Former Netburst Celeron's vs todays Celeron's / Pentium's

[Compman55]

Back in the netburst days, socket 478 to be precise, I remember a P4 northwood vs. Celeron non hyperthread felt very close in real world use. After windows XP got improved and newer games came out, I and those on the interent felt the Celerons were very inferior. Affordable AMD chips would leave the Celerons in the dust.

So fast forward to today, I have a Haswell E3-1245V3, and the Ivy G2020 Pentium. Let me tell you that in windows 7 I can feel no difference. The G2020 is so impressive for what it is. I feel if I put a good video card in it, the thing would rock. Today when comparing low end Celeron / Pentiums, is there as much difference as back in the day?

 

[VirtualLarry]

They used to primarily differentiate based on cache sizes. Pentiums had barely enough to run semi-efficiently, Celerons didn't have enough. The performance differences were noticeable.

Now, Celerons and Pentium still have slightly less cache memory, but it's generally "enough", and single-threaded performance for most tasks is similiar per clockspeed, between Celeron, Pentium, and Core. The biggest differentiation these days is based on HyperThreading, Turbo, and instruction-set extensions.

 

[zir_blazer]

Cache itself doesn't seems to be as important as it was, specifically for the NetBurst architecture (Willamate and Northwood, Prescott not soo much). There are a whole bunch of latency hiding techniques that have improved a lot from that time (Integrated Memory Controller, Branch Predictors, overally faster RAM, and I think some more) that may have diminished how much Cache individually weights in performance.

Overally, Cache L2 was NetBurst inherent weakness. At the same time you also had AMD K7 (Durons, Athlons XP), and there wasn't a major difference in performance between Durons Applebreads with 64 KB Cache L2 vs Thoroughbreds and Bartons with 256 and 512 KB Cache L2, respectively. They were faster, but not by the massive difference that there was between Celerons Willamate and Northwoods and Pentium 4s with the full Cache L2 (A 2.6 GHz Celeron Northwood was around 25% slower than a Pentium 4 1.8 GHz).
Also, this somehow changed with Prescott, NetBurst third generation. Prescott wasn't as Cache hungry. The difference between Celeron D with 256 KB Cache L2 and the 1 MB Pentium 4 was much less. And Prescott 2M, with twice the Cache L2, was around the same performance than the regular Prescott (Possibily because Cache Latency increase at the same time, but still...).

 

[Sequences]

If you're not using either chips at capacity, there is no point in comparing them. A Honda Civic will drive at 20 mph just like a BMW, but once you start pushing the performance barrier, one will shine through.

 

[2is]

If you can't tell the difference between those two processors, then you aren't doing much above and beyond basic navigating through windows, browsing the forums and watching youtube. Which makes perfect sense since the lower end CPU's are designed to do just those types of tasks well.

The Netburst Celerons, in addition to being clocked lower and having less cache than their Pentium counterparts also had a slower FSB and no HyperThreading.

 

[dave_the_nerd]

If you're not using either chips at capacity, there is no point in comparing them. A Honda Civic will drive at 20 mph just like a BMW, but once you start pushing the performance barrier, one will shine through.

Yeah... the Civic.

 

[Compman55]

No $hit. I would have never guessed.

Asking about todays CPU's. I already know about the cache, FSB, and lack of HT.

 

[TreVader]

Difference between current celeron and pentium is 2MB vs 3MB L3 Cache. The Haswell Celeron has the same 512KB L2 Cache as a haswell i3.



Netburst Celerons had like 128kb of L2 Cache vs 512 with the Northwood P4 and 1M for Prescott.

 

[2is]

No $hit. I would have never guessed.

Asking about todays CPU's. I already know about the cache, FSB, and lack of HT.

Likewise, I'd never have guessed when you typed the word "former" in your title post. You actually meant "todays"

At any rate, what you're asking seems rather obvious doesn't it? You don't notice the difference between [todays] higher end and lower end processors.

You can conclude one of two things.

1) You aren't doing anything very processor intensive
2) There is no difference and Intel is selling you the same processor under a different sku.

Which of those would you guess it is?

 

[Compman55]

I run Dx8 games, lots of VM's, and at least 50 tabs of internet at any given time. So no.......I do not take advantage of every performance aspect of a processor. Around 20GB ram out of 32GB in use as well.

What I am asking simply, is that back in the day, Celerons felt inferior once windows XP matured. Windows 7 is about as mature as it gets and I still see that low end CPU's perform very well unlike the days of the netburst. Looking for opinions in that matter.

 

[zir_blazer]

What I am asking simply, is that back in the day, Celerons felt inferior once windows XP matured. Windows 7 is about as mature as it gets and I still see that low end CPU's perform very well unlike the days of the netburst. Looking for opinions in that matter.

Because even low end Processors are already superior to most people daily processing needs. You can throw a faster Processor, and the difference may not be big or perceptible because you already have something fast enough for what you do most of the time. So for most task of the typical Desktop user, we're deep into diminishing returns.

Back at that time, Multitasking relied on context switching in a single Core Processor, with all the Threads running on the computer wanting a piece of the cake. During the Dual Core era, the single most important thing, which was OS responsiveness (Check "The Experience", any oldschooler should remember that), drastically increased. This was pretty much because during W9x/WXP reing with Single Core Processors, a CPU heavy application could put the sole Core at Full Load and it was hard to do other things simultaneously because the entire GUI lagged (This was worse with applications that crashed, as opening Task Manager to kill that Thread that was eating CPU resources was also slow and unresponsive). The same behavior on a multicore Processor doesn't causes havoc on the responsiveness of your system. And even if an individual application takes some time to complete its task, you can always Alt + Tab and do something else on another application, something than at that era wasn't easier because any CPU heavy application usually demanded exclusivity. That's why a faster Processor always speeded up everything, you could deal with CPU intensive applications faster to resume everyday tasks, while now you can do both at the same time.

 

[pantsaregood]

Actually, modern CPUs are stratified much more heavily than CPUs from the Netburst era. Pentium 4s from that era were usually between 0%-15% faster than their Celeron counterparts - the difference almost never exceeded 15% once Celeron D was released, as it wasn't as cache-starved as earlier revisions. It was relatively easy to buy a Celeron and overclock it to achieve performance equal to a Pentium.

Modern Celerons don't have any hope of competing with Core CPUs. A Celeron G470 has no hope of comparing to an i7 4960, but a Celeron D 2.8 could pretty reliably hit 4 GHz and match high-end Pentium 4s.

Pentium 4s can run Windows 7 without issue, and a G470 outperforms them by quite a bit. So it isn't a case of Celerons being closer to high-end CPUs, it's just that CPUs are powerful enough to run most things well now.

 

[sm625]

This debate was pretty clear cut until they started releasing pentium-branded atoms. Now we're back to square one.

 

[Arkaign]

Actually, modern CPUs are stratified much more heavily than CPUs from the Netburst era. Pentium 4s from that era were usually between 0%-15% faster than their Celeron counterparts - the difference almost never exceeded 15% once Celeron D was released, as it wasn't as cache-starved as earlier revisions. It was relatively easy to buy a Celeron and overclock it to achieve performance equal to a Pentium.

Modern Celerons don't have any hope of competing with Core CPUs. A Celeron G470 has no hope of comparing to an i7 4960, but a Celeron D 2.8 could pretty reliably hit 4 GHz and match high-end Pentium 4s.

Pentium 4s can run Windows 7 without issue, and a G470 outperforms them by quite a bit. So it isn't a case of Celerons being closer to high-end CPUs, it's just that CPUs are powerful enough to run most things well now.

Wat? Celeron performance was abysmal during the netburst era. Perhaps you're remembering the P3 Celeron era.

http://www.xbitlabs.com/articles/cpu/display/celeron-d_8.html#sect2

Celeron D needs 3.8Ghz to get close to P4 2.8Ghz.

Ancient P4 2.4A (super slow bus) is faster than Celeron D @ 2.8Ghz

And so on. One could say they were a good value in many cases, I know I had a few. But in the Socket 478 era, the P4 was dominantly faster per-clock. And since basically any Northwood P4 could do ~3Ghz+, that really put the gap on display.

 

[pantsaregood]

Wat? Celeron performance was abysmal during the netburst era. Perhaps you're remembering the P3 Celeron era.

http://www.xbitlabs.com/articles/cpu/display/celeron-d_8.html#sect2

Celeron D needs 3.8Ghz to get close to P4 2.8Ghz.

Ancient P4 2.4A (super slow bus) is faster than Celeron D @ 2.8Ghz

And so on. One could say they were a good value in many cases, I know I had a few. But in the Socket 478 era, the P4 was dominantly faster per-clock. And since basically any Northwood P4 could do ~3Ghz+, that really put the gap on display.

The 2.4A (that's not a super slow bus, it's 533 MHz, just like the Celeron D) being about the same speed as a 2.8 GHz Celeron D puts the Pentium 4 at about 17% faster than the Celeron D at the same clock speed. Generally, a 15% increase in clock speed seems to be enough to make up for the decrease in cache - the Celeron D at 3.8 GHz only comparing with a 2.8E looks to be a result of some sort of scaling issue with NetBurst, as the 14% increase in clock speed looks like it is frequently yielding performance improvements under 5%.

Call it abysmal if you'd like, but the absolute performance difference peaked at around 15%.

 

[Insert_Nickname]

Generally, a 15% increase in clock speed seems to be enough to make up for the decrease in cache - the Celeron D at 3.8 GHz only comparing with a 2.8E looks to be a result of some sort of scaling issue with NetBurst, as the 14% increase in clock speed looks like it is frequently yielding performance improvements under 5%.

Call it abysmal if you'd like, but the absolute performance difference peaked at around 15%.

If you try running modern software on that thing, no amount of clockspeed will make for the pathetic 256KB L2 of the Celeron. Even given top performing memory, the Celeron is just too slow to run modern software...

I remember restoring an old Celeron D 360 (Cedar Mill/3.46GHz/512KB L2). Was that thing slow. It made my C60-based XPbox seem lightning fast in comparison. That's no small "achievement".

The cache size was less an issue back then, but the problem with many Celerons was that they where sold with too little memory, and that excuse for integrated graphics called Intel "Extreme" Graphics. The only "extreme" thing about those IGPs was how bad they where. What was worse was that they used the CPU for handling some of the graphics pipeline. Throw that on top of a memory limited Celeron, and you had a recipe for abysmal performance.

 

[pantsaregood]

Actually, my grandparents have a Celeron D 355 (3.33 GHz Prescott) running Windows 7. The only "modern software" it has issues with is Adobe Flash, which it seems can choke any CPU.

Also, if an AMD C-60 is outperforming a Celeron D 360, something is going wrong with that Celeron D. The C-60 is a considerably weaker CPU.

Also, top performing memory? The importance of memory bandwidth is greatly overstated in this topic. The difference in FSB speeds during the NetBurst era resulted in minimal performance gains.

NetBurst Celerons were slower than comparably clocked Pentium 4s, but not to the degree that modern Celerons are slower than Core i7s. I can run modern software with L2/L3 cache disabled entirely - it performs rather poorly, but it does function. Increasing cache often results in diminishing returns - that's why Prescott "F" barely saw performance improvements over Prescott "E" and the different cache levels on Core 2 CPUs had a minimal effect on performance.

 

[nwo]

I remember Celerons back in the day were absolutely terrible. I got burned a few times with a Celeron and afterwards avoided them like the plague and just used AMD Athlon XP, Sempron, & 64 x2 based desktop CPUs. I actually had a sempron server rig which would host source based games and it was absolutely amazing for the $$.

I recently put together an Ivy Celeron G1620 build for my office rig. It was only $34.99 at my local Micro Center so I figured why not give it a shot. It was a ridiculously huge upgrade from a 2GHz Core 2 Duo. Words cannot even describe what a difference it makes in simple everyday office activities such as web browsing, database entry programs, watching youtube, and other audio/video playback. The Core 2 Duo would struggle at any of the above tasks while the Celeron has no trouble flying through them.

I also feel like it would be an adequate CPU for certain gaming needs. Blizzard games in particular since they only use 2 cores. In fact, I recently put testing Diablo 3 gameplay on the Celeron rig on my to-do list. I recently put a Radeon 270 in there (for mining) so it will be comparable to the 7850 and 7870 I use on my other two rigs. :biggrin:

 

[Arkaign]

The 2.4A (that's not a super slow bus, it's 533 MHz, just like the Celeron D) being about the same speed as a 2.8 GHz Celeron D puts the Pentium 4 at about 17% faster than the Celeron D at the same clock speed. Generally, a 15% increase in clock speed seems to be enough to make up for the decrease in cache - the Celeron D at 3.8 GHz only comparing with a 2.8E looks to be a result of some sort of scaling issue with NetBurst, as the 14% increase in clock speed looks like it is frequently yielding performance improvements under 5%.

Call it abysmal if you'd like, but the absolute performance difference peaked at around 15%.

Eh, I also remember a Celeron 2.8 benching about identically to a 1.8Ghz 400FSB P4 Willamette. Which really shouldn't happen. Celeron D improved, but the contemporary Celerons to Northwoods were complete garbage.

By the time Celeron D came on the scene, we had cheap lower-clock P4s that would massively outperform the Celeron D's when overclocked.

For example, this is WAY more than 15% off in terms of IPC :

http://hexus.net/tech/reviews/cpu/720-intel-celeron-28ghz/?page=7

That's abysmal any way you cut it.

For example, the Celeron 2.8Ghz has a 16.6% higher clock speed than the P4 2.4. Adjusting the P4 score to 2.8Ghz nets a 3dmark score of 19413 at equal clocks for a rough IPC comparison, or 159%+ of the performance of the Celeron at the same clock speed.

And even at 2.4, it offers over 135% of the performance of the 2.8 Celeron.

 

[zir_blazer]

NetBurst Celerons were slower than comparably clocked Pentium 4s, but not to the degree that modern Celerons are slower than Core i7s. I can run modern software with L2/L3 cache disabled entirely - it performs rather poorly, but it does function. Increasing cache often results in diminishing returns - that's why Prescott "F" barely saw performance improvements over Prescott "E" and the different cache levels on Core 2 CPUs had a minimal effect on performance.

Is true that current budget Processors are a mere fraction in raw processing power than mainstream and high end ones, but even then, they are good enough for everyday performance anyways. That's the point. It didn't used to be like that, as NetBurst Celerons (No Celerons D, those were at least usable) were absolute pieces of trash and the entire machine felt sluggish. They also were very expensive - most models had prices comparable to high end Athlons XP, which mopped the floor with them.

The reason why you can disable the Cache L2 and L3 and still have a usable computer should be thanks to the Integrated Memory Controller mostly. Also, Prescott 2M, while it had twice the Cache L2 than Prescott, it also had increased Cache Latency to compensate. Overally performance was around the same, but on some scenarios it could be higher or lower. Is hard to compare the importance of Cache L2 itself because it wasn't isolated. One would have expected Prescott to react very positively to Cache, but it didn't. Prescott scaling was somehow flawed, as more Cache, faster Bus, more Memory Bandwidth, etc, didn't made the massive performance differences observable during the Northwood era.

 

[piasabird]

My Haswell i3 4330 has a 4 Meg Cache, runs at 3.5 Ghz and has Intel HD 4600 Graphics. Celerons and Pentiums come in so many varieties today.

When it came to Celerons, the P II level Celerons were a lot weaker than the PIII Celerons that came out before the P4. The 1.2 Ghz - 1.4 ghz Celerons were just as good as the First P4 processors that came out. They had 256kb of L2 Cache and were quite fast. In fact the mobile P4 was based on the P3 architecture.

 

[zir_blazer]

In fact the mobile P4 was based on the P3 architecture.

The P6 based was the Pentium M Banias and Dothan. There were two versions of Pentiums 4 for Mobiles, one was the Pentium 4-M and the other the Mobile Pentium 4, which was for DTR (Desktop Replacement). Those were NetBurst.

 

[SPBHM]

Celeron was much closer to high end CPUs in the past, but also it was more expensive, it was more or less the i3/i5 of the time, current Celeron is priced way lower, Pentium was the i5/i7.

Prescott to Northwood was a good improvement for Celeron, not so much for P4...
so a $100 Celeron D with some overclock was not to far off some extreme $1K p4, or at least the $200-400 P4s,

it's impossible for a current i3 or Celeron to get as close to a 4770k/4960x.
on anything using 4c+

the Pentium II era Celeron, the second version with ondie 64K of l2 was also pretty impressive, almost the same as a PII when overclocked... but again it was nothing like a $50 CPU, CPUs used to be way more expensive on average.

The P6 based was the Pentium M Banias and Dothan. There were two versions of Pentiums 4 for Mobiles, one was the Pentium 4-M and the other the Mobile Pentium 4, which was for DTR (Desktop Replacement). Those were NetBurst.

yes, I've dealt with Northwood and Willamette based laptops, Pentium and Celeron.

 

[VirtualLarry]

This thread makes me want to find a S478 board for a P4 3.2 Northwood chip I think I still have stashed somewhere, just so I can benchmark it.

 

[Insert_Nickname]

Also, if an AMD C-60 is outperforming a Celeron D 360, something is going wrong with that Celeron D. The C-60 is a considerably weaker CPU.

Not outperforming as in raw CPU power, but system responsiveness was like night and day. Properly because the C60 has that second core to play with. The Celeron doesn't even have HT.

Tested with one of the last boards (i945-based?) to support Netburst and with DDR2-667.

Also, top performing memory? The importance of memory bandwidth is greatly overstated in this topic. The difference in FSB speeds during the NetBurst era resulted in minimal performance gains.

I don't know what platform you've been using, but going from 533MHz FSB to 800MHz+ FSB was defiantly noticeable in benchmarks. Paired with DDR2-533 or higher.

P4's where always bandwidth and cache hungry. If you don't believe me, look up some of the old articles on what effect using SDR vs RDRAM, or going from single channel DDR i845 to dual channel DDR on the i865 had on performance...

the different cache levels on Core 2 CPUs had a minimal effect on performance.

But increasing/decreasing the size of the L2 cache had a 5-10% effect on performance.

http://www.anandtech.com/show/2045/4