which processor is faster?

[bambam]

I know this is old school - but my brother says a Pentium3 550mhz is faster than a Celeron 1.8 for stuff like rendering - is this true? I thought the newer faster celerons were faster that the older Pentium 3s?

thanks

[Dr Mordrid]

Not even close.

The Celeron 1.8 will be faster because they are nothing but one of the original P4 Williamettes in drag with a 128k cache instead of 256k.

The smaller cache will hurt the Celeron 1.8's performance a bit in most 3D apps, but the clock speed being over 3x faster plus its ability to use SSE2 in supporting apps mitigate that, and not just a little. The PIII's only support SSE and not SSE2.

TMPGEnc is one of those apps that use SSE2. So is MediaStudio Pro 6.5 and most of Uleads other video apps.

SSE2 was at one time the orphan child of software, but many of the advances in terms of faster codecs and software realtime are because of it.

Dr. Mordrid

[bambam]

Many thanks He owes a beer.

[ZokesPro]

Easy bet.

[UtwigMU]

Celeron won't be significantly faster.

Note that there are 2 Pentium III cores: The 550 Katmai (250nm process) with 512kB of half speed (@275MHz) cache and 550E Coppermine (180nm process) with 128kB of full speed cache.

Generally the beefier cache version performed better with professional applications, while lower cache versions performed better at gaming.

Not all 3D software is SSE2 optimized and SSE CPUs also benefit from optimized code.

This goes only for Lightwave. In 3D studio MAX and Maya Athlon which has only SSE performs on par or outperforms Pentium 4 (SSE2) with same MHz as Athlon's PR as can be seen in this article on aceshardware.

Now let's take a look at pure numbers from this test in tom's hardware.

Celeron 1.8 vs athlon, percentage:
Q3: 173.9 / 211.6 / 82.18%
Comanche4: 20.35 / 36.71 / 55.43
3D mark: 8338 / 10253 / 81.32
sysmark: (this is the benchmark that got accused of being optimized for P4) 198 / 195 / 101.54%
PC mark: 4232 / 4281 / 98.85%
Memory Bench: 3769 / 3048 / 123.65%
XMPEG 4.5 151 / 133 / 88.08%
MP3 Lame 212 / 203 / 95.75

Average 90.85 or if we omit sysmark, PC Mark and Memory bench and 3D Mark which is more videocard benchmark 80.36

Also note that according to this benchmark 2.0GHz Willamette with SDRAM performs @67.48 of AthlonXP 2000+ with DDR and 2.0/DDR performs @ 69.4 (both test averaged).

If we assume that 1.8 GHz celeron performs at 80% of Athlon 1600+ (which performs similarly to Willamette 1.7 in general applications) we can conclude that celeron performs at 80% of 1.7 Willamette which translates to Willamette at 1.36 GHz.

If the Willamette performs at 70% of Athlon with similar PR rating ve can assume that Willamette running @1.36GHz would perform at 70% Athlon 1360+which would translate to Athlon at 868 MHz. (60*0.66 = 39.6 1360 = 1500-200+60 -> 1333.3-66.7-66.7+39,9 = 1239.9 1239.9x0.70 = 867.39)

This conclusion is based on several assumptions, but I'd say that 1.8 Celeron would not be more than 2 times better at Max and Maya than a PIII@550MHz.

If we were to find a benchmark severly dependant on cache...

Another interesting thing would be double blind test with both CPUs.

Cleary Willamete sucks in 3D apps compared to Athlon and Celeron should only perform worse, so If you are looking for CPU for 3D applications consider either AthlonXP or Northwood Pentium 4.

[Evildead666]

PIII 550 is the slot one chip?
100MHz bus, the 133 bus were flipchips...fcpga.
but there were some 100mhz bus in fcpga too.

The only good thing, is that they usually accept a 133mhz bus, with little extra cooling, and there that would give you (its late, calculator...)733mhz.
that should be quite zippy.
i have a PIII 450@500 running my server (proxy etc...)and its very fast for what it has to do. (beats the shit out of the P233MMX that was in there b4 )
just give it plenty of memory (i have 512Mb on it, pc133 cas 3)

If you take video encoding for example, its clocks that count. The Celery 4 2.0 was tested at Toms, and for a cheap price, gave 3.06mhz of power. that was all it was good at tho. it was behind the competition in every othr test bar the mp3 one.

[Ribbit]

Originally posted by UtwigMU
Note that there are 2 Pentium III cores: The 550 Katmai (250nm process) with 512kB of half speed (@275MHz) cache and 550E Coppermine (180nm process) with 128kB of full speed cache.

Just a tiny nitpick, Coppermine had 256KB L2 cache. It was a certain PII/PIII Celeron (Celeron A? I don't recall) which had 128KB full-speed.

[funky-d-munky]

I believe all Celerons before 1gig were 128kb cache, and 256kb 1 gig and above.

As far as the question the 550(100 bus pIII) has no chance against a 1.8 celeron.
But I would prefer a 1.4gig Celeron(essentially the same chip as the pIII) over the 1.8 celeron(128kb cache p4)
The Celeron at 1.4(the last of the pIII cores) was a far better processor.
But ultimately I would take the Celeron 1.7 if I could use it with DDR and clock it up. Running 100fsb quad, gives you lots of room. The P4 still needs big time cache to be any good, but at least in that case it could beat the 1.4 Celeron, nevermind that 550 pIII.

[Dr Mordrid]

The question was how would they compare when "rendering"...which basically means encoding.

Given that most all MPEG encoders now support SSE2 the faster clock and SSE2 functions of a Celeron or P4, even one with a smaller cache, would be insurmountable.

This advantage would be further enhanced by an editing program that also supports SSE2 throughout. MediaStudio6.5 and VideoStudio6 do, and with every new version other companies (save for Adobe Premiere) are doing likewise.

This support is rapidly being adopted in DirectShow filters and VfW codecs, so presuming that the Celeron's SSE2 would not matter is not a presumption based in reality.

Dr. Mordrid

[Byock]

Originally posted by Ribbit
Just a tiny nitpick, Coppermine had 256KB L2 cache. It was a certain PII/PIII Celeron (Celeron A? I don't recall) which had 128KB full-speed.

Yup. The original Celery had no L2 cache. The Celery A had 128KB full clock L2.

[Kooldino]

The only way the P3 would be faster would be "clock for clock". Or, like stated above, MAYBE if you wrote a specific program that deals w/ cache a CERTAIN way...but even that's debatable...