Blazing 800Mhz Pentium3 vs Celeron s.370

[Ningenpo]

Hi,

I have a problem, I have an old computer that used to house a cel 800Mhz proc, a tnt2 and 256mb ram. The only thing I did was to exchange the cpu to a Pentium3 800Mhz, same frequency that is. Now here's whats odd. Everything works alot smoother with the P3, opening windows and progs. (yeah I know, been a while since you thought about a computer that you actually could see having to think just to open my computer). But I thought I would do a comparison just for fun, so I got 3dmark 2000 and ran everything default with both procs.

Believe this:
Cel 800; tot 2106, cpu 214
P3 800; tot 1970, cpu 242

WTF???

How can I get a lower total score when the only thing I changed was the proc, and the cpu score is higher with the P3, and also the test apear to run smoother on the P3?

Anyone care to explain?

(You don't have to tell me, I will stick with the P3 because it works better, but I just want to know for the fun of it)

Thanks in advance!

 

[joefriday]

hmm....that is interesting. Just goes to show that you can't rely on synthetic benchmarks. I am surprised you were able to really discern a difference between a PIII 800 E and a Celeron 800, as both are 100 bus. A 766 MHz Celeron would have been a very evident difference.

A Celeron 800 was in the first computer I purchased, an HP 6835. That was a nice rig.

Thanks for the nostalgia

 

[kitchenshark]

Hmmm...well. That's only a difference of 36 points for the total score. A number of factors could have influenced it. To start with, every benchmark run will be slightly different anyway. Also if there was even one extra task running somewhere during the run, that could have influenced the result just that much.

EDIT: DOH!! A difference of 136 points! ARGH!!! Guess I need more sleep. >.< Anyhow, there still could have been an extra process somewhere sucking up a bit of resources.

 

[linux_0]

hmm....that is interesting. Just goes to show that you can't rely on synthetic benchmarks. I am surprised you were able to really discern a difference between a PIII 800 E and a Celeron 800, as both are 100 bus. A 766 MHz Celeron would have been a very evident difference.

A Celeron 800 was in the first computer I purchased, an HP 6835. That was a nice rig.

Thanks for the nostalgia

VERY true.

Synthetic benchmarks are notoriously unreliable.

 

[apache_lives]

Hi,

I have a problem, I have an old computer that used to house a cel 800Mhz proc, a tnt2 and 256mb ram. The only thing I did was to exchange the cpu to a Pentium3 800Mhz, same frequency that is. Now here's whats odd. Everything works alot smoother with the P3, opening windows and progs. (yeah I know, been a while since you thought about a computer that you actually could see having to think just to open my computer). But I thought I would do a comparison just for fun, so I got 3dmark 2000 and ran everything default with both procs.

Believe this:
Cel 800; tot 2106, cpu 214
P3 800; tot 1970, cpu 242

WTF???

How can I get a lower total score when the only thing I changed was the proc, and the cpu score is higher with the P3, and also the test apear to run smoother on the P3?

Anyone care to explain?

(You don't have to tell me, I will stick with the P3 because it works better, but I just want to know for the fun of it)

Thanks in advance!

Run CPUZ and check that its not a P3-800 EB (FSB133) running at FSB 100 giving you 600mhz, and benchmark it using another app like superpi.

 

[TabrisDarkPeace]

All other things being equal: The less cache there is the faster it can be searched through.

This gives the Celeron (all other things being equal, including FSB, architecture [both P3 die], HyperThreading, memory configuration [inc dual-channel], etc) a slight advantage in some areas. It is hinted that Intel changed the Celeron L2 cache latency slightly in some revisions to stop them looking as good for the dollar.

However, the CPU test no doubt benefits from the large quantity of the cache that the Pentium 3 has.

Synthetic tests are the best to learn from, as if you can't explain something you need to research it and think about it some more. Although I am personally no fan of 3DMark... at least not the Futuremark versions.

If someone can not explain why synthetic and real-life benchmarks differ, in technical detail, that is a good indication they don't really know what is going on 'under the hood'. No doubt a decent poll of a large number of University students would confirm this.

In real-life / game situations (not time demos) the cache hit rate will be lower, as the machine can not really predict what a human player is going to do. ATI especially noticed this in 3DMark 05/06, as they made the hit rates of their new 512-bit ring bus memory controller public, and in timedemos (including 3DMark, vs actual interactive game play) they did an amazingly good job.

Just a little slice of history, and a more recent version of the same thing really. 8)

 

[Ningenpo]

Hi again,

I checked with cpu-z, it's 100 * 8, so it's working at 800Mhz. I don't really care since it works way better than the Celeron. I just wanted to still my curiosity.

It must be some kind of bug with that prog, I ran it with my X2 4800 and got 1191... Just kidding.

But there was a guy saying that the cashe might be the culprit, that might be just so. Best explaination so far atleast.

I know you get different scores each time, but to minimize differences I (I use ME) rebooted the computer and terminated all non essential processes leaving 4, then ran the test in a loop 10 times with both. It's the average that's 2106 and 1970.

Now I'm thinking of upgrading and (this comp is for my wife doing progs like Word) have a chance to buy a Profet 4500 128Mb ram graphics card or a 64Mb Geforce 2 MX. What should I go with? This is only done because she might want to connect the comp to the Tv.

 

[Ningenpo]

Hi again,

I checked with cpu-z, it's 100 * 8, so it's working at 800Mhz. I don't really care since it works way better than the Celeron. I just wanted to still my curiosity.

It must be some kind of bug with that prog, I ran it with my X2 4800 and got 1191... Just kidding.

But there was a guy saying that the cashe might be the culprit, that might be just so. Best explaination so far atleast.

I know you get different scores each time, but to minimize differences I (I use ME) rebooted the computer and terminated all non essential processes leaving 4, then ran the test in a loop 10 times with both. It's the average that's 2106 and 1970.

Now I'm thinking of upgrading and (this comp is for my wife doing progs like Word) have a chance to buy a Profet 4500 128Mb ram graphics card or a 64Mb Geforce 2 MX. What should I go with? This is only done because she might want to connect the comp to the Tv.

 

[TabrisDarkPeace]

The Kryo (Guillemot / Hercules rebranded) Prophet 4500 is not a 'bad' card, however getting drivers for it in this day and age is almost impossible. Assuming it has TV-out it should be nice... if you can get it working.

The GeForce 2 MX, TV-out quality is acceptable, and drivers have come a long way in that area since. Also getting the drivers will be far easier.

Overall, the GeForce 2 MX is going to work with a larger quantity of software, and it will work 'easier' (less work will be involved).

If you want info on the 'rare' Kryo Prophet 4500 cards, then do a search on TomsHardware main page for articles containing Kryo & 4500 , they lacked H/W accelerated T&L, but their performance in FSAA for the time was worlds ahead of the pack. Of course the Radeon 9800 Pro changed any future they might have had by being good at pretty much everything (in comparison, back then).

Also, the spelling may be Kryo or Kyro, I can't recall if they diliberately changed the word and pronunciation to make it a (tm) or something.