I chose to plonk in there. It's way faster than my AThlon 64 2.5ghz rig BTW.

I'll add that I was a Netburst hater back in the day (P3 and s939 user from back when), but having owned and used my various P4 rigs recently has led me to dramatically rethink my previous stands.

Try overclocking the Tualatin or underclocking the Willamette, though, and it changes. Clock for clock a Tualatin is faster than Willamette. That was one of the big complaints when the P4 was first released was at the same clock speeds, the Tualatin was faster. You can easily test this by finding some of the slowest Willamettes. They made them as slow as 1.3ghz at one point, though those may be hard to find, so doing an apples to apples comparison at the same speed is possible. You might have to resort to using a mobile version on a P4 motherboard to get the desired speed, but it is possible.

I had a Gateway OEM board (Intel D845GV or something like that) and a P4 Willamette 1.5 with PC133 CL2. It was so slow... I swear a Coppermine 700 is similar when installing XP and just web browsing and such.

P4's performance varies wildly. If the app works well with its long pipeline and loves-to-stall design, it can be very fast. But most apps aren't that optimized and so it performs pretty bad. Benchmarks of course tend to be based on optimized apps.

Nobody ever brings up Pentium M, probably because it's a notebook CPU, but that chip is a P4 destroyer. For 30W of power, a 2.1 GHz Pentium M performs like a 90W P4 3.2 GHz or so. It also rivals Athlon 64 for that matter.

I had a Gateway OEM board (Intel D845GV or something like that) and a P4 Willamette 1.5 with PC133 CL2. It was so slow... I swear a Coppermine 700 is similar when installing XP and just web browsing and such.

Not surprising... the PC133 version of the 845 was easily one of the shittiest chipsets ever made, as far as performance goes. P4s need memory bandwidth, and that chipset didn't have any. Even with DDR, the 845 wasn't great, but with SDRAM it choked the P4 so badly to be near useless. And using the onboard graphics made it even worse, since it shared the main memory.

swaaye wrote:

I had a Gateway OEM board (Intel D845GV or something like that) and a P4 Willamette 1.5 with PC133 CL2. It was so slow... I swear a Coppermine 700 is similar when installing XP and just web browsing and such.

P4's performance varies wildly. If the app works well with its long pipeline and loves-to-stall design, it can be very fast. But most apps aren't that optimized and so it performs pretty bad. Benchmarks of course tend to be based on optimized apps.

Nobody ever brings up Pentium M, probably because it's a notebook CPU, but that chip is a P4 destroyer. For 30W of power, a 2.1 GHz Pentium M performs like a 90W P4 3.2 GHz or so. It also rivals Athlon 64 for that matter.

Pentium M and Core processors are derived from the Tualatin core. Intel actually went backwards when they designed those and threw away most of the Netburst architecture otherwise they would have had the same scaling issues with multi-core Netbursts over 3ghz as they had with single core ones. They went from Tualatin to Netburst because they enthusiastically overestimated how far it would scale. They thought they would be good all the way to 10ghz with Netburst and Tualatin couldn't go that high. Even all these years later with all the design improvements that have taken place we have yet to see a 10ghz CPU. (And please, nobody bring up any supercooling experiments with liquid nitrogen or any one off supercomputer processors which may exist because those will never sell in large numbers or are impractical for mainstream use).

Problem with Preshot is, they are so darn hot!

Has there ever been a consumer processor released that runs @ 4Ghz btw?

No, the fastest P4 was on Socket 775 and it was running at 3.8GHz.

Tetrium wrote:

Problem with Preshot is, they are so darn hot!

Has there ever been a consumer processor released that runs @ 4Ghz btw?

It was also difficult to control the voltage inside the processor as well, and that was the real issue. I can't think of a more elegant or technically correct way of putting it, but essentially the flow of electrons wouldn't stay on their designated pathways above 3.8ghz and would spill over onto nearby pathways. Intel hadn't foreseen this possibility and couldn't correct it so a new architecture was needed.

What I don't like about P4 is that I need to give it a better PSU, and I'd rather put the same PSU in a system that's more powerful. P3 can do fine with a 300W PSU (250W is fine, as long as the PSU is of a good make and you don't put extraordinarily high-end expansion cards in it).

If I didn't have several high watt recent ATX PSUs around that might annoy me too 🤣

Regarding P4 and the semiconductor physics failure that it was, if you consider that each CPU spends several years in development, you can probably safely say that Core 2 was in development when Northwood came out. I don't think Intel was ever really unsure of where to go next. Obviously they did believe back with Willamette that they would make it to 6ghz or whatever but there was other research going on over there.

For example the recently announced trigate transistors were figured out way back in 2002 but they've been working on making them mass produceable.

sliderider wrote:

Try overclocking the Tualatin or underclocking the Willamette, though, and it changes. Clock for clock a Tualatin is faster than Willamette. That was one of the big complaints when the P4 was first released was at the same clock speeds, the Tualatin was faster. You can easily test this by finding some of the slowest Willamettes. They made them as slow as 1.3ghz at one point, though those may be hard to find, so doing an apples to apples comparison at the same speed is possible. You might have to resort to using a mobile version on a P4 motherboard to get the desired speed, but it is possible.

Yes, it's well known that the P4 architecture is the slowest clock-for-clock among any of the modern lines. But the whole point to Netburst is to run at such high clock speeds as to negate that disadvantage. Obviously, they hadn't foreseen scaling issues at higher clock speeds, which made going much over 4ghz near impossible. But at their commonly attainable 3+ ghz speeds, both Northwoods and Prescotts are faster than any o/c'ed Tualatin or Athlon XP and compare well with Athlon 64s.

Heat and power consumption are very high but Prescott's are not alone in this. Try an Athlon 64 6000/ 6400+ or Phenom 9850/ 9950 for high heat/ power consumption. I still wouldn't call them lousy chips- my take on this is a bit more care in parts selection and building will still give excellent results - and this applies to Netburst, Prescotts etc.

swaaye wrote:

I had a Gateway OEM board (Intel D845GV or something like that) and a P4 Willamette 1.5 with PC133 CL2. It was so slow... I swear a Coppermine 700 is similar when installing XP and just web browsing and such.

P4's performance varies wildly. If the app works well with its long pipeline and loves-to-stall design, it can be very fast. But most apps aren't that optimized and so it performs pretty bad. Benchmarks of course tend to be based on optimized apps.

Nobody ever brings up Pentium M, probably because it's a notebook CPU, but that chip is a P4 destroyer. For 30W of power, a 2.1 GHz Pentium M performs like a 90W P4 3.2 GHz or so. It also rivals Athlon 64 for that matter.

Intel should never have released a SDRAM compatible chipset for the P4, and should have just gone straight to DDR, but they were trying to force people into Rambus. P4 systems with SDRAM were sluggish for certain.

For the most part, anything that really required a lot of CPU horsepower was optimized though. My P4 ran games really well, which is all I cared about, although I was also lucky I didn't have to pay for it. Maybe if I had I would have bought an Athlon instead. There was a 2-300 dollar price gap between the two platforms.

I had a laptop later on with a 1.8ghz Pentium M. It felt about the same speed wise as my (2 year old) 2.8 ghz Northwood (except in games as it had a doodoo graphics card). The Pentium M was always a glorfied PIII, and quite good. I agree that Intel shouldn't have gone with Netburst, but it really didn't start sucking until the competition caught up with it.

sgt76 wrote:

Yes, it's well known that the P4 architecture is the slowest clock-for-clock among any of the modern lines. But the whole point to Netburst is to run at such high clock speeds as to negate that disadvantage. Obviously, they hadn't foreseen scaling issues at higher clock speeds, which made going much over 4ghz near impossible. But at their commonly attainable 3+ ghz speeds, both Northwoods and Prescotts are faster than any o/c'ed Tualatin or Athlon XP and compare well with Athlon 64s.

Netburst does perform very well in some cases. It excels at media encoding for example. It probably has to do with the process fitting nicely in its caches and not stalling the pipeline as much as general applications. Also, if SSE or SSE2 is heavily used it helps a ton because P4 benefits hugely from its clock speed there.

Intel also, as usual, had an advantage with cache and memory performance over AMD. Between P4's clock speed and the L2 being 256-bits wide, and I believe lower latency as well(!), Intel destroyed AMD here. And Intel chipsets had better memory controllers. AMD was saddled with subpar chipsets until nForce 2 came along. But by then the K7's FSB was also far behind Intel's FSB in bandwidth.

The difference between K7 and K8 was mostly a massive improvement in memory performance. Gives an idea of how bottlenecked K7 was there.

sgt76 wrote:

But at their commonly attainable 3+ ghz speeds, both Northwoods and Prescotts are faster than any o/c'ed Tualatin or Athlon XP and compare well with Athlon 64s.

Even with a Barton OC'd at, say, 2.6Ghz?

If we compare a late generation K7 like Barton with a later generation Netburst, we can see that for PIII level code they are about equal. I base that off of SANDRA's marks for both processors, and is the reason why I stand that Athlons are optimal for older gaming, all the way up to the DX8/9 era. This paradigm does change however, if you have code that is being optimized for longer pipelines, streaming extensions like SSE2, and in a big way, memory bandwidth. That last point becomes a big deal as graphic complexity increases. I was looking at a review where they compared 3200+ with the first 800mhz bus P4, a 3.0ghz northwood. They basically trade blows, but if you factor in non-gaming tasks the P4 looks superior. Now figure that you can get a factory P4 at 3.8, and that getting a Barton to 2.6ghz is a task requiring some skill and luck; clearly the P4 is the easiest to obtain. Granted, the 3.8 specimen is actually a Cedar Bridge; basically a Prescott with a die shrink with lager cache. But it isn't overclocked, so allowing this constraint I would believe that the P4 would ultimately be the performance leader, since you have the a memory bandwidth when you need it and the raw clock speed will pull it up even in legacy applications. It is worth noting that particular processor uses socket 775, so one will end up with PCI-E graphics unless a few certain motherboards are used.

That being said, I still personally use an Athlon since they work just as well for most stuff and use a lot less energy.

prophase_j wrote:

If we compare a late generation K7 like Barton with a later generation Netburst, we can see that for PIII level code they are about equal. I base that off of SANDRA's marks for both processors, and is the reason why I stand that Athlons are optimal for older gaming, all the way up to the DX8/9 era. This paradigm does change however, if you have code that is being optimized for longer pipelines, streaming extensions like SSE2, and in a big way, memory bandwidth. That last point becomes a big deal as graphic complexity increases. I was looking at a review where they compared 3200+ with the first 800mhz bus P4, a 3.0ghz northwood. They basically trade blows, but if you factor in non-gaming tasks the P4 looks superior. Now figure that you can get a factory P4 at 3.8, and that getting a Barton to 2.6ghz is a task requiring some skill and luck; clearly the P4 is the easiest to obtain. Granted, the 3.8 specimen is actually a Cedar Bridge; basically a Prescott with a die shrink with lager cache. But it isn't overclocked, so allowing this constraint I would believe that the P4 would ultimately be the performance leader, since you have the a memory bandwidth when you need it and the raw clock speed will pull it up even in legacy applications. It is worth noting that particular processor uses socket 775, so one will end up with PCI-E graphics unless a few certain motherboards are used.

That being said, I still personally use an Athlon since they work just as well for most stuff and use a lot less energy.

FX-55 kills any Netburst including EE.

Guys, have you ever seen or heard about some non-intel, overclock-friendly i840 boards? Would it not be awesome to bench such a system with, lets say, 2x512 RDRAM and a 1.4 (preferably overclocked) Tualatin? 😀

I know that many of you will start with the usual "the PIII CPUs were saturated already by SDRAM" but I would still give it a shot, just to make sure 😀

elfuego wrote:

Guys, have you ever seen or heard about some non-intel, overclock-friendly i840 boards? Would it not be awesome to bench such a system with, lets say, 2x512 RDRAM and a 1.4 (preferably overclocked) Tualatin? 😀

I know that many of you will start with the usual "the PIII CPUs were saturated already by SDRAM" but I would still give it a shot, just to make sure 😀

Tyan Thunder LE S2510. Only one I could find. It probably wasn't a popular chipset outside of Intel due to the RDRAM requirement.

That board also has Rage XL onboard video and only 2 PCI-X slots for expansion, so you won't be able to use a decent video card with it. I'm not sure you'll be able to do a fair comparison. You'd have to use a Rage Pro video card in whatever machine you test it against as that's the nearest dedicated video card you'll find to an onboard Rage XL.

Two variants of the board also use onboard SCSI controllers, so you'd have to use SCSI hard drives in whatever system you test against as well.